xref: /openbsd-src/sys/kern/kern_fork.c (revision 18914fdf6bf023a7446c7d0d18ef21f99bafd895) 
1 /*	$OpenBSD: kern_fork.c,v 1.142 2012/08/02 03:18:48 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_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(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_NOZOMBIE | FORK_SHAREFILES;
139 
140 	return (fork1(p, 0, flags, param.tf_stack, param.tf_tid,
141 	    tfork_child_return, param.tf_tcb, retval, NULL));
142 }
143 
144 #ifdef COMPAT_O51
145 int
146 compat_o51_sys___tfork(struct proc *p, void *v, register_t *retval)
147 {
148 	struct compat_o51_sys___tfork_args /* {
149 		syscallarg(struct __tfork51) *param;
150 	} */ *uap = v;
151 	struct __tfork51 param;
152 	int flags;
153 	int error;
154 
155 	if ((error = copyin(SCARG(uap, param), &param, sizeof(param))))
156 		return (error);
157 
158 	if (param.tf_flags != 0)
159 		return (EINVAL);
160 
161 	flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM
162 	    | FORK_NOZOMBIE | FORK_SHAREFILES;
163 
164 	return (fork1(p, 0, flags, NULL, param.tf_tid, tfork_child_return,
165 	    param.tf_tcb, retval, NULL));
166 }
167 #endif
168 
169 void
170 tfork_child_return(void *arg)
171 {
172 	struct proc *p = curproc;
173 
174 	TCB_SET(p, arg);
175 	child_return(p);
176 }
177 
178 /*
179  * Allocate and initialize a new process.
180  */
181 void
182 process_new(struct proc *p, struct process *parent)
183 {
184 	struct process *pr;
185 
186 	pr = pool_get(&process_pool, PR_WAITOK);
187 	pr->ps_mainproc = p;
188 
189 	TAILQ_INIT(&pr->ps_threads);
190 	TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
191 	pr->ps_pptr = parent;
192 	LIST_INIT(&pr->ps_children);
193 	pr->ps_refcnt = 1;
194 
195 	/*
196 	 * Make a process structure for the new process.
197 	 * Start by zeroing the section of proc that is zero-initialized,
198 	 * then copy the section that is copied directly from the parent.
199 	 */
200 	bzero(&pr->ps_startzero,
201 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
202 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
203 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
204 
205 	/* post-copy fixups */
206 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
207 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
208 	crhold(parent->ps_cred->pc_ucred);
209 	pr->ps_limit->p_refcnt++;
210 
211 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
212 	timeout_set(&pr->ps_virt_to, virttimer_trampoline, pr);
213 	timeout_set(&pr->ps_prof_to, proftimer_trampoline, pr);
214 
215 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
216 	if (parent->ps_session->s_ttyvp != NULL &&
217 	    parent->ps_flags & PS_CONTROLT)
218 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
219 
220 	p->p_p = pr;
221 }
222 
223 /* print the 'table full' message once per 10 seconds */
224 struct timeval fork_tfmrate = { 10, 0 };
225 
226 int
227 fork1(struct proc *curp, int exitsig, int flags, void *stack, pid_t *tidptr,
228     void (*func)(void *), void *arg, register_t *retval,
229     struct proc **rnewprocp)
230 {
231 	struct process *curpr = curp->p_p;
232 	struct process *pr;
233 	struct proc *p;
234 	uid_t uid;
235 	struct vmspace *vm;
236 	int count;
237 	vaddr_t uaddr;
238 	int s;
239 	struct  ptrace_state *newptstat = NULL;
240 #if NSYSTRACE > 0
241 	void *newstrp = NULL;
242 #endif
243 
244 	/* sanity check some flag combinations */
245 	if (flags & FORK_THREAD) {
246 		if (!rthreads_enabled)
247 			return (ENOTSUP);
248 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
249 		    (FORK_SIGHAND | FORK_NOZOMBIE))
250 			return (EINVAL);
251 	}
252 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
253 		return (EINVAL);
254 
255 	/*
256 	 * Although process entries are dynamically created, we still keep
257 	 * a global limit on the maximum number we will create. We reserve
258 	 * the last 5 processes to root. The variable nprocesses is the
259 	 * current number of processes, maxprocess is the limit.  Similar
260 	 * rules for threads (struct proc): we reserve the last 5 to root;
261 	 * the variable nthreads is the current number of procs, maxthread is
262 	 * the limit.
263 	 */
264 	uid = curp->p_cred->p_ruid;
265 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
266 		static struct timeval lasttfm;
267 
268 		if (ratecheck(&lasttfm, &fork_tfmrate))
269 			tablefull("proc");
270 		return (EAGAIN);
271 	}
272 	nthreads++;
273 
274 	if ((flags & FORK_THREAD) == 0) {
275 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
276 		    nprocesses >= maxprocess) {
277 			static struct timeval lasttfm;
278 
279 			if (ratecheck(&lasttfm, &fork_tfmrate))
280 				tablefull("process");
281 			nthreads--;
282 			return (EAGAIN);
283 		}
284 		nprocesses++;
285 
286 		/*
287 		 * Increment the count of processes running with
288 		 * this uid.  Don't allow a nonprivileged user to
289 		 * exceed their current limit.
290 		 */
291 		count = chgproccnt(uid, 1);
292 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
293 			(void)chgproccnt(uid, -1);
294 			nprocesses--;
295 			nthreads--;
296 			return (EAGAIN);
297 		}
298 	}
299 
300 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
301 	    USPACE_ALIGN, UVM_KMF_ZERO,
302 	    no_constraint.ucr_low, no_constraint.ucr_high,
303 	    0, 0, USPACE/PAGE_SIZE);
304 	if (uaddr == 0) {
305 		chgproccnt(uid, -1);
306 		nprocesses--;
307 		nthreads--;
308 		return (ENOMEM);
309 	}
310 
311 	/*
312 	 * From now on, we're committed to the fork and cannot fail.
313 	 */
314 
315 	/* Allocate new proc. */
316 	p = pool_get(&proc_pool, PR_WAITOK);
317 
318 	p->p_stat = SIDL;			/* protect against others */
319 	p->p_exitsig = exitsig;
320 	p->p_flag = 0;
321 	p->p_xstat = 0;
322 
323 	if (flags & FORK_THREAD) {
324 		atomic_setbits_int(&p->p_flag, P_THREAD);
325 		p->p_p = pr = curpr;
326 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
327 		pr->ps_refcnt++;
328 	} else {
329 		process_new(p, curpr);
330 		pr = p->p_p;
331 	}
332 
333 	/*
334 	 * Make a proc table entry for the new process.
335 	 * Start by zeroing the section of proc that is zero-initialized,
336 	 * then copy the section that is copied directly from the parent.
337 	 */
338 	bzero(&p->p_startzero,
339 	    (unsigned) ((caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero));
340 	bcopy(&curp->p_startcopy, &p->p_startcopy,
341 	    (unsigned) ((caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy));
342 
343 	/*
344 	 * Initialize the timeouts.
345 	 */
346 	timeout_set(&p->p_sleep_to, endtsleep, p);
347 
348 	/*
349 	 * Duplicate sub-structures as needed.
350 	 * Increase reference counts on shared objects.
351 	 */
352 	if ((flags & FORK_THREAD) == 0) {
353 		if (curpr->ps_flags & PS_PROFIL)
354 			startprofclock(pr);
355 		if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
356 			atomic_setbits_int(&pr->ps_flags, PS_TRACED);
357 	}
358 
359 	/* bump references to the text vnode (for procfs) */
360 	p->p_textvp = curp->p_textvp;
361 	if (p->p_textvp)
362 		vref(p->p_textvp);
363 
364 	if (flags & FORK_SHAREFILES)
365 		p->p_fd = fdshare(curp);
366 	else
367 		p->p_fd = fdcopy(curp);
368 
369 	if (flags & FORK_PPWAIT) {
370 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
371 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
372 	}
373 	if (flags & FORK_NOZOMBIE)
374 		atomic_setbits_int(&p->p_flag, P_NOZOMBIE);
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 	/*
394 	 * Create signal actions for the child process.
395 	 */
396 	if (flags & FORK_SIGHAND)
397 		p->p_sigacts = sigactsshare(curp);
398 	else
399 		p->p_sigacts = sigactsinit(curp);
400 	if (flags & FORK_THREAD)
401 		sigstkinit(&p->p_sigstk);
402 
403 	/*
404 	 * If emulation has process fork hook, call it now.
405 	 */
406 	if (p->p_emul->e_proc_fork)
407 		(*p->p_emul->e_proc_fork)(p, curp);
408 
409 	p->p_addr = (struct user *)uaddr;
410 
411 	/*
412 	 * Finish creating the child process.  It will return through a
413 	 * different path later.
414 	 */
415 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
416 	    0, func ? func : child_return, arg ? arg : p);
417 
418 	vm = p->p_vmspace;
419 
420 	if (flags & FORK_FORK) {
421 		forkstat.cntfork++;
422 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
423 	} else if (flags & FORK_VFORK) {
424 		forkstat.cntvfork++;
425 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
426 	} else if (flags & FORK_TFORK) {
427 		forkstat.cnttfork++;
428 	} else {
429 		forkstat.cntkthread++;
430 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
431 	}
432 
433 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
434 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
435 #if NSYSTRACE > 0
436 	if (ISSET(curp->p_flag, P_SYSTRACE))
437 		newstrp = systrace_getproc();
438 #endif
439 
440 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
441 	do {
442 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
443 	} while (pidtaken(lastpid));
444 	p->p_pid = lastpid;
445 
446 	LIST_INSERT_HEAD(&allproc, p, p_list);
447 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
448 	if ((flags & FORK_THREAD) == 0) {
449 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
450 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
451 
452 		if (pr->ps_flags & PS_TRACED) {
453 			pr->ps_oppid = curpr->ps_pid;
454 			if (pr->ps_pptr != curpr->ps_pptr)
455 				proc_reparent(pr, curpr->ps_pptr);
456 
457 			/*
458 			 * Set ptrace status.
459 			 */
460 			if (flags & FORK_FORK) {
461 				pr->ps_ptstat = newptstat;
462 				newptstat = NULL;
463 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
464 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
465 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
466 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
467 			}
468 		}
469 	}
470 
471 #if NSYSTRACE > 0
472 	if (newstrp)
473 		systrace_fork(curp, p, newstrp);
474 #endif
475 
476 	if (tidptr != NULL) {
477 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
478 
479 		if (copyout(&pid, tidptr, sizeof(pid)))
480 			psignal(curp, SIGSEGV);
481 	}
482 
483 	/*
484 	 * For new processes, set accounting bits
485 	 */
486 	if ((flags & FORK_THREAD) == 0) {
487 		getmicrotime(&pr->ps_start);
488 		pr->ps_acflag = AFORK;
489 	}
490 
491 	/*
492 	 * Make child runnable and add to run queue.
493 	 */
494 	SCHED_LOCK(s);
495 	p->p_stat = SRUN;
496 	p->p_cpu = sched_choosecpu_fork(curp, flags);
497 	setrunqueue(p);
498 	SCHED_UNLOCK(s);
499 
500 	if (newptstat)
501 		free(newptstat, M_SUBPROC);
502 
503 	/*
504 	 * Notify any interested parties about the new process.
505 	 */
506 	if ((flags & FORK_THREAD) == 0)
507 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
508 
509 	/*
510 	 * Update stats now that we know the fork was successful.
511 	 */
512 	uvmexp.forks++;
513 	if (flags & FORK_PPWAIT)
514 		uvmexp.forks_ppwait++;
515 	if (flags & FORK_SHAREVM)
516 		uvmexp.forks_sharevm++;
517 
518 	/*
519 	 * Pass a pointer to the new process to the caller.
520 	 */
521 	if (rnewprocp != NULL)
522 		*rnewprocp = p;
523 
524 	/*
525 	 * Preserve synchronization semantics of vfork.  If waiting for
526 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
527 	 * on ourselves, and sleep on our process for the latter flag
528 	 * to go away.
529 	 * XXX Need to stop other rthreads in the parent
530 	 */
531 	if (flags & FORK_PPWAIT)
532 		while (curpr->ps_flags & PS_ISPWAIT)
533 			tsleep(curpr, PWAIT, "ppwait", 0);
534 
535 	/*
536 	 * If we're tracing the child, alert the parent too.
537 	 */
538 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
539 		psignal(curp, SIGTRAP);
540 
541 	/*
542 	 * Return child pid to parent process,
543 	 * marking us as parent via retval[1].
544 	 */
545 	if (retval != NULL) {
546 		retval[0] = p->p_pid +
547 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
548 		retval[1] = 0;
549 	}
550 	return (0);
551 }
552 
553 /*
554  * Checks for current use of a pid, either as a pid or pgid.
555  */
556 int
557 pidtaken(pid_t pid)
558 {
559 	struct proc *p;
560 
561 	if (pfind(pid) != NULL)
562 		return (1);
563 	if (pgfind(pid) != NULL)
564 		return (1);
565 	LIST_FOREACH(p, &zombproc, p_list) {
566 		if (p->p_pid == pid || (p->p_p->ps_pgrp && p->p_p->ps_pgrp->pg_id == pid))
567 			return (1);
568 	}
569 	return (0);
570 }
571 
572 #if defined(MULTIPROCESSOR)
573 /*
574  * XXX This is a slight hack to get newly-formed processes to
575  * XXX acquire the kernel lock as soon as they run.
576  */
577 void
578 proc_trampoline_mp(void)
579 {
580 	struct proc *p;
581 
582 	p = curproc;
583 
584 	SCHED_ASSERT_LOCKED();
585 	__mp_unlock(&sched_lock);
586 	spl0();
587 	SCHED_ASSERT_UNLOCKED();
588 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
589 
590 	KERNEL_LOCK();
591 }
592 #endif
593