xref: /openbsd-src/sys/kern/kern_fork.c (revision d559b8cb6cdd6f912edd28712213aec79fb0b7bc) 
1 /*	$OpenBSD: kern_fork.c,v 1.162 2014/03/30 21:54:48 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.h>
66 
67 #ifdef __HAVE_MD_TCB
68 # include <machine/tcb.h>
69 #endif
70 
71 int	nprocesses = 1;		/* process 0 */
72 int	nthreads = 1;		/* proc 0 */
73 int	randompid;		/* when set to 1, pid's go random */
74 struct	forkstat forkstat;
75 
76 void fork_return(void *);
77 void tfork_child_return(void *);
78 int pidtaken(pid_t);
79 
80 void process_new(struct proc *, struct process *, int);
81 
82 void
83 fork_return(void *arg)
84 {
85 	struct proc *p = (struct proc *)arg;
86 
87 	if (p->p_p->ps_flags & PS_TRACED)
88 		psignal(p, SIGTRAP);
89 
90 	child_return(p);
91 }
92 
93 /*ARGSUSED*/
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 /*ARGSUSED*/
106 int
107 sys_vfork(struct proc *p, void *v, register_t *retval)
108 {
109 	return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL,
110 	    NULL, retval, NULL));
111 }
112 
113 int
114 sys___tfork(struct proc *p, void *v, register_t *retval)
115 {
116 	struct sys___tfork_args /* {
117 		syscallarg(const struct __tfork) *param;
118 		syscallarg(size_t) psize;
119 	} */ *uap = v;
120 	size_t psize = SCARG(uap, psize);
121 	struct __tfork param = { 0 };
122 	int flags;
123 	int error;
124 
125 	if (psize == 0 || psize > sizeof(param))
126 		return (EINVAL);
127 	if ((error = copyin(SCARG(uap, param), &param, psize)))
128 		return (error);
129 #ifdef KTRACE
130 	if (KTRPOINT(p, KTR_STRUCT))
131 		ktrstruct(p, "tfork", &param, sizeof(param));
132 #endif
133 
134 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
135 	    | FORK_SHAREFILES;
136 
137 	return (fork1(p, flags, param.tf_stack, param.tf_tid,
138 	    tfork_child_return, param.tf_tcb, retval, NULL));
139 }
140 
141 void
142 tfork_child_return(void *arg)
143 {
144 	struct proc *p = curproc;
145 
146 	TCB_SET(p, arg);
147 	child_return(p);
148 }
149 
150 /*
151  * Allocate and initialize a new process.
152  */
153 void
154 process_new(struct proc *p, struct process *parent, int flags)
155 {
156 	struct process *pr;
157 
158 	pr = pool_get(&process_pool, PR_WAITOK);
159 	pr->ps_mainproc = p;
160 
161 	TAILQ_INIT(&pr->ps_threads);
162 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
163 	pr->ps_pptr = parent;
164 	LIST_INIT(&pr->ps_children);
165 	pr->ps_refcnt = 1;
166 
167 	/*
168 	 * Make a process structure for the new process.
169 	 * Start by zeroing the section of proc that is zero-initialized,
170 	 * then copy the section that is copied directly from the parent.
171 	 */
172 	memset(&pr->ps_startzero, 0,
173 	    (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero);
174 	memcpy(&pr->ps_startcopy, &parent->ps_startcopy,
175 	    (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy);
176 
177 	/* post-copy fixups */
178 	pr->ps_ucred = parent->ps_ucred;
179 	crhold(pr->ps_ucred);
180 	pr->ps_limit->p_refcnt++;
181 
182 	/* bump references to the text vnode (for procfs) */
183 	pr->ps_textvp = parent->ps_textvp;
184 	if (pr->ps_textvp)
185 		vref(pr->ps_textvp);
186 
187 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
188 
189 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
190 	if (parent->ps_session->s_ttyvp != NULL &&
191 	    parent->ps_flags & PS_CONTROLT)
192 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
193 
194 	p->p_p = pr;
195 
196 	/*
197 	 * Create signal actions for the child process.
198 	 */
199 	if (flags & FORK_SIGHAND)
200 		pr->ps_sigacts = sigactsshare(parent);
201 	else
202 		pr->ps_sigacts = sigactsinit(parent);
203 
204 	if (parent->ps_flags & PS_PROFIL)
205 		startprofclock(pr);
206 	if ((flags & FORK_PTRACE) && (parent->ps_flags & PS_TRACED))
207 		atomic_setbits_int(&pr->ps_flags, PS_TRACED);
208 	if (flags & FORK_NOZOMBIE)
209 		atomic_setbits_int(&pr->ps_flags, PS_NOZOMBIE);
210 
211 	/* it's sufficiently inited to be globally visible */
212 	LIST_INSERT_HEAD(&allprocess, pr, ps_list);
213 }
214 
215 /* print the 'table full' message once per 10 seconds */
216 struct timeval fork_tfmrate = { 10, 0 };
217 
218 int
219 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr,
220     void (*func)(void *), void *arg, register_t *retval,
221     struct proc **rnewprocp)
222 {
223 	struct process *curpr = curp->p_p;
224 	struct process *pr;
225 	struct proc *p;
226 	uid_t uid;
227 	struct vmspace *vm;
228 	int count;
229 	vaddr_t uaddr;
230 	int s;
231 	struct  ptrace_state *newptstat = NULL;
232 #if NSYSTRACE > 0
233 	void *newstrp = NULL;
234 #endif
235 
236 	/* sanity check some flag combinations */
237 	if (flags & FORK_THREAD) {
238 		if ((flags & FORK_SIGHAND) == 0)
239 			return (EINVAL);
240 	}
241 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
242 		return (EINVAL);
243 
244 	/*
245 	 * Although process entries are dynamically created, we still keep
246 	 * a global limit on the maximum number we will create. We reserve
247 	 * the last 5 processes to root. The variable nprocesses is the
248 	 * current number of processes, maxprocess is the limit.  Similar
249 	 * rules for threads (struct proc): we reserve the last 5 to root;
250 	 * the variable nthreads is the current number of procs, maxthread is
251 	 * the limit.
252 	 */
253 	uid = curp->p_ucred->cr_ruid;
254 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
255 		static struct timeval lasttfm;
256 
257 		if (ratecheck(&lasttfm, &fork_tfmrate))
258 			tablefull("proc");
259 		return (EAGAIN);
260 	}
261 	nthreads++;
262 
263 	if ((flags & FORK_THREAD) == 0) {
264 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
265 		    nprocesses >= maxprocess) {
266 			static struct timeval lasttfm;
267 
268 			if (ratecheck(&lasttfm, &fork_tfmrate))
269 				tablefull("process");
270 			nthreads--;
271 			return (EAGAIN);
272 		}
273 		nprocesses++;
274 
275 		/*
276 		 * Increment the count of processes running with
277 		 * this uid.  Don't allow a nonprivileged user to
278 		 * exceed their current limit.
279 		 */
280 		count = chgproccnt(uid, 1);
281 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
282 			(void)chgproccnt(uid, -1);
283 			nprocesses--;
284 			nthreads--;
285 			return (EAGAIN);
286 		}
287 	}
288 
289 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
290 	    USPACE_ALIGN, UVM_KMF_ZERO,
291 	    no_constraint.ucr_low, no_constraint.ucr_high,
292 	    0, 0, USPACE/PAGE_SIZE);
293 	if (uaddr == 0) {
294 		if ((flags & FORK_THREAD) == 0) {
295 			(void)chgproccnt(uid, -1);
296 			nprocesses--;
297 		}
298 		nthreads--;
299 		return (ENOMEM);
300 	}
301 
302 	/*
303 	 * From now on, we're committed to the fork and cannot fail.
304 	 */
305 
306 	/* Allocate new proc. */
307 	p = pool_get(&proc_pool, PR_WAITOK);
308 
309 	p->p_stat = SIDL;			/* protect against others */
310 	p->p_flag = 0;
311 
312 	/*
313 	 * Make a proc table entry for the new process.
314 	 * Start by zeroing the section of proc that is zero-initialized,
315 	 * then copy the section that is copied directly from the parent.
316 	 */
317 	memset(&p->p_startzero, 0,
318 	    (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero);
319 	memcpy(&p->p_startcopy, &curp->p_startcopy,
320 	    (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy);
321 
322 	/*
323 	 * Initialize the timeouts.
324 	 */
325 	timeout_set(&p->p_sleep_to, endtsleep, p);
326 
327 	if (flags & FORK_THREAD) {
328 		atomic_setbits_int(&p->p_flag, P_THREAD);
329 		p->p_p = pr = curpr;
330 		pr->ps_refcnt++;
331 	} else {
332 		process_new(p, curpr, flags);
333 		pr = p->p_p;
334 	}
335 
336 	/*
337 	 * Duplicate sub-structures as needed.
338 	 * Increase reference counts on shared objects.
339 	 */
340 	if (flags & FORK_SHAREFILES)
341 		p->p_fd = fdshare(curp);
342 	else
343 		p->p_fd = fdcopy(curp);
344 
345 	if (flags & FORK_PPWAIT) {
346 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
347 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
348 	}
349 
350 #ifdef KTRACE
351 	/*
352 	 * Copy traceflag and tracefile if enabled.
353 	 * If not inherited, these were zeroed above.
354 	 */
355 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
356 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
357 		    curpr->ps_tracecred);
358 #endif
359 
360 	/*
361 	 * set priority of child to be that of parent
362 	 * XXX should move p_estcpu into the region of struct proc which gets
363 	 * copied.
364 	 */
365 	scheduler_fork_hook(curp, p);
366 
367 	if (flags & FORK_THREAD)
368 		sigstkinit(&p->p_sigstk);
369 
370 	/*
371 	 * If emulation has thread fork hook, call it now.
372 	 */
373 	if (pr->ps_emul->e_proc_fork)
374 		(*pr->ps_emul->e_proc_fork)(p, curp);
375 
376 	p->p_addr = (struct user *)uaddr;
377 
378 	/*
379 	 * Finish creating the child process.  It will return through a
380 	 * different path later.
381 	 */
382 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
383 	    0, func ? func : child_return, arg ? arg : p);
384 
385 	vm = p->p_vmspace;
386 
387 	if (flags & FORK_FORK) {
388 		forkstat.cntfork++;
389 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
390 	} else if (flags & FORK_VFORK) {
391 		forkstat.cntvfork++;
392 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
393 	} else if (flags & FORK_TFORK) {
394 		forkstat.cnttfork++;
395 	} else {
396 		forkstat.cntkthread++;
397 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
398 	}
399 
400 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
401 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
402 #if NSYSTRACE > 0
403 	if (ISSET(curp->p_flag, P_SYSTRACE))
404 		newstrp = systrace_getproc();
405 #endif
406 
407 	p->p_pid = allocpid();
408 
409 	LIST_INSERT_HEAD(&allproc, p, p_list);
410 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
411 	if ((flags & FORK_THREAD) == 0) {
412 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
413 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
414 
415 		if (pr->ps_flags & PS_TRACED) {
416 			pr->ps_oppid = curpr->ps_pid;
417 			if (pr->ps_pptr != curpr->ps_pptr)
418 				proc_reparent(pr, curpr->ps_pptr);
419 
420 			/*
421 			 * Set ptrace status.
422 			 */
423 			if (flags & FORK_FORK) {
424 				pr->ps_ptstat = newptstat;
425 				newptstat = NULL;
426 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
427 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
428 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
429 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
430 			}
431 		}
432 	} else {
433 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
434 		/*
435 		 * if somebody else wants to take us to single threaded mode,
436 		 * count ourselves in.
437 		 */
438 		if (pr->ps_single) {
439 			curpr->ps_singlecount++;
440 			atomic_setbits_int(&p->p_flag, P_SUSPSINGLE);
441 		}
442 	}
443 
444 #if NSYSTRACE > 0
445 	if (newstrp)
446 		systrace_fork(curp, p, newstrp);
447 #endif
448 
449 	if (tidptr != NULL) {
450 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
451 
452 		if (copyout(&pid, tidptr, sizeof(pid)))
453 			psignal(curp, SIGSEGV);
454 	}
455 
456 	/*
457 	 * For new processes, set accounting bits
458 	 */
459 	if ((flags & FORK_THREAD) == 0) {
460 		getnanotime(&pr->ps_start);
461 		pr->ps_acflag = AFORK;
462 	}
463 
464 	/*
465 	 * Make child runnable and add to run queue.
466 	 */
467 	if ((flags & FORK_IDLE) == 0) {
468 		SCHED_LOCK(s);
469 		p->p_stat = SRUN;
470 		p->p_cpu = sched_choosecpu_fork(curp, flags);
471 		setrunqueue(p);
472 		SCHED_UNLOCK(s);
473 	} else
474 		p->p_cpu = arg;
475 
476 	if (newptstat)
477 		free(newptstat, M_SUBPROC);
478 
479 	/*
480 	 * Notify any interested parties about the new process.
481 	 */
482 	if ((flags & FORK_THREAD) == 0)
483 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
484 
485 	/*
486 	 * Update stats now that we know the fork was successful.
487 	 */
488 	uvmexp.forks++;
489 	if (flags & FORK_PPWAIT)
490 		uvmexp.forks_ppwait++;
491 	if (flags & FORK_SHAREVM)
492 		uvmexp.forks_sharevm++;
493 
494 	/*
495 	 * Pass a pointer to the new process to the caller.
496 	 */
497 	if (rnewprocp != NULL)
498 		*rnewprocp = p;
499 
500 	/*
501 	 * Preserve synchronization semantics of vfork.  If waiting for
502 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
503 	 * on ourselves, and sleep on our process for the latter flag
504 	 * to go away.
505 	 * XXX Need to stop other rthreads in the parent
506 	 */
507 	if (flags & FORK_PPWAIT)
508 		while (curpr->ps_flags & PS_ISPWAIT)
509 			tsleep(curpr, PWAIT, "ppwait", 0);
510 
511 	/*
512 	 * If we're tracing the child, alert the parent too.
513 	 */
514 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
515 		psignal(curp, SIGTRAP);
516 
517 	/*
518 	 * Return child pid to parent process,
519 	 * marking us as parent via retval[1].
520 	 */
521 	if (retval != NULL) {
522 		retval[0] = p->p_pid +
523 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
524 		retval[1] = 0;
525 	}
526 	return (0);
527 }
528 
529 /*
530  * Checks for current use of a pid, either as a pid or pgid.
531  */
532 pid_t oldpids[100];
533 int
534 ispidtaken(pid_t pid)
535 {
536 	uint32_t i;
537 	struct process *pr;
538 
539 	for (i = 0; i < nitems(oldpids); i++)
540 		if (pid == oldpids[i])
541 			return (1);
542 
543 	if (pfind(pid) != NULL)
544 		return (1);
545 	if (pgfind(pid) != NULL)
546 		return (1);
547 	LIST_FOREACH(pr, &zombprocess, ps_list) {
548 		if (pr->ps_pid == pid ||
549 		    (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid))
550 			return (1);
551 	}
552 	return (0);
553 }
554 
555 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
556 pid_t
557 allocpid(void)
558 {
559 	static pid_t lastpid;
560 	pid_t pid;
561 
562 	if (!randompid) {
563 		/* only used early on for system processes */
564 		pid = ++lastpid;
565 	} else {
566 		do {
567 			pid = 1 + arc4random_uniform(PID_MAX);
568 		} while (ispidtaken(pid));
569 	}
570 
571 	return pid;
572 }
573 
574 void
575 freepid(pid_t pid)
576 {
577 	static uint32_t idx;
578 
579 	oldpids[idx++ % nitems(oldpids)] = pid;
580 }
581 
582 #if defined(MULTIPROCESSOR)
583 /*
584  * XXX This is a slight hack to get newly-formed processes to
585  * XXX acquire the kernel lock as soon as they run.
586  */
587 void
588 proc_trampoline_mp(void)
589 {
590 	struct proc *p;
591 
592 	p = curproc;
593 
594 	SCHED_ASSERT_LOCKED();
595 	__mp_unlock(&sched_lock);
596 	spl0();
597 	SCHED_ASSERT_UNLOCKED();
598 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
599 
600 	KERNEL_LOCK();
601 }
602 #endif
603