xref: /openbsd-src/sys/kern/kern_fork.c (revision 590b2eb9e19bae0b4044768999e242671dc0397f) 
1 /*	$OpenBSD: kern_fork.c,v 1.144 2013/03/02 07:05:17 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 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)
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 	bzero(&pr->ps_startzero,
176 	    (unsigned) ((caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero));
177 	bcopy(&parent->ps_startcopy, &pr->ps_startcopy,
178 	    (unsigned) ((caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy));
179 
180 	/* post-copy fixups */
181 	pr->ps_cred = pool_get(&pcred_pool, PR_WAITOK);
182 	bcopy(parent->ps_cred, pr->ps_cred, sizeof(*pr->ps_cred));
183 	crhold(parent->ps_cred->pc_ucred);
184 	pr->ps_limit->p_refcnt++;
185 
186 	timeout_set(&pr->ps_realit_to, realitexpire, pr);
187 	timeout_set(&pr->ps_virt_to, virttimer_trampoline, pr);
188 	timeout_set(&pr->ps_prof_to, proftimer_trampoline, pr);
189 
190 	pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC);
191 	if (parent->ps_session->s_ttyvp != NULL &&
192 	    parent->ps_flags & PS_CONTROLT)
193 		atomic_setbits_int(&pr->ps_flags, PS_CONTROLT);
194 
195 	p->p_p = pr;
196 }
197 
198 /* print the 'table full' message once per 10 seconds */
199 struct timeval fork_tfmrate = { 10, 0 };
200 
201 int
202 fork1(struct proc *curp, int exitsig, int flags, void *stack, pid_t *tidptr,
203     void (*func)(void *), void *arg, register_t *retval,
204     struct proc **rnewprocp)
205 {
206 	struct process *curpr = curp->p_p;
207 	struct process *pr;
208 	struct proc *p;
209 	uid_t uid;
210 	struct vmspace *vm;
211 	int count;
212 	vaddr_t uaddr;
213 	int s;
214 	struct  ptrace_state *newptstat = NULL;
215 #if NSYSTRACE > 0
216 	void *newstrp = NULL;
217 #endif
218 
219 	/* sanity check some flag combinations */
220 	if (flags & FORK_THREAD) {
221 		if (!rthreads_enabled)
222 			return (ENOTSUP);
223 		if ((flags & (FORK_SIGHAND | FORK_NOZOMBIE)) !=
224 		    (FORK_SIGHAND | FORK_NOZOMBIE))
225 			return (EINVAL);
226 	}
227 	if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0)
228 		return (EINVAL);
229 
230 	/*
231 	 * Although process entries are dynamically created, we still keep
232 	 * a global limit on the maximum number we will create. We reserve
233 	 * the last 5 processes to root. The variable nprocesses is the
234 	 * current number of processes, maxprocess is the limit.  Similar
235 	 * rules for threads (struct proc): we reserve the last 5 to root;
236 	 * the variable nthreads is the current number of procs, maxthread is
237 	 * the limit.
238 	 */
239 	uid = curp->p_cred->p_ruid;
240 	if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) {
241 		static struct timeval lasttfm;
242 
243 		if (ratecheck(&lasttfm, &fork_tfmrate))
244 			tablefull("proc");
245 		return (EAGAIN);
246 	}
247 	nthreads++;
248 
249 	if ((flags & FORK_THREAD) == 0) {
250 		if ((nprocesses >= maxprocess - 5 && uid != 0) ||
251 		    nprocesses >= maxprocess) {
252 			static struct timeval lasttfm;
253 
254 			if (ratecheck(&lasttfm, &fork_tfmrate))
255 				tablefull("process");
256 			nthreads--;
257 			return (EAGAIN);
258 		}
259 		nprocesses++;
260 
261 		/*
262 		 * Increment the count of processes running with
263 		 * this uid.  Don't allow a nonprivileged user to
264 		 * exceed their current limit.
265 		 */
266 		count = chgproccnt(uid, 1);
267 		if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
268 			(void)chgproccnt(uid, -1);
269 			nprocesses--;
270 			nthreads--;
271 			return (EAGAIN);
272 		}
273 	}
274 
275 	uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE,
276 	    USPACE_ALIGN, UVM_KMF_ZERO,
277 	    no_constraint.ucr_low, no_constraint.ucr_high,
278 	    0, 0, USPACE/PAGE_SIZE);
279 	if (uaddr == 0) {
280 		if ((flags & FORK_THREAD) == 0) {
281 			(void)chgproccnt(uid, -1);
282 			nprocesses--;
283 		}
284 		nthreads--;
285 		return (ENOMEM);
286 	}
287 
288 	/*
289 	 * From now on, we're committed to the fork and cannot fail.
290 	 */
291 
292 	/* Allocate new proc. */
293 	p = pool_get(&proc_pool, PR_WAITOK);
294 
295 	p->p_stat = SIDL;			/* protect against others */
296 	p->p_exitsig = exitsig;
297 	p->p_flag = 0;
298 	p->p_xstat = 0;
299 
300 	if (flags & FORK_THREAD) {
301 		atomic_setbits_int(&p->p_flag, P_THREAD);
302 		p->p_p = pr = curpr;
303 		TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link);
304 		pr->ps_refcnt++;
305 	} else {
306 		process_new(p, curpr);
307 		pr = p->p_p;
308 	}
309 
310 	/*
311 	 * Make a proc table entry for the new process.
312 	 * Start by zeroing the section of proc that is zero-initialized,
313 	 * then copy the section that is copied directly from the parent.
314 	 */
315 	bzero(&p->p_startzero,
316 	    (unsigned) ((caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero));
317 	bcopy(&curp->p_startcopy, &p->p_startcopy,
318 	    (unsigned) ((caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy));
319 
320 	/*
321 	 * Initialize the timeouts.
322 	 */
323 	timeout_set(&p->p_sleep_to, endtsleep, p);
324 
325 	/*
326 	 * Duplicate sub-structures as needed.
327 	 * Increase reference counts on shared objects.
328 	 */
329 	if ((flags & FORK_THREAD) == 0) {
330 		if (curpr->ps_flags & PS_PROFIL)
331 			startprofclock(pr);
332 		if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
333 			atomic_setbits_int(&pr->ps_flags, PS_TRACED);
334 	}
335 
336 	/* bump references to the text vnode (for procfs) */
337 	p->p_textvp = curp->p_textvp;
338 	if (p->p_textvp)
339 		vref(p->p_textvp);
340 
341 	if (flags & FORK_SHAREFILES)
342 		p->p_fd = fdshare(curp);
343 	else
344 		p->p_fd = fdcopy(curp);
345 
346 	if (flags & FORK_PPWAIT) {
347 		atomic_setbits_int(&pr->ps_flags, PS_PPWAIT);
348 		atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT);
349 	}
350 	if (flags & FORK_NOZOMBIE)
351 		atomic_setbits_int(&p->p_flag, P_NOZOMBIE);
352 
353 #ifdef KTRACE
354 	/*
355 	 * Copy traceflag and tracefile if enabled.
356 	 * If not inherited, these were zeroed above.
357 	 */
358 	if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT)
359 		ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp,
360 		    curpr->ps_tracecred);
361 #endif
362 
363 	/*
364 	 * set priority of child to be that of parent
365 	 * XXX should move p_estcpu into the region of struct proc which gets
366 	 * copied.
367 	 */
368 	scheduler_fork_hook(curp, p);
369 
370 	/*
371 	 * Create signal actions for the child process.
372 	 */
373 	if (flags & FORK_SIGHAND)
374 		p->p_sigacts = sigactsshare(curp);
375 	else
376 		p->p_sigacts = sigactsinit(curp);
377 	if (flags & FORK_THREAD)
378 		sigstkinit(&p->p_sigstk);
379 
380 	/*
381 	 * If emulation has process fork hook, call it now.
382 	 */
383 	if (p->p_emul->e_proc_fork)
384 		(*p->p_emul->e_proc_fork)(p, curp);
385 
386 	p->p_addr = (struct user *)uaddr;
387 
388 	/*
389 	 * Finish creating the child process.  It will return through a
390 	 * different path later.
391 	 */
392 	uvm_fork(curp, p, ((flags & FORK_SHAREVM) ? TRUE : FALSE), stack,
393 	    0, func ? func : child_return, arg ? arg : p);
394 
395 	vm = p->p_vmspace;
396 
397 	if (flags & FORK_FORK) {
398 		forkstat.cntfork++;
399 		forkstat.sizfork += vm->vm_dsize + vm->vm_ssize;
400 	} else if (flags & FORK_VFORK) {
401 		forkstat.cntvfork++;
402 		forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize;
403 	} else if (flags & FORK_TFORK) {
404 		forkstat.cnttfork++;
405 	} else {
406 		forkstat.cntkthread++;
407 		forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize;
408 	}
409 
410 	if (pr->ps_flags & PS_TRACED && flags & FORK_FORK)
411 		newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK);
412 #if NSYSTRACE > 0
413 	if (ISSET(curp->p_flag, P_SYSTRACE))
414 		newstrp = systrace_getproc();
415 #endif
416 
417 	/* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */
418 	do {
419 		lastpid = 1 + (randompid ? arc4random() : lastpid) % PID_MAX;
420 	} while (pidtaken(lastpid));
421 	p->p_pid = lastpid;
422 
423 	LIST_INSERT_HEAD(&allproc, p, p_list);
424 	LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash);
425 	if ((flags & FORK_THREAD) == 0) {
426 		LIST_INSERT_AFTER(curpr, pr, ps_pglist);
427 		LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling);
428 
429 		if (pr->ps_flags & PS_TRACED) {
430 			pr->ps_oppid = curpr->ps_pid;
431 			if (pr->ps_pptr != curpr->ps_pptr)
432 				proc_reparent(pr, curpr->ps_pptr);
433 
434 			/*
435 			 * Set ptrace status.
436 			 */
437 			if (flags & FORK_FORK) {
438 				pr->ps_ptstat = newptstat;
439 				newptstat = NULL;
440 				curpr->ps_ptstat->pe_report_event = PTRACE_FORK;
441 				pr->ps_ptstat->pe_report_event = PTRACE_FORK;
442 				curpr->ps_ptstat->pe_other_pid = pr->ps_pid;
443 				pr->ps_ptstat->pe_other_pid = curpr->ps_pid;
444 			}
445 		}
446 	}
447 
448 #if NSYSTRACE > 0
449 	if (newstrp)
450 		systrace_fork(curp, p, newstrp);
451 #endif
452 
453 	if (tidptr != NULL) {
454 		pid_t	pid = p->p_pid + THREAD_PID_OFFSET;
455 
456 		if (copyout(&pid, tidptr, sizeof(pid)))
457 			psignal(curp, SIGSEGV);
458 	}
459 
460 	/*
461 	 * For new processes, set accounting bits
462 	 */
463 	if ((flags & FORK_THREAD) == 0) {
464 		getmicrotime(&pr->ps_start);
465 		pr->ps_acflag = AFORK;
466 	}
467 
468 	/*
469 	 * Make child runnable and add to run queue.
470 	 */
471 	SCHED_LOCK(s);
472 	p->p_stat = SRUN;
473 	p->p_cpu = sched_choosecpu_fork(curp, flags);
474 	setrunqueue(p);
475 	SCHED_UNLOCK(s);
476 
477 	if (newptstat)
478 		free(newptstat, M_SUBPROC);
479 
480 	/*
481 	 * Notify any interested parties about the new process.
482 	 */
483 	if ((flags & FORK_THREAD) == 0)
484 		KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid);
485 
486 	/*
487 	 * Update stats now that we know the fork was successful.
488 	 */
489 	uvmexp.forks++;
490 	if (flags & FORK_PPWAIT)
491 		uvmexp.forks_ppwait++;
492 	if (flags & FORK_SHAREVM)
493 		uvmexp.forks_sharevm++;
494 
495 	/*
496 	 * Pass a pointer to the new process to the caller.
497 	 */
498 	if (rnewprocp != NULL)
499 		*rnewprocp = p;
500 
501 	/*
502 	 * Preserve synchronization semantics of vfork.  If waiting for
503 	 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT
504 	 * on ourselves, and sleep on our process for the latter flag
505 	 * to go away.
506 	 * XXX Need to stop other rthreads in the parent
507 	 */
508 	if (flags & FORK_PPWAIT)
509 		while (curpr->ps_flags & PS_ISPWAIT)
510 			tsleep(curpr, PWAIT, "ppwait", 0);
511 
512 	/*
513 	 * If we're tracing the child, alert the parent too.
514 	 */
515 	if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED))
516 		psignal(curp, SIGTRAP);
517 
518 	/*
519 	 * Return child pid to parent process,
520 	 * marking us as parent via retval[1].
521 	 */
522 	if (retval != NULL) {
523 		retval[0] = p->p_pid +
524 		    (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0);
525 		retval[1] = 0;
526 	}
527 	return (0);
528 }
529 
530 /*
531  * Checks for current use of a pid, either as a pid or pgid.
532  */
533 int
534 pidtaken(pid_t pid)
535 {
536 	struct proc *p;
537 
538 	if (pfind(pid) != NULL)
539 		return (1);
540 	if (pgfind(pid) != NULL)
541 		return (1);
542 	LIST_FOREACH(p, &zombproc, p_list) {
543 		if (p->p_pid == pid || (p->p_p->ps_pgrp && p->p_p->ps_pgrp->pg_id == pid))
544 			return (1);
545 	}
546 	return (0);
547 }
548 
549 #if defined(MULTIPROCESSOR)
550 /*
551  * XXX This is a slight hack to get newly-formed processes to
552  * XXX acquire the kernel lock as soon as they run.
553  */
554 void
555 proc_trampoline_mp(void)
556 {
557 	struct proc *p;
558 
559 	p = curproc;
560 
561 	SCHED_ASSERT_LOCKED();
562 	__mp_unlock(&sched_lock);
563 	spl0();
564 	SCHED_ASSERT_UNLOCKED();
565 	KASSERT(__mp_lock_held(&kernel_lock) == 0);
566 
567 	KERNEL_LOCK();
568 }
569 #endif
570