xref: /freebsd/sys/kern/kern_resource.c (revision eacee0ff7ec955b32e09515246bd97b6edcd2b0f)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_resource.c	8.5 (Berkeley) 1/21/94
39  * $FreeBSD$
40  */
41 
42 #include "opt_compat.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/file.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/sx.h>
55 #include <sys/time.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_param.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_map.h>
61 
62 static int donice __P((struct proc *curp, struct proc *chgp, int n));
63 
64 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
65 #define	UIHASH(uid)	(&uihashtbl[(uid) & uihash])
66 static struct mtx uihashtbl_mtx;
67 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
68 static u_long uihash;		/* size of hash table - 1 */
69 
70 static struct uidinfo	*uilookup __P((uid_t uid));
71 
72 /*
73  * Resource controls and accounting.
74  */
75 
76 #ifndef _SYS_SYSPROTO_H_
77 struct getpriority_args {
78 	int	which;
79 	int	who;
80 };
81 #endif
82 /*
83  * MPSAFE
84  */
85 int
86 getpriority(td, uap)
87 	struct thread *td;
88 	register struct getpriority_args *uap;
89 {
90 	struct proc *curp = td->td_proc;
91 	register struct proc *p;
92 	register int low = PRIO_MAX + 1;
93 	int error = 0;
94 
95 	mtx_lock(&Giant);
96 
97 	switch (uap->which) {
98 	case PRIO_PROCESS:
99 		if (uap->who == 0)
100 			low = td->td_ksegrp->kg_nice;
101 		else {
102 			p = pfind(uap->who);
103 			if (p == NULL)
104 				break;
105 			if (p_cansee(curp, p) == 0)
106 				low = p->p_ksegrp.kg_nice /* XXXKSE */ ;
107 			PROC_UNLOCK(p);
108 		}
109 		break;
110 
111 	case PRIO_PGRP: {
112 		register struct pgrp *pg;
113 
114 		if (uap->who == 0)
115 			pg = curp->p_pgrp;
116 		else if ((pg = pgfind(uap->who)) == NULL)
117 			break;
118 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
119 			if (!p_cansee(curp, p) && p->p_ksegrp.kg_nice /* XXXKSE */  < low)
120 				low = p->p_ksegrp.kg_nice /* XXXKSE */ ;
121 		}
122 		break;
123 	}
124 
125 	case PRIO_USER:
126 		if (uap->who == 0)
127 			uap->who = curp->p_ucred->cr_uid;
128 		sx_slock(&allproc_lock);
129 		LIST_FOREACH(p, &allproc, p_list)
130 			if (!p_cansee(curp, p) &&
131 			    p->p_ucred->cr_uid == uap->who &&
132 			    p->p_ksegrp.kg_nice /* XXXKSE */  < low)
133 				low = p->p_ksegrp.kg_nice /* XXXKSE */ ;
134 		sx_sunlock(&allproc_lock);
135 		break;
136 
137 	default:
138 		error = EINVAL;
139 		break;
140 	}
141 	if (low == PRIO_MAX + 1 && error == 0)
142 		error = ESRCH;
143 	td->td_retval[0] = low;
144 	mtx_unlock(&Giant);
145 	return (error);
146 }
147 
148 #ifndef _SYS_SYSPROTO_H_
149 struct setpriority_args {
150 	int	which;
151 	int	who;
152 	int	prio;
153 };
154 #endif
155 /*
156  * MPSAFE
157  */
158 /* ARGSUSED */
159 int
160 setpriority(td, uap)
161 	struct thread *td;
162 	register struct setpriority_args *uap;
163 {
164 	struct proc *curp = td->td_proc;
165 	register struct proc *p;
166 	int found = 0, error = 0;
167 
168 	mtx_lock(&Giant);
169 
170 	switch (uap->which) {
171 	case PRIO_PROCESS:
172 		if (uap->who == 0)
173 			error = donice(curp, curp, uap->prio);
174 		else {
175 			p = pfind(uap->who);
176 			if (p == 0)
177 				break;
178 			if (p_cansee(curp, p) == 0)
179 				error = donice(curp, p, uap->prio);
180 			PROC_UNLOCK(p);
181 		}
182 		found++;
183 		break;
184 
185 	case PRIO_PGRP: {
186 		register struct pgrp *pg;
187 
188 		if (uap->who == 0)
189 			pg = curp->p_pgrp;
190 		else if ((pg = pgfind(uap->who)) == NULL)
191 			break;
192 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
193 			if (!p_cansee(curp, p)) {
194 				error = donice(curp, p, uap->prio);
195 				found++;
196 			}
197 		}
198 		break;
199 	}
200 
201 	case PRIO_USER:
202 		if (uap->who == 0)
203 			uap->who = curp->p_ucred->cr_uid;
204 		sx_slock(&allproc_lock);
205 		FOREACH_PROC_IN_SYSTEM(p) {
206 			if (p->p_ucred->cr_uid == uap->who &&
207 			    !p_cansee(curp, p)) {
208 				error = donice(curp, p, uap->prio);
209 				found++;
210 			}
211 		}
212 		sx_sunlock(&allproc_lock);
213 		break;
214 
215 	default:
216 		error = EINVAL;
217 		break;
218 	}
219 	if (found == 0 && error == 0)
220 		error = ESRCH;
221 	mtx_unlock(&Giant);
222 	return (error);
223 }
224 
225 static int
226 donice(curp, chgp, n)
227 	register struct proc *curp, *chgp;
228 	register int n;
229 {
230 	int	error;
231 
232 	if ((error = p_cansched(curp, chgp)))
233 		return (error);
234 	if (n > PRIO_MAX)
235 		n = PRIO_MAX;
236 	if (n < PRIO_MIN)
237 		n = PRIO_MIN;
238 	if (n < chgp->p_ksegrp.kg_nice /* XXXKSE */  &&
239 	    suser_xxx(curp->p_ucred, NULL, 0))
240 		return (EACCES);
241 	chgp->p_ksegrp.kg_nice /* XXXKSE */  = n;
242 	(void)resetpriority(&chgp->p_ksegrp); /* XXXKSE */
243 	return (0);
244 }
245 
246 /* rtprio system call */
247 #ifndef _SYS_SYSPROTO_H_
248 struct rtprio_args {
249 	int		function;
250 	pid_t		pid;
251 	struct rtprio	*rtp;
252 };
253 #endif
254 
255 /*
256  * Set realtime priority
257  */
258 
259 /*
260  * MPSAFE
261  */
262 /* ARGSUSED */
263 int
264 rtprio(td, uap)
265 	struct thread *td;
266 	register struct rtprio_args *uap;
267 {
268 	struct proc *curp = td->td_proc;
269 	register struct proc *p;
270 	struct rtprio rtp;
271 	int error;
272 
273 	mtx_lock(&Giant);
274 
275 	if (uap->pid == 0) {
276 		p = curp;
277 		PROC_LOCK(p);
278 	} else {
279 		p = pfind(uap->pid);
280 	}
281 
282 	if (p == NULL) {
283 		error = ESRCH;
284 		goto done2;
285 	}
286 
287 	switch (uap->function) {
288 	case RTP_LOOKUP:
289 		if ((error = p_cansee(curp, p)))
290 			break;
291 		mtx_lock_spin(&sched_lock);
292 		pri_to_rtp(&p->p_ksegrp /* XXXKSE */ , &rtp);
293 		mtx_unlock_spin(&sched_lock);
294 		error = copyout(&rtp, uap->rtp, sizeof(struct rtprio));
295 		break;
296 	case RTP_SET:
297 		if ((error = p_cansched(curp, p)) ||
298 		    (error = copyin(uap->rtp, &rtp, sizeof(struct rtprio))))
299 			break;
300 		/* disallow setting rtprio in most cases if not superuser */
301 		if (suser_xxx(curp->p_ucred, NULL, 0) != 0) {
302 			/* can't set someone else's */
303 			if (uap->pid) {
304 				error = EPERM;
305 				break;
306 			}
307 			/* can't set realtime priority */
308 /*
309  * Realtime priority has to be restricted for reasons which should be
310  * obvious. However, for idle priority, there is a potential for
311  * system deadlock if an idleprio process gains a lock on a resource
312  * that other processes need (and the idleprio process can't run
313  * due to a CPU-bound normal process). Fix me! XXX
314  */
315 #if 0
316  			if (RTP_PRIO_IS_REALTIME(rtp.type))
317 #endif
318 			if (rtp.type != RTP_PRIO_NORMAL) {
319 				error = EPERM;
320 				break;
321 			}
322 		}
323 		mtx_lock_spin(&sched_lock);
324 		error = rtp_to_pri(&rtp, &p->p_ksegrp);
325 		mtx_unlock_spin(&sched_lock);
326 		break;
327 	default:
328 		error = EINVAL;
329 		break;
330 	}
331 	PROC_UNLOCK(p);
332 done2:
333 	mtx_unlock(&Giant);
334 	return (error);
335 }
336 
337 int
338 rtp_to_pri(struct rtprio *rtp, struct ksegrp *kg)
339 {
340 
341 	if (rtp->prio > RTP_PRIO_MAX)
342 		return (EINVAL);
343 	switch (RTP_PRIO_BASE(rtp->type)) {
344 	case RTP_PRIO_REALTIME:
345 		kg->kg_user_pri = PRI_MIN_REALTIME + rtp->prio;
346 		break;
347 	case RTP_PRIO_NORMAL:
348 		kg->kg_user_pri = PRI_MIN_TIMESHARE + rtp->prio;
349 		break;
350 	case RTP_PRIO_IDLE:
351 		kg->kg_user_pri = PRI_MIN_IDLE + rtp->prio;
352 		break;
353 	default:
354 		return (EINVAL);
355 	}
356 	kg->kg_pri_class = rtp->type;
357 	if (curthread->td_ksegrp == kg) {
358 		curthread->td_base_pri = kg->kg_user_pri;
359 		curthread->td_priority = kg->kg_user_pri; /* XXX dubious */
360 	}
361 	return (0);
362 }
363 
364 void
365 pri_to_rtp(struct ksegrp *kg, struct rtprio *rtp)
366 {
367 
368 	switch (PRI_BASE(kg->kg_pri_class)) {
369 	case PRI_REALTIME:
370 		rtp->prio = kg->kg_user_pri - PRI_MIN_REALTIME;
371 		break;
372 	case PRI_TIMESHARE:
373 		rtp->prio = kg->kg_user_pri - PRI_MIN_TIMESHARE;
374 		break;
375 	case PRI_IDLE:
376 		rtp->prio = kg->kg_user_pri - PRI_MIN_IDLE;
377 		break;
378 	default:
379 		break;
380 	}
381 	rtp->type = kg->kg_pri_class;
382 }
383 
384 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
385 #ifndef _SYS_SYSPROTO_H_
386 struct osetrlimit_args {
387 	u_int	which;
388 	struct	orlimit *rlp;
389 };
390 #endif
391 /*
392  * MPSAFE
393  */
394 /* ARGSUSED */
395 int
396 osetrlimit(td, uap)
397 	struct thread *td;
398 	register struct osetrlimit_args *uap;
399 {
400 	struct orlimit olim;
401 	struct rlimit lim;
402 	int error;
403 
404 	if ((error =
405 	    copyin((caddr_t)uap->rlp, (caddr_t)&olim, sizeof(struct orlimit))))
406 		return (error);
407 	lim.rlim_cur = olim.rlim_cur;
408 	lim.rlim_max = olim.rlim_max;
409 	mtx_lock(&Giant);
410 	error = dosetrlimit(td, uap->which, &lim);
411 	mtx_unlock(&Giant);
412 	return (error);
413 }
414 
415 #ifndef _SYS_SYSPROTO_H_
416 struct ogetrlimit_args {
417 	u_int	which;
418 	struct	orlimit *rlp;
419 };
420 #endif
421 /*
422  * MPSAFE
423  */
424 /* ARGSUSED */
425 int
426 ogetrlimit(td, uap)
427 	struct thread *td;
428 	register struct ogetrlimit_args *uap;
429 {
430 	struct proc *p = td->td_proc;
431 	struct orlimit olim;
432 	int error;
433 
434 	if (uap->which >= RLIM_NLIMITS)
435 		return (EINVAL);
436 	mtx_lock(&Giant);
437 	olim.rlim_cur = p->p_rlimit[uap->which].rlim_cur;
438 	if (olim.rlim_cur == -1)
439 		olim.rlim_cur = 0x7fffffff;
440 	olim.rlim_max = p->p_rlimit[uap->which].rlim_max;
441 	if (olim.rlim_max == -1)
442 		olim.rlim_max = 0x7fffffff;
443 	error = copyout((caddr_t)&olim, (caddr_t)uap->rlp, sizeof(olim));
444 	mtx_unlock(&Giant);
445 	return (error);
446 }
447 #endif /* COMPAT_43 || COMPAT_SUNOS */
448 
449 #ifndef _SYS_SYSPROTO_H_
450 struct __setrlimit_args {
451 	u_int	which;
452 	struct	rlimit *rlp;
453 };
454 #endif
455 /*
456  * MPSAFE
457  */
458 /* ARGSUSED */
459 int
460 setrlimit(td, uap)
461 	struct thread *td;
462 	register struct __setrlimit_args *uap;
463 {
464 	struct rlimit alim;
465 	int error;
466 
467 	if ((error =
468 	    copyin((caddr_t)uap->rlp, (caddr_t)&alim, sizeof (struct rlimit))))
469 		return (error);
470 	mtx_lock(&Giant);
471 	error = dosetrlimit(td, uap->which, &alim);
472 	mtx_unlock(&Giant);
473 	return (error);
474 }
475 
476 int
477 dosetrlimit(td, which, limp)
478 	struct thread *td;
479 	u_int which;
480 	struct rlimit *limp;
481 {
482 	struct proc *p = td->td_proc;
483 	register struct rlimit *alimp;
484 	int error;
485 
486 	GIANT_REQUIRED;
487 
488 	if (which >= RLIM_NLIMITS)
489 		return (EINVAL);
490 	alimp = &p->p_rlimit[which];
491 
492 	/*
493 	 * Preserve historical bugs by treating negative limits as unsigned.
494 	 */
495 	if (limp->rlim_cur < 0)
496 		limp->rlim_cur = RLIM_INFINITY;
497 	if (limp->rlim_max < 0)
498 		limp->rlim_max = RLIM_INFINITY;
499 
500 	if (limp->rlim_cur > alimp->rlim_max ||
501 	    limp->rlim_max > alimp->rlim_max)
502 		if ((error = suser_xxx(0, p, PRISON_ROOT)))
503 			return (error);
504 	if (limp->rlim_cur > limp->rlim_max)
505 		limp->rlim_cur = limp->rlim_max;
506 	if (p->p_limit->p_refcnt > 1 &&
507 	    (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
508 		p->p_limit->p_refcnt--;
509 		p->p_limit = limcopy(p->p_limit);
510 		alimp = &p->p_rlimit[which];
511 	}
512 
513 	switch (which) {
514 
515 	case RLIMIT_CPU:
516 		if (limp->rlim_cur > RLIM_INFINITY / (rlim_t)1000000)
517 			p->p_limit->p_cpulimit = RLIM_INFINITY;
518 		else
519 			p->p_limit->p_cpulimit =
520 			    (rlim_t)1000000 * limp->rlim_cur;
521 		break;
522 	case RLIMIT_DATA:
523 		if (limp->rlim_cur > maxdsiz)
524 			limp->rlim_cur = maxdsiz;
525 		if (limp->rlim_max > maxdsiz)
526 			limp->rlim_max = maxdsiz;
527 		break;
528 
529 	case RLIMIT_STACK:
530 		if (limp->rlim_cur > maxssiz)
531 			limp->rlim_cur = maxssiz;
532 		if (limp->rlim_max > maxssiz)
533 			limp->rlim_max = maxssiz;
534 		/*
535 		 * Stack is allocated to the max at exec time with only
536 		 * "rlim_cur" bytes accessible.  If stack limit is going
537 		 * up make more accessible, if going down make inaccessible.
538 		 */
539 		if (limp->rlim_cur != alimp->rlim_cur) {
540 			vm_offset_t addr;
541 			vm_size_t size;
542 			vm_prot_t prot;
543 
544 			if (limp->rlim_cur > alimp->rlim_cur) {
545 				prot = VM_PROT_ALL;
546 				size = limp->rlim_cur - alimp->rlim_cur;
547 				addr = USRSTACK - limp->rlim_cur;
548 			} else {
549 				prot = VM_PROT_NONE;
550 				size = alimp->rlim_cur - limp->rlim_cur;
551 				addr = USRSTACK - alimp->rlim_cur;
552 			}
553 			addr = trunc_page(addr);
554 			size = round_page(size);
555 			(void) vm_map_protect(&p->p_vmspace->vm_map,
556 					      addr, addr+size, prot, FALSE);
557 		}
558 		break;
559 
560 	case RLIMIT_NOFILE:
561 		if (limp->rlim_cur > maxfilesperproc)
562 			limp->rlim_cur = maxfilesperproc;
563 		if (limp->rlim_max > maxfilesperproc)
564 			limp->rlim_max = maxfilesperproc;
565 		break;
566 
567 	case RLIMIT_NPROC:
568 		if (limp->rlim_cur > maxprocperuid)
569 			limp->rlim_cur = maxprocperuid;
570 		if (limp->rlim_max > maxprocperuid)
571 			limp->rlim_max = maxprocperuid;
572 		if (limp->rlim_cur < 1)
573 			limp->rlim_cur = 1;
574 		if (limp->rlim_max < 1)
575 			limp->rlim_max = 1;
576 		break;
577 	}
578 	*alimp = *limp;
579 	return (0);
580 }
581 
582 #ifndef _SYS_SYSPROTO_H_
583 struct __getrlimit_args {
584 	u_int	which;
585 	struct	rlimit *rlp;
586 };
587 #endif
588 /*
589  * MPSAFE
590  */
591 /* ARGSUSED */
592 int
593 getrlimit(td, uap)
594 	struct thread *td;
595 	register struct __getrlimit_args *uap;
596 {
597 	int error;
598 	struct proc *p = td->td_proc;
599 
600 	if (uap->which >= RLIM_NLIMITS)
601 		return (EINVAL);
602 	mtx_lock(&Giant);
603 	error = copyout((caddr_t)&p->p_rlimit[uap->which], (caddr_t)uap->rlp,
604 		    sizeof (struct rlimit));
605 	mtx_unlock(&Giant);
606 	return(error);
607 }
608 
609 /*
610  * Transform the running time and tick information in proc p into user,
611  * system, and interrupt time usage.
612  */
613 void
614 calcru(p, up, sp, ip)
615 	struct proc *p;
616 	struct timeval *up;
617 	struct timeval *sp;
618 	struct timeval *ip;
619 {
620 	/* {user, system, interrupt, total} {ticks, usec}; previous tu: */
621 	u_int64_t ut, uu, st, su, it, iu, tt, tu, ptu;
622 	u_int64_t uut = 0, sut = 0, iut = 0;
623 	int s;
624 	struct timeval tv;
625 	struct kse *ke;
626 	struct ksegrp *kg;
627 
628 	mtx_assert(&sched_lock, MA_OWNED);
629 	/* XXX: why spl-protect ?  worst case is an off-by-one report */
630 
631 	FOREACH_KSEGRP_IN_PROC(p, kg) {
632 		/* we could accumulate per ksegrp and per process here*/
633 		FOREACH_KSE_IN_GROUP(kg, ke) {
634 			s = splstatclock();
635 			ut = ke->ke_uticks;
636 			st = ke->ke_sticks;
637 			it = ke->ke_iticks;
638 			splx(s);
639 
640 			tt = ut + st + it;
641 			if (tt == 0) {
642 				st = 1;
643 				tt = 1;
644 			}
645 
646 			tu = p->p_runtime;
647 			if (ke == curthread->td_kse) {
648 		/*
649 		 * Adjust for the current time slice.  This is actually fairly
650 		 * important since the error here is on the order of a time
651 		 * quantum, which is much greater than the sampling error.
652 		 * XXXKSE use a different test due to threads on other
653 		 * processors also being 'current'.
654 		 */
655 				microuptime(&tv);
656 				if (timevalcmp(&tv, PCPU_PTR(switchtime), <))
657 					printf("microuptime() went backwards (%ld.%06ld -> %ld.%06ld)\n",
658 			    		    (long)PCPU_GET(switchtime.tv_sec),
659 					    PCPU_GET(switchtime.tv_usec),
660 			    		    (long)tv.tv_sec, tv.tv_usec);
661 				else
662 					tu += (tv.tv_usec
663 						- PCPU_GET(switchtime.tv_usec))
664 						+ (tv.tv_sec
665 						  - PCPU_GET(switchtime.tv_sec))
666 					      * (int64_t)1000000;
667 			}
668 			ptu = ke->ke_uu + ke->ke_su + ke->ke_iu;
669 			if (tu < ptu || (int64_t)tu < 0) {
670 				/* XXX no %qd in kernel.  Truncate. */
671 				printf("calcru: negative time of %ld usec for pid %d (%s)\n",
672 		       		(long)tu, p->p_pid, p->p_comm);
673 				tu = ptu;
674 			}
675 
676 			/* Subdivide tu. */
677 			uu = (tu * ut) / tt;
678 			su = (tu * st) / tt;
679 			iu = tu - uu - su;
680 
681 			/* Enforce monotonicity. */
682 			if (uu < ke->ke_uu || su < ke->ke_su || iu < ke->ke_iu) {
683 				if (uu < ke->ke_uu)
684 					uu = ke->ke_uu;
685 				else if (uu + ke->ke_su + ke->ke_iu > tu)
686 					uu = tu - ke->ke_su - ke->ke_iu;
687 				if (st == 0)
688 					su = ke->ke_su;
689 				else {
690 					su = ((tu - uu) * st) / (st + it);
691 					if (su < ke->ke_su)
692 						su = ke->ke_su;
693 					else if (uu + su + ke->ke_iu > tu)
694 						su = tu - uu - ke->ke_iu;
695 				}
696 				KASSERT(uu + su + ke->ke_iu <= tu,
697 		    		("calcru: monotonisation botch 1"));
698 				iu = tu - uu - su;
699 				KASSERT(iu >= ke->ke_iu,
700 		    		("calcru: monotonisation botch 2"));
701 			}
702 			ke->ke_uu = uu;
703 			ke->ke_su = su;
704 			ke->ke_iu = iu;
705 			uut += uu;
706 			sut += su;
707 			iut += iu;
708 
709 		} /* end kse loop */
710 	} /* end kseg loop */
711 	up->tv_sec = uut / 1000000;
712 	up->tv_usec = uut % 1000000;
713 	sp->tv_sec = sut / 1000000;
714 	sp->tv_usec = sut % 1000000;
715 	if (ip != NULL) {
716 		ip->tv_sec = iut / 1000000;
717 		ip->tv_usec = iut % 1000000;
718 	}
719 }
720 
721 #ifndef _SYS_SYSPROTO_H_
722 struct getrusage_args {
723 	int	who;
724 	struct	rusage *rusage;
725 };
726 #endif
727 /*
728  * MPSAFE
729  */
730 /* ARGSUSED */
731 int
732 getrusage(td, uap)
733 	register struct thread *td;
734 	register struct getrusage_args *uap;
735 {
736 	struct proc *p = td->td_proc;
737 	register struct rusage *rup;
738 	int error = 0;
739 
740 	mtx_lock(&Giant);
741 
742 	switch (uap->who) {
743 	case RUSAGE_SELF:
744 		rup = &p->p_stats->p_ru;
745 		mtx_lock_spin(&sched_lock);
746 		calcru(p, &rup->ru_utime, &rup->ru_stime, NULL);
747 		mtx_unlock_spin(&sched_lock);
748 		break;
749 
750 	case RUSAGE_CHILDREN:
751 		rup = &p->p_stats->p_cru;
752 		break;
753 
754 	default:
755 		rup = NULL;
756 		error = EINVAL;
757 		break;
758 	}
759 	mtx_unlock(&Giant);
760 	if (error == 0) {
761 		error = copyout((caddr_t)rup, (caddr_t)uap->rusage,
762 		    sizeof (struct rusage));
763 	}
764 	return(error);
765 }
766 
767 void
768 ruadd(ru, ru2)
769 	register struct rusage *ru, *ru2;
770 {
771 	register long *ip, *ip2;
772 	register int i;
773 
774 	timevaladd(&ru->ru_utime, &ru2->ru_utime);
775 	timevaladd(&ru->ru_stime, &ru2->ru_stime);
776 	if (ru->ru_maxrss < ru2->ru_maxrss)
777 		ru->ru_maxrss = ru2->ru_maxrss;
778 	ip = &ru->ru_first; ip2 = &ru2->ru_first;
779 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
780 		*ip++ += *ip2++;
781 }
782 
783 /*
784  * Make a copy of the plimit structure.
785  * We share these structures copy-on-write after fork,
786  * and copy when a limit is changed.
787  */
788 struct plimit *
789 limcopy(lim)
790 	struct plimit *lim;
791 {
792 	register struct plimit *copy;
793 
794 	MALLOC(copy, struct plimit *, sizeof(struct plimit),
795 	    M_SUBPROC, M_WAITOK);
796 	bcopy(lim->pl_rlimit, copy->pl_rlimit, sizeof(struct plimit));
797 	copy->p_lflags = 0;
798 	copy->p_refcnt = 1;
799 	return (copy);
800 }
801 
802 /*
803  * Find the uidinfo structure for a uid.  This structure is used to
804  * track the total resource consumption (process count, socket buffer
805  * size, etc.) for the uid and impose limits.
806  */
807 void
808 uihashinit()
809 {
810 
811 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
812 	mtx_init(&uihashtbl_mtx, "uidinfo hash", MTX_DEF);
813 }
814 
815 /*
816  * lookup a uidinfo struct for the parameter uid.
817  * uihashtbl_mtx must be locked.
818  */
819 static struct uidinfo *
820 uilookup(uid)
821 	uid_t uid;
822 {
823 	struct	uihashhead *uipp;
824 	struct	uidinfo *uip;
825 
826 	mtx_assert(&uihashtbl_mtx, MA_OWNED);
827 	uipp = UIHASH(uid);
828 	LIST_FOREACH(uip, uipp, ui_hash)
829 		if (uip->ui_uid == uid)
830 			break;
831 
832 	return (uip);
833 }
834 
835 /*
836  * Find or allocate a struct uidinfo for a particular uid.
837  * Increase refcount on uidinfo struct returned.
838  * uifree() should be called on a struct uidinfo when released.
839  */
840 struct uidinfo *
841 uifind(uid)
842 	uid_t uid;
843 {
844 	struct	uidinfo *uip;
845 
846 	mtx_lock(&uihashtbl_mtx);
847 	uip = uilookup(uid);
848 	if (uip == NULL) {
849 		struct  uidinfo *old_uip;
850 
851 		mtx_unlock(&uihashtbl_mtx);
852 		uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
853 		mtx_lock(&uihashtbl_mtx);
854 		/*
855 		 * There's a chance someone created our uidinfo while we
856 		 * were in malloc and not holding the lock, so we have to
857 		 * make sure we don't insert a duplicate uidinfo
858 		 */
859 		if ((old_uip = uilookup(uid)) != NULL) {
860 			/* someone else beat us to it */
861 			free(uip, M_UIDINFO);
862 			uip = old_uip;
863 		} else {
864 			uip->ui_mtxp = mtx_pool_alloc();
865 			uip->ui_uid = uid;
866 			LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
867 		}
868 	}
869 	uihold(uip);
870 	mtx_unlock(&uihashtbl_mtx);
871 	return (uip);
872 }
873 
874 /*
875  * Place another refcount on a uidinfo struct.
876  */
877 void
878 uihold(uip)
879 	struct uidinfo *uip;
880 {
881 
882 	UIDINFO_LOCK(uip);
883 	uip->ui_ref++;
884 	UIDINFO_UNLOCK(uip);
885 }
886 
887 /*-
888  * Since uidinfo structs have a long lifetime, we use an
889  * opportunistic refcounting scheme to avoid locking the lookup hash
890  * for each release.
891  *
892  * If the refcount hits 0, we need to free the structure,
893  * which means we need to lock the hash.
894  * Optimal case:
895  *   After locking the struct and lowering the refcount, if we find
896  *   that we don't need to free, simply unlock and return.
897  * Suboptimal case:
898  *   If refcount lowering results in need to free, bump the count
899  *   back up, loose the lock and aquire the locks in the proper
900  *   order to try again.
901  */
902 void
903 uifree(uip)
904 	struct uidinfo *uip;
905 {
906 
907 	/* Prepare for optimal case. */
908 	UIDINFO_LOCK(uip);
909 
910 	if (--uip->ui_ref != 0) {
911 		UIDINFO_UNLOCK(uip);
912 		return;
913 	}
914 
915 	/* Prepare for suboptimal case. */
916 	uip->ui_ref++;
917 	UIDINFO_UNLOCK(uip);
918 	mtx_lock(&uihashtbl_mtx);
919 	UIDINFO_LOCK(uip);
920 
921 	/*
922 	 * We must subtract one from the count again because we backed out
923 	 * our initial subtraction before dropping the lock.
924 	 * Since another thread may have added a reference after we dropped the
925 	 * initial lock we have to test for zero again.
926 	 */
927 	if (--uip->ui_ref == 0) {
928 		LIST_REMOVE(uip, ui_hash);
929 		mtx_unlock(&uihashtbl_mtx);
930 		if (uip->ui_sbsize != 0)
931 			/* XXX no %qd in kernel.  Truncate. */
932 			printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
933 			    uip->ui_uid, (long)uip->ui_sbsize);
934 		if (uip->ui_proccnt != 0)
935 			printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
936 			    uip->ui_uid, uip->ui_proccnt);
937 		UIDINFO_UNLOCK(uip);
938 		FREE(uip, M_UIDINFO);
939 		return;
940 	}
941 
942 	mtx_unlock(&uihashtbl_mtx);
943 	UIDINFO_UNLOCK(uip);
944 }
945 
946 /*
947  * Change the count associated with number of processes
948  * a given user is using.  When 'max' is 0, don't enforce a limit
949  */
950 int
951 chgproccnt(uip, diff, max)
952 	struct	uidinfo	*uip;
953 	int	diff;
954 	int	max;
955 {
956 
957 	UIDINFO_LOCK(uip);
958 	/* don't allow them to exceed max, but allow subtraction */
959 	if (diff > 0 && uip->ui_proccnt + diff > max && max != 0) {
960 		UIDINFO_UNLOCK(uip);
961 		return (0);
962 	}
963 	uip->ui_proccnt += diff;
964 	if (uip->ui_proccnt < 0)
965 		printf("negative proccnt for uid = %d\n", uip->ui_uid);
966 	UIDINFO_UNLOCK(uip);
967 	return (1);
968 }
969 
970 /*
971  * Change the total socket buffer size a user has used.
972  */
973 int
974 chgsbsize(uip, hiwat, to, max)
975 	struct	uidinfo	*uip;
976 	u_long *hiwat;
977 	u_long	to;
978 	rlim_t	max;
979 {
980 	rlim_t new;
981 	int s;
982 
983 	s = splnet();
984 	UIDINFO_LOCK(uip);
985 	new = uip->ui_sbsize + to - *hiwat;
986 	/* don't allow them to exceed max, but allow subtraction */
987 	if (to > *hiwat && new > max) {
988 		splx(s);
989 		UIDINFO_UNLOCK(uip);
990 		return (0);
991 	}
992 	uip->ui_sbsize = new;
993 	*hiwat = to;
994 	if (uip->ui_sbsize < 0)
995 		printf("negative sbsize for uid = %d\n", uip->ui_uid);
996 	splx(s);
997 	UIDINFO_UNLOCK(uip);
998 	return (1);
999 }
1000