xref: /openbsd/sys/kern/kern_fork.c (revision a6445c1d)
1 /*	$OpenBSD: kern_fork.c,v 1.177 2014/11/18 02:37:31 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/vmmeter.h>
52 #include <sys/file.h>
53 #include <sys/acct.h>
54 #include <sys/ktrace.h>
55 #include <sys/sched.h>
56 #include <sys/sysctl.h>
57 #include <sys/pool.h>
58 #include <sys/mman.h>
59 #include <sys/ptrace.h>
60 #include <sys/atomic.h>
61 #include <sys/unistd.h>
62 
63 #include <sys/syscallargs.h>
64 
65 #include "systrace.h"
66 #include <dev/systrace.h>
67 
68 #include <uvm/uvm.h>
69 
70 #ifdef __HAVE_MD_TCB
71 # include <machine/tcb.h>
72 #endif
73 
74 int	nprocesses = 1;		/* process 0 */
75 int	nthreads = 1;		/* proc 0 */
76 int	randompid;		/* when set to 1, pid's go random */
77 struct	forkstat forkstat;
78 
79 void fork_return(void *);
80 void tfork_child_return(void *);
81 int pidtaken(pid_t);
82 
83 void process_new(struct proc *, struct process *, int);
84 
85 void
86 fork_return(void *arg)
87 {
88 	struct proc *p = (struct proc *)arg;
89 
90 	if (p->p_p->ps_flags & PS_TRACED)
91 		psignal(p, SIGTRAP);
92 
93 	child_return(p);
94 }
95 
96 /*ARGSUSED*/
97 int
98 sys_fork(struct proc *p, void *v, register_t *retval)
99 {
100 	int flags;
101 
102 	flags = FORK_FORK;
103 	if (p->p_p->ps_ptmask & PTRACE_FORK)
104 		flags |= FORK_PTRACE;
105 	return (fork1(p, flags, NULL, 0, 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, 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(const struct __tfork) *param;
121 		syscallarg(size_t) psize;
122 	} */ *uap = v;
123 	size_t psize = SCARG(uap, psize);
124 	struct __tfork param = { 0 };
125 	int flags;
126 	int error;
127 
128 	if (psize == 0 || psize > sizeof(param))
129 		return (EINVAL);
130 	if ((error = copyin(SCARG(uap, param), &param, psize)))
131 		return (error);
132 #ifdef KTRACE
133 	if (KTRPOINT(p, KTR_STRUCT))
134 		ktrstruct(p, "tfork", &param, sizeof(param));
135 #endif
136 
137 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
138 	    | FORK_SHAREFILES;
139 
140 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
141 	    tfork_child_return, param.tf_tcb, retval, NULL));
142 }
143 
144 void
145 tfork_child_return(void *arg)
146 {
147 	struct proc *p = curproc;
148 
149 	TCB_SET(p, arg);
150 	child_return(p);
151 }
152 
153 /*
154  * Allocate and initialize a new process.
155  */
156 void
157 process_new(struct proc *p, struct process *parent, int flags)
158 {
159 	struct process *pr;
160 
161 	pr = pool_get(&process_pool, PR_WAITOK);
162 	pr->ps_mainproc = p;
163 
164 	TAILQ_INIT(&pr->ps_threads);
165 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
166 	pr->ps_pptr = parent;
167 	LIST_INIT(&pr->ps_children);
168 	pr->ps_refcnt = 1;
169 
170 	/*
171 	 * Make a process structure for the new process.
172 	 * Start by zeroing the section of proc that is zero-initialized,
173 	 * then copy the section that is copied directly from the parent.
174 	 */
175 	memset(&pr->ps_startzero, 0,
176 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
177 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
178 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
179 
180 	/* post-copy fixups */
181 	pr->ps_ucred = p->p_ucred;
182 	crhold(pr->ps_ucred);
183 	KASSERT(p->p_ucred->cr_ref >= 3); /* fork thr, new thr, new process */
184 	pr->ps_limit->p_refcnt++;
185 
186 	/* bump references to the text vnode (for sysctl) */
187 	pr->ps_textvp = parent->ps_textvp;
188 	if (pr->ps_textvp)
189 		vref(pr->ps_textvp);
190 
191 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
192 
193 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
194 	if (parent->ps_session->s_ttyvp != NULL)
195 		pr->ps_flags |= parent->ps_flags & PS_CONTROLT;
196 
197 	p->p_p = pr;
198 
199 	/*
200 	 * Duplicate sub-structures as needed.
201 	 * Increase reference counts on shared objects.
202 	 */
203 	if (flags & FORK_SHAREFILES)
204 		pr->ps_fd = fdshare(parent);
205 	else
206 		pr->ps_fd = fdcopy(parent);
207 	if (flags & FORK_SIGHAND)
208 		pr->ps_sigacts = sigactsshare(parent);
209 	else
210 		pr->ps_sigacts = sigactsinit(parent);
211 	if (flags & FORK_SHAREVM)
212 		pr->ps_vmspace = uvmspace_share(parent);
213 	else
214 		pr->ps_vmspace = uvmspace_fork(parent);
215 
216 	if (parent->ps_flags & PS_PROFIL)
217 		startprofclock(pr);
218 	if (flags & FORK_PTRACE)
219 		pr->ps_flags |= parent->ps_flags & PS_TRACED;
220 	if (flags & FORK_NOZOMBIE)
221 		pr->ps_flags |= PS_NOZOMBIE;
222 	if (flags & FORK_SYSTEM)
223 		pr->ps_flags |= PS_SYSTEM;
224 
225 	/* mark as embryo to protect against others */
226 	pr->ps_flags |= PS_EMBRYO;
227 
228 	/* Force visibility of all of the above changes */
229 	membar_producer();
230 
231 	/* it's sufficiently inited to be globally visible */
232 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
233 }
234 
235 /* print the 'table full' message once per 10 seconds */
236 struct timeval fork_tfmrate = { 10, 0 };
237 
238 int
239 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
240     void (*func)(void *), void *arg, register_t *retval,
241     struct proc **rnewprocp)
242 {
243 	struct process *curpr = curp->p_p;
244 	struct process *pr;
245 	struct proc *p;
246 	uid_t uid;
247 	struct vmspace *vm;
248 	int count;
249 	vaddr_t uaddr;
250 	int s;
251 	struct  ptrace_state *newptstat = NULL;
252 #if NSYSTRACE > 0
253 	void *newstrp = NULL;
254 #endif
255 
256 	/* sanity check some flag combinations */
257 	if (flags & FORK_THREAD) {
258 		if ((flags & FORK_SHAREFILES) == 0 ||
259 		    (flags & FORK_SIGHAND) == 0 ||
260 		    (flags & FORK_SYSTEM) != 0)
261 			return (EINVAL);
262 	}
263 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
264 		return (EINVAL);
265 
266 	/*
267 	 * Although process entries are dynamically created, we still keep
268 	 * a global limit on the maximum number we will create. We reserve
269 	 * the last 5 processes to root. The variable nprocesses is the
270 	 * current number of processes, maxprocess is the limit.  Similar
271 	 * rules for threads (struct proc): we reserve the last 5 to root;
272 	 * the variable nthreads is the current number of procs, maxthread is
273 	 * the limit.
274 	 */
275 	uid = curp->p_ucred->cr_ruid;
276 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
277 		static struct timeval lasttfm;
278 
279 		if (ratecheck(&lasttfm, &fork_tfmrate))
280 			tablefull("proc");
281 		return (EAGAIN);
282 	}
283 	nthreads++;
284 
285 	if ((flags & FORK_THREAD) == 0) {
286 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
287 		    nprocesses >= maxprocess) {
288 			static struct timeval lasttfm;
289 
290 			if (ratecheck(&lasttfm, &fork_tfmrate))
291 				tablefull("process");
292 			nthreads--;
293 			return (EAGAIN);
294 		}
295 		nprocesses++;
296 
297 		/*
298 		 * Increment the count of processes running with
299 		 * this uid.  Don't allow a nonprivileged user to
300 		 * exceed their current limit.
301 		 */
302 		count = chgproccnt(uid, 1);
303 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
304 			(void)chgproccnt(uid, -1);
305 			nprocesses--;
306 			nthreads--;
307 			return (EAGAIN);
308 		}
309 	}
310 
311 	uaddr = uvm_uarea_alloc();
312 	if (uaddr == 0) {
313 		if ((flags & FORK_THREAD) == 0) {
314 			(void)chgproccnt(uid, -1);
315 			nprocesses--;
316 		}
317 		nthreads--;
318 		return (ENOMEM);
319 	}
320 
321 	/*
322 	 * From now on, we're committed to the fork and cannot fail.
323 	 */
324 
325 	/* Allocate new proc. */
326 	p = pool_get(&proc_pool, PR_WAITOK);
327 
328 	p->p_stat = SIDL;			/* protect against others */
329 	p->p_flag = 0;
330 
331 	/*
332 	 * Make a proc table entry for the new process.
333 	 * Start by zeroing the section of proc that is zero-initialized,
334 	 * then copy the section that is copied directly from the parent.
335 	 */
336 	memset(&p->p_startzero, 0,
337 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
338 	memcpy(&p->p_startcopy, &curp->p_startcopy,
339 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
340 	crhold(p->p_ucred);
341 
342 	/*
343 	 * Initialize the timeouts.
344 	 */
345 	timeout_set(&p->p_sleep_to, endtsleep, p);
346 
347 	if (flags & FORK_THREAD) {
348 		atomic_setbits_int(&p->p_flag, P_THREAD);
349 		p->p_p = pr = curpr;
350 		pr->ps_refcnt++;
351 	} else {
352 		process_new(p, curpr, flags);
353 		pr = p->p_p;
354 	}
355 	p->p_fd		= pr->ps_fd;
356 	p->p_vmspace	= pr->ps_vmspace;
357 	if (pr->ps_flags & PS_SYSTEM)
358 		atomic_setbits_int(&p->p_flag, P_SYSTEM);
359 
360 	if (flags & FORK_PPWAIT) {
361 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
362 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
363 	}
364 
365 #ifdef KTRACE
366 	/*
367 	 * Copy traceflag and tracefile if enabled.
368 	 * If not inherited, these were zeroed above.
369 	 */
370 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
371 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
372 		    curpr->ps_tracecred);
373 #endif
374 
375 	/*
376 	 * set priority of child to be that of parent
377 	 * XXX should move p_estcpu into the region of struct proc which gets
378 	 * copied.
379 	 */
380 	scheduler_fork_hook(curp, p);
381 
382 	if (flags & FORK_THREAD)
383 		sigstkinit(&p->p_sigstk);
384 
385 	/*
386 	 * If emulation has thread fork hook, call it now.
387 	 */
388 	if (pr->ps_emul->e_proc_fork)
389 		(*pr->ps_emul->e_proc_fork)(p, curp);
390 
391 	p->p_addr = (struct user *)uaddr;
392 
393 	/*
394 	 * Finish creating the child thread.  cpu_fork() will copy
395 	 * and update the pcb and make the child ready to run.  If
396 	 * this is a normal user fork, the child will exit directly
397 	 * to user mode via child_return() on its first time slice
398 	 * and will not return here.  If this is a kernel thread,
399 	 * the specified entry point will be executed.
400 	 */
401 	cpu_fork(curp, p, stack, 0, func ? func : child_return, arg ? arg : p);
402 
403 	vm = pr->ps_vmspace;
404 
405 	if (flags & FORK_FORK) {
406 		forkstat.cntfork++;
407 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
408 	} else if (flags & FORK_VFORK) {
409 		forkstat.cntvfork++;
410 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
411 	} else if (flags & FORK_TFORK) {
412 		forkstat.cnttfork++;
413 	} else {
414 		forkstat.cntkthread++;
415 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
416 	}
417 
418 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
419 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
420 #if NSYSTRACE > 0
421 	if (ISSET(curp->p_flag, P_SYSTRACE))
422 		newstrp = systrace_getproc();
423 #endif
424 
425 	p->p_pid = allocpid();
426 
427 	LIST_INSERT_HEAD(&allproc, p, p_list);
428 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
429 	if ((flags & FORK_THREAD) == 0) {
430 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
431 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
432 
433 		if (pr->ps_flags & PS_TRACED) {
434 			pr->ps_oppid = curpr->ps_pid;
435 			if (pr->ps_pptr != curpr->ps_pptr)
436 				proc_reparent(pr, curpr->ps_pptr);
437 
438 			/*
439 			 * Set ptrace status.
440 			 */
441 			if (flags & FORK_FORK) {
442 				pr->ps_ptstat = newptstat;
443 				newptstat = NULL;
444 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
445 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
446 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
447 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
448 			}
449 		}
450 	} else {
451 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
452 		/*
453 		 * if somebody else wants to take us to single threaded mode,
454 		 * count ourselves in.
455 		 */
456 		if (pr->ps_single) {
457 			curpr->ps_singlecount++;
458 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
459 		}
460 	}
461 
462 #if NSYSTRACE > 0
463 	if (newstrp)
464 		systrace_fork(curp, p, newstrp);
465 #endif
466 
467 	if (tidptr != NULL) {
468 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
469 
470 		if (copyout(&pid, tidptr, sizeof(pid)))
471 			psignal(curp, SIGSEGV);
472 	}
473 
474 	/*
475 	 * For new processes, set accounting bits and mark as complete.
476 	 */
477 	if ((flags & FORK_THREAD) == 0) {
478 		getnanotime(&pr->ps_start);
479 		pr->ps_acflag = AFORK;
480 		atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO);
481 	}
482 
483 	/*
484 	 * Make child runnable and add to run queue.
485 	 */
486 	if ((flags & FORK_IDLE) == 0) {
487 		SCHED_LOCK(s);
488 		p->p_stat = SRUN;
489 		p->p_cpu = sched_choosecpu_fork(curp, flags);
490 		setrunqueue(p);
491 		SCHED_UNLOCK(s);
492 	} else
493 		p->p_cpu = arg;
494 
495 	if (newptstat)
496 		free(newptstat, M_SUBPROC, sizeof(*newptstat));
497 
498 	/*
499 	 * Notify any interested parties about the new process.
500 	 */
501 	if ((flags & FORK_THREAD) == 0)
502 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
503 
504 	/*
505 	 * Update stats now that we know the fork was successful.
506 	 */
507 	uvmexp.forks++;
508 	if (flags & FORK_PPWAIT)
509 		uvmexp.forks_ppwait++;
510 	if (flags & FORK_SHAREVM)
511 		uvmexp.forks_sharevm++;
512 
513 	/*
514 	 * Pass a pointer to the new process to the caller.
515 	 */
516 	if (rnewprocp != NULL)
517 		*rnewprocp = p;
518 
519 	/*
520 	 * Preserve synchronization semantics of vfork.  If waiting for
521 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
522 	 * on ourselves, and sleep on our process for the latter flag
523 	 * to go away.
524 	 * XXX Need to stop other rthreads in the parent
525 	 */
526 	if (flags & FORK_PPWAIT)
527 		while (curpr->ps_flags & PS_ISPWAIT)
528 			tsleep(curpr, PWAIT, "ppwait", 0);
529 
530 	/*
531 	 * If we're tracing the child, alert the parent too.
532 	 */
533 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
534 		psignal(curp, SIGTRAP);
535 
536 	/*
537 	 * Return child pid to parent process,
538 	 * marking us as parent via retval[1].
539 	 */
540 	if (retval != NULL) {
541 		retval[0] = p->p_pid +
542 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
543 		retval[1] = 0;
544 	}
545 	return (0);
546 }
547 
548 /*
549  * Checks for current use of a pid, either as a pid or pgid.
550  */
551 pid_t oldpids[100];
552 int
553 ispidtaken(pid_t pid)
554 {
555 	uint32_t i;
556 	struct process *pr;
557 
558 	for (i = 0; i < nitems(oldpids); i++)
559 		if (pid == oldpids[i])
560 			return (1);
561 
562 	if (pfind(pid) != NULL)
563 		return (1);
564 	if (pgfind(pid) != NULL)
565 		return (1);
566 	LIST_FOREACH(pr, &zombprocess, ps_list) {
567 		if (pr->ps_pid == pid ||
568 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
569 			return (1);
570 	}
571 	return (0);
572 }
573 
574 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
575 pid_t
576 allocpid(void)
577 {
578 	static pid_t lastpid;
579 	pid_t pid;
580 
581 	if (!randompid) {
582 		/* only used early on for system processes */
583 		pid = ++lastpid;
584 	} else {
585 		do {
586 			pid = 1 + arc4random_uniform(PID_MAX);
587 		} while (ispidtaken(pid));
588 	}
589 
590 	return pid;
591 }
592 
593 void
594 freepid(pid_t pid)
595 {
596 	static uint32_t idx;
597 
598 	oldpids[idx++ % nitems(oldpids)] = pid;
599 }
600 
601 #if defined(MULTIPROCESSOR)
602 /*
603  * XXX This is a slight hack to get newly-formed processes to
604  * XXX acquire the kernel lock as soon as they run.
605  */
606 void
607 proc_trampoline_mp(void)
608 {
609 	struct proc *p;
610 
611 	p = curproc;
612 
613 	SCHED_ASSERT_LOCKED();
614 	__mp_unlock(&sched_lock);
615 	spl0();
616 	SCHED_ASSERT_UNLOCKED();
617 	KERNEL_ASSERT_UNLOCKED();
618 
619 	KERNEL_LOCK();
620 }
621 #endif
622