xref: /netbsd-src/sys/kern/kern_fork.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: kern_fork.c,v 1.66 2000/05/31 05:02:32 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_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 <vm/vm.h>
67 #include <vm/vm_kern.h>
68 
69 #include <uvm/uvm_extern.h>
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, SIGCHLD, NULL, 0, NULL, NULL, 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, SIGCHLD, NULL, 0, NULL, NULL,
97 	    retval, NULL));
98 }
99 
100 /*
101  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
102  * semantics.  Address space is shared, and parent is blocked until child exit.
103  */
104 /*ARGSUSED*/
105 int
106 sys___vfork14(p, v, retval)
107 	struct proc *p;
108 	void *v;
109 	register_t *retval;
110 {
111 
112 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
113 	    NULL, NULL, retval, NULL));
114 }
115 
116 int
117 fork1(p1, flags, exitsig, stack, stacksize, func, arg, retval, rnewprocp)
118 	struct proc *p1;
119 	int flags;
120 	int exitsig;
121 	void *stack;
122 	size_t stacksize;
123 	void (*func) __P((void *));
124 	void *arg;
125 	register_t *retval;
126 	struct proc **rnewprocp;
127 {
128 	struct proc *p2;
129 	uid_t uid;
130 	struct proc *newproc;
131 	int count, s;
132 	vaddr_t uaddr;
133 	static int nextpid, pidchecked = 0;
134 
135 	/*
136 	 * Although process entries are dynamically created, we still keep
137 	 * a global limit on the maximum number we will create.  Don't allow
138 	 * a nonprivileged user to use the last process; don't let root
139 	 * exceed the limit. The variable nprocs is the current number of
140 	 * processes, maxproc is the limit.
141 	 */
142 	uid = p1->p_cred->p_ruid;
143 	if (__predict_false((nprocs >= maxproc - 1 && uid != 0) ||
144 			    nprocs >= maxproc)) {
145 		tablefull("proc");
146 		return (EAGAIN);
147 	}
148 
149 	/*
150 	 * Increment the count of procs running with this uid. Don't allow
151 	 * a nonprivileged user to exceed their current limit.
152 	 */
153 	count = chgproccnt(uid, 1);
154 	if (__predict_false(uid != 0 && count >
155 			    p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
156 		(void)chgproccnt(uid, -1);
157 		return (EAGAIN);
158 	}
159 
160 	/*
161 	 * Allocate virtual address space for the U-area now, while it
162 	 * is still easy to abort the fork operation if we're out of
163 	 * kernel virtual address space.  The actual U-area pages will
164 	 * be allocated and wired in vm_fork().
165 	 */
166 	uaddr = uvm_km_valloc(kernel_map, USPACE);
167 	if (__predict_false(uaddr == 0)) {
168 		(void)chgproccnt(uid, -1);
169 		return (ENOMEM);
170 	}
171 
172 	/*
173 	 * We are now committed to the fork.  From here on, we may
174 	 * block on resources, but resource allocation may NOT fail.
175 	 */
176 
177 	/* Allocate new proc. */
178 	newproc = pool_get(&proc_pool, PR_WAITOK);
179 
180 	/*
181 	 * BEGIN PID ALLOCATION.
182 	 */
183 	s = proclist_lock_write();
184 
185 	/*
186 	 * Find an unused process ID.  We remember a range of unused IDs
187 	 * ready to use (from nextpid+1 through pidchecked-1).
188 	 */
189 	nextpid++;
190 retry:
191 	/*
192 	 * If the process ID prototype has wrapped around,
193 	 * restart somewhat above 0, as the low-numbered procs
194 	 * tend to include daemons that don't exit.
195 	 */
196 	if (nextpid >= PID_MAX) {
197 		nextpid = 100;
198 		pidchecked = 0;
199 	}
200 	if (nextpid >= pidchecked) {
201 		const struct proclist_desc *pd;
202 
203 		pidchecked = PID_MAX;
204 		/*
205 		 * Scan the process lists to check whether this pid
206 		 * is in use.  Remember the lowest pid that's greater
207 		 * than nextpid, so we can avoid checking for a while.
208 		 */
209 		pd = proclists;
210 again:
211 		for (p2 = LIST_FIRST(pd->pd_list); p2 != 0;
212 		     p2 = LIST_NEXT(p2, p_list)) {
213 			while (p2->p_pid == nextpid ||
214 			    p2->p_pgrp->pg_id == nextpid ||
215 			    p2->p_session->s_sid == nextpid) {
216 				nextpid++;
217 				if (nextpid >= pidchecked)
218 					goto retry;
219 			}
220 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
221 				pidchecked = p2->p_pid;
222 
223 			if (p2->p_pgrp->pg_id > nextpid &&
224 			    pidchecked > p2->p_pgrp->pg_id)
225 				pidchecked = p2->p_pgrp->pg_id;
226 
227 			if (p2->p_session->s_sid > nextpid &&
228 			    pidchecked > p2->p_session->s_sid)
229 				pidchecked = p2->p_session->s_sid;
230 		}
231 
232 		/*
233 		 * If there's another list, scan it.  If we have checked
234 		 * them all, we've found one!
235 		 */
236 		pd++;
237 		if (pd->pd_list != NULL)
238 			goto again;
239 	}
240 
241 	nprocs++;
242 	p2 = newproc;
243 
244 	/* Record the pid we've allocated. */
245 	p2->p_pid = nextpid;
246 
247 	/* Record the signal to be delivered to the parent on exit. */
248 	p2->p_exitsig = exitsig;
249 
250 	/*
251 	 * Put the proc on allproc before unlocking PID allocation
252 	 * so that waiters won't grab it as soon as we unlock.
253 	 */
254 
255 	p2->p_stat = SIDL;			/* protect against others */
256 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
257 
258 	LIST_INSERT_HEAD(&allproc, p2, p_list);
259 
260 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
261 
262 	/*
263 	 * END PID ALLOCATION.
264 	 */
265 	proclist_unlock_write(s);
266 
267 	/*
268 	 * Make a proc table entry for the new process.
269 	 * Start by zeroing the section of proc that is zero-initialized,
270 	 * then copy the section that is copied directly from the parent.
271 	 */
272 	memset(&p2->p_startzero, 0,
273 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
274 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
275 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
276 
277 #if !defined(MULTIPROCESSOR)
278 	/*
279 	 * In the single-processor case, all processes will always run
280 	 * on the same CPU.  So, initialize the child's CPU to the parent's
281 	 * now.  In the multiprocessor case, the child's CPU will be
282 	 * initialized in the low-level context switch code when the
283 	 * process runs.
284 	 */
285 	p2->p_cpu = p1->p_cpu;
286 #endif /* ! MULTIPROCESSOR */
287 
288 	/*
289 	 * Duplicate sub-structures as needed.
290 	 * Increase reference counts on shared objects.
291 	 * The p_stats and p_sigacts substructs are set in vm_fork.
292 	 */
293 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
294 	p2->p_emul = p1->p_emul;
295 	if (p1->p_flag & P_PROFIL)
296 		startprofclock(p2);
297 	p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
298 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
299 	p2->p_cred->p_refcnt = 1;
300 	crhold(p1->p_ucred);
301 
302 	/* bump references to the text vnode (for procfs) */
303 	p2->p_textvp = p1->p_textvp;
304 	if (p2->p_textvp)
305 		VREF(p2->p_textvp);
306 
307 	if (flags & FORK_SHAREFILES)
308 		fdshare(p1, p2);
309 	else
310 		p2->p_fd = fdcopy(p1);
311 
312 	if (flags & FORK_SHARECWD)
313 		cwdshare(p1, p2);
314 	else
315 		p2->p_cwdi = cwdinit(p1);
316 
317 	/*
318 	 * If p_limit is still copy-on-write, bump refcnt,
319 	 * otherwise get a copy that won't be modified.
320 	 * (If PL_SHAREMOD is clear, the structure is shared
321 	 * copy-on-write.)
322 	 */
323 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
324 		p2->p_limit = limcopy(p1->p_limit);
325 	else {
326 		p2->p_limit = p1->p_limit;
327 		p2->p_limit->p_refcnt++;
328 	}
329 
330 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
331 		p2->p_flag |= P_CONTROLT;
332 	if (flags & FORK_PPWAIT)
333 		p2->p_flag |= P_PPWAIT;
334 	LIST_INSERT_AFTER(p1, p2, p_pglist);
335 	p2->p_pptr = p1;
336 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
337 	LIST_INIT(&p2->p_children);
338 
339 	callout_init(&p2->p_realit_ch);
340 	callout_init(&p2->p_tsleep_ch);
341 
342 #ifdef KTRACE
343 	/*
344 	 * Copy traceflag and tracefile if enabled.
345 	 * If not inherited, these were zeroed above.
346 	 */
347 	if (p1->p_traceflag&KTRFAC_INHERIT) {
348 		p2->p_traceflag = p1->p_traceflag;
349 		if ((p2->p_tracep = p1->p_tracep) != NULL)
350 			ktradref(p2);
351 	}
352 #endif
353 	scheduler_fork_hook(p1, p2);
354 
355 	/*
356 	 * Create signal actions for the child process.
357 	 */
358 	if (flags & FORK_SHARESIGS)
359 		sigactsshare(p1, p2);
360 	else
361 		p2->p_sigacts = sigactsinit(p1);
362 
363 	/*
364 	 * This begins the section where we must prevent the parent
365 	 * from being swapped.
366 	 */
367 	PHOLD(p1);
368 
369 	/*
370 	 * Finish creating the child process.  It will return through a
371 	 * different path later.
372 	 */
373 	p2->p_addr = (struct user *)uaddr;
374 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE,
375 	    stack, stacksize,
376 	    (func != NULL) ? func : child_return,
377 	    (arg != NULL) ? arg : p2);
378 
379 	/*
380 	 * Make child runnable, set start time, and add to run queue.
381 	 */
382 	s = splstatclock();
383 	p2->p_stats->p_start = time;
384 	p2->p_acflag = AFORK;
385 	p2->p_stat = SRUN;
386 	setrunqueue(p2);
387 	splx(s);
388 
389 	/*
390 	 * Now can be swapped.
391 	 */
392 	PRELE(p1);
393 
394 	/*
395 	 * Update stats now that we know the fork was successful.
396 	 */
397 	uvmexp.forks++;
398 	if (flags & FORK_PPWAIT)
399 		uvmexp.forks_ppwait++;
400 	if (flags & FORK_SHAREVM)
401 		uvmexp.forks_sharevm++;
402 
403 	/*
404 	 * Pass a pointer to the new process to the caller.
405 	 */
406 	if (rnewprocp != NULL)
407 		*rnewprocp = p2;
408 
409 	/*
410 	 * Preserve synchronization semantics of vfork.  If waiting for
411 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
412 	 * proc (in case of exit).
413 	 */
414 	if (flags & FORK_PPWAIT)
415 		while (p2->p_flag & P_PPWAIT)
416 			tsleep(p1, PWAIT, "ppwait", 0);
417 
418 	/*
419 	 * Return child pid to parent process,
420 	 * marking us as parent via retval[1].
421 	 */
422 	if (retval != NULL) {
423 		retval[0] = p2->p_pid;
424 		retval[1] = 0;
425 	}
426 	return (0);
427 }
428