xref: /freebsd/sys/kern/kern_resource.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_resource.c	8.5 (Berkeley) 1/21/94
37  */
38 
39 #include <sys/cdefs.h>
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/file.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mutex.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/refcount.h>
51 #include <sys/racct.h>
52 #include <sys/resourcevar.h>
53 #include <sys/rwlock.h>
54 #include <sys/sched.h>
55 #include <sys/sx.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/time.h>
60 #include <sys/umtxvar.h>
61 
62 #include <vm/vm.h>
63 #include <vm/vm_param.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_map.h>
66 
67 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
68 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
69 #define	UIHASH(uid)	(&uihashtbl[(uid) & uihash])
70 static struct rwlock uihashtbl_lock;
71 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
72 static u_long uihash;		/* size of hash table - 1 */
73 
74 static void	calcru1(struct proc *p, struct rusage_ext *ruxp,
75 		    struct timeval *up, struct timeval *sp);
76 static int	donice(struct thread *td, struct proc *chgp, int n);
77 static struct uidinfo *uilookup(uid_t uid);
78 static void	ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td);
79 
80 /*
81  * Resource controls and accounting.
82  */
83 #ifndef _SYS_SYSPROTO_H_
84 struct getpriority_args {
85 	int	which;
86 	int	who;
87 };
88 #endif
89 int
90 sys_getpriority(struct thread *td, struct getpriority_args *uap)
91 {
92 
93 	return (kern_getpriority(td, uap->which, uap->who));
94 }
95 
96 int
97 kern_getpriority(struct thread *td, int which, int who)
98 {
99 	struct proc *p;
100 	struct pgrp *pg;
101 	int error, low;
102 
103 	error = 0;
104 	low = PRIO_MAX + 1;
105 	switch (which) {
106 	case PRIO_PROCESS:
107 		if (who == 0)
108 			low = td->td_proc->p_nice;
109 		else {
110 			p = pfind(who);
111 			if (p == NULL)
112 				break;
113 			if (p_cansee(td, p) == 0)
114 				low = p->p_nice;
115 			PROC_UNLOCK(p);
116 		}
117 		break;
118 
119 	case PRIO_PGRP:
120 		sx_slock(&proctree_lock);
121 		if (who == 0) {
122 			pg = td->td_proc->p_pgrp;
123 			PGRP_LOCK(pg);
124 		} else {
125 			pg = pgfind(who);
126 			if (pg == NULL) {
127 				sx_sunlock(&proctree_lock);
128 				break;
129 			}
130 		}
131 		sx_sunlock(&proctree_lock);
132 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
133 			PROC_LOCK(p);
134 			if (p->p_state == PRS_NORMAL &&
135 			    p_cansee(td, p) == 0) {
136 				if (p->p_nice < low)
137 					low = p->p_nice;
138 			}
139 			PROC_UNLOCK(p);
140 		}
141 		PGRP_UNLOCK(pg);
142 		break;
143 
144 	case PRIO_USER:
145 		if (who == 0)
146 			who = td->td_ucred->cr_uid;
147 		sx_slock(&allproc_lock);
148 		FOREACH_PROC_IN_SYSTEM(p) {
149 			PROC_LOCK(p);
150 			if (p->p_state == PRS_NORMAL &&
151 			    p_cansee(td, p) == 0 &&
152 			    p->p_ucred->cr_uid == who) {
153 				if (p->p_nice < low)
154 					low = p->p_nice;
155 			}
156 			PROC_UNLOCK(p);
157 		}
158 		sx_sunlock(&allproc_lock);
159 		break;
160 
161 	default:
162 		error = EINVAL;
163 		break;
164 	}
165 	if (low == PRIO_MAX + 1 && error == 0)
166 		error = ESRCH;
167 	td->td_retval[0] = low;
168 	return (error);
169 }
170 
171 #ifndef _SYS_SYSPROTO_H_
172 struct setpriority_args {
173 	int	which;
174 	int	who;
175 	int	prio;
176 };
177 #endif
178 int
179 sys_setpriority(struct thread *td, struct setpriority_args *uap)
180 {
181 
182 	return (kern_setpriority(td, uap->which, uap->who, uap->prio));
183 }
184 
185 int
186 kern_setpriority(struct thread *td, int which, int who, int prio)
187 {
188 	struct proc *curp, *p;
189 	struct pgrp *pg;
190 	int found = 0, error = 0;
191 
192 	curp = td->td_proc;
193 	switch (which) {
194 	case PRIO_PROCESS:
195 		if (who == 0) {
196 			PROC_LOCK(curp);
197 			error = donice(td, curp, prio);
198 			PROC_UNLOCK(curp);
199 		} else {
200 			p = pfind(who);
201 			if (p == NULL)
202 				break;
203 			error = p_cansee(td, p);
204 			if (error == 0)
205 				error = donice(td, p, prio);
206 			PROC_UNLOCK(p);
207 		}
208 		found++;
209 		break;
210 
211 	case PRIO_PGRP:
212 		sx_slock(&proctree_lock);
213 		if (who == 0) {
214 			pg = curp->p_pgrp;
215 			PGRP_LOCK(pg);
216 		} else {
217 			pg = pgfind(who);
218 			if (pg == NULL) {
219 				sx_sunlock(&proctree_lock);
220 				break;
221 			}
222 		}
223 		sx_sunlock(&proctree_lock);
224 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
225 			PROC_LOCK(p);
226 			if (p->p_state == PRS_NORMAL &&
227 			    p_cansee(td, p) == 0) {
228 				error = donice(td, p, prio);
229 				found++;
230 			}
231 			PROC_UNLOCK(p);
232 		}
233 		PGRP_UNLOCK(pg);
234 		break;
235 
236 	case PRIO_USER:
237 		if (who == 0)
238 			who = td->td_ucred->cr_uid;
239 		sx_slock(&allproc_lock);
240 		FOREACH_PROC_IN_SYSTEM(p) {
241 			PROC_LOCK(p);
242 			if (p->p_state == PRS_NORMAL &&
243 			    p->p_ucred->cr_uid == who &&
244 			    p_cansee(td, p) == 0) {
245 				error = donice(td, p, prio);
246 				found++;
247 			}
248 			PROC_UNLOCK(p);
249 		}
250 		sx_sunlock(&allproc_lock);
251 		break;
252 
253 	default:
254 		error = EINVAL;
255 		break;
256 	}
257 	if (found == 0 && error == 0)
258 		error = ESRCH;
259 	return (error);
260 }
261 
262 /*
263  * Set "nice" for a (whole) process.
264  */
265 static int
266 donice(struct thread *td, struct proc *p, int n)
267 {
268 	int error;
269 
270 	PROC_LOCK_ASSERT(p, MA_OWNED);
271 	if ((error = p_cansched(td, p)))
272 		return (error);
273 	if (n > PRIO_MAX)
274 		n = PRIO_MAX;
275 	if (n < PRIO_MIN)
276 		n = PRIO_MIN;
277 	if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
278 		return (EACCES);
279 	sched_nice(p, n);
280 	return (0);
281 }
282 
283 static int unprivileged_idprio;
284 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
285     &unprivileged_idprio, 0,
286     "Allow non-root users to set an idle priority (deprecated)");
287 
288 /*
289  * Set realtime priority for LWP.
290  */
291 #ifndef _SYS_SYSPROTO_H_
292 struct rtprio_thread_args {
293 	int		function;
294 	lwpid_t		lwpid;
295 	struct rtprio	*rtp;
296 };
297 #endif
298 int
299 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
300 {
301 	struct proc *p;
302 	struct rtprio rtp;
303 	struct thread *td1;
304 	int cierror, error;
305 
306 	/* Perform copyin before acquiring locks if needed. */
307 	if (uap->function == RTP_SET)
308 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
309 	else
310 		cierror = 0;
311 
312 	if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
313 		p = td->td_proc;
314 		td1 = td;
315 		PROC_LOCK(p);
316 	} else {
317 		td1 = tdfind(uap->lwpid, -1);
318 		if (td1 == NULL)
319 			return (ESRCH);
320 		p = td1->td_proc;
321 	}
322 
323 	switch (uap->function) {
324 	case RTP_LOOKUP:
325 		if ((error = p_cansee(td, p)))
326 			break;
327 		pri_to_rtp(td1, &rtp);
328 		PROC_UNLOCK(p);
329 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
330 	case RTP_SET:
331 		if ((error = p_cansched(td, p)) || (error = cierror))
332 			break;
333 
334 		/* Disallow setting rtprio in most cases if not superuser. */
335 
336 		/*
337 		 * Realtime priority has to be restricted for reasons which
338 		 * should be obvious.  However, for idleprio processes, there is
339 		 * a potential for system deadlock if an idleprio process gains
340 		 * a lock on a resource that other processes need (and the
341 		 * idleprio process can't run due to a CPU-bound normal
342 		 * process).  Fix me!  XXX
343 		 *
344 		 * This problem is not only related to idleprio process.
345 		 * A user level program can obtain a file lock and hold it
346 		 * indefinitely.  Additionally, without idleprio processes it is
347 		 * still conceivable that a program with low priority will never
348 		 * get to run.  In short, allowing this feature might make it
349 		 * easier to lock a resource indefinitely, but it is not the
350 		 * only thing that makes it possible.
351 		 */
352 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
353 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
354 			break;
355 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
356 		    unprivileged_idprio == 0 &&
357 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
358 			break;
359 		error = rtp_to_pri(&rtp, td1);
360 		break;
361 	default:
362 		error = EINVAL;
363 		break;
364 	}
365 	PROC_UNLOCK(p);
366 	return (error);
367 }
368 
369 /*
370  * Set realtime priority.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct rtprio_args {
374 	int		function;
375 	pid_t		pid;
376 	struct rtprio	*rtp;
377 };
378 #endif
379 int
380 sys_rtprio(struct thread *td, struct rtprio_args *uap)
381 {
382 	struct proc *p;
383 	struct thread *tdp;
384 	struct rtprio rtp;
385 	int cierror, error;
386 
387 	/* Perform copyin before acquiring locks if needed. */
388 	if (uap->function == RTP_SET)
389 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
390 	else
391 		cierror = 0;
392 
393 	if (uap->pid == 0) {
394 		p = td->td_proc;
395 		PROC_LOCK(p);
396 	} else {
397 		p = pfind(uap->pid);
398 		if (p == NULL)
399 			return (ESRCH);
400 	}
401 
402 	switch (uap->function) {
403 	case RTP_LOOKUP:
404 		if ((error = p_cansee(td, p)))
405 			break;
406 		/*
407 		 * Return OUR priority if no pid specified,
408 		 * or if one is, report the highest priority
409 		 * in the process.  There isn't much more you can do as
410 		 * there is only room to return a single priority.
411 		 * Note: specifying our own pid is not the same
412 		 * as leaving it zero.
413 		 */
414 		if (uap->pid == 0) {
415 			pri_to_rtp(td, &rtp);
416 		} else {
417 			struct rtprio rtp2;
418 
419 			rtp.type = RTP_PRIO_IDLE;
420 			rtp.prio = RTP_PRIO_MAX;
421 			FOREACH_THREAD_IN_PROC(p, tdp) {
422 				pri_to_rtp(tdp, &rtp2);
423 				if (rtp2.type <  rtp.type ||
424 				    (rtp2.type == rtp.type &&
425 				    rtp2.prio < rtp.prio)) {
426 					rtp.type = rtp2.type;
427 					rtp.prio = rtp2.prio;
428 				}
429 			}
430 		}
431 		PROC_UNLOCK(p);
432 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
433 	case RTP_SET:
434 		if ((error = p_cansched(td, p)) || (error = cierror))
435 			break;
436 
437 		/*
438 		 * Disallow setting rtprio in most cases if not superuser.
439 		 * See the comment in sys_rtprio_thread about idprio
440 		 * threads holding a lock.
441 		 */
442 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
443 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
444 			break;
445 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
446 		    unprivileged_idprio == 0 &&
447 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
448 			break;
449 
450 		/*
451 		 * If we are setting our own priority, set just our
452 		 * thread but if we are doing another process,
453 		 * do all the threads on that process. If we
454 		 * specify our own pid we do the latter.
455 		 */
456 		if (uap->pid == 0) {
457 			error = rtp_to_pri(&rtp, td);
458 		} else {
459 			FOREACH_THREAD_IN_PROC(p, td) {
460 				if ((error = rtp_to_pri(&rtp, td)) != 0)
461 					break;
462 			}
463 		}
464 		break;
465 	default:
466 		error = EINVAL;
467 		break;
468 	}
469 	PROC_UNLOCK(p);
470 	return (error);
471 }
472 
473 int
474 rtp_to_pri(struct rtprio *rtp, struct thread *td)
475 {
476 	u_char  newpri, oldclass, oldpri;
477 
478 	switch (RTP_PRIO_BASE(rtp->type)) {
479 	case RTP_PRIO_REALTIME:
480 		if (rtp->prio > RTP_PRIO_MAX)
481 			return (EINVAL);
482 		newpri = PRI_MIN_REALTIME + rtp->prio;
483 		break;
484 	case RTP_PRIO_NORMAL:
485 		if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
486 			return (EINVAL);
487 		newpri = PRI_MIN_TIMESHARE + rtp->prio;
488 		break;
489 	case RTP_PRIO_IDLE:
490 		if (rtp->prio > RTP_PRIO_MAX)
491 			return (EINVAL);
492 		newpri = PRI_MIN_IDLE + rtp->prio;
493 		break;
494 	default:
495 		return (EINVAL);
496 	}
497 
498 	thread_lock(td);
499 	oldclass = td->td_pri_class;
500 	sched_class(td, rtp->type);	/* XXX fix */
501 	oldpri = td->td_user_pri;
502 	sched_user_prio(td, newpri);
503 	if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
504 	    td->td_pri_class != RTP_PRIO_NORMAL))
505 		sched_prio(td, td->td_user_pri);
506 	if (TD_ON_UPILOCK(td) && oldpri != newpri) {
507 		critical_enter();
508 		thread_unlock(td);
509 		umtx_pi_adjust(td, oldpri);
510 		critical_exit();
511 	} else
512 		thread_unlock(td);
513 	return (0);
514 }
515 
516 void
517 pri_to_rtp(struct thread *td, struct rtprio *rtp)
518 {
519 
520 	thread_lock(td);
521 	switch (PRI_BASE(td->td_pri_class)) {
522 	case PRI_REALTIME:
523 		rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
524 		break;
525 	case PRI_TIMESHARE:
526 		rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
527 		break;
528 	case PRI_IDLE:
529 		rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
530 		break;
531 	default:
532 		break;
533 	}
534 	rtp->type = td->td_pri_class;
535 	thread_unlock(td);
536 }
537 
538 #if defined(COMPAT_43)
539 #ifndef _SYS_SYSPROTO_H_
540 struct osetrlimit_args {
541 	u_int	which;
542 	struct	orlimit *rlp;
543 };
544 #endif
545 int
546 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
547 {
548 	struct orlimit olim;
549 	struct rlimit lim;
550 	int error;
551 
552 	if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
553 		return (error);
554 	lim.rlim_cur = olim.rlim_cur;
555 	lim.rlim_max = olim.rlim_max;
556 	error = kern_setrlimit(td, uap->which, &lim);
557 	return (error);
558 }
559 
560 #ifndef _SYS_SYSPROTO_H_
561 struct ogetrlimit_args {
562 	u_int	which;
563 	struct	orlimit *rlp;
564 };
565 #endif
566 int
567 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
568 {
569 	struct orlimit olim;
570 	struct rlimit rl;
571 	int error;
572 
573 	if (uap->which >= RLIM_NLIMITS)
574 		return (EINVAL);
575 	lim_rlimit(td, uap->which, &rl);
576 
577 	/*
578 	 * XXX would be more correct to convert only RLIM_INFINITY to the
579 	 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
580 	 * values.  Most 64->32 and 32->16 conversions, including not
581 	 * unimportant ones of uids are even more broken than what we
582 	 * do here (they blindly truncate).  We don't do this correctly
583 	 * here since we have little experience with EOVERFLOW yet.
584 	 * Elsewhere, getuid() can't fail...
585 	 */
586 	olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
587 	olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
588 	error = copyout(&olim, uap->rlp, sizeof(olim));
589 	return (error);
590 }
591 #endif /* COMPAT_43 */
592 
593 #ifndef _SYS_SYSPROTO_H_
594 struct setrlimit_args {
595 	u_int	which;
596 	struct	rlimit *rlp;
597 };
598 #endif
599 int
600 sys_setrlimit(struct thread *td, struct setrlimit_args *uap)
601 {
602 	struct rlimit alim;
603 	int error;
604 
605 	if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
606 		return (error);
607 	error = kern_setrlimit(td, uap->which, &alim);
608 	return (error);
609 }
610 
611 static void
612 lim_cb(void *arg)
613 {
614 	struct rlimit rlim;
615 	struct thread *td;
616 	struct proc *p;
617 
618 	p = arg;
619 	PROC_LOCK_ASSERT(p, MA_OWNED);
620 	/*
621 	 * Check if the process exceeds its cpu resource allocation.  If
622 	 * it reaches the max, arrange to kill the process in ast().
623 	 */
624 	if (p->p_cpulimit == RLIM_INFINITY)
625 		return;
626 	PROC_STATLOCK(p);
627 	FOREACH_THREAD_IN_PROC(p, td) {
628 		ruxagg(p, td);
629 	}
630 	PROC_STATUNLOCK(p);
631 	if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
632 		lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
633 		if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
634 			killproc(p, "exceeded maximum CPU limit");
635 		} else {
636 			if (p->p_cpulimit < rlim.rlim_max)
637 				p->p_cpulimit += 5;
638 			kern_psignal(p, SIGXCPU);
639 		}
640 	}
641 	if ((p->p_flag & P_WEXIT) == 0)
642 		callout_reset_sbt(&p->p_limco, SBT_1S, 0,
643 		    lim_cb, p, C_PREL(1));
644 }
645 
646 int
647 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
648 {
649 
650 	return (kern_proc_setrlimit(td, td->td_proc, which, limp));
651 }
652 
653 int
654 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
655     struct rlimit *limp)
656 {
657 	struct plimit *newlim, *oldlim, *oldlim_td;
658 	struct rlimit *alimp;
659 	struct rlimit oldssiz;
660 	int error;
661 
662 	if (which >= RLIM_NLIMITS)
663 		return (EINVAL);
664 
665 	/*
666 	 * Preserve historical bugs by treating negative limits as unsigned.
667 	 */
668 	if (limp->rlim_cur < 0)
669 		limp->rlim_cur = RLIM_INFINITY;
670 	if (limp->rlim_max < 0)
671 		limp->rlim_max = RLIM_INFINITY;
672 
673 	oldssiz.rlim_cur = 0;
674 	newlim = lim_alloc();
675 	PROC_LOCK(p);
676 	oldlim = p->p_limit;
677 	alimp = &oldlim->pl_rlimit[which];
678 	if (limp->rlim_cur > alimp->rlim_max ||
679 	    limp->rlim_max > alimp->rlim_max)
680 		if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
681 			PROC_UNLOCK(p);
682 			lim_free(newlim);
683 			return (error);
684 		}
685 	if (limp->rlim_cur > limp->rlim_max)
686 		limp->rlim_cur = limp->rlim_max;
687 	lim_copy(newlim, oldlim);
688 	alimp = &newlim->pl_rlimit[which];
689 
690 	switch (which) {
691 	case RLIMIT_CPU:
692 		if (limp->rlim_cur != RLIM_INFINITY &&
693 		    p->p_cpulimit == RLIM_INFINITY)
694 			callout_reset_sbt(&p->p_limco, SBT_1S, 0,
695 			    lim_cb, p, C_PREL(1));
696 		p->p_cpulimit = limp->rlim_cur;
697 		break;
698 	case RLIMIT_DATA:
699 		if (limp->rlim_cur > maxdsiz)
700 			limp->rlim_cur = maxdsiz;
701 		if (limp->rlim_max > maxdsiz)
702 			limp->rlim_max = maxdsiz;
703 		break;
704 
705 	case RLIMIT_STACK:
706 		if (limp->rlim_cur > maxssiz)
707 			limp->rlim_cur = maxssiz;
708 		if (limp->rlim_max > maxssiz)
709 			limp->rlim_max = maxssiz;
710 		oldssiz = *alimp;
711 		if (p->p_sysent->sv_fixlimit != NULL)
712 			p->p_sysent->sv_fixlimit(&oldssiz,
713 			    RLIMIT_STACK);
714 		break;
715 
716 	case RLIMIT_NOFILE:
717 		if (limp->rlim_cur > maxfilesperproc)
718 			limp->rlim_cur = maxfilesperproc;
719 		if (limp->rlim_max > maxfilesperproc)
720 			limp->rlim_max = maxfilesperproc;
721 		break;
722 
723 	case RLIMIT_NPROC:
724 		if (limp->rlim_cur > maxprocperuid)
725 			limp->rlim_cur = maxprocperuid;
726 		if (limp->rlim_max > maxprocperuid)
727 			limp->rlim_max = maxprocperuid;
728 		if (limp->rlim_cur < 1)
729 			limp->rlim_cur = 1;
730 		if (limp->rlim_max < 1)
731 			limp->rlim_max = 1;
732 		break;
733 	}
734 	if (p->p_sysent->sv_fixlimit != NULL)
735 		p->p_sysent->sv_fixlimit(limp, which);
736 	*alimp = *limp;
737 	p->p_limit = newlim;
738 	PROC_UPDATE_COW(p);
739 	oldlim_td = NULL;
740 	if (td == curthread && PROC_COW_CHANGECOUNT(td, p) == 1) {
741 		oldlim_td = lim_cowsync();
742 		thread_cow_synced(td);
743 	}
744 	PROC_UNLOCK(p);
745 	if (oldlim_td != NULL) {
746 		MPASS(oldlim_td == oldlim);
747 		lim_freen(oldlim, 2);
748 	} else {
749 		lim_free(oldlim);
750 	}
751 
752 	if (which == RLIMIT_STACK &&
753 	    /*
754 	     * Skip calls from exec_new_vmspace(), done when stack is
755 	     * not mapped yet.
756 	     */
757 	    (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
758 		/*
759 		 * Stack is allocated to the max at exec time with only
760 		 * "rlim_cur" bytes accessible.  If stack limit is going
761 		 * up make more accessible, if going down make inaccessible.
762 		 */
763 		if (limp->rlim_cur != oldssiz.rlim_cur) {
764 			vm_offset_t addr;
765 			vm_size_t size;
766 			vm_prot_t prot;
767 
768 			if (limp->rlim_cur > oldssiz.rlim_cur) {
769 				prot = p->p_sysent->sv_stackprot;
770 				size = limp->rlim_cur - oldssiz.rlim_cur;
771 				addr = round_page(p->p_vmspace->vm_stacktop) -
772 				    limp->rlim_cur;
773 			} else {
774 				prot = VM_PROT_NONE;
775 				size = oldssiz.rlim_cur - limp->rlim_cur;
776 				addr = round_page(p->p_vmspace->vm_stacktop) -
777 				    oldssiz.rlim_cur;
778 			}
779 			addr = trunc_page(addr);
780 			size = round_page(size);
781 			(void)vm_map_protect(&p->p_vmspace->vm_map,
782 			    addr, addr + size, prot, 0,
783 			    VM_MAP_PROTECT_SET_PROT);
784 		}
785 	}
786 
787 	return (0);
788 }
789 
790 #ifndef _SYS_SYSPROTO_H_
791 struct getrlimit_args {
792 	u_int	which;
793 	struct	rlimit *rlp;
794 };
795 #endif
796 /* ARGSUSED */
797 int
798 sys_getrlimit(struct thread *td, struct getrlimit_args *uap)
799 {
800 	struct rlimit rlim;
801 	int error;
802 
803 	if (uap->which >= RLIM_NLIMITS)
804 		return (EINVAL);
805 	lim_rlimit(td, uap->which, &rlim);
806 	error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
807 	return (error);
808 }
809 
810 /*
811  * Transform the running time and tick information for children of proc p
812  * into user and system time usage.
813  */
814 void
815 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
816 {
817 
818 	PROC_LOCK_ASSERT(p, MA_OWNED);
819 	calcru1(p, &p->p_crux, up, sp);
820 }
821 
822 /*
823  * Transform the running time and tick information in proc p into user
824  * and system time usage.  If appropriate, include the current time slice
825  * on this CPU.
826  */
827 void
828 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
829 {
830 	struct thread *td;
831 	uint64_t runtime, u;
832 
833 	PROC_LOCK_ASSERT(p, MA_OWNED);
834 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
835 	/*
836 	 * If we are getting stats for the current process, then add in the
837 	 * stats that this thread has accumulated in its current time slice.
838 	 * We reset the thread and CPU state as if we had performed a context
839 	 * switch right here.
840 	 */
841 	td = curthread;
842 	if (td->td_proc == p) {
843 		u = cpu_ticks();
844 		runtime = u - PCPU_GET(switchtime);
845 		td->td_runtime += runtime;
846 		td->td_incruntime += runtime;
847 		PCPU_SET(switchtime, u);
848 	}
849 	/* Make sure the per-thread stats are current. */
850 	FOREACH_THREAD_IN_PROC(p, td) {
851 		if (td->td_incruntime == 0)
852 			continue;
853 		ruxagg(p, td);
854 	}
855 	calcru1(p, &p->p_rux, up, sp);
856 }
857 
858 /* Collect resource usage for a single thread. */
859 void
860 rufetchtd(struct thread *td, struct rusage *ru)
861 {
862 	struct proc *p;
863 	uint64_t runtime, u;
864 
865 	p = td->td_proc;
866 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
867 	THREAD_LOCK_ASSERT(td, MA_OWNED);
868 	/*
869 	 * If we are getting stats for the current thread, then add in the
870 	 * stats that this thread has accumulated in its current time slice.
871 	 * We reset the thread and CPU state as if we had performed a context
872 	 * switch right here.
873 	 */
874 	if (td == curthread) {
875 		u = cpu_ticks();
876 		runtime = u - PCPU_GET(switchtime);
877 		td->td_runtime += runtime;
878 		td->td_incruntime += runtime;
879 		PCPU_SET(switchtime, u);
880 	}
881 	ruxagg_locked(p, td);
882 	*ru = td->td_ru;
883 	calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
884 }
885 
886 static uint64_t
887 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
888 {
889 	uint64_t acc, bh, bl;
890 	int i, s, sa, sb;
891 
892 	/*
893 	 * Calculate (a * b) / c accurately enough without overflowing.  c
894 	 * must be nonzero, and its top bit must be 0.  a or b must be
895 	 * <= c, and the implementation is tuned for b <= c.
896 	 *
897 	 * The comments about times are for use in calcru1() with units of
898 	 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
899 	 *
900 	 * Let n be the number of top zero bits in c.  Each iteration
901 	 * either returns, or reduces b by right shifting it by at least n.
902 	 * The number of iterations is at most 1 + 64 / n, and the error is
903 	 * at most the number of iterations.
904 	 *
905 	 * It is very unusual to need even 2 iterations.  Previous
906 	 * implementations overflowed essentially by returning early in the
907 	 * first iteration, with n = 38 giving overflow at 105+ hours and
908 	 * n = 32 giving overlow at at 388+ days despite a more careful
909 	 * calculation.  388 days is a reasonable uptime, and the calculation
910 	 * needs to work for the uptime times the number of CPUs since 'a'
911 	 * is per-process.
912 	 */
913 	if (a >= (uint64_t)1 << 63)
914 		return (0);		/* Unsupported arg -- can't happen. */
915 	acc = 0;
916 	for (i = 0; i < 128; i++) {
917 		sa = flsll(a);
918 		sb = flsll(b);
919 		if (sa + sb <= 64)
920 			/* Up to 105 hours on first iteration. */
921 			return (acc + (a * b) / c);
922 		if (a >= c) {
923 			/*
924 			 * This reduction is based on a = q * c + r, with the
925 			 * remainder r < c.  'a' may be large to start, and
926 			 * moving bits from b into 'a' at the end of the loop
927 			 * sets the top bit of 'a', so the reduction makes
928 			 * significant progress.
929 			 */
930 			acc += (a / c) * b;
931 			a %= c;
932 			sa = flsll(a);
933 			if (sa + sb <= 64)
934 				/* Up to 388 days on first iteration. */
935 				return (acc + (a * b) / c);
936 		}
937 
938 		/*
939 		 * This step writes a * b as a * ((bh << s) + bl) =
940 		 * a * (bh << s) + a * bl = (a << s) * bh + a * bl.  The 2
941 		 * additive terms are handled separately.  Splitting in
942 		 * this way is linear except for rounding errors.
943 		 *
944 		 * s = 64 - sa is the maximum such that a << s fits in 64
945 		 * bits.  Since a < c and c has at least 1 zero top bit,
946 		 * sa < 64 and s > 0.  Thus this step makes progress by
947 		 * reducing b (it increases 'a', but taking remainders on
948 		 * the next iteration completes the reduction).
949 		 *
950 		 * Finally, the choice for s is just what is needed to keep
951 		 * a * bl from overflowing, so we don't need complications
952 		 * like a recursive call mul64_by_fraction(a, bl, c) to
953 		 * handle the second additive term.
954 		 */
955 		s = 64 - sa;
956 		bh = b >> s;
957 		bl = b - (bh << s);
958 		acc += (a * bl) / c;
959 		a <<= s;
960 		b = bh;
961 	}
962 	return (0);		/* Algorithm failure -- can't happen. */
963 }
964 
965 static void
966 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
967     struct timeval *sp)
968 {
969 	/* {user, system, interrupt, total} {ticks, usec}: */
970 	uint64_t ut, uu, st, su, it, tt, tu;
971 
972 	ut = ruxp->rux_uticks;
973 	st = ruxp->rux_sticks;
974 	it = ruxp->rux_iticks;
975 	tt = ut + st + it;
976 	if (tt == 0) {
977 		/* Avoid divide by zero */
978 		st = 1;
979 		tt = 1;
980 	}
981 	tu = cputick2usec(ruxp->rux_runtime);
982 	if ((int64_t)tu < 0) {
983 		/* XXX: this should be an assert /phk */
984 		printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
985 		    (intmax_t)tu, p->p_pid, p->p_comm);
986 		tu = ruxp->rux_tu;
987 	}
988 
989 	/* Subdivide tu.  Avoid overflow in the multiplications. */
990 	if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
991 		/* Up to 76 hours when stathz is 128. */
992 		uu = (tu * ut) / tt;
993 		su = (tu * st) / tt;
994 	} else {
995 		uu = mul64_by_fraction(tu, ut, tt);
996 		su = mul64_by_fraction(tu, st, tt);
997 	}
998 
999 	if (tu >= ruxp->rux_tu) {
1000 		/*
1001 		 * The normal case, time increased.
1002 		 * Enforce monotonicity of bucketed numbers.
1003 		 */
1004 		if (uu < ruxp->rux_uu)
1005 			uu = ruxp->rux_uu;
1006 		if (su < ruxp->rux_su)
1007 			su = ruxp->rux_su;
1008 	} else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1009 		/*
1010 		 * When we calibrate the cputicker, it is not uncommon to
1011 		 * see the presumably fixed frequency increase slightly over
1012 		 * time as a result of thermal stabilization and NTP
1013 		 * discipline (of the reference clock).  We therefore ignore
1014 		 * a bit of backwards slop because we  expect to catch up
1015 		 * shortly.  We use a 3 microsecond limit to catch low
1016 		 * counts and a 1% limit for high counts.
1017 		 */
1018 		uu = ruxp->rux_uu;
1019 		su = ruxp->rux_su;
1020 		tu = ruxp->rux_tu;
1021 	} else if (vm_guest == VM_GUEST_NO) {  /* tu < ruxp->rux_tu */
1022 		/*
1023 		 * What happened here was likely that a laptop, which ran at
1024 		 * a reduced clock frequency at boot, kicked into high gear.
1025 		 * The wisdom of spamming this message in that case is
1026 		 * dubious, but it might also be indicative of something
1027 		 * serious, so lets keep it and hope laptops can be made
1028 		 * more truthful about their CPU speed via ACPI.
1029 		 */
1030 		printf("calcru: runtime went backwards from %ju usec "
1031 		    "to %ju usec for pid %d (%s)\n",
1032 		    (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1033 		    p->p_pid, p->p_comm);
1034 	}
1035 
1036 	ruxp->rux_uu = uu;
1037 	ruxp->rux_su = su;
1038 	ruxp->rux_tu = tu;
1039 
1040 	up->tv_sec = uu / 1000000;
1041 	up->tv_usec = uu % 1000000;
1042 	sp->tv_sec = su / 1000000;
1043 	sp->tv_usec = su % 1000000;
1044 }
1045 
1046 #ifndef _SYS_SYSPROTO_H_
1047 struct getrusage_args {
1048 	int	who;
1049 	struct	rusage *rusage;
1050 };
1051 #endif
1052 int
1053 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1054 {
1055 	struct rusage ru;
1056 	int error;
1057 
1058 	error = kern_getrusage(td, uap->who, &ru);
1059 	if (error == 0)
1060 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1061 	return (error);
1062 }
1063 
1064 int
1065 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1066 {
1067 	struct proc *p;
1068 	int error;
1069 
1070 	error = 0;
1071 	p = td->td_proc;
1072 	PROC_LOCK(p);
1073 	switch (who) {
1074 	case RUSAGE_SELF:
1075 		rufetchcalc(p, rup, &rup->ru_utime,
1076 		    &rup->ru_stime);
1077 		break;
1078 
1079 	case RUSAGE_CHILDREN:
1080 		*rup = p->p_stats->p_cru;
1081 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1082 		break;
1083 
1084 	case RUSAGE_THREAD:
1085 		PROC_STATLOCK(p);
1086 		thread_lock(td);
1087 		rufetchtd(td, rup);
1088 		thread_unlock(td);
1089 		PROC_STATUNLOCK(p);
1090 		break;
1091 
1092 	default:
1093 		error = EINVAL;
1094 	}
1095 	PROC_UNLOCK(p);
1096 	return (error);
1097 }
1098 
1099 void
1100 rucollect(struct rusage *ru, struct rusage *ru2)
1101 {
1102 	long *ip, *ip2;
1103 	int i;
1104 
1105 	if (ru->ru_maxrss < ru2->ru_maxrss)
1106 		ru->ru_maxrss = ru2->ru_maxrss;
1107 	ip = &ru->ru_first;
1108 	ip2 = &ru2->ru_first;
1109 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1110 		*ip++ += *ip2++;
1111 }
1112 
1113 void
1114 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1115     struct rusage_ext *rux2)
1116 {
1117 
1118 	rux->rux_runtime += rux2->rux_runtime;
1119 	rux->rux_uticks += rux2->rux_uticks;
1120 	rux->rux_sticks += rux2->rux_sticks;
1121 	rux->rux_iticks += rux2->rux_iticks;
1122 	rux->rux_uu += rux2->rux_uu;
1123 	rux->rux_su += rux2->rux_su;
1124 	rux->rux_tu += rux2->rux_tu;
1125 	rucollect(ru, ru2);
1126 }
1127 
1128 /*
1129  * Aggregate tick counts into the proc's rusage_ext.
1130  */
1131 static void
1132 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1133 {
1134 
1135 	rux->rux_runtime += td->td_incruntime;
1136 	rux->rux_uticks += td->td_uticks;
1137 	rux->rux_sticks += td->td_sticks;
1138 	rux->rux_iticks += td->td_iticks;
1139 }
1140 
1141 void
1142 ruxagg_locked(struct proc *p, struct thread *td)
1143 {
1144 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1145 	PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1146 
1147 	ruxagg_ext_locked(&p->p_rux, td);
1148 	ruxagg_ext_locked(&td->td_rux, td);
1149 	td->td_incruntime = 0;
1150 	td->td_uticks = 0;
1151 	td->td_iticks = 0;
1152 	td->td_sticks = 0;
1153 }
1154 
1155 void
1156 ruxagg(struct proc *p, struct thread *td)
1157 {
1158 
1159 	thread_lock(td);
1160 	ruxagg_locked(p, td);
1161 	thread_unlock(td);
1162 }
1163 
1164 /*
1165  * Update the rusage_ext structure and fetch a valid aggregate rusage
1166  * for proc p if storage for one is supplied.
1167  */
1168 void
1169 rufetch(struct proc *p, struct rusage *ru)
1170 {
1171 	struct thread *td;
1172 
1173 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
1174 
1175 	*ru = p->p_ru;
1176 	if (p->p_numthreads > 0)  {
1177 		FOREACH_THREAD_IN_PROC(p, td) {
1178 			ruxagg(p, td);
1179 			rucollect(ru, &td->td_ru);
1180 		}
1181 	}
1182 }
1183 
1184 /*
1185  * Atomically perform a rufetch and a calcru together.
1186  * Consumers, can safely assume the calcru is executed only once
1187  * rufetch is completed.
1188  */
1189 void
1190 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1191     struct timeval *sp)
1192 {
1193 
1194 	PROC_STATLOCK(p);
1195 	rufetch(p, ru);
1196 	calcru(p, up, sp);
1197 	PROC_STATUNLOCK(p);
1198 }
1199 
1200 /*
1201  * Allocate a new resource limits structure and initialize its
1202  * reference count and mutex pointer.
1203  */
1204 struct plimit *
1205 lim_alloc(void)
1206 {
1207 	struct plimit *limp;
1208 
1209 	limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1210 	refcount_init(&limp->pl_refcnt, 1);
1211 	return (limp);
1212 }
1213 
1214 struct plimit *
1215 lim_hold(struct plimit *limp)
1216 {
1217 
1218 	refcount_acquire(&limp->pl_refcnt);
1219 	return (limp);
1220 }
1221 
1222 struct plimit *
1223 lim_cowsync(void)
1224 {
1225 	struct thread *td;
1226 	struct proc *p;
1227 	struct plimit *oldlimit;
1228 
1229 	td = curthread;
1230 	p = td->td_proc;
1231 	PROC_LOCK_ASSERT(p, MA_OWNED);
1232 
1233 	if (td->td_limit == p->p_limit)
1234 		return (NULL);
1235 
1236 	oldlimit = td->td_limit;
1237 	td->td_limit = lim_hold(p->p_limit);
1238 
1239 	return (oldlimit);
1240 }
1241 
1242 void
1243 lim_fork(struct proc *p1, struct proc *p2)
1244 {
1245 
1246 	PROC_LOCK_ASSERT(p1, MA_OWNED);
1247 	PROC_LOCK_ASSERT(p2, MA_OWNED);
1248 
1249 	p2->p_limit = lim_hold(p1->p_limit);
1250 	callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1251 	if (p1->p_cpulimit != RLIM_INFINITY)
1252 		callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1253 		    lim_cb, p2, C_PREL(1));
1254 }
1255 
1256 void
1257 lim_free(struct plimit *limp)
1258 {
1259 
1260 	if (refcount_release(&limp->pl_refcnt))
1261 		free((void *)limp, M_PLIMIT);
1262 }
1263 
1264 void
1265 lim_freen(struct plimit *limp, int n)
1266 {
1267 
1268 	if (refcount_releasen(&limp->pl_refcnt, n))
1269 		free((void *)limp, M_PLIMIT);
1270 }
1271 
1272 void
1273 limbatch_add(struct limbatch *lb, struct thread *td)
1274 {
1275 	struct plimit *limp;
1276 
1277 	MPASS(td->td_limit != NULL);
1278 	limp = td->td_limit;
1279 
1280 	if (lb->limp != limp) {
1281 		if (lb->count != 0) {
1282 			lim_freen(lb->limp, lb->count);
1283 			lb->count = 0;
1284 		}
1285 		lb->limp = limp;
1286 	}
1287 
1288 	lb->count++;
1289 }
1290 
1291 void
1292 limbatch_final(struct limbatch *lb)
1293 {
1294 
1295 	MPASS(lb->count != 0);
1296 	lim_freen(lb->limp, lb->count);
1297 }
1298 
1299 /*
1300  * Make a copy of the plimit structure.
1301  * We share these structures copy-on-write after fork.
1302  */
1303 void
1304 lim_copy(struct plimit *dst, struct plimit *src)
1305 {
1306 
1307 	KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1308 	bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1309 }
1310 
1311 /*
1312  * Return the hard limit for a particular system resource.  The
1313  * which parameter specifies the index into the rlimit array.
1314  */
1315 rlim_t
1316 lim_max(struct thread *td, int which)
1317 {
1318 	struct rlimit rl;
1319 
1320 	lim_rlimit(td, which, &rl);
1321 	return (rl.rlim_max);
1322 }
1323 
1324 rlim_t
1325 lim_max_proc(struct proc *p, int which)
1326 {
1327 	struct rlimit rl;
1328 
1329 	lim_rlimit_proc(p, which, &rl);
1330 	return (rl.rlim_max);
1331 }
1332 
1333 /*
1334  * Return the current (soft) limit for a particular system resource.
1335  * The which parameter which specifies the index into the rlimit array
1336  */
1337 rlim_t
1338 (lim_cur)(struct thread *td, int which)
1339 {
1340 	struct rlimit rl;
1341 
1342 	lim_rlimit(td, which, &rl);
1343 	return (rl.rlim_cur);
1344 }
1345 
1346 rlim_t
1347 lim_cur_proc(struct proc *p, int which)
1348 {
1349 	struct rlimit rl;
1350 
1351 	lim_rlimit_proc(p, which, &rl);
1352 	return (rl.rlim_cur);
1353 }
1354 
1355 /*
1356  * Return a copy of the entire rlimit structure for the system limit
1357  * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1358  */
1359 void
1360 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1361 {
1362 	struct proc *p = td->td_proc;
1363 
1364 	MPASS(td == curthread);
1365 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1366 	    ("request for invalid resource limit"));
1367 	*rlp = td->td_limit->pl_rlimit[which];
1368 	if (p->p_sysent->sv_fixlimit != NULL)
1369 		p->p_sysent->sv_fixlimit(rlp, which);
1370 }
1371 
1372 void
1373 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1374 {
1375 
1376 	PROC_LOCK_ASSERT(p, MA_OWNED);
1377 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1378 	    ("request for invalid resource limit"));
1379 	*rlp = p->p_limit->pl_rlimit[which];
1380 	if (p->p_sysent->sv_fixlimit != NULL)
1381 		p->p_sysent->sv_fixlimit(rlp, which);
1382 }
1383 
1384 void
1385 uihashinit(void)
1386 {
1387 
1388 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1389 	rw_init(&uihashtbl_lock, "uidinfo hash");
1390 }
1391 
1392 /*
1393  * Look up a uidinfo struct for the parameter uid.
1394  * uihashtbl_lock must be locked.
1395  * Increase refcount on uidinfo struct returned.
1396  */
1397 static struct uidinfo *
1398 uilookup(uid_t uid)
1399 {
1400 	struct uihashhead *uipp;
1401 	struct uidinfo *uip;
1402 
1403 	rw_assert(&uihashtbl_lock, RA_LOCKED);
1404 	uipp = UIHASH(uid);
1405 	LIST_FOREACH(uip, uipp, ui_hash)
1406 		if (uip->ui_uid == uid) {
1407 			uihold(uip);
1408 			break;
1409 		}
1410 
1411 	return (uip);
1412 }
1413 
1414 /*
1415  * Find or allocate a struct uidinfo for a particular uid.
1416  * Returns with uidinfo struct referenced.
1417  * uifree() should be called on a struct uidinfo when released.
1418  */
1419 struct uidinfo *
1420 uifind(uid_t uid)
1421 {
1422 	struct uidinfo *new_uip, *uip;
1423 	struct ucred *cred;
1424 
1425 	cred = curthread->td_ucred;
1426 	if (cred->cr_uidinfo->ui_uid == uid) {
1427 		uip = cred->cr_uidinfo;
1428 		uihold(uip);
1429 		return (uip);
1430 	} else if (cred->cr_ruidinfo->ui_uid == uid) {
1431 		uip = cred->cr_ruidinfo;
1432 		uihold(uip);
1433 		return (uip);
1434 	}
1435 
1436 	rw_rlock(&uihashtbl_lock);
1437 	uip = uilookup(uid);
1438 	rw_runlock(&uihashtbl_lock);
1439 	if (uip != NULL)
1440 		return (uip);
1441 
1442 	new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1443 	racct_create(&new_uip->ui_racct);
1444 	refcount_init(&new_uip->ui_ref, 1);
1445 	new_uip->ui_uid = uid;
1446 
1447 	rw_wlock(&uihashtbl_lock);
1448 	/*
1449 	 * There's a chance someone created our uidinfo while we
1450 	 * were in malloc and not holding the lock, so we have to
1451 	 * make sure we don't insert a duplicate uidinfo.
1452 	 */
1453 	if ((uip = uilookup(uid)) == NULL) {
1454 		LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1455 		rw_wunlock(&uihashtbl_lock);
1456 		uip = new_uip;
1457 	} else {
1458 		rw_wunlock(&uihashtbl_lock);
1459 		racct_destroy(&new_uip->ui_racct);
1460 		free(new_uip, M_UIDINFO);
1461 	}
1462 	return (uip);
1463 }
1464 
1465 /*
1466  * Place another refcount on a uidinfo struct.
1467  */
1468 void
1469 uihold(struct uidinfo *uip)
1470 {
1471 
1472 	refcount_acquire(&uip->ui_ref);
1473 }
1474 
1475 /*-
1476  * Since uidinfo structs have a long lifetime, we use an
1477  * opportunistic refcounting scheme to avoid locking the lookup hash
1478  * for each release.
1479  *
1480  * If the refcount hits 0, we need to free the structure,
1481  * which means we need to lock the hash.
1482  * Optimal case:
1483  *   After locking the struct and lowering the refcount, if we find
1484  *   that we don't need to free, simply unlock and return.
1485  * Suboptimal case:
1486  *   If refcount lowering results in need to free, bump the count
1487  *   back up, lose the lock and acquire the locks in the proper
1488  *   order to try again.
1489  */
1490 void
1491 uifree(struct uidinfo *uip)
1492 {
1493 
1494 	if (refcount_release_if_not_last(&uip->ui_ref))
1495 		return;
1496 
1497 	rw_wlock(&uihashtbl_lock);
1498 	if (refcount_release(&uip->ui_ref) == 0) {
1499 		rw_wunlock(&uihashtbl_lock);
1500 		return;
1501 	}
1502 
1503 	racct_destroy(&uip->ui_racct);
1504 	LIST_REMOVE(uip, ui_hash);
1505 	rw_wunlock(&uihashtbl_lock);
1506 
1507 	if (uip->ui_sbsize != 0)
1508 		printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1509 		    uip->ui_uid, uip->ui_sbsize);
1510 	if (uip->ui_proccnt != 0)
1511 		printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1512 		    uip->ui_uid, uip->ui_proccnt);
1513 	if (uip->ui_vmsize != 0)
1514 		printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1515 		    uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1516 	free(uip, M_UIDINFO);
1517 }
1518 
1519 #ifdef RACCT
1520 void
1521 ui_racct_foreach(void (*callback)(struct racct *racct,
1522     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1523     void *arg2, void *arg3)
1524 {
1525 	struct uidinfo *uip;
1526 	struct uihashhead *uih;
1527 
1528 	rw_rlock(&uihashtbl_lock);
1529 	if (pre != NULL)
1530 		(pre)();
1531 	for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1532 		LIST_FOREACH(uip, uih, ui_hash) {
1533 			(callback)(uip->ui_racct, arg2, arg3);
1534 		}
1535 	}
1536 	if (post != NULL)
1537 		(post)();
1538 	rw_runlock(&uihashtbl_lock);
1539 }
1540 #endif
1541 
1542 static inline int
1543 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1544 {
1545 	long new;
1546 
1547 	/* Don't allow them to exceed max, but allow subtraction. */
1548 	new = atomic_fetchadd_long(limit, (long)diff) + diff;
1549 	if (diff > 0 && max != 0) {
1550 		if (new < 0 || new > max) {
1551 			atomic_subtract_long(limit, (long)diff);
1552 			return (0);
1553 		}
1554 	} else if (new < 0)
1555 		printf("negative %s for uid = %d\n", name, uip->ui_uid);
1556 	return (1);
1557 }
1558 
1559 /*
1560  * Change the count associated with number of processes
1561  * a given user is using.  When 'max' is 0, don't enforce a limit
1562  */
1563 int
1564 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1565 {
1566 
1567 	return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1568 }
1569 
1570 /*
1571  * Change the total socket buffer size a user has used.
1572  */
1573 int
1574 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1575 {
1576 	int diff, rv;
1577 
1578 	diff = to - *hiwat;
1579 	if (diff > 0 && max == 0) {
1580 		rv = 0;
1581 	} else {
1582 		rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1583 		if (rv != 0)
1584 			*hiwat = to;
1585 	}
1586 	return (rv);
1587 }
1588 
1589 /*
1590  * Change the count associated with number of pseudo-terminals
1591  * a given user is using.  When 'max' is 0, don't enforce a limit
1592  */
1593 int
1594 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1595 {
1596 
1597 	return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1598 }
1599 
1600 int
1601 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1602 {
1603 
1604 	return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1605 }
1606 
1607 int
1608 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1609 {
1610 
1611 	return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));
1612 }
1613