xref: /openbsd-src/sys/kern/kern_fork.c (revision cf1e490fb54cdfefb5a907fe39e03744593142e1)
1 /*	$OpenBSD: kern_fork.c,v 1.180 2015/03/14 07:33:42 jsg 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 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, flags, NULL, 0, fork_return, NULL, retval, NULL));
105 }
106 
107 int
108 sys_vfork(struct proc *p, void *v, register_t *retval)
109 {
110 	return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
111 	    NULL, retval, NULL));
112 }
113 
114 int
115 sys___tfork(struct proc *p, void *v, register_t *retval)
116 {
117 	struct sys___tfork_args /* {
118 		syscallarg(const struct __tfork) *param;
119 		syscallarg(size_t) psize;
120 	} */ *uap = v;
121 	size_t psize = SCARG(uap, psize);
122 	struct __tfork param = { 0 };
123 	int flags;
124 	int error;
125 
126 	if (psize == 0 || psize > sizeof(param))
127 		return (EINVAL);
128 	if ((error = copyin(SCARG(uap, param), &param, psize)))
129 		return (error);
130 #ifdef KTRACE
131 	if (KTRPOINT(p, KTR_STRUCT))
132 		ktrstruct(p, "tfork", &param, sizeof(param));
133 #endif
134 
135 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
136 	    | FORK_SHAREFILES;
137 
138 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
139 	    tfork_child_return, param.tf_tcb, retval, NULL));
140 }
141 
142 void
143 tfork_child_return(void *arg)
144 {
145 	struct proc *p = curproc;
146 
147 	TCB_SET(p, arg);
148 	child_return(p);
149 }
150 
151 /*
152  * Initialize common bits of a process structure, given the initial thread.
153  */
154 void
155 process_initialize(struct process *pr, struct proc *p)
156 {
157 	/* initialize the thread links */
158 	pr->ps_mainproc = p;
159 	TAILQ_INIT(&pr->ps_threads);
160 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
161 	pr->ps_refcnt = 1;
162 	p->p_p = pr;
163 
164 	/* give the process the same creds as the initial thread */
165 	pr->ps_ucred = p->p_ucred;
166 	crhold(pr->ps_ucred);
167 	KASSERT(p->p_ucred->cr_ref >= 2);	/* new thread and new process */
168 
169 	LIST_INIT(&pr->ps_children);
170 
171 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
172 }
173 
174 
175 /*
176  * Allocate and initialize a new process.
177  */
178 void
179 process_new(struct proc *p, struct process *parent, int flags)
180 {
181 	struct process *pr;
182 
183 	pr = pool_get(&process_pool, PR_WAITOK);
184 
185 	/*
186 	 * Make a process structure for the new process.
187 	 * Start by zeroing the section of proc that is zero-initialized,
188 	 * then copy the section that is copied directly from the parent.
189 	 */
190 	memset(&pr->ps_startzero, 0,
191 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
192 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
193 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
194 
195 	process_initialize(pr, p);
196 
197 	/* post-copy fixups */
198 	pr->ps_pptr = parent;
199 	pr->ps_limit->p_refcnt++;
200 
201 	/* bump references to the text vnode (for sysctl) */
202 	pr->ps_textvp = parent->ps_textvp;
203 	if (pr->ps_textvp)
204 		vref(pr->ps_textvp);
205 
206 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
207 	if (parent->ps_session->s_ttyvp != NULL)
208 		pr->ps_flags |= parent->ps_flags & PS_CONTROLT;
209 
210 	/*
211 	 * Duplicate sub-structures as needed.
212 	 * Increase reference counts on shared objects.
213 	 */
214 	if (flags & FORK_SHAREFILES)
215 		pr->ps_fd = fdshare(parent);
216 	else
217 		pr->ps_fd = fdcopy(parent);
218 	if (flags & FORK_SIGHAND)
219 		pr->ps_sigacts = sigactsshare(parent);
220 	else
221 		pr->ps_sigacts = sigactsinit(parent);
222 	if (flags & FORK_SHAREVM)
223 		pr->ps_vmspace = uvmspace_share(parent);
224 	else
225 		pr->ps_vmspace = uvmspace_fork(parent);
226 
227 	if (parent->ps_flags & PS_PROFIL)
228 		startprofclock(pr);
229 	if (flags & FORK_PTRACE)
230 		pr->ps_flags |= parent->ps_flags & PS_TRACED;
231 	if (flags & FORK_NOZOMBIE)
232 		pr->ps_flags |= PS_NOZOMBIE;
233 	if (flags & FORK_SYSTEM)
234 		pr->ps_flags |= PS_SYSTEM;
235 
236 	/* mark as embryo to protect against others */
237 	pr->ps_flags |= PS_EMBRYO;
238 
239 	/* Force visibility of all of the above changes */
240 	membar_producer();
241 
242 	/* it's sufficiently inited to be globally visible */
243 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
244 }
245 
246 /* print the 'table full' message once per 10 seconds */
247 struct timeval fork_tfmrate = { 10, 0 };
248 
249 int
250 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
251     void (*func)(void *), void *arg, register_t *retval,
252     struct proc **rnewprocp)
253 {
254 	struct process *curpr = curp->p_p;
255 	struct process *pr;
256 	struct proc *p;
257 	uid_t uid;
258 	struct vmspace *vm;
259 	int count;
260 	vaddr_t uaddr;
261 	int s;
262 	struct  ptrace_state *newptstat = NULL;
263 #if NSYSTRACE > 0
264 	void *newstrp = NULL;
265 #endif
266 
267 	/* sanity check some flag combinations */
268 	if (flags & FORK_THREAD) {
269 		if ((flags & FORK_SHAREFILES) == 0 ||
270 		    (flags & FORK_SIGHAND) == 0 ||
271 		    (flags & FORK_SYSTEM) != 0)
272 			return (EINVAL);
273 	}
274 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
275 		return (EINVAL);
276 
277 	/*
278 	 * Although process entries are dynamically created, we still keep
279 	 * a global limit on the maximum number we will create. We reserve
280 	 * the last 5 processes to root. The variable nprocesses is the
281 	 * current number of processes, maxprocess is the limit.  Similar
282 	 * rules for threads (struct proc): we reserve the last 5 to root;
283 	 * the variable nthreads is the current number of procs, maxthread is
284 	 * the limit.
285 	 */
286 	uid = curp->p_ucred->cr_ruid;
287 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
288 		static struct timeval lasttfm;
289 
290 		if (ratecheck(&lasttfm, &fork_tfmrate))
291 			tablefull("proc");
292 		return (EAGAIN);
293 	}
294 	nthreads++;
295 
296 	if ((flags & FORK_THREAD) == 0) {
297 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
298 		    nprocesses >= maxprocess) {
299 			static struct timeval lasttfm;
300 
301 			if (ratecheck(&lasttfm, &fork_tfmrate))
302 				tablefull("process");
303 			nthreads--;
304 			return (EAGAIN);
305 		}
306 		nprocesses++;
307 
308 		/*
309 		 * Increment the count of processes running with
310 		 * this uid.  Don't allow a nonprivileged user to
311 		 * exceed their current limit.
312 		 */
313 		count = chgproccnt(uid, 1);
314 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
315 			(void)chgproccnt(uid, -1);
316 			nprocesses--;
317 			nthreads--;
318 			return (EAGAIN);
319 		}
320 	}
321 
322 	uaddr = uvm_uarea_alloc();
323 	if (uaddr == 0) {
324 		if ((flags & FORK_THREAD) == 0) {
325 			(void)chgproccnt(uid, -1);
326 			nprocesses--;
327 		}
328 		nthreads--;
329 		return (ENOMEM);
330 	}
331 
332 	/*
333 	 * From now on, we're committed to the fork and cannot fail.
334 	 */
335 
336 	/* Allocate new proc. */
337 	p = pool_get(&proc_pool, PR_WAITOK);
338 
339 	p->p_stat = SIDL;			/* protect against others */
340 	p->p_flag = 0;
341 
342 	/*
343 	 * Make a proc table entry for the new process.
344 	 * Start by zeroing the section of proc that is zero-initialized,
345 	 * then copy the section that is copied directly from the parent.
346 	 */
347 	memset(&p->p_startzero, 0,
348 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
349 	memcpy(&p->p_startcopy, &curp->p_startcopy,
350 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
351 	crhold(p->p_ucred);
352 
353 	/*
354 	 * Initialize the timeouts.
355 	 */
356 	timeout_set(&p->p_sleep_to, endtsleep, p);
357 
358 	if (flags & FORK_THREAD) {
359 		atomic_setbits_int(&p->p_flag, P_THREAD);
360 		p->p_p = pr = curpr;
361 		pr->ps_refcnt++;
362 	} else {
363 		process_new(p, curpr, flags);
364 		pr = p->p_p;
365 	}
366 	p->p_fd		= pr->ps_fd;
367 	p->p_vmspace	= pr->ps_vmspace;
368 	if (pr->ps_flags & PS_SYSTEM)
369 		atomic_setbits_int(&p->p_flag, P_SYSTEM);
370 
371 	if (flags & FORK_PPWAIT) {
372 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
373 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
374 	}
375 
376 #ifdef KTRACE
377 	/*
378 	 * Copy traceflag and tracefile if enabled.
379 	 * If not inherited, these were zeroed above.
380 	 */
381 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
382 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
383 		    curpr->ps_tracecred);
384 #endif
385 
386 	/*
387 	 * set priority of child to be that of parent
388 	 * XXX should move p_estcpu into the region of struct proc which gets
389 	 * copied.
390 	 */
391 	scheduler_fork_hook(curp, p);
392 
393 	if (flags & FORK_THREAD)
394 		sigstkinit(&p->p_sigstk);
395 
396 	/*
397 	 * If emulation has thread fork hook, call it now.
398 	 */
399 	if (pr->ps_emul->e_proc_fork)
400 		(*pr->ps_emul->e_proc_fork)(p, curp);
401 
402 	p->p_addr = (struct user *)uaddr;
403 
404 	/*
405 	 * Finish creating the child thread.  cpu_fork() will copy
406 	 * and update the pcb and make the child ready to run.  If
407 	 * this is a normal user fork, the child will exit directly
408 	 * to user mode via child_return() on its first time slice
409 	 * and will not return here.  If this is a kernel thread,
410 	 * the specified entry point will be executed.
411 	 */
412 	cpu_fork(curp, p, stack, 0, func ? func : child_return, arg ? arg : p);
413 
414 	vm = pr->ps_vmspace;
415 
416 	if (flags & FORK_FORK) {
417 		forkstat.cntfork++;
418 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
419 	} else if (flags & FORK_VFORK) {
420 		forkstat.cntvfork++;
421 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
422 	} else if (flags & FORK_TFORK) {
423 		forkstat.cnttfork++;
424 	} else {
425 		forkstat.cntkthread++;
426 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
427 	}
428 
429 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
430 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
431 #if NSYSTRACE > 0
432 	if (ISSET(curp->p_flag, P_SYSTRACE))
433 		newstrp = systrace_getproc();
434 #endif
435 
436 	p->p_pid = allocpid();
437 
438 	LIST_INSERT_HEAD(&allproc, p, p_list);
439 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
440 	if ((flags & FORK_THREAD) == 0) {
441 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
442 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
443 
444 		if (pr->ps_flags & PS_TRACED) {
445 			pr->ps_oppid = curpr->ps_pid;
446 			if (pr->ps_pptr != curpr->ps_pptr)
447 				proc_reparent(pr, curpr->ps_pptr);
448 
449 			/*
450 			 * Set ptrace status.
451 			 */
452 			if (flags & FORK_FORK) {
453 				pr->ps_ptstat = newptstat;
454 				newptstat = NULL;
455 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
456 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
457 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
458 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
459 			}
460 		}
461 	} else {
462 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
463 		/*
464 		 * if somebody else wants to take us to single threaded mode,
465 		 * count ourselves in.
466 		 */
467 		if (pr->ps_single) {
468 			curpr->ps_singlecount++;
469 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
470 		}
471 	}
472 
473 #if NSYSTRACE > 0
474 	if (newstrp)
475 		systrace_fork(curp, p, newstrp);
476 #endif
477 
478 	if (tidptr != NULL) {
479 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
480 
481 		if (copyout(&pid, tidptr, sizeof(pid)))
482 			psignal(curp, SIGSEGV);
483 	}
484 
485 	/*
486 	 * For new processes, set accounting bits and mark as complete.
487 	 */
488 	if ((flags & FORK_THREAD) == 0) {
489 		getnanotime(&pr->ps_start);
490 		pr->ps_acflag = AFORK;
491 		atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO);
492 	}
493 
494 	/*
495 	 * Make child runnable and add to run queue.
496 	 */
497 	if ((flags & FORK_IDLE) == 0) {
498 		SCHED_LOCK(s);
499 		p->p_stat = SRUN;
500 		p->p_cpu = sched_choosecpu_fork(curp, flags);
501 		setrunqueue(p);
502 		SCHED_UNLOCK(s);
503 	} else
504 		p->p_cpu = arg;
505 
506 	if (newptstat)
507 		free(newptstat, M_SUBPROC, sizeof(*newptstat));
508 
509 	/*
510 	 * Notify any interested parties about the new process.
511 	 */
512 	if ((flags & FORK_THREAD) == 0)
513 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
514 
515 	/*
516 	 * Update stats now that we know the fork was successful.
517 	 */
518 	uvmexp.forks++;
519 	if (flags & FORK_PPWAIT)
520 		uvmexp.forks_ppwait++;
521 	if (flags & FORK_SHAREVM)
522 		uvmexp.forks_sharevm++;
523 
524 	/*
525 	 * Pass a pointer to the new process to the caller.
526 	 */
527 	if (rnewprocp != NULL)
528 		*rnewprocp = p;
529 
530 	/*
531 	 * Preserve synchronization semantics of vfork.  If waiting for
532 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
533 	 * on ourselves, and sleep on our process for the latter flag
534 	 * to go away.
535 	 * XXX Need to stop other rthreads in the parent
536 	 */
537 	if (flags & FORK_PPWAIT)
538 		while (curpr->ps_flags & PS_ISPWAIT)
539 			tsleep(curpr, PWAIT, "ppwait", 0);
540 
541 	/*
542 	 * If we're tracing the child, alert the parent too.
543 	 */
544 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
545 		psignal(curp, SIGTRAP);
546 
547 	/*
548 	 * Return child pid to parent process,
549 	 * marking us as parent via retval[1].
550 	 */
551 	if (retval != NULL) {
552 		retval[0] = p->p_pid +
553 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
554 		retval[1] = 0;
555 	}
556 	return (0);
557 }
558 
559 /*
560  * Checks for current use of a pid, either as a pid or pgid.
561  */
562 pid_t oldpids[100];
563 int
564 ispidtaken(pid_t pid)
565 {
566 	uint32_t i;
567 	struct process *pr;
568 
569 	for (i = 0; i < nitems(oldpids); i++)
570 		if (pid == oldpids[i])
571 			return (1);
572 
573 	if (pfind(pid) != NULL)
574 		return (1);
575 	if (pgfind(pid) != NULL)
576 		return (1);
577 	LIST_FOREACH(pr, &zombprocess, ps_list) {
578 		if (pr->ps_pid == pid ||
579 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
580 			return (1);
581 	}
582 	return (0);
583 }
584 
585 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
586 pid_t
587 allocpid(void)
588 {
589 	static pid_t lastpid;
590 	pid_t pid;
591 
592 	if (!randompid) {
593 		/* only used early on for system processes */
594 		pid = ++lastpid;
595 	} else {
596 		do {
597 			pid = 1 + arc4random_uniform(PID_MAX);
598 		} while (ispidtaken(pid));
599 	}
600 
601 	return pid;
602 }
603 
604 void
605 freepid(pid_t pid)
606 {
607 	static uint32_t idx;
608 
609 	oldpids[idx++ % nitems(oldpids)] = pid;
610 }
611 
612 #if defined(MULTIPROCESSOR)
613 /*
614  * XXX This is a slight hack to get newly-formed processes to
615  * XXX acquire the kernel lock as soon as they run.
616  */
617 void
618 proc_trampoline_mp(void)
619 {
620 	struct proc *p;
621 
622 	p = curproc;
623 
624 	SCHED_ASSERT_LOCKED();
625 	__mp_unlock(&sched_lock);
626 	spl0();
627 	SCHED_ASSERT_UNLOCKED();
628 	KERNEL_ASSERT_UNLOCKED();
629 
630 	KERNEL_LOCK();
631 }
632 #endif
633