xref: /netbsd-src/sys/kern/kern_fork.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: kern_fork.c,v 1.69 2000/07/04 15:33:30 jdolecek 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_multiprocessor.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 #include <sys/sched.h>
62 #include <sys/signalvar.h>
63 
64 #include <sys/syscallargs.h>
65 
66 #include <uvm/uvm_extern.h>
67 
68 int	nprocs = 1;		/* process 0 */
69 
70 /*ARGSUSED*/
71 int
72 sys_fork(p, v, retval)
73 	struct proc *p;
74 	void *v;
75 	register_t *retval;
76 {
77 
78 	return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
79 }
80 
81 /*
82  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
83  * Address space is not shared, but parent is blocked until child exit.
84  */
85 /*ARGSUSED*/
86 int
87 sys_vfork(p, v, retval)
88 	struct proc *p;
89 	void *v;
90 	register_t *retval;
91 {
92 
93 	return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
94 	    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, SIGCHLD, NULL, 0,
110 	    NULL, NULL, retval, NULL));
111 }
112 
113 int
114 fork1(p1, flags, exitsig, stack, stacksize, func, arg, retval, rnewprocp)
115 	struct proc *p1;
116 	int flags;
117 	int exitsig;
118 	void *stack;
119 	size_t stacksize;
120 	void (*func) __P((void *));
121 	void *arg;
122 	register_t *retval;
123 	struct proc **rnewprocp;
124 {
125 	struct proc *p2;
126 	uid_t uid;
127 	struct proc *newproc;
128 	int count, s;
129 	vaddr_t uaddr;
130 	static int nextpid, pidchecked = 0;
131 
132 	/*
133 	 * Although process entries are dynamically created, we still keep
134 	 * a global limit on the maximum number we will create.  Don't allow
135 	 * a nonprivileged user to use the last process; don't let root
136 	 * exceed the limit. The variable nprocs is the current number of
137 	 * processes, maxproc is the limit.
138 	 */
139 	uid = p1->p_cred->p_ruid;
140 	if (__predict_false((nprocs >= maxproc - 1 && uid != 0) ||
141 			    nprocs >= maxproc)) {
142 		tablefull("proc", "increase kern.maxproc or NPROC");
143 		return (EAGAIN);
144 	}
145 
146 	/*
147 	 * Increment the count of procs running with this uid. Don't allow
148 	 * a nonprivileged user to exceed their current limit.
149 	 */
150 	count = chgproccnt(uid, 1);
151 	if (__predict_false(uid != 0 && count >
152 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
153 		(void)chgproccnt(uid, -1);
154 		return (EAGAIN);
155 	}
156 
157 	/*
158 	 * Allocate virtual address space for the U-area now, while it
159 	 * is still easy to abort the fork operation if we're out of
160 	 * kernel virtual address space.  The actual U-area pages will
161 	 * be allocated and wired in vm_fork().
162 	 */
163 	uaddr = uvm_km_valloc(kernel_map, USPACE);
164 	if (__predict_false(uaddr == 0)) {
165 		(void)chgproccnt(uid, -1);
166 		return (ENOMEM);
167 	}
168 
169 	/*
170 	 * We are now committed to the fork.  From here on, we may
171 	 * block on resources, but resource allocation may NOT fail.
172 	 */
173 
174 	/* Allocate new proc. */
175 	newproc = pool_get(&proc_pool, PR_WAITOK);
176 
177 	/*
178 	 * BEGIN PID ALLOCATION.
179 	 */
180 	s = proclist_lock_write();
181 
182 	/*
183 	 * Find an unused process ID.  We remember a range of unused IDs
184 	 * ready to use (from nextpid+1 through pidchecked-1).
185 	 */
186 	nextpid++;
187 retry:
188 	/*
189 	 * If the process ID prototype has wrapped around,
190 	 * restart somewhat above 0, as the low-numbered procs
191 	 * tend to include daemons that don't exit.
192 	 */
193 	if (nextpid >= PID_MAX) {
194 		nextpid = 100;
195 		pidchecked = 0;
196 	}
197 	if (nextpid >= pidchecked) {
198 		const struct proclist_desc *pd;
199 
200 		pidchecked = PID_MAX;
201 		/*
202 		 * Scan the process lists to check whether this pid
203 		 * is in use.  Remember the lowest pid that's greater
204 		 * than nextpid, so we can avoid checking for a while.
205 		 */
206 		pd = proclists;
207 again:
208 		for (p2 = LIST_FIRST(pd->pd_list); p2 != 0;
209 		     p2 = LIST_NEXT(p2, p_list)) {
210 			while (p2->p_pid == nextpid ||
211 			    p2->p_pgrp->pg_id == nextpid ||
212 			    p2->p_session->s_sid == nextpid) {
213 				nextpid++;
214 				if (nextpid >= pidchecked)
215 					goto retry;
216 			}
217 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
218 				pidchecked = p2->p_pid;
219 
220 			if (p2->p_pgrp->pg_id > nextpid &&
221 			    pidchecked > p2->p_pgrp->pg_id)
222 				pidchecked = p2->p_pgrp->pg_id;
223 
224 			if (p2->p_session->s_sid > nextpid &&
225 			    pidchecked > p2->p_session->s_sid)
226 				pidchecked = p2->p_session->s_sid;
227 		}
228 
229 		/*
230 		 * If there's another list, scan it.  If we have checked
231 		 * them all, we've found one!
232 		 */
233 		pd++;
234 		if (pd->pd_list != NULL)
235 			goto again;
236 	}
237 
238 	nprocs++;
239 	p2 = newproc;
240 
241 	/* Record the pid we've allocated. */
242 	p2->p_pid = nextpid;
243 
244 	/* Record the signal to be delivered to the parent on exit. */
245 	p2->p_exitsig = exitsig;
246 
247 	/*
248 	 * Put the proc on allproc before unlocking PID allocation
249 	 * so that waiters won't grab it as soon as we unlock.
250 	 */
251 
252 	p2->p_stat = SIDL;			/* protect against others */
253 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
254 
255 	LIST_INSERT_HEAD(&allproc, p2, p_list);
256 
257 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
258 
259 	/*
260 	 * END PID ALLOCATION.
261 	 */
262 	proclist_unlock_write(s);
263 
264 	/*
265 	 * Make a proc table entry for the new process.
266 	 * Start by zeroing the section of proc that is zero-initialized,
267 	 * then copy the section that is copied directly from the parent.
268 	 */
269 	memset(&p2->p_startzero, 0,
270 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
271 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
272 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
273 
274 #if !defined(MULTIPROCESSOR)
275 	/*
276 	 * In the single-processor case, all processes will always run
277 	 * on the same CPU.  So, initialize the child's CPU to the parent's
278 	 * now.  In the multiprocessor case, the child's CPU will be
279 	 * initialized in the low-level context switch code when the
280 	 * process runs.
281 	 */
282 	p2->p_cpu = p1->p_cpu;
283 #endif /* ! MULTIPROCESSOR */
284 
285 	/*
286 	 * Duplicate sub-structures as needed.
287 	 * Increase reference counts on shared objects.
288 	 * The p_stats and p_sigacts substructs are set in vm_fork.
289 	 */
290 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
291 	p2->p_emul = p1->p_emul;
292 	if (p1->p_flag & P_PROFIL)
293 		startprofclock(p2);
294 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
295 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
296 	p2->p_cred->p_refcnt = 1;
297 	crhold(p1->p_ucred);
298 
299 	/* bump references to the text vnode (for procfs) */
300 	p2->p_textvp = p1->p_textvp;
301 	if (p2->p_textvp)
302 		VREF(p2->p_textvp);
303 
304 	if (flags & FORK_SHAREFILES)
305 		fdshare(p1, p2);
306 	else
307 		p2->p_fd = fdcopy(p1);
308 
309 	if (flags & FORK_SHARECWD)
310 		cwdshare(p1, p2);
311 	else
312 		p2->p_cwdi = cwdinit(p1);
313 
314 	/*
315 	 * If p_limit is still copy-on-write, bump refcnt,
316 	 * otherwise get a copy that won't be modified.
317 	 * (If PL_SHAREMOD is clear, the structure is shared
318 	 * copy-on-write.)
319 	 */
320 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
321 		p2->p_limit = limcopy(p1->p_limit);
322 	else {
323 		p2->p_limit = p1->p_limit;
324 		p2->p_limit->p_refcnt++;
325 	}
326 
327 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
328 		p2->p_flag |= P_CONTROLT;
329 	if (flags & FORK_PPWAIT)
330 		p2->p_flag |= P_PPWAIT;
331 	LIST_INSERT_AFTER(p1, p2, p_pglist);
332 	p2->p_pptr = p1;
333 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
334 	LIST_INIT(&p2->p_children);
335 
336 	callout_init(&p2->p_realit_ch);
337 	callout_init(&p2->p_tsleep_ch);
338 
339 #ifdef KTRACE
340 	/*
341 	 * Copy traceflag and tracefile if enabled.
342 	 * If not inherited, these were zeroed above.
343 	 */
344 	if (p1->p_traceflag&KTRFAC_INHERIT) {
345 		p2->p_traceflag = p1->p_traceflag;
346 		if ((p2->p_tracep = p1->p_tracep) != NULL)
347 			ktradref(p2);
348 	}
349 #endif
350 	scheduler_fork_hook(p1, p2);
351 
352 	/*
353 	 * Create signal actions for the child process.
354 	 */
355 	if (flags & FORK_SHARESIGS)
356 		sigactsshare(p1, p2);
357 	else
358 		p2->p_sigacts = sigactsinit(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 	 * Make child runnable, set start time, and add to run queue.
378 	 */
379 	s = splstatclock();
380 	p2->p_stats->p_start = time;
381 	p2->p_acflag = AFORK;
382 	p2->p_stat = SRUN;
383 	setrunqueue(p2);
384 	splx(s);
385 
386 	/*
387 	 * Now can be swapped.
388 	 */
389 	PRELE(p1);
390 
391 	/*
392 	 * Update stats now that we know the fork was successful.
393 	 */
394 	uvmexp.forks++;
395 	if (flags & FORK_PPWAIT)
396 		uvmexp.forks_ppwait++;
397 	if (flags & FORK_SHAREVM)
398 		uvmexp.forks_sharevm++;
399 
400 	/*
401 	 * Pass a pointer to the new process to the caller.
402 	 */
403 	if (rnewprocp != NULL)
404 		*rnewprocp = p2;
405 
406 	/*
407 	 * Preserve synchronization semantics of vfork.  If waiting for
408 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
409 	 * proc (in case of exit).
410 	 */
411 	if (flags & FORK_PPWAIT)
412 		while (p2->p_flag & P_PPWAIT)
413 			tsleep(p1, PWAIT, "ppwait", 0);
414 
415 	/*
416 	 * Return child pid to parent process,
417 	 * marking us as parent via retval[1].
418 	 */
419 	if (retval != NULL) {
420 		retval[0] = p2->p_pid;
421 		retval[1] = 0;
422 	}
423 	return (0);
424 }
425