xref: /openbsd-src/sys/kern/kern_fork.c (revision 9babc541bce666a7097e5c15b118b349eae5d803) 
1 /*	$OpenBSD: kern_fork.c,v 1.189 2016/09/03 14:29:05 jca 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/pledge.h>
62 #include <sys/unistd.h>
63 
64 #include <sys/syscallargs.h>
65 
66 #include <uvm/uvm.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 struct	forkstat forkstat;
76 
77 void fork_return(void *);
78 void tfork_child_return(void *);
79 int pidtaken(pid_t);
80 
81 void process_new(struct proc *, struct process *, int);
82 
83 void
84 fork_return(void *arg)
85 {
86 	struct proc *p = (struct proc *)arg;
87 
88 	if (p->p_p->ps_flags & PS_TRACED)
89 		psignal(p, SIGTRAP);
90 
91 	child_return(p);
92 }
93 
94 int
95 sys_fork(struct proc *p, void *v, register_t *retval)
96 {
97 	int flags;
98 
99 	flags = FORK_FORK;
100 	if (p->p_p->ps_ptmask & PTRACE_FORK)
101 		flags |= FORK_PTRACE;
102 	return (fork1(p, flags, NULL, 0, fork_return, NULL, retval, NULL));
103 }
104 
105 int
106 sys_vfork(struct proc *p, void *v, register_t *retval)
107 {
108 	return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
109 	    NULL, retval, NULL));
110 }
111 
112 int
113 sys___tfork(struct proc *p, void *v, register_t *retval)
114 {
115 	struct sys___tfork_args /* {
116 		syscallarg(const struct __tfork) *param;
117 		syscallarg(size_t) psize;
118 	} */ *uap = v;
119 	size_t psize = SCARG(uap, psize);
120 	struct __tfork param = { 0 };
121 	int flags;
122 	int error;
123 
124 	if (psize == 0 || psize > sizeof(param))
125 		return (EINVAL);
126 	if ((error = copyin(SCARG(uap, param), &param, psize)))
127 		return (error);
128 #ifdef KTRACE
129 	if (KTRPOINT(p, KTR_STRUCT))
130 		ktrstruct(p, "tfork", &param, sizeof(param));
131 #endif
132 
133 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
134 	    | FORK_SHAREFILES;
135 
136 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
137 	    tfork_child_return, param.tf_tcb, retval, NULL));
138 }
139 
140 void
141 tfork_child_return(void *arg)
142 {
143 	struct proc *p = curproc;
144 
145 	TCB_SET(p, arg);
146 	child_return(p);
147 }
148 
149 /*
150  * Initialize common bits of a process structure, given the initial thread.
151  */
152 void
153 process_initialize(struct process *pr, struct proc *p)
154 {
155 	/* initialize the thread links */
156 	pr->ps_mainproc = p;
157 	TAILQ_INIT(&pr->ps_threads);
158 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
159 	pr->ps_refcnt = 1;
160 	p->p_p = pr;
161 
162 	/* give the process the same creds as the initial thread */
163 	pr->ps_ucred = p->p_ucred;
164 	crhold(pr->ps_ucred);
165 	KASSERT(p->p_ucred->cr_ref >= 2);	/* new thread and new process */
166 
167 	LIST_INIT(&pr->ps_children);
168 
169 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
170 }
171 
172 
173 /*
174  * Allocate and initialize a new process.
175  */
176 void
177 process_new(struct proc *p, struct process *parent, int flags)
178 {
179 	struct process *pr;
180 
181 	pr = pool_get(&process_pool, PR_WAITOK);
182 
183 	/*
184 	 * Make a process structure for the new process.
185 	 * Start by zeroing the section of proc that is zero-initialized,
186 	 * then copy the section that is copied directly from the parent.
187 	 */
188 	memset(&pr->ps_startzero, 0,
189 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
190 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
191 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
192 
193 	process_initialize(pr, p);
194 
195 	/* post-copy fixups */
196 	pr->ps_pptr = parent;
197 	pr->ps_limit->p_refcnt++;
198 
199 	/* bump references to the text vnode (for sysctl) */
200 	pr->ps_textvp = parent->ps_textvp;
201 	if (pr->ps_textvp)
202 		vref(pr->ps_textvp);
203 
204 	pr->ps_flags = parent->ps_flags &
205 	    (PS_SUGID | PS_SUGIDEXEC | PS_PLEDGE | PS_WXNEEDED);
206 	if (parent->ps_session->s_ttyvp != NULL)
207 		pr->ps_flags |= parent->ps_flags & PS_CONTROLT;
208 
209 	/*
210 	 * Duplicate sub-structures as needed.
211 	 * Increase reference counts on shared objects.
212 	 */
213 	if (flags & FORK_SHAREFILES)
214 		pr->ps_fd = fdshare(parent);
215 	else
216 		pr->ps_fd = fdcopy(parent);
217 	if (flags & FORK_SIGHAND)
218 		pr->ps_sigacts = sigactsshare(parent);
219 	else
220 		pr->ps_sigacts = sigactsinit(parent);
221 	if (flags & FORK_SHAREVM)
222 		pr->ps_vmspace = uvmspace_share(parent);
223 	else
224 		pr->ps_vmspace = uvmspace_fork(parent);
225 
226 	if (pr->ps_pledgepaths)
227 		pr->ps_pledgepaths->wl_ref++;
228 
229 	if (parent->ps_flags & PS_PROFIL)
230 		startprofclock(pr);
231 	if (flags & FORK_PTRACE)
232 		pr->ps_flags |= parent->ps_flags & PS_TRACED;
233 	if (flags & FORK_NOZOMBIE)
234 		pr->ps_flags |= PS_NOZOMBIE;
235 	if (flags & FORK_SYSTEM)
236 		pr->ps_flags |= PS_SYSTEM;
237 
238 	/* mark as embryo to protect against others */
239 	pr->ps_flags |= PS_EMBRYO;
240 
241 	/* Force visibility of all of the above changes */
242 	membar_producer();
243 
244 	/* it's sufficiently inited to be globally visible */
245 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
246 }
247 
248 /* print the 'table full' message once per 10 seconds */
249 struct timeval fork_tfmrate = { 10, 0 };
250 
251 int
252 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
253     void (*func)(void *), void *arg, register_t *retval,
254     struct proc **rnewprocp)
255 {
256 	struct process *curpr = curp->p_p;
257 	struct process *pr;
258 	struct proc *p;
259 	uid_t uid;
260 	struct vmspace *vm;
261 	int count;
262 	vaddr_t uaddr;
263 	int s;
264 	struct  ptrace_state *newptstat = NULL;
265 
266 	/* sanity check some flag combinations */
267 	if (flags & FORK_THREAD) {
268 		if ((flags & FORK_SHAREFILES) == 0 ||
269 		    (flags & FORK_SIGHAND) == 0 ||
270 		    (flags & FORK_SYSTEM) != 0)
271 			return (EINVAL);
272 	}
273 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
274 		return (EINVAL);
275 
276 	/*
277 	 * Although process entries are dynamically created, we still keep
278 	 * a global limit on the maximum number we will create. We reserve
279 	 * the last 5 processes to root. The variable nprocesses is the
280 	 * current number of processes, maxprocess is the limit.  Similar
281 	 * rules for threads (struct proc): we reserve the last 5 to root;
282 	 * the variable nthreads is the current number of procs, maxthread is
283 	 * the limit.
284 	 */
285 	uid = curp->p_ucred->cr_ruid;
286 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
287 		static struct timeval lasttfm;
288 
289 		if (ratecheck(&lasttfm, &fork_tfmrate))
290 			tablefull("proc");
291 		return (EAGAIN);
292 	}
293 	nthreads++;
294 
295 	if ((flags & FORK_THREAD) == 0) {
296 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
297 		    nprocesses >= maxprocess) {
298 			static struct timeval lasttfm;
299 
300 			if (ratecheck(&lasttfm, &fork_tfmrate))
301 				tablefull("process");
302 			nthreads--;
303 			return (EAGAIN);
304 		}
305 		nprocesses++;
306 
307 		/*
308 		 * Increment the count of processes running with
309 		 * this uid.  Don't allow a nonprivileged user to
310 		 * exceed their current limit.
311 		 */
312 		count = chgproccnt(uid, 1);
313 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
314 			(void)chgproccnt(uid, -1);
315 			nprocesses--;
316 			nthreads--;
317 			return (EAGAIN);
318 		}
319 	}
320 
321 	uaddr = uvm_uarea_alloc();
322 	if (uaddr == 0) {
323 		if ((flags & FORK_THREAD) == 0) {
324 			(void)chgproccnt(uid, -1);
325 			nprocesses--;
326 		}
327 		nthreads--;
328 		return (ENOMEM);
329 	}
330 
331 	/*
332 	 * From now on, we're committed to the fork and cannot fail.
333 	 */
334 
335 	/* Allocate new proc. */
336 	p = pool_get(&proc_pool, PR_WAITOK);
337 
338 	p->p_stat = SIDL;			/* protect against others */
339 	p->p_flag = 0;
340 
341 	/*
342 	 * Make a proc table entry for the new process.
343 	 * Start by zeroing the section of proc that is zero-initialized,
344 	 * then copy the section that is copied directly from the parent.
345 	 */
346 	memset(&p->p_startzero, 0,
347 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
348 	memcpy(&p->p_startcopy, &curp->p_startcopy,
349 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
350 	crhold(p->p_ucred);
351 
352 	/*
353 	 * Initialize the timeouts.
354 	 */
355 	timeout_set(&p->p_sleep_to, endtsleep, p);
356 
357 	if (flags & FORK_THREAD) {
358 		atomic_setbits_int(&p->p_flag, P_THREAD);
359 		p->p_p = pr = curpr;
360 		pr->ps_refcnt++;
361 	} else {
362 		process_new(p, curpr, flags);
363 		pr = p->p_p;
364 	}
365 	p->p_fd		= pr->ps_fd;
366 	p->p_vmspace	= pr->ps_vmspace;
367 	if (pr->ps_flags & PS_SYSTEM)
368 		atomic_setbits_int(&p->p_flag, P_SYSTEM);
369 
370 	if (flags & FORK_PPWAIT) {
371 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
372 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
373 	}
374 
375 #ifdef KTRACE
376 	/*
377 	 * Copy traceflag and tracefile if enabled.
378 	 * If not inherited, these were zeroed above.
379 	 */
380 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
381 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
382 		    curpr->ps_tracecred);
383 #endif
384 
385 	/*
386 	 * set priority of child to be that of parent
387 	 * XXX should move p_estcpu into the region of struct proc which gets
388 	 * copied.
389 	 */
390 	scheduler_fork_hook(curp, p);
391 
392 	if (flags & FORK_THREAD)
393 		sigstkinit(&p->p_sigstk);
394 
395 	/*
396 	 * If emulation has thread fork hook, call it now.
397 	 */
398 	if (pr->ps_emul->e_proc_fork)
399 		(*pr->ps_emul->e_proc_fork)(p, curp);
400 
401 	p->p_addr = (struct user *)uaddr;
402 
403 	/*
404 	 * Finish creating the child thread.  cpu_fork() will copy
405 	 * and update the pcb and make the child ready to run.  If
406 	 * this is a normal user fork, the child will exit directly
407 	 * to user mode via child_return() on its first time slice
408 	 * and will not return here.  If this is a kernel thread,
409 	 * the specified entry point will be executed.
410 	 */
411 	cpu_fork(curp, p, stack, 0, func ? func : child_return, arg ? arg : p);
412 
413 	vm = pr->ps_vmspace;
414 
415 	if (flags & FORK_FORK) {
416 		forkstat.cntfork++;
417 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
418 	} else if (flags & FORK_VFORK) {
419 		forkstat.cntvfork++;
420 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
421 	} else if (flags & FORK_TFORK) {
422 		forkstat.cnttfork++;
423 	} else {
424 		forkstat.cntkthread++;
425 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
426 	}
427 
428 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
429 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
430 
431 	p->p_pid = allocpid();
432 
433 	LIST_INSERT_HEAD(&allproc, p, p_list);
434 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
435 	if ((flags & FORK_THREAD) == 0) {
436 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
437 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
438 
439 		if (pr->ps_flags & PS_TRACED) {
440 			pr->ps_oppid = curpr->ps_pid;
441 			if (pr->ps_pptr != curpr->ps_pptr)
442 				proc_reparent(pr, curpr->ps_pptr);
443 
444 			/*
445 			 * Set ptrace status.
446 			 */
447 			if (flags & FORK_FORK) {
448 				pr->ps_ptstat = newptstat;
449 				newptstat = NULL;
450 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
451 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
452 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
453 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
454 			}
455 		}
456 	} else {
457 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
458 		/*
459 		 * if somebody else wants to take us to single threaded mode,
460 		 * count ourselves in.
461 		 */
462 		if (pr->ps_single) {
463 			curpr->ps_singlecount++;
464 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
465 		}
466 	}
467 
468 	if (tidptr != NULL) {
469 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
470 
471 		if (copyout(&pid, tidptr, sizeof(pid)))
472 			psignal(curp, SIGSEGV);
473 	}
474 
475 	/*
476 	 * For new processes, set accounting bits and mark as complete.
477 	 */
478 	if ((flags & FORK_THREAD) == 0) {
479 		getnanotime(&pr->ps_start);
480 		pr->ps_acflag = AFORK;
481 		atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO);
482 	}
483 
484 	/*
485 	 * Make child runnable and add to run queue.
486 	 */
487 	if ((flags & FORK_IDLE) == 0) {
488 		SCHED_LOCK(s);
489 		p->p_stat = SRUN;
490 		p->p_cpu = sched_choosecpu_fork(curp, flags);
491 		setrunqueue(p);
492 		SCHED_UNLOCK(s);
493 	} else
494 		p->p_cpu = arg;
495 
496 	if (newptstat)
497 		free(newptstat, M_SUBPROC, sizeof(*newptstat));
498 
499 	/*
500 	 * Notify any interested parties about the new process.
501 	 */
502 	if ((flags & FORK_THREAD) == 0)
503 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
504 
505 	/*
506 	 * Update stats now that we know the fork was successful.
507 	 */
508 	uvmexp.forks++;
509 	if (flags & FORK_PPWAIT)
510 		uvmexp.forks_ppwait++;
511 	if (flags & FORK_SHAREVM)
512 		uvmexp.forks_sharevm++;
513 
514 	/*
515 	 * Pass a pointer to the new process to the caller.
516 	 */
517 	if (rnewprocp != NULL)
518 		*rnewprocp = p;
519 
520 	/*
521 	 * Preserve synchronization semantics of vfork.  If waiting for
522 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
523 	 * on ourselves, and sleep on our process for the latter flag
524 	 * to go away.
525 	 * XXX Need to stop other rthreads in the parent
526 	 */
527 	if (flags & FORK_PPWAIT)
528 		while (curpr->ps_flags & PS_ISPWAIT)
529 			tsleep(curpr, PWAIT, "ppwait", 0);
530 
531 	/*
532 	 * If we're tracing the child, alert the parent too.
533 	 */
534 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
535 		psignal(curp, SIGTRAP);
536 
537 	/*
538 	 * Return child pid to parent process,
539 	 * marking us as parent via retval[1].
540 	 */
541 	if (retval != NULL) {
542 		retval[0] = p->p_pid +
543 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
544 		retval[1] = 0;
545 	}
546 	return (0);
547 }
548 
549 /*
550  * Checks for current use of a pid, either as a pid or pgid.
551  */
552 pid_t oldpids[128];
553 int
554 ispidtaken(pid_t pid)
555 {
556 	uint32_t i;
557 	struct process *pr;
558 
559 	for (i = 0; i < nitems(oldpids); i++)
560 		if (pid == oldpids[i])
561 			return (1);
562 
563 	if (pfind(pid) != NULL)
564 		return (1);
565 	if (pgfind(pid) != NULL)
566 		return (1);
567 	LIST_FOREACH(pr, &zombprocess, ps_list) {
568 		if (pr->ps_pid == pid ||
569 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
570 			return (1);
571 	}
572 	return (0);
573 }
574 
575 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
576 pid_t
577 allocpid(void)
578 {
579 	static pid_t lastpid;
580 	pid_t pid;
581 
582 	if (!randompid) {
583 		/* only used early on for system processes */
584 		pid = ++lastpid;
585 	} else {
586 		do {
587 			pid = 1 + arc4random_uniform(PID_MAX);
588 		} while (ispidtaken(pid));
589 	}
590 
591 	return pid;
592 }
593 
594 void
595 freepid(pid_t pid)
596 {
597 	static uint32_t idx;
598 
599 	oldpids[idx++ % nitems(oldpids)] = pid;
600 }
601 
602 #if defined(MULTIPROCESSOR)
603 /*
604  * XXX This is a slight hack to get newly-formed processes to
605  * XXX acquire the kernel lock as soon as they run.
606  */
607 void
608 proc_trampoline_mp(void)
609 {
610 	SCHED_ASSERT_LOCKED();
611 	__mp_unlock(&sched_lock);
612 	spl0();
613 	SCHED_ASSERT_UNLOCKED();
614 	KERNEL_ASSERT_UNLOCKED();
615 
616 	KERNEL_LOCK();
617 }
618 #endif
619