xref: /openbsd-src/sys/kern/kern_fork.c (revision 0303b08324c895da9001caa5e6b5e4140157dac4) 
1 /*	$OpenBSD: kern_fork.c,v 1.140 2012/05/10 05:01:23 guenther 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 #ifdef __HAVE_MD_TCB
69 # include <machine/tcb.h>
70 #endif
71 
72 int	nprocesses = 1;		/* process 0 */
73 int	nthreads = 1;		/* proc 0 */
74 int	randompid;		/* when set to 1, pid's go random */
75 pid_t	lastpid;
76 struct	forkstat forkstat;
77 
78 void fork_return(void *);
79 void tfork_child_return(void *);
80 int pidtaken(pid_t);
81 
82 void process_new(struct proc *, struct process *);
83 
84 void
85 fork_return(void *arg)
86 {
87 	struct proc *p = (struct proc *)arg;
88 
89 	if (p->p_p->ps_flags & PS_TRACED)
90 		psignal(p, SIGTRAP);
91 
92 	child_return(p);
93 }
94 
95 /*ARGSUSED*/
96 int
97 sys_fork(struct proc *p, void *v, register_t *retval)
98 {
99 	int flags;
100 
101 	flags = FORK_FORK;
102 	if (p->p_p->ps_ptmask & PTRACE_FORK)
103 		flags |= FORK_PTRACE;
104 	return (fork1(p, SIGCHLD, flags, NULL, 0,
105 	    fork_return, NULL, retval, NULL));
106 }
107 
108 /*ARGSUSED*/
109 int
110 sys_vfork(struct proc *p, void *v, register_t *retval)
111 {
112 	return (fork1(p, SIGCHLD, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
113 	    NULL, retval, NULL));
114 }
115 
116 int
117 sys___tfork(struct proc *p, void *v, register_t *retval)
118 {
119 	struct sys___tfork_args /* {
120 		syscallarg(struct __tfork) *param;
121 	} */ *uap = v;
122 	struct __tfork param;
123 	int flags;
124 	int error;
125 
126 	if ((error = copyin(SCARG(uap, param), &param, sizeof(param))))
127 		return (error);
128 
129 	if (param.tf_flags != 0)
130 		return (EINVAL);
131 
132 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
133 	    | FORK_NOZOMBIE | FORK_SHAREFILES;
134 
135 	return (fork1(p, 0, flags, NULL, param.tf_tid, tfork_child_return,
136 	    param.tf_tcb, retval, NULL));
137 }
138 
139 void
140 tfork_child_return(void *arg)
141 {
142 	struct proc *p = curproc;
143 
144 	TCB_SET(p, arg);
145 	child_return(p);
146 }
147 
148 /*
149  * Allocate and initialize a new process.
150  */
151 void
152 process_new(struct proc *p, struct process *parent)
153 {
154 	struct process *pr;
155 
156 	pr = pool_get(&process_pool, PR_WAITOK);
157 	pr->ps_mainproc = p;
158 
159 	TAILQ_INIT(&pr->ps_threads);
160 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
161 	pr->ps_pptr = parent;
162 	LIST_INIT(&pr->ps_children);
163 	pr->ps_refcnt = 1;
164 
165 	/*
166 	 * Make a process structure for the new process.
167 	 * Start by zeroing the section of proc that is zero-initialized,
168 	 * then copy the section that is copied directly from the parent.
169 	 */
170 	bzero(&pr->ps_startzero,
171 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
172 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
173 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
174 
175 	/* post-copy fixups */
176 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
177 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
178 	crhold(parent->ps_cred->pc_ucred);
179 	pr->ps_limit->p_refcnt++;
180 
181 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
182 	timeout_set(&pr->ps_virt_to, virttimer_trampoline, pr);
183 	timeout_set(&pr->ps_prof_to, proftimer_trampoline, pr);
184 
185 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
186 	if (parent->ps_session->s_ttyvp != NULL &&
187 	    parent->ps_flags & PS_CONTROLT)
188 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
189 
190 	p->p_p = pr;
191 }
192 
193 /* print the 'table full' message once per 10 seconds */
194 struct timeval fork_tfmrate = { 10, 0 };
195 
196 int
197 fork1(struct proc *curp, int exitsig, int flags, void *stack, pid_t *tidptr,
198     void (*func)(void *), void *arg, register_t *retval,
199     struct proc **rnewprocp)
200 {
201 	struct process *curpr = curp->p_p;
202 	struct process *pr;
203 	struct proc *p;
204 	uid_t uid;
205 	struct vmspace *vm;
206 	int count;
207 	vaddr_t uaddr;
208 	int s;
209 	struct  ptrace_state *newptstat = NULL;
210 #if NSYSTRACE > 0
211 	void *newstrp = NULL;
212 #endif
213 
214 	/* sanity check some flag combinations */
215 	if (flags & FORK_THREAD) {
216 		if (!rthreads_enabled)
217 			return (ENOTSUP);
218 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
219 		    (FORK_SIGHAND | FORK_NOZOMBIE))
220 			return (EINVAL);
221 	}
222 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
223 		return (EINVAL);
224 
225 	/*
226 	 * Although process entries are dynamically created, we still keep
227 	 * a global limit on the maximum number we will create. We reserve
228 	 * the last 5 processes to root. The variable nprocesses is the
229 	 * current number of processes, maxprocess is the limit.  Similar
230 	 * rules for threads (struct proc): we reserve the last 5 to root;
231 	 * the variable nthreads is the current number of procs, maxthread is
232 	 * the limit.
233 	 */
234 	uid = curp->p_cred->p_ruid;
235 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
236 		static struct timeval lasttfm;
237 
238 		if (ratecheck(&lasttfm, &fork_tfmrate))
239 			tablefull("proc");
240 		return (EAGAIN);
241 	}
242 	nthreads++;
243 
244 	if ((flags & FORK_THREAD) == 0) {
245 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
246 		    nprocesses >= maxprocess) {
247 			static struct timeval lasttfm;
248 
249 			if (ratecheck(&lasttfm, &fork_tfmrate))
250 				tablefull("process");
251 			nthreads--;
252 			return (EAGAIN);
253 		}
254 		nprocesses++;
255 
256 		/*
257 		 * Increment the count of processes running with
258 		 * this uid.  Don't allow a nonprivileged user to
259 		 * exceed their current limit.
260 		 */
261 		count = chgproccnt(uid, 1);
262 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
263 			(void)chgproccnt(uid, -1);
264 			nprocesses--;
265 			nthreads--;
266 			return (EAGAIN);
267 		}
268 	}
269 
270 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
271 	    USPACE_ALIGN, UVM_KMF_ZERO,
272 	    no_constraint.ucr_low, no_constraint.ucr_high,
273 	    0, 0, USPACE/PAGE_SIZE);
274 	if (uaddr == 0) {
275 		chgproccnt(uid, -1);
276 		nprocesses--;
277 		nthreads--;
278 		return (ENOMEM);
279 	}
280 
281 	/*
282 	 * From now on, we're committed to the fork and cannot fail.
283 	 */
284 
285 	/* Allocate new proc. */
286 	p = pool_get(&proc_pool, PR_WAITOK);
287 
288 	p->p_stat = SIDL;			/* protect against others */
289 	p->p_exitsig = exitsig;
290 	p->p_flag = 0;
291 	p->p_xstat = 0;
292 
293 	if (flags & FORK_THREAD) {
294 		atomic_setbits_int(&p->p_flag, P_THREAD);
295 		p->p_p = pr = curpr;
296 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
297 		pr->ps_refcnt++;
298 	} else {
299 		process_new(p, curpr);
300 		pr = p->p_p;
301 	}
302 
303 	/*
304 	 * Make a proc table entry for the new process.
305 	 * Start by zeroing the section of proc that is zero-initialized,
306 	 * then copy the section that is copied directly from the parent.
307 	 */
308 	bzero(&p->p_startzero,
309 	    (unsigned) ((caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero));
310 	bcopy(&curp->p_startcopy, &p->p_startcopy,
311 	    (unsigned) ((caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy));
312 
313 	/*
314 	 * Initialize the timeouts.
315 	 */
316 	timeout_set(&p->p_sleep_to, endtsleep, p);
317 
318 	/*
319 	 * Duplicate sub-structures as needed.
320 	 * Increase reference counts on shared objects.
321 	 */
322 	if (curp->p_flag & P_PROFIL)
323 		startprofclock(p);
324 	if (flags & FORK_PTRACE)
325 		atomic_setbits_int(&pr->ps_flags, curpr->ps_flags & PS_TRACED);
326 
327 	/* bump references to the text vnode (for procfs) */
328 	p->p_textvp = curp->p_textvp;
329 	if (p->p_textvp)
330 		vref(p->p_textvp);
331 
332 	if (flags & FORK_SHAREFILES)
333 		p->p_fd = fdshare(curp);
334 	else
335 		p->p_fd = fdcopy(curp);
336 
337 	if (flags & FORK_PPWAIT) {
338 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
339 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
340 	}
341 	if (flags & FORK_NOZOMBIE)
342 		atomic_setbits_int(&p->p_flag, P_NOZOMBIE);
343 
344 #ifdef KTRACE
345 	/*
346 	 * Copy traceflag and tracefile if enabled.
347 	 * If not inherited, these were zeroed above.
348 	 */
349 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
350 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
351 		    curpr->ps_tracecred);
352 #endif
353 
354 	/*
355 	 * set priority of child to be that of parent
356 	 * XXX should move p_estcpu into the region of struct proc which gets
357 	 * copied.
358 	 */
359 	scheduler_fork_hook(curp, p);
360 
361 	/*
362 	 * Create signal actions for the child process.
363 	 */
364 	if (flags & FORK_SIGHAND)
365 		p->p_sigacts = sigactsshare(curp);
366 	else
367 		p->p_sigacts = sigactsinit(curp);
368 	if (flags & FORK_THREAD)
369 		sigstkinit(&p->p_sigstk);
370 
371 	/*
372 	 * If emulation has process fork hook, call it now.
373 	 */
374 	if (p->p_emul->e_proc_fork)
375 		(*p->p_emul->e_proc_fork)(p, curp);
376 
377 	p->p_addr = (struct user *)uaddr;
378 
379 	/*
380 	 * Finish creating the child process.  It will return through a
381 	 * different path later.
382 	 */
383 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
384 	    0, func ? func : child_return, arg ? arg : p);
385 
386 	vm = p->p_vmspace;
387 
388 	if (flags & FORK_FORK) {
389 		forkstat.cntfork++;
390 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
391 	} else if (flags & FORK_VFORK) {
392 		forkstat.cntvfork++;
393 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
394 	} else if (flags & FORK_TFORK) {
395 		forkstat.cnttfork++;
396 	} else {
397 		forkstat.cntkthread++;
398 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
399 	}
400 
401 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
402 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
403 #if NSYSTRACE > 0
404 	if (ISSET(curp->p_flag, P_SYSTRACE))
405 		newstrp = systrace_getproc();
406 #endif
407 
408 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
409 	do {
410 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
411 	} while (pidtaken(lastpid));
412 	p->p_pid = lastpid;
413 
414 	LIST_INSERT_HEAD(&allproc, p, p_list);
415 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
416 	if ((flags & FORK_THREAD) == 0) {
417 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
418 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
419 
420 		if (pr->ps_flags & PS_TRACED) {
421 			pr->ps_oppid = curpr->ps_pid;
422 			if (pr->ps_pptr != curpr->ps_pptr)
423 				proc_reparent(pr, curpr->ps_pptr);
424 
425 			/*
426 			 * Set ptrace status.
427 			 */
428 			if (flags & FORK_FORK) {
429 				pr->ps_ptstat = newptstat;
430 				newptstat = NULL;
431 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
432 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
433 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
434 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
435 			}
436 		}
437 	}
438 
439 #if NSYSTRACE > 0
440 	if (newstrp)
441 		systrace_fork(curp, p, newstrp);
442 #endif
443 
444 	if (tidptr != NULL) {
445 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
446 
447 		if (copyout(&pid, tidptr, sizeof(pid)))
448 			psignal(curp, SIGSEGV);
449 	}
450 
451 	/*
452 	 * For new processes, set accounting bits
453 	 */
454 	if ((flags & FORK_THREAD) == 0) {
455 		getmicrotime(&pr->ps_start);
456 		pr->ps_acflag = AFORK;
457 	}
458 
459 	/*
460 	 * Make child runnable and add to run queue.
461 	 */
462 	SCHED_LOCK(s);
463 	p->p_stat = SRUN;
464 	p->p_cpu = sched_choosecpu_fork(curp, flags);
465 	setrunqueue(p);
466 	SCHED_UNLOCK(s);
467 
468 	if (newptstat)
469 		free(newptstat, M_SUBPROC);
470 
471 	/*
472 	 * Notify any interested parties about the new process.
473 	 */
474 	if ((flags & FORK_THREAD) == 0)
475 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
476 
477 	/*
478 	 * Update stats now that we know the fork was successful.
479 	 */
480 	uvmexp.forks++;
481 	if (flags & FORK_PPWAIT)
482 		uvmexp.forks_ppwait++;
483 	if (flags & FORK_SHAREVM)
484 		uvmexp.forks_sharevm++;
485 
486 	/*
487 	 * Pass a pointer to the new process to the caller.
488 	 */
489 	if (rnewprocp != NULL)
490 		*rnewprocp = p;
491 
492 	/*
493 	 * Preserve synchronization semantics of vfork.  If waiting for
494 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
495 	 * on ourselves, and sleep on our process for the latter flag
496 	 * to go away.
497 	 * XXX Need to stop other rthreads in the parent
498 	 */
499 	if (flags & FORK_PPWAIT)
500 		while (curpr->ps_flags & PS_ISPWAIT)
501 			tsleep(curpr, PWAIT, "ppwait", 0);
502 
503 	/*
504 	 * If we're tracing the child, alert the parent too.
505 	 */
506 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
507 		psignal(curp, SIGTRAP);
508 
509 	/*
510 	 * Return child pid to parent process,
511 	 * marking us as parent via retval[1].
512 	 */
513 	if (retval != NULL) {
514 		retval[0] = p->p_pid +
515 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
516 		retval[1] = 0;
517 	}
518 	return (0);
519 }
520 
521 /*
522  * Checks for current use of a pid, either as a pid or pgid.
523  */
524 int
525 pidtaken(pid_t pid)
526 {
527 	struct proc *p;
528 
529 	if (pfind(pid) != NULL)
530 		return (1);
531 	if (pgfind(pid) != NULL)
532 		return (1);
533 	LIST_FOREACH(p, &zombproc, p_list) {
534 		if (p->p_pid == pid || (p->p_p->ps_pgrp && p->p_p->ps_pgrp->pg_id == pid))
535 			return (1);
536 	}
537 	return (0);
538 }
539 
540 #if defined(MULTIPROCESSOR)
541 /*
542  * XXX This is a slight hack to get newly-formed processes to
543  * XXX acquire the kernel lock as soon as they run.
544  */
545 void
546 proc_trampoline_mp(void)
547 {
548 	struct proc *p;
549 
550 	p = curproc;
551 
552 	SCHED_ASSERT_LOCKED();
553 	__mp_unlock(&sched_lock);
554 	spl0();
555 	SCHED_ASSERT_UNLOCKED();
556 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
557 
558 	KERNEL_LOCK();
559 }
560 #endif
561