xref: /openbsd-src/sys/kern/kern_fork.c (revision 0c5bca190bd7b51928def75cadfcc7274c101a8d) 
1 /*	$OpenBSD: kern_fork.c,v 1.95 2008/05/11 23:54:40 tedu Exp $	*/
2 /*	$NetBSD: kern_fork.c,v 1.29 1996/02/09 18:59:34 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1989, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_fork.c	8.6 (Berkeley) 4/8/94
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/filedesc.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/proc.h>
47 #include <sys/exec.h>
48 #include <sys/resourcevar.h>
49 #include <sys/signalvar.h>
50 #include <sys/vnode.h>
51 #include <sys/file.h>
52 #include <sys/acct.h>
53 #include <sys/ktrace.h>
54 #include <sys/sched.h>
55 #include <dev/rndvar.h>
56 #include <sys/pool.h>
57 #include <sys/mman.h>
58 #include <sys/ptrace.h>
59 
60 #include <sys/syscallargs.h>
61 
62 #include "systrace.h"
63 #include <dev/systrace.h>
64 
65 #include <uvm/uvm_extern.h>
66 #include <uvm/uvm_map.h>
67 
68 int	nprocs = 1;		/* process 0 */
69 int	randompid;		/* when set to 1, pid's go random */
70 pid_t	lastpid;
71 struct	forkstat forkstat;
72 
73 void fork_return(void *);
74 int pidtaken(pid_t);
75 
76 void process_new(struct proc *, struct proc *);
77 
78 void
79 fork_return(void *arg)
80 {
81 	struct proc *p = (struct proc *)arg;
82 
83 	if (p->p_flag & P_TRACED)
84 		psignal(p, SIGTRAP);
85 
86 	child_return(p);
87 }
88 
89 /*ARGSUSED*/
90 int
91 sys_fork(struct proc *p, void *v, register_t *retval)
92 {
93 	int flags;
94 
95 	flags = FORK_FORK;
96 	if (p->p_ptmask & PTRACE_FORK)
97 		flags |= FORK_PTRACE;
98 	return (fork1(p, SIGCHLD, flags, NULL, 0,
99 	    fork_return, NULL, retval, NULL));
100 }
101 
102 /*ARGSUSED*/
103 int
104 sys_vfork(struct proc *p, void *v, register_t *retval)
105 {
106 	return (fork1(p, SIGCHLD, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
107 	    NULL, retval, NULL));
108 }
109 
110 int
111 sys_rfork(struct proc *p, void *v, register_t *retval)
112 {
113 	struct sys_rfork_args /* {
114 		syscallarg(int) flags;
115 	} */ *uap = v;
116 
117 	int rforkflags;
118 	int flags;
119 
120 	flags = FORK_RFORK;
121 	rforkflags = SCARG(uap, flags);
122 
123 	if ((rforkflags & RFPROC) == 0)
124 		return (EINVAL);
125 
126 	switch(rforkflags & (RFFDG|RFCFDG)) {
127 	case (RFFDG|RFCFDG):
128 		return EINVAL;
129 	case RFCFDG:
130 		flags |= FORK_CLEANFILES;
131 		break;
132 	case RFFDG:
133 		break;
134 	default:
135 		flags |= FORK_SHAREFILES;
136 		break;
137 	}
138 
139 	if (rforkflags & RFNOWAIT)
140 		flags |= FORK_NOZOMBIE;
141 
142 	if (rforkflags & RFMEM)
143 		flags |= FORK_SHAREVM;
144 #ifdef RTHREADS
145 	if (rforkflags & RFTHREAD)
146 		flags |= FORK_THREAD | FORK_SIGHAND;
147 #endif
148 
149 	return (fork1(p, SIGCHLD, flags, NULL, 0, NULL, NULL, retval, NULL));
150 }
151 
152 /*
153  * Allocate and initialize a new process.
154  */
155 void
156 process_new(struct proc *newproc, struct proc *parent)
157 {
158 	struct process *pr;
159 
160 	pr = pool_get(&process_pool, PR_WAITOK);
161 	pr->ps_mainproc = newproc;
162 	TAILQ_INIT(&pr->ps_threads);
163 	TAILQ_INSERT_TAIL(&pr->ps_threads, newproc, p_thr_link);
164 	newproc->p_p = pr;
165 }
166 
167 /* print the 'table full' message once per 10 seconds */
168 struct timeval fork_tfmrate = { 10, 0 };
169 
170 int
171 fork1(struct proc *p1, int exitsig, int flags, void *stack, size_t stacksize,
172     void (*func)(void *), void *arg, register_t *retval,
173     struct proc **rnewprocp)
174 {
175 	struct proc *p2;
176 	uid_t uid;
177 	struct vmspace *vm;
178 	int count;
179 	vaddr_t uaddr;
180 	int s;
181 	extern void endtsleep(void *);
182 	extern void realitexpire(void *);
183 
184 	/*
185 	 * Although process entries are dynamically created, we still keep
186 	 * a global limit on the maximum number we will create. We reserve
187 	 * the last 5 processes to root. The variable nprocs is the current
188 	 * number of processes, maxproc is the limit.
189 	 */
190 	uid = p1->p_cred->p_ruid;
191 	if ((nprocs >= maxproc - 5 && uid != 0) || nprocs >= maxproc) {
192 		static struct timeval lasttfm;
193 
194 		if (ratecheck(&lasttfm, &fork_tfmrate))
195 			tablefull("proc");
196 		return (EAGAIN);
197 	}
198 	nprocs++;
199 
200 	/*
201 	 * Increment the count of procs running with this uid. Don't allow
202 	 * a nonprivileged user to exceed their current limit.
203 	 */
204 	count = chgproccnt(uid, 1);
205 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
206 		(void)chgproccnt(uid, -1);
207 		nprocs--;
208 		return (EAGAIN);
209 	}
210 
211 	uaddr = uvm_km_alloc1(kernel_map, USPACE, USPACE_ALIGN, 1);
212 	if (uaddr == 0) {
213 		chgproccnt(uid, -1);
214 		nprocs--;
215 		return (ENOMEM);
216 	}
217 
218 	/*
219 	 * From now on, we're committed to the fork and cannot fail.
220 	 */
221 
222 	/* Allocate new proc. */
223 	p2 = pool_get(&proc_pool, PR_WAITOK);
224 
225 	p2->p_stat = SIDL;			/* protect against others */
226 	p2->p_exitsig = exitsig;
227 	p2->p_flag = 0;
228 
229 #ifdef RTHREADS
230 	if (flags & FORK_THREAD) {
231 		atomic_setbits_int(&p2->p_flag, P_THREAD);
232 		p2->p_p = p1->p_p;
233 		TAILQ_INSERT_TAIL(&p2->p_p->ps_threads, p2, p_thr_link);
234 	} else {
235 		process_new(p2, p1);
236 	}
237 #else
238 	process_new(p2, p1);
239 #endif
240 
241 	/*
242 	 * Make a proc table entry for the new process.
243 	 * Start by zeroing the section of proc that is zero-initialized,
244 	 * then copy the section that is copied directly from the parent.
245 	 */
246 	bzero(&p2->p_startzero,
247 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
248 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
249 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
250 
251 	/*
252 	 * Initialize the timeouts.
253 	 */
254 	timeout_set(&p2->p_sleep_to, endtsleep, p2);
255 	timeout_set(&p2->p_realit_to, realitexpire, p2);
256 
257 	p2->p_cpu = p1->p_cpu;
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_emul = p1->p_emul;
265 	if (p1->p_flag & P_PROFIL)
266 		startprofclock(p2);
267 	atomic_setbits_int(&p2->p_flag, p1->p_flag & (P_SUGID | P_SUGIDEXEC));
268 	if (flags & FORK_PTRACE)
269 		atomic_setbits_int(&p2->p_flag, p1->p_flag & P_TRACED);
270 #ifdef RTHREADS
271 	if (flags & FORK_THREAD) {
272 		/* nothing */
273 	} else
274 #endif
275 	{
276 		p2->p_p->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
277 		bcopy(p1->p_p->ps_cred, p2->p_p->ps_cred, sizeof(*p2->p_p->ps_cred));
278 		p2->p_p->ps_cred->p_refcnt = 1;
279 		crhold(p1->p_ucred);
280 	}
281 
282 	/* bump references to the text vnode (for procfs) */
283 	p2->p_textvp = p1->p_textvp;
284 	if (p2->p_textvp)
285 		VREF(p2->p_textvp);
286 
287 	if (flags & FORK_CLEANFILES)
288 		p2->p_fd = fdinit(p1);
289 	else if (flags & FORK_SHAREFILES)
290 		p2->p_fd = fdshare(p1);
291 	else
292 		p2->p_fd = fdcopy(p1);
293 
294 	/*
295 	 * If ps_limit is still copy-on-write, bump refcnt,
296 	 * otherwise get a copy that won't be modified.
297 	 * (If PL_SHAREMOD is clear, the structure is shared
298 	 * copy-on-write.)
299 	 */
300 #ifdef RTHREADS
301 	if (flags & FORK_THREAD) {
302 		/* nothing */
303 	} else
304 #endif
305 	{
306 		if (p1->p_p->ps_limit->p_lflags & PL_SHAREMOD)
307 			p2->p_p->ps_limit = limcopy(p1->p_p->ps_limit);
308 		else {
309 			p2->p_p->ps_limit = p1->p_p->ps_limit;
310 			p2->p_p->ps_limit->p_refcnt++;
311 		}
312 	}
313 
314 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
315 		atomic_setbits_int(&p2->p_flag, P_CONTROLT);
316 	if (flags & FORK_PPWAIT)
317 		atomic_setbits_int(&p2->p_flag, P_PPWAIT);
318 	p2->p_pptr = p1;
319 	if (flags & FORK_NOZOMBIE)
320 		atomic_setbits_int(&p2->p_flag, P_NOZOMBIE);
321 	LIST_INIT(&p2->p_children);
322 
323 #ifdef KTRACE
324 	/*
325 	 * Copy traceflag and tracefile if enabled.
326 	 * If not inherited, these were zeroed above.
327 	 */
328 	if (p1->p_traceflag & KTRFAC_INHERIT) {
329 		p2->p_traceflag = p1->p_traceflag;
330 		if ((p2->p_tracep = p1->p_tracep) != NULL)
331 			VREF(p2->p_tracep);
332 	}
333 #endif
334 
335 	/*
336 	 * set priority of child to be that of parent
337 	 * XXX should move p_estcpu into the region of struct proc which gets
338 	 * copied.
339 	 */
340 	scheduler_fork_hook(p1, p2);
341 
342 	/*
343 	 * Create signal actions for the child process.
344 	 */
345 	if (flags & FORK_SIGHAND)
346 		sigactsshare(p1, p2);
347 	else
348 		p2->p_sigacts = sigactsinit(p1);
349 
350 	/*
351 	 * If emulation has process fork hook, call it now.
352 	 */
353 	if (p2->p_emul->e_proc_fork)
354 		(*p2->p_emul->e_proc_fork)(p2, p1);
355 
356 	p2->p_addr = (struct user *)uaddr;
357 
358 	/*
359 	 * Finish creating the child process.  It will return through a
360 	 * different path later.
361 	 */
362 	uvm_fork(p1, p2, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
363 	    stacksize, func ? func : child_return, arg ? arg : p2);
364 
365 	timeout_set(&p2->p_stats->p_virt_to, virttimer_trampoline, p2);
366 	timeout_set(&p2->p_stats->p_prof_to, proftimer_trampoline, p2);
367 
368 	vm = p2->p_vmspace;
369 
370 	if (flags & FORK_FORK) {
371 		forkstat.cntfork++;
372 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
373 	} else if (flags & FORK_VFORK) {
374 		forkstat.cntvfork++;
375 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
376 	} else if (flags & FORK_RFORK) {
377 		forkstat.cntrfork++;
378 		forkstat.sizrfork += vm->vm_dsize + vm->vm_ssize;
379 	} else {
380 		forkstat.cntkthread++;
381 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
382 	}
383 
384 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
385 	do {
386 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
387 	} while (pidtaken(lastpid));
388 	p2->p_pid = lastpid;
389 
390 	LIST_INSERT_HEAD(&allproc, p2, p_list);
391 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
392 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
393 	LIST_INSERT_AFTER(p1, p2, p_pglist);
394 	if (p2->p_flag & P_TRACED) {
395 		p2->p_oppid = p1->p_pid;
396 		if (p2->p_pptr != p1->p_pptr)
397 			proc_reparent(p2, p1->p_pptr);
398 
399 		/*
400 		 * Set ptrace status.
401 		 */
402 		if (flags & FORK_FORK) {
403 			p2->p_ptstat = malloc(sizeof(*p2->p_ptstat),
404 			    M_SUBPROC, M_WAITOK);
405 			p1->p_ptstat->pe_report_event = PTRACE_FORK;
406 			p2->p_ptstat->pe_report_event = PTRACE_FORK;
407 			p1->p_ptstat->pe_other_pid = p2->p_pid;
408 			p2->p_ptstat->pe_other_pid = p1->p_pid;
409 		}
410 	}
411 
412 #if NSYSTRACE > 0
413 	if (ISSET(p1->p_flag, P_SYSTRACE))
414 		systrace_fork(p1, p2);
415 #endif
416 
417 	/*
418 	 * Make child runnable, set start time, and add to run queue.
419 	 */
420 	SCHED_LOCK(s);
421  	getmicrotime(&p2->p_stats->p_start);
422 	p2->p_acflag = AFORK;
423 	p2->p_stat = SRUN;
424 	setrunqueue(p2);
425 	SCHED_UNLOCK(s);
426 
427 	/*
428 	 * Notify any interested parties about the new process.
429 	 */
430 	KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
431 
432 	/*
433 	 * Update stats now that we know the fork was successfull.
434 	 */
435 	uvmexp.forks++;
436 	if (flags & FORK_PPWAIT)
437 		uvmexp.forks_ppwait++;
438 	if (flags & FORK_SHAREVM)
439 		uvmexp.forks_sharevm++;
440 
441 	/*
442 	 * Pass a pointer to the new process to the caller.
443 	 */
444 	if (rnewprocp != NULL)
445 		*rnewprocp = p2;
446 
447 	/*
448 	 * Preserve synchronization semantics of vfork.  If waiting for
449 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
450 	 * proc (in case of exit).
451 	 */
452 	if (flags & FORK_PPWAIT)
453 		while (p2->p_flag & P_PPWAIT)
454 			tsleep(p1, PWAIT, "ppwait", 0);
455 
456 	/*
457 	 * If we're tracing the child, alert the parent too.
458 	 */
459 	if ((flags & FORK_PTRACE) && (p1->p_flag & P_TRACED))
460 		psignal(p1, SIGTRAP);
461 
462 	/*
463 	 * Return child pid to parent process,
464 	 * marking us as parent via retval[1].
465 	 */
466 	if (retval != NULL) {
467 		retval[0] = p2->p_pid;
468 		retval[1] = 0;
469 	}
470 	return (0);
471 }
472 
473 /*
474  * Checks for current use of a pid, either as a pid or pgid.
475  */
476 int
477 pidtaken(pid_t pid)
478 {
479 	struct proc *p;
480 
481 	if (pfind(pid) != NULL)
482 		return (1);
483 	if (pgfind(pid) != NULL)
484 		return (1);
485 	LIST_FOREACH(p, &zombproc, p_list)
486 		if (p->p_pid == pid || p->p_pgid == pid)
487 			return (1);
488 	return (0);
489 }
490 
491 #if defined(MULTIPROCESSOR)
492 /*
493  * XXX This is a slight hack to get newly-formed processes to
494  * XXX acquire the kernel lock as soon as they run.
495  */
496 void
497 proc_trampoline_mp(void)
498 {
499 	struct proc *p;
500 
501 	p = curproc;
502 
503 	SCHED_ASSERT_LOCKED();
504 	__mp_unlock(&sched_lock);
505 	spl0();
506 	SCHED_ASSERT_UNLOCKED();
507 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
508 
509 	KERNEL_PROC_LOCK(p);
510 }
511 #endif
512