xref: /freebsd/sys/kern/kern_resource.c (revision 089eb4ead8d22d90100a2ad0923794daee31777d)
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 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/sysproto.h>
45 #include <sys/file.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mutex.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/refcount.h>
53 #include <sys/racct.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
56 #include <sys/sched.h>
57 #include <sys/sx.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
61 #include <sys/time.h>
62 #include <sys/umtxvar.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_param.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 
69 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
70 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
71 #define	UIHASH(uid)	(&uihashtbl[(uid) & uihash])
72 static struct rwlock uihashtbl_lock;
73 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
74 static u_long uihash;		/* size of hash table - 1 */
75 
76 static void	calcru1(struct proc *p, struct rusage_ext *ruxp,
77 		    struct timeval *up, struct timeval *sp);
78 static int	donice(struct thread *td, struct proc *chgp, int n);
79 static struct uidinfo *uilookup(uid_t uid);
80 static void	ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td);
81 
82 /*
83  * Resource controls and accounting.
84  */
85 #ifndef _SYS_SYSPROTO_H_
86 struct getpriority_args {
87 	int	which;
88 	int	who;
89 };
90 #endif
91 int
92 sys_getpriority(struct thread *td, struct getpriority_args *uap)
93 {
94 
95 	return (kern_getpriority(td, uap->which, uap->who));
96 }
97 
98 int
99 kern_getpriority(struct thread *td, int which, int who)
100 {
101 	struct proc *p;
102 	struct pgrp *pg;
103 	int error, low;
104 
105 	error = 0;
106 	low = PRIO_MAX + 1;
107 	switch (which) {
108 	case PRIO_PROCESS:
109 		if (who == 0)
110 			low = td->td_proc->p_nice;
111 		else {
112 			p = pfind(who);
113 			if (p == NULL)
114 				break;
115 			if (p_cansee(td, p) == 0)
116 				low = p->p_nice;
117 			PROC_UNLOCK(p);
118 		}
119 		break;
120 
121 	case PRIO_PGRP:
122 		sx_slock(&proctree_lock);
123 		if (who == 0) {
124 			pg = td->td_proc->p_pgrp;
125 			PGRP_LOCK(pg);
126 		} else {
127 			pg = pgfind(who);
128 			if (pg == NULL) {
129 				sx_sunlock(&proctree_lock);
130 				break;
131 			}
132 		}
133 		sx_sunlock(&proctree_lock);
134 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
135 			PROC_LOCK(p);
136 			if (p->p_state == PRS_NORMAL &&
137 			    p_cansee(td, p) == 0) {
138 				if (p->p_nice < low)
139 					low = p->p_nice;
140 			}
141 			PROC_UNLOCK(p);
142 		}
143 		PGRP_UNLOCK(pg);
144 		break;
145 
146 	case PRIO_USER:
147 		if (who == 0)
148 			who = td->td_ucred->cr_uid;
149 		sx_slock(&allproc_lock);
150 		FOREACH_PROC_IN_SYSTEM(p) {
151 			PROC_LOCK(p);
152 			if (p->p_state == PRS_NORMAL &&
153 			    p_cansee(td, p) == 0 &&
154 			    p->p_ucred->cr_uid == who) {
155 				if (p->p_nice < low)
156 					low = p->p_nice;
157 			}
158 			PROC_UNLOCK(p);
159 		}
160 		sx_sunlock(&allproc_lock);
161 		break;
162 
163 	default:
164 		error = EINVAL;
165 		break;
166 	}
167 	if (low == PRIO_MAX + 1 && error == 0)
168 		error = ESRCH;
169 	td->td_retval[0] = low;
170 	return (error);
171 }
172 
173 #ifndef _SYS_SYSPROTO_H_
174 struct setpriority_args {
175 	int	which;
176 	int	who;
177 	int	prio;
178 };
179 #endif
180 int
181 sys_setpriority(struct thread *td, struct setpriority_args *uap)
182 {
183 
184 	return (kern_setpriority(td, uap->which, uap->who, uap->prio));
185 }
186 
187 int
188 kern_setpriority(struct thread *td, int which, int who, int prio)
189 {
190 	struct proc *curp, *p;
191 	struct pgrp *pg;
192 	int found = 0, error = 0;
193 
194 	curp = td->td_proc;
195 	switch (which) {
196 	case PRIO_PROCESS:
197 		if (who == 0) {
198 			PROC_LOCK(curp);
199 			error = donice(td, curp, prio);
200 			PROC_UNLOCK(curp);
201 		} else {
202 			p = pfind(who);
203 			if (p == NULL)
204 				break;
205 			error = p_cansee(td, p);
206 			if (error == 0)
207 				error = donice(td, p, prio);
208 			PROC_UNLOCK(p);
209 		}
210 		found++;
211 		break;
212 
213 	case PRIO_PGRP:
214 		sx_slock(&proctree_lock);
215 		if (who == 0) {
216 			pg = curp->p_pgrp;
217 			PGRP_LOCK(pg);
218 		} else {
219 			pg = pgfind(who);
220 			if (pg == NULL) {
221 				sx_sunlock(&proctree_lock);
222 				break;
223 			}
224 		}
225 		sx_sunlock(&proctree_lock);
226 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
227 			PROC_LOCK(p);
228 			if (p->p_state == PRS_NORMAL &&
229 			    p_cansee(td, p) == 0) {
230 				error = donice(td, p, prio);
231 				found++;
232 			}
233 			PROC_UNLOCK(p);
234 		}
235 		PGRP_UNLOCK(pg);
236 		break;
237 
238 	case PRIO_USER:
239 		if (who == 0)
240 			who = td->td_ucred->cr_uid;
241 		sx_slock(&allproc_lock);
242 		FOREACH_PROC_IN_SYSTEM(p) {
243 			PROC_LOCK(p);
244 			if (p->p_state == PRS_NORMAL &&
245 			    p->p_ucred->cr_uid == who &&
246 			    p_cansee(td, p) == 0) {
247 				error = donice(td, p, prio);
248 				found++;
249 			}
250 			PROC_UNLOCK(p);
251 		}
252 		sx_sunlock(&allproc_lock);
253 		break;
254 
255 	default:
256 		error = EINVAL;
257 		break;
258 	}
259 	if (found == 0 && error == 0)
260 		error = ESRCH;
261 	return (error);
262 }
263 
264 /*
265  * Set "nice" for a (whole) process.
266  */
267 static int
268 donice(struct thread *td, struct proc *p, int n)
269 {
270 	int error;
271 
272 	PROC_LOCK_ASSERT(p, MA_OWNED);
273 	if ((error = p_cansched(td, p)))
274 		return (error);
275 	if (n > PRIO_MAX)
276 		n = PRIO_MAX;
277 	if (n < PRIO_MIN)
278 		n = PRIO_MIN;
279 	if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
280 		return (EACCES);
281 	sched_nice(p, n);
282 	return (0);
283 }
284 
285 static int unprivileged_idprio;
286 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
287     &unprivileged_idprio, 0,
288     "Allow non-root users to set an idle priority (deprecated)");
289 
290 /*
291  * Set realtime priority for LWP.
292  */
293 #ifndef _SYS_SYSPROTO_H_
294 struct rtprio_thread_args {
295 	int		function;
296 	lwpid_t		lwpid;
297 	struct rtprio	*rtp;
298 };
299 #endif
300 int
301 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
302 {
303 	struct proc *p;
304 	struct rtprio rtp;
305 	struct thread *td1;
306 	int cierror, error;
307 
308 	/* Perform copyin before acquiring locks if needed. */
309 	if (uap->function == RTP_SET)
310 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
311 	else
312 		cierror = 0;
313 
314 	if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
315 		p = td->td_proc;
316 		td1 = td;
317 		PROC_LOCK(p);
318 	} else {
319 		td1 = tdfind(uap->lwpid, -1);
320 		if (td1 == NULL)
321 			return (ESRCH);
322 		p = td1->td_proc;
323 	}
324 
325 	switch (uap->function) {
326 	case RTP_LOOKUP:
327 		if ((error = p_cansee(td, p)))
328 			break;
329 		pri_to_rtp(td1, &rtp);
330 		PROC_UNLOCK(p);
331 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
332 	case RTP_SET:
333 		if ((error = p_cansched(td, p)) || (error = cierror))
334 			break;
335 
336 		/* Disallow setting rtprio in most cases if not superuser. */
337 
338 		/*
339 		 * Realtime priority has to be restricted for reasons which
340 		 * should be obvious.  However, for idleprio processes, there is
341 		 * a potential for system deadlock if an idleprio process gains
342 		 * a lock on a resource that other processes need (and the
343 		 * idleprio process can't run due to a CPU-bound normal
344 		 * process).  Fix me!  XXX
345 		 *
346 		 * This problem is not only related to idleprio process.
347 		 * A user level program can obtain a file lock and hold it
348 		 * indefinitely.  Additionally, without idleprio processes it is
349 		 * still conceivable that a program with low priority will never
350 		 * get to run.  In short, allowing this feature might make it
351 		 * easier to lock a resource indefinitely, but it is not the
352 		 * only thing that makes it possible.
353 		 */
354 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
355 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
356 			break;
357 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
358 		    unprivileged_idprio == 0 &&
359 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
360 			break;
361 		error = rtp_to_pri(&rtp, td1);
362 		break;
363 	default:
364 		error = EINVAL;
365 		break;
366 	}
367 	PROC_UNLOCK(p);
368 	return (error);
369 }
370 
371 /*
372  * Set realtime priority.
373  */
374 #ifndef _SYS_SYSPROTO_H_
375 struct rtprio_args {
376 	int		function;
377 	pid_t		pid;
378 	struct rtprio	*rtp;
379 };
380 #endif
381 int
382 sys_rtprio(struct thread *td, struct rtprio_args *uap)
383 {
384 	struct proc *p;
385 	struct thread *tdp;
386 	struct rtprio rtp;
387 	int cierror, error;
388 
389 	/* Perform copyin before acquiring locks if needed. */
390 	if (uap->function == RTP_SET)
391 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
392 	else
393 		cierror = 0;
394 
395 	if (uap->pid == 0) {
396 		p = td->td_proc;
397 		PROC_LOCK(p);
398 	} else {
399 		p = pfind(uap->pid);
400 		if (p == NULL)
401 			return (ESRCH);
402 	}
403 
404 	switch (uap->function) {
405 	case RTP_LOOKUP:
406 		if ((error = p_cansee(td, p)))
407 			break;
408 		/*
409 		 * Return OUR priority if no pid specified,
410 		 * or if one is, report the highest priority
411 		 * in the process.  There isn't much more you can do as
412 		 * there is only room to return a single priority.
413 		 * Note: specifying our own pid is not the same
414 		 * as leaving it zero.
415 		 */
416 		if (uap->pid == 0) {
417 			pri_to_rtp(td, &rtp);
418 		} else {
419 			struct rtprio rtp2;
420 
421 			rtp.type = RTP_PRIO_IDLE;
422 			rtp.prio = RTP_PRIO_MAX;
423 			FOREACH_THREAD_IN_PROC(p, tdp) {
424 				pri_to_rtp(tdp, &rtp2);
425 				if (rtp2.type <  rtp.type ||
426 				    (rtp2.type == rtp.type &&
427 				    rtp2.prio < rtp.prio)) {
428 					rtp.type = rtp2.type;
429 					rtp.prio = rtp2.prio;
430 				}
431 			}
432 		}
433 		PROC_UNLOCK(p);
434 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
435 	case RTP_SET:
436 		if ((error = p_cansched(td, p)) || (error = cierror))
437 			break;
438 
439 		/*
440 		 * Disallow setting rtprio in most cases if not superuser.
441 		 * See the comment in sys_rtprio_thread about idprio
442 		 * threads holding a lock.
443 		 */
444 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
445 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
446 			break;
447 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
448 		    unprivileged_idprio == 0 &&
449 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
450 			break;
451 
452 		/*
453 		 * If we are setting our own priority, set just our
454 		 * thread but if we are doing another process,
455 		 * do all the threads on that process. If we
456 		 * specify our own pid we do the latter.
457 		 */
458 		if (uap->pid == 0) {
459 			error = rtp_to_pri(&rtp, td);
460 		} else {
461 			FOREACH_THREAD_IN_PROC(p, td) {
462 				if ((error = rtp_to_pri(&rtp, td)) != 0)
463 					break;
464 			}
465 		}
466 		break;
467 	default:
468 		error = EINVAL;
469 		break;
470 	}
471 	PROC_UNLOCK(p);
472 	return (error);
473 }
474 
475 int
476 rtp_to_pri(struct rtprio *rtp, struct thread *td)
477 {
478 	u_char  newpri, oldclass, oldpri;
479 
480 	switch (RTP_PRIO_BASE(rtp->type)) {
481 	case RTP_PRIO_REALTIME:
482 		if (rtp->prio > RTP_PRIO_MAX)
483 			return (EINVAL);
484 		newpri = PRI_MIN_REALTIME + rtp->prio;
485 		break;
486 	case RTP_PRIO_NORMAL:
487 		if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
488 			return (EINVAL);
489 		newpri = PRI_MIN_TIMESHARE + rtp->prio;
490 		break;
491 	case RTP_PRIO_IDLE:
492 		if (rtp->prio > RTP_PRIO_MAX)
493 			return (EINVAL);
494 		newpri = PRI_MIN_IDLE + rtp->prio;
495 		break;
496 	default:
497 		return (EINVAL);
498 	}
499 
500 	thread_lock(td);
501 	oldclass = td->td_pri_class;
502 	sched_class(td, rtp->type);	/* XXX fix */
503 	oldpri = td->td_user_pri;
504 	sched_user_prio(td, newpri);
505 	if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
506 	    td->td_pri_class != RTP_PRIO_NORMAL))
507 		sched_prio(td, td->td_user_pri);
508 	if (TD_ON_UPILOCK(td) && oldpri != newpri) {
509 		critical_enter();
510 		thread_unlock(td);
511 		umtx_pi_adjust(td, oldpri);
512 		critical_exit();
513 	} else
514 		thread_unlock(td);
515 	return (0);
516 }
517 
518 void
519 pri_to_rtp(struct thread *td, struct rtprio *rtp)
520 {
521 
522 	thread_lock(td);
523 	switch (PRI_BASE(td->td_pri_class)) {
524 	case PRI_REALTIME:
525 		rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
526 		break;
527 	case PRI_TIMESHARE:
528 		rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
529 		break;
530 	case PRI_IDLE:
531 		rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
532 		break;
533 	default:
534 		break;
535 	}
536 	rtp->type = td->td_pri_class;
537 	thread_unlock(td);
538 }
539 
540 #if defined(COMPAT_43)
541 #ifndef _SYS_SYSPROTO_H_
542 struct osetrlimit_args {
543 	u_int	which;
544 	struct	orlimit *rlp;
545 };
546 #endif
547 int
548 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
549 {
550 	struct orlimit olim;
551 	struct rlimit lim;
552 	int error;
553 
554 	if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
555 		return (error);
556 	lim.rlim_cur = olim.rlim_cur;
557 	lim.rlim_max = olim.rlim_max;
558 	error = kern_setrlimit(td, uap->which, &lim);
559 	return (error);
560 }
561 
562 #ifndef _SYS_SYSPROTO_H_
563 struct ogetrlimit_args {
564 	u_int	which;
565 	struct	orlimit *rlp;
566 };
567 #endif
568 int
569 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
570 {
571 	struct orlimit olim;
572 	struct rlimit rl;
573 	int error;
574 
575 	if (uap->which >= RLIM_NLIMITS)
576 		return (EINVAL);
577 	lim_rlimit(td, uap->which, &rl);
578 
579 	/*
580 	 * XXX would be more correct to convert only RLIM_INFINITY to the
581 	 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
582 	 * values.  Most 64->32 and 32->16 conversions, including not
583 	 * unimportant ones of uids are even more broken than what we
584 	 * do here (they blindly truncate).  We don't do this correctly
585 	 * here since we have little experience with EOVERFLOW yet.
586 	 * Elsewhere, getuid() can't fail...
587 	 */
588 	olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
589 	olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
590 	error = copyout(&olim, uap->rlp, sizeof(olim));
591 	return (error);
592 }
593 #endif /* COMPAT_43 */
594 
595 #ifndef _SYS_SYSPROTO_H_
596 struct setrlimit_args {
597 	u_int	which;
598 	struct	rlimit *rlp;
599 };
600 #endif
601 int
602 sys_setrlimit(struct thread *td, struct setrlimit_args *uap)
603 {
604 	struct rlimit alim;
605 	int error;
606 
607 	if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
608 		return (error);
609 	error = kern_setrlimit(td, uap->which, &alim);
610 	return (error);
611 }
612 
613 static void
614 lim_cb(void *arg)
615 {
616 	struct rlimit rlim;
617 	struct thread *td;
618 	struct proc *p;
619 
620 	p = arg;
621 	PROC_LOCK_ASSERT(p, MA_OWNED);
622 	/*
623 	 * Check if the process exceeds its cpu resource allocation.  If
624 	 * it reaches the max, arrange to kill the process in ast().
625 	 */
626 	if (p->p_cpulimit == RLIM_INFINITY)
627 		return;
628 	PROC_STATLOCK(p);
629 	FOREACH_THREAD_IN_PROC(p, td) {
630 		ruxagg(p, td);
631 	}
632 	PROC_STATUNLOCK(p);
633 	if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
634 		lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
635 		if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
636 			killproc(p, "exceeded maximum CPU limit");
637 		} else {
638 			if (p->p_cpulimit < rlim.rlim_max)
639 				p->p_cpulimit += 5;
640 			kern_psignal(p, SIGXCPU);
641 		}
642 	}
643 	if ((p->p_flag & P_WEXIT) == 0)
644 		callout_reset_sbt(&p->p_limco, SBT_1S, 0,
645 		    lim_cb, p, C_PREL(1));
646 }
647 
648 int
649 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
650 {
651 
652 	return (kern_proc_setrlimit(td, td->td_proc, which, limp));
653 }
654 
655 int
656 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
657     struct rlimit *limp)
658 {
659 	struct plimit *newlim, *oldlim, *oldlim_td;
660 	struct rlimit *alimp;
661 	struct rlimit oldssiz;
662 	int error;
663 
664 	if (which >= RLIM_NLIMITS)
665 		return (EINVAL);
666 
667 	/*
668 	 * Preserve historical bugs by treating negative limits as unsigned.
669 	 */
670 	if (limp->rlim_cur < 0)
671 		limp->rlim_cur = RLIM_INFINITY;
672 	if (limp->rlim_max < 0)
673 		limp->rlim_max = RLIM_INFINITY;
674 
675 	oldssiz.rlim_cur = 0;
676 	newlim = lim_alloc();
677 	PROC_LOCK(p);
678 	oldlim = p->p_limit;
679 	alimp = &oldlim->pl_rlimit[which];
680 	if (limp->rlim_cur > alimp->rlim_max ||
681 	    limp->rlim_max > alimp->rlim_max)
682 		if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
683 			PROC_UNLOCK(p);
684 			lim_free(newlim);
685 			return (error);
686 		}
687 	if (limp->rlim_cur > limp->rlim_max)
688 		limp->rlim_cur = limp->rlim_max;
689 	lim_copy(newlim, oldlim);
690 	alimp = &newlim->pl_rlimit[which];
691 
692 	switch (which) {
693 	case RLIMIT_CPU:
694 		if (limp->rlim_cur != RLIM_INFINITY &&
695 		    p->p_cpulimit == RLIM_INFINITY)
696 			callout_reset_sbt(&p->p_limco, SBT_1S, 0,
697 			    lim_cb, p, C_PREL(1));
698 		p->p_cpulimit = limp->rlim_cur;
699 		break;
700 	case RLIMIT_DATA:
701 		if (limp->rlim_cur > maxdsiz)
702 			limp->rlim_cur = maxdsiz;
703 		if (limp->rlim_max > maxdsiz)
704 			limp->rlim_max = maxdsiz;
705 		break;
706 
707 	case RLIMIT_STACK:
708 		if (limp->rlim_cur > maxssiz)
709 			limp->rlim_cur = maxssiz;
710 		if (limp->rlim_max > maxssiz)
711 			limp->rlim_max = maxssiz;
712 		oldssiz = *alimp;
713 		if (p->p_sysent->sv_fixlimit != NULL)
714 			p->p_sysent->sv_fixlimit(&oldssiz,
715 			    RLIMIT_STACK);
716 		break;
717 
718 	case RLIMIT_NOFILE:
719 		if (limp->rlim_cur > maxfilesperproc)
720 			limp->rlim_cur = maxfilesperproc;
721 		if (limp->rlim_max > maxfilesperproc)
722 			limp->rlim_max = maxfilesperproc;
723 		break;
724 
725 	case RLIMIT_NPROC:
726 		if (limp->rlim_cur > maxprocperuid)
727 			limp->rlim_cur = maxprocperuid;
728 		if (limp->rlim_max > maxprocperuid)
729 			limp->rlim_max = maxprocperuid;
730 		if (limp->rlim_cur < 1)
731 			limp->rlim_cur = 1;
732 		if (limp->rlim_max < 1)
733 			limp->rlim_max = 1;
734 		break;
735 	}
736 	if (p->p_sysent->sv_fixlimit != NULL)
737 		p->p_sysent->sv_fixlimit(limp, which);
738 	*alimp = *limp;
739 	p->p_limit = newlim;
740 	PROC_UPDATE_COW(p);
741 	oldlim_td = NULL;
742 	if (td == curthread && PROC_COW_CHANGECOUNT(td, p) == 1) {
743 		oldlim_td = lim_cowsync();
744 		thread_cow_synced(td);
745 	}
746 	PROC_UNLOCK(p);
747 	if (oldlim_td != NULL) {
748 		MPASS(oldlim_td == oldlim);
749 		lim_freen(oldlim, 2);
750 	} else {
751 		lim_free(oldlim);
752 	}
753 
754 	if (which == RLIMIT_STACK &&
755 	    /*
756 	     * Skip calls from exec_new_vmspace(), done when stack is
757 	     * not mapped yet.
758 	     */
759 	    (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
760 		/*
761 		 * Stack is allocated to the max at exec time with only
762 		 * "rlim_cur" bytes accessible.  If stack limit is going
763 		 * up make more accessible, if going down make inaccessible.
764 		 */
765 		if (limp->rlim_cur != oldssiz.rlim_cur) {
766 			vm_offset_t addr;
767 			vm_size_t size;
768 			vm_prot_t prot;
769 
770 			if (limp->rlim_cur > oldssiz.rlim_cur) {
771 				prot = p->p_sysent->sv_stackprot;
772 				size = limp->rlim_cur - oldssiz.rlim_cur;
773 				addr = round_page(p->p_vmspace->vm_stacktop) -
774 				    limp->rlim_cur;
775 			} else {
776 				prot = VM_PROT_NONE;
777 				size = oldssiz.rlim_cur - limp->rlim_cur;
778 				addr = round_page(p->p_vmspace->vm_stacktop) -
779 				    oldssiz.rlim_cur;
780 			}
781 			addr = trunc_page(addr);
782 			size = round_page(size);
783 			(void)vm_map_protect(&p->p_vmspace->vm_map,
784 			    addr, addr + size, prot, 0,
785 			    VM_MAP_PROTECT_SET_PROT);
786 		}
787 	}
788 
789 	return (0);
790 }
791 
792 #ifndef _SYS_SYSPROTO_H_
793 struct getrlimit_args {
794 	u_int	which;
795 	struct	rlimit *rlp;
796 };
797 #endif
798 /* ARGSUSED */
799 int
800 sys_getrlimit(struct thread *td, struct getrlimit_args *uap)
801 {
802 	struct rlimit rlim;
803 	int error;
804 
805 	if (uap->which >= RLIM_NLIMITS)
806 		return (EINVAL);
807 	lim_rlimit(td, uap->which, &rlim);
808 	error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
809 	return (error);
810 }
811 
812 /*
813  * Transform the running time and tick information for children of proc p
814  * into user and system time usage.
815  */
816 void
817 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
818 {
819 
820 	PROC_LOCK_ASSERT(p, MA_OWNED);
821 	calcru1(p, &p->p_crux, up, sp);
822 }
823 
824 /*
825  * Transform the running time and tick information in proc p into user
826  * and system time usage.  If appropriate, include the current time slice
827  * on this CPU.
828  */
829 void
830 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
831 {
832 	struct thread *td;
833 	uint64_t runtime, u;
834 
835 	PROC_LOCK_ASSERT(p, MA_OWNED);
836 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
837 	/*
838 	 * If we are getting stats for the current process, then add in the
839 	 * stats that this thread has accumulated in its current time slice.
840 	 * We reset the thread and CPU state as if we had performed a context
841 	 * switch right here.
842 	 */
843 	td = curthread;
844 	if (td->td_proc == p) {
845 		u = cpu_ticks();
846 		runtime = u - PCPU_GET(switchtime);
847 		td->td_runtime += runtime;
848 		td->td_incruntime += runtime;
849 		PCPU_SET(switchtime, u);
850 	}
851 	/* Make sure the per-thread stats are current. */
852 	FOREACH_THREAD_IN_PROC(p, td) {
853 		if (td->td_incruntime == 0)
854 			continue;
855 		ruxagg(p, td);
856 	}
857 	calcru1(p, &p->p_rux, up, sp);
858 }
859 
860 /* Collect resource usage for a single thread. */
861 void
862 rufetchtd(struct thread *td, struct rusage *ru)
863 {
864 	struct proc *p;
865 	uint64_t runtime, u;
866 
867 	p = td->td_proc;
868 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
869 	THREAD_LOCK_ASSERT(td, MA_OWNED);
870 	/*
871 	 * If we are getting stats for the current thread, then add in the
872 	 * stats that this thread has accumulated in its current time slice.
873 	 * We reset the thread and CPU state as if we had performed a context
874 	 * switch right here.
875 	 */
876 	if (td == curthread) {
877 		u = cpu_ticks();
878 		runtime = u - PCPU_GET(switchtime);
879 		td->td_runtime += runtime;
880 		td->td_incruntime += runtime;
881 		PCPU_SET(switchtime, u);
882 	}
883 	ruxagg_locked(p, td);
884 	*ru = td->td_ru;
885 	calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
886 }
887 
888 /* XXX: the MI version is too slow to use: */
889 #ifndef __HAVE_INLINE_FLSLL
890 #define	flsll(x)	(fls((x) >> 32) != 0 ? fls((x) >> 32) + 32 : fls(x))
891 #endif
892 
893 static uint64_t
894 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
895 {
896 	uint64_t acc, bh, bl;
897 	int i, s, sa, sb;
898 
899 	/*
900 	 * Calculate (a * b) / c accurately enough without overflowing.  c
901 	 * must be nonzero, and its top bit must be 0.  a or b must be
902 	 * <= c, and the implementation is tuned for b <= c.
903 	 *
904 	 * The comments about times are for use in calcru1() with units of
905 	 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
906 	 *
907 	 * Let n be the number of top zero bits in c.  Each iteration
908 	 * either returns, or reduces b by right shifting it by at least n.
909 	 * The number of iterations is at most 1 + 64 / n, and the error is
910 	 * at most the number of iterations.
911 	 *
912 	 * It is very unusual to need even 2 iterations.  Previous
913 	 * implementations overflowed essentially by returning early in the
914 	 * first iteration, with n = 38 giving overflow at 105+ hours and
915 	 * n = 32 giving overlow at at 388+ days despite a more careful
916 	 * calculation.  388 days is a reasonable uptime, and the calculation
917 	 * needs to work for the uptime times the number of CPUs since 'a'
918 	 * is per-process.
919 	 */
920 	if (a >= (uint64_t)1 << 63)
921 		return (0);		/* Unsupported arg -- can't happen. */
922 	acc = 0;
923 	for (i = 0; i < 128; i++) {
924 		sa = flsll(a);
925 		sb = flsll(b);
926 		if (sa + sb <= 64)
927 			/* Up to 105 hours on first iteration. */
928 			return (acc + (a * b) / c);
929 		if (a >= c) {
930 			/*
931 			 * This reduction is based on a = q * c + r, with the
932 			 * remainder r < c.  'a' may be large to start, and
933 			 * moving bits from b into 'a' at the end of the loop
934 			 * sets the top bit of 'a', so the reduction makes
935 			 * significant progress.
936 			 */
937 			acc += (a / c) * b;
938 			a %= c;
939 			sa = flsll(a);
940 			if (sa + sb <= 64)
941 				/* Up to 388 days on first iteration. */
942 				return (acc + (a * b) / c);
943 		}
944 
945 		/*
946 		 * This step writes a * b as a * ((bh << s) + bl) =
947 		 * a * (bh << s) + a * bl = (a << s) * bh + a * bl.  The 2
948 		 * additive terms are handled separately.  Splitting in
949 		 * this way is linear except for rounding errors.
950 		 *
951 		 * s = 64 - sa is the maximum such that a << s fits in 64
952 		 * bits.  Since a < c and c has at least 1 zero top bit,
953 		 * sa < 64 and s > 0.  Thus this step makes progress by
954 		 * reducing b (it increases 'a', but taking remainders on
955 		 * the next iteration completes the reduction).
956 		 *
957 		 * Finally, the choice for s is just what is needed to keep
958 		 * a * bl from overflowing, so we don't need complications
959 		 * like a recursive call mul64_by_fraction(a, bl, c) to
960 		 * handle the second additive term.
961 		 */
962 		s = 64 - sa;
963 		bh = b >> s;
964 		bl = b - (bh << s);
965 		acc += (a * bl) / c;
966 		a <<= s;
967 		b = bh;
968 	}
969 	return (0);		/* Algorithm failure -- can't happen. */
970 }
971 
972 static void
973 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
974     struct timeval *sp)
975 {
976 	/* {user, system, interrupt, total} {ticks, usec}: */
977 	uint64_t ut, uu, st, su, it, tt, tu;
978 
979 	ut = ruxp->rux_uticks;
980 	st = ruxp->rux_sticks;
981 	it = ruxp->rux_iticks;
982 	tt = ut + st + it;
983 	if (tt == 0) {
984 		/* Avoid divide by zero */
985 		st = 1;
986 		tt = 1;
987 	}
988 	tu = cputick2usec(ruxp->rux_runtime);
989 	if ((int64_t)tu < 0) {
990 		/* XXX: this should be an assert /phk */
991 		printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
992 		    (intmax_t)tu, p->p_pid, p->p_comm);
993 		tu = ruxp->rux_tu;
994 	}
995 
996 	/* Subdivide tu.  Avoid overflow in the multiplications. */
997 	if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
998 		/* Up to 76 hours when stathz is 128. */
999 		uu = (tu * ut) / tt;
1000 		su = (tu * st) / tt;
1001 	} else {
1002 		uu = mul64_by_fraction(tu, ut, tt);
1003 		su = mul64_by_fraction(tu, st, tt);
1004 	}
1005 
1006 	if (tu >= ruxp->rux_tu) {
1007 		/*
1008 		 * The normal case, time increased.
1009 		 * Enforce monotonicity of bucketed numbers.
1010 		 */
1011 		if (uu < ruxp->rux_uu)
1012 			uu = ruxp->rux_uu;
1013 		if (su < ruxp->rux_su)
1014 			su = ruxp->rux_su;
1015 	} else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1016 		/*
1017 		 * When we calibrate the cputicker, it is not uncommon to
1018 		 * see the presumably fixed frequency increase slightly over
1019 		 * time as a result of thermal stabilization and NTP
1020 		 * discipline (of the reference clock).  We therefore ignore
1021 		 * a bit of backwards slop because we  expect to catch up
1022 		 * shortly.  We use a 3 microsecond limit to catch low
1023 		 * counts and a 1% limit for high counts.
1024 		 */
1025 		uu = ruxp->rux_uu;
1026 		su = ruxp->rux_su;
1027 		tu = ruxp->rux_tu;
1028 	} else if (vm_guest == VM_GUEST_NO) {  /* tu < ruxp->rux_tu */
1029 		/*
1030 		 * What happened here was likely that a laptop, which ran at
1031 		 * a reduced clock frequency at boot, kicked into high gear.
1032 		 * The wisdom of spamming this message in that case is
1033 		 * dubious, but it might also be indicative of something
1034 		 * serious, so lets keep it and hope laptops can be made
1035 		 * more truthful about their CPU speed via ACPI.
1036 		 */
1037 		printf("calcru: runtime went backwards from %ju usec "
1038 		    "to %ju usec for pid %d (%s)\n",
1039 		    (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1040 		    p->p_pid, p->p_comm);
1041 	}
1042 
1043 	ruxp->rux_uu = uu;
1044 	ruxp->rux_su = su;
1045 	ruxp->rux_tu = tu;
1046 
1047 	up->tv_sec = uu / 1000000;
1048 	up->tv_usec = uu % 1000000;
1049 	sp->tv_sec = su / 1000000;
1050 	sp->tv_usec = su % 1000000;
1051 }
1052 
1053 #ifndef _SYS_SYSPROTO_H_
1054 struct getrusage_args {
1055 	int	who;
1056 	struct	rusage *rusage;
1057 };
1058 #endif
1059 int
1060 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1061 {
1062 	struct rusage ru;
1063 	int error;
1064 
1065 	error = kern_getrusage(td, uap->who, &ru);
1066 	if (error == 0)
1067 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1068 	return (error);
1069 }
1070 
1071 int
1072 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1073 {
1074 	struct proc *p;
1075 	int error;
1076 
1077 	error = 0;
1078 	p = td->td_proc;
1079 	PROC_LOCK(p);
1080 	switch (who) {
1081 	case RUSAGE_SELF:
1082 		rufetchcalc(p, rup, &rup->ru_utime,
1083 		    &rup->ru_stime);
1084 		break;
1085 
1086 	case RUSAGE_CHILDREN:
1087 		*rup = p->p_stats->p_cru;
1088 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1089 		break;
1090 
1091 	case RUSAGE_THREAD:
1092 		PROC_STATLOCK(p);
1093 		thread_lock(td);
1094 		rufetchtd(td, rup);
1095 		thread_unlock(td);
1096 		PROC_STATUNLOCK(p);
1097 		break;
1098 
1099 	default:
1100 		error = EINVAL;
1101 	}
1102 	PROC_UNLOCK(p);
1103 	return (error);
1104 }
1105 
1106 void
1107 rucollect(struct rusage *ru, struct rusage *ru2)
1108 {
1109 	long *ip, *ip2;
1110 	int i;
1111 
1112 	if (ru->ru_maxrss < ru2->ru_maxrss)
1113 		ru->ru_maxrss = ru2->ru_maxrss;
1114 	ip = &ru->ru_first;
1115 	ip2 = &ru2->ru_first;
1116 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1117 		*ip++ += *ip2++;
1118 }
1119 
1120 void
1121 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1122     struct rusage_ext *rux2)
1123 {
1124 
1125 	rux->rux_runtime += rux2->rux_runtime;
1126 	rux->rux_uticks += rux2->rux_uticks;
1127 	rux->rux_sticks += rux2->rux_sticks;
1128 	rux->rux_iticks += rux2->rux_iticks;
1129 	rux->rux_uu += rux2->rux_uu;
1130 	rux->rux_su += rux2->rux_su;
1131 	rux->rux_tu += rux2->rux_tu;
1132 	rucollect(ru, ru2);
1133 }
1134 
1135 /*
1136  * Aggregate tick counts into the proc's rusage_ext.
1137  */
1138 static void
1139 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1140 {
1141 
1142 	rux->rux_runtime += td->td_incruntime;
1143 	rux->rux_uticks += td->td_uticks;
1144 	rux->rux_sticks += td->td_sticks;
1145 	rux->rux_iticks += td->td_iticks;
1146 }
1147 
1148 void
1149 ruxagg_locked(struct proc *p, struct thread *td)
1150 {
1151 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1152 	PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1153 
1154 	ruxagg_ext_locked(&p->p_rux, td);
1155 	ruxagg_ext_locked(&td->td_rux, td);
1156 	td->td_incruntime = 0;
1157 	td->td_uticks = 0;
1158 	td->td_iticks = 0;
1159 	td->td_sticks = 0;
1160 }
1161 
1162 void
1163 ruxagg(struct proc *p, struct thread *td)
1164 {
1165 
1166 	thread_lock(td);
1167 	ruxagg_locked(p, td);
1168 	thread_unlock(td);
1169 }
1170 
1171 /*
1172  * Update the rusage_ext structure and fetch a valid aggregate rusage
1173  * for proc p if storage for one is supplied.
1174  */
1175 void
1176 rufetch(struct proc *p, struct rusage *ru)
1177 {
1178 	struct thread *td;
1179 
1180 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
1181 
1182 	*ru = p->p_ru;
1183 	if (p->p_numthreads > 0)  {
1184 		FOREACH_THREAD_IN_PROC(p, td) {
1185 			ruxagg(p, td);
1186 			rucollect(ru, &td->td_ru);
1187 		}
1188 	}
1189 }
1190 
1191 /*
1192  * Atomically perform a rufetch and a calcru together.
1193  * Consumers, can safely assume the calcru is executed only once
1194  * rufetch is completed.
1195  */
1196 void
1197 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1198     struct timeval *sp)
1199 {
1200 
1201 	PROC_STATLOCK(p);
1202 	rufetch(p, ru);
1203 	calcru(p, up, sp);
1204 	PROC_STATUNLOCK(p);
1205 }
1206 
1207 /*
1208  * Allocate a new resource limits structure and initialize its
1209  * reference count and mutex pointer.
1210  */
1211 struct plimit *
1212 lim_alloc()
1213 {
1214 	struct plimit *limp;
1215 
1216 	limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1217 	refcount_init(&limp->pl_refcnt, 1);
1218 	return (limp);
1219 }
1220 
1221 struct plimit *
1222 lim_hold(struct plimit *limp)
1223 {
1224 
1225 	refcount_acquire(&limp->pl_refcnt);
1226 	return (limp);
1227 }
1228 
1229 struct plimit *
1230 lim_cowsync(void)
1231 {
1232 	struct thread *td;
1233 	struct proc *p;
1234 	struct plimit *oldlimit;
1235 
1236 	td = curthread;
1237 	p = td->td_proc;
1238 	PROC_LOCK_ASSERT(p, MA_OWNED);
1239 
1240 	if (td->td_limit == p->p_limit)
1241 		return (NULL);
1242 
1243 	oldlimit = td->td_limit;
1244 	td->td_limit = lim_hold(p->p_limit);
1245 
1246 	return (oldlimit);
1247 }
1248 
1249 void
1250 lim_fork(struct proc *p1, struct proc *p2)
1251 {
1252 
1253 	PROC_LOCK_ASSERT(p1, MA_OWNED);
1254 	PROC_LOCK_ASSERT(p2, MA_OWNED);
1255 
1256 	p2->p_limit = lim_hold(p1->p_limit);
1257 	callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1258 	if (p1->p_cpulimit != RLIM_INFINITY)
1259 		callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1260 		    lim_cb, p2, C_PREL(1));
1261 }
1262 
1263 void
1264 lim_free(struct plimit *limp)
1265 {
1266 
1267 	if (refcount_release(&limp->pl_refcnt))
1268 		free((void *)limp, M_PLIMIT);
1269 }
1270 
1271 void
1272 lim_freen(struct plimit *limp, int n)
1273 {
1274 
1275 	if (refcount_releasen(&limp->pl_refcnt, n))
1276 		free((void *)limp, M_PLIMIT);
1277 }
1278 
1279 /*
1280  * Make a copy of the plimit structure.
1281  * We share these structures copy-on-write after fork.
1282  */
1283 void
1284 lim_copy(struct plimit *dst, struct plimit *src)
1285 {
1286 
1287 	KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1288 	bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1289 }
1290 
1291 /*
1292  * Return the hard limit for a particular system resource.  The
1293  * which parameter specifies the index into the rlimit array.
1294  */
1295 rlim_t
1296 lim_max(struct thread *td, int which)
1297 {
1298 	struct rlimit rl;
1299 
1300 	lim_rlimit(td, which, &rl);
1301 	return (rl.rlim_max);
1302 }
1303 
1304 rlim_t
1305 lim_max_proc(struct proc *p, int which)
1306 {
1307 	struct rlimit rl;
1308 
1309 	lim_rlimit_proc(p, which, &rl);
1310 	return (rl.rlim_max);
1311 }
1312 
1313 /*
1314  * Return the current (soft) limit for a particular system resource.
1315  * The which parameter which specifies the index into the rlimit array
1316  */
1317 rlim_t
1318 (lim_cur)(struct thread *td, int which)
1319 {
1320 	struct rlimit rl;
1321 
1322 	lim_rlimit(td, which, &rl);
1323 	return (rl.rlim_cur);
1324 }
1325 
1326 rlim_t
1327 lim_cur_proc(struct proc *p, int which)
1328 {
1329 	struct rlimit rl;
1330 
1331 	lim_rlimit_proc(p, which, &rl);
1332 	return (rl.rlim_cur);
1333 }
1334 
1335 /*
1336  * Return a copy of the entire rlimit structure for the system limit
1337  * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1338  */
1339 void
1340 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1341 {
1342 	struct proc *p = td->td_proc;
1343 
1344 	MPASS(td == curthread);
1345 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1346 	    ("request for invalid resource limit"));
1347 	*rlp = td->td_limit->pl_rlimit[which];
1348 	if (p->p_sysent->sv_fixlimit != NULL)
1349 		p->p_sysent->sv_fixlimit(rlp, which);
1350 }
1351 
1352 void
1353 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1354 {
1355 
1356 	PROC_LOCK_ASSERT(p, MA_OWNED);
1357 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1358 	    ("request for invalid resource limit"));
1359 	*rlp = p->p_limit->pl_rlimit[which];
1360 	if (p->p_sysent->sv_fixlimit != NULL)
1361 		p->p_sysent->sv_fixlimit(rlp, which);
1362 }
1363 
1364 void
1365 uihashinit()
1366 {
1367 
1368 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1369 	rw_init(&uihashtbl_lock, "uidinfo hash");
1370 }
1371 
1372 /*
1373  * Look up a uidinfo struct for the parameter uid.
1374  * uihashtbl_lock must be locked.
1375  * Increase refcount on uidinfo struct returned.
1376  */
1377 static struct uidinfo *
1378 uilookup(uid_t uid)
1379 {
1380 	struct uihashhead *uipp;
1381 	struct uidinfo *uip;
1382 
1383 	rw_assert(&uihashtbl_lock, RA_LOCKED);
1384 	uipp = UIHASH(uid);
1385 	LIST_FOREACH(uip, uipp, ui_hash)
1386 		if (uip->ui_uid == uid) {
1387 			uihold(uip);
1388 			break;
1389 		}
1390 
1391 	return (uip);
1392 }
1393 
1394 /*
1395  * Find or allocate a struct uidinfo for a particular uid.
1396  * Returns with uidinfo struct referenced.
1397  * uifree() should be called on a struct uidinfo when released.
1398  */
1399 struct uidinfo *
1400 uifind(uid_t uid)
1401 {
1402 	struct uidinfo *new_uip, *uip;
1403 	struct ucred *cred;
1404 
1405 	cred = curthread->td_ucred;
1406 	if (cred->cr_uidinfo->ui_uid == uid) {
1407 		uip = cred->cr_uidinfo;
1408 		uihold(uip);
1409 		return (uip);
1410 	} else if (cred->cr_ruidinfo->ui_uid == uid) {
1411 		uip = cred->cr_ruidinfo;
1412 		uihold(uip);
1413 		return (uip);
1414 	}
1415 
1416 	rw_rlock(&uihashtbl_lock);
1417 	uip = uilookup(uid);
1418 	rw_runlock(&uihashtbl_lock);
1419 	if (uip != NULL)
1420 		return (uip);
1421 
1422 	new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1423 	racct_create(&new_uip->ui_racct);
1424 	refcount_init(&new_uip->ui_ref, 1);
1425 	new_uip->ui_uid = uid;
1426 
1427 	rw_wlock(&uihashtbl_lock);
1428 	/*
1429 	 * There's a chance someone created our uidinfo while we
1430 	 * were in malloc and not holding the lock, so we have to
1431 	 * make sure we don't insert a duplicate uidinfo.
1432 	 */
1433 	if ((uip = uilookup(uid)) == NULL) {
1434 		LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1435 		rw_wunlock(&uihashtbl_lock);
1436 		uip = new_uip;
1437 	} else {
1438 		rw_wunlock(&uihashtbl_lock);
1439 		racct_destroy(&new_uip->ui_racct);
1440 		free(new_uip, M_UIDINFO);
1441 	}
1442 	return (uip);
1443 }
1444 
1445 /*
1446  * Place another refcount on a uidinfo struct.
1447  */
1448 void
1449 uihold(struct uidinfo *uip)
1450 {
1451 
1452 	refcount_acquire(&uip->ui_ref);
1453 }
1454 
1455 /*-
1456  * Since uidinfo structs have a long lifetime, we use an
1457  * opportunistic refcounting scheme to avoid locking the lookup hash
1458  * for each release.
1459  *
1460  * If the refcount hits 0, we need to free the structure,
1461  * which means we need to lock the hash.
1462  * Optimal case:
1463  *   After locking the struct and lowering the refcount, if we find
1464  *   that we don't need to free, simply unlock and return.
1465  * Suboptimal case:
1466  *   If refcount lowering results in need to free, bump the count
1467  *   back up, lose the lock and acquire the locks in the proper
1468  *   order to try again.
1469  */
1470 void
1471 uifree(struct uidinfo *uip)
1472 {
1473 
1474 	if (refcount_release_if_not_last(&uip->ui_ref))
1475 		return;
1476 
1477 	rw_wlock(&uihashtbl_lock);
1478 	if (refcount_release(&uip->ui_ref) == 0) {
1479 		rw_wunlock(&uihashtbl_lock);
1480 		return;
1481 	}
1482 
1483 	racct_destroy(&uip->ui_racct);
1484 	LIST_REMOVE(uip, ui_hash);
1485 	rw_wunlock(&uihashtbl_lock);
1486 
1487 	if (uip->ui_sbsize != 0)
1488 		printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1489 		    uip->ui_uid, uip->ui_sbsize);
1490 	if (uip->ui_proccnt != 0)
1491 		printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1492 		    uip->ui_uid, uip->ui_proccnt);
1493 	if (uip->ui_vmsize != 0)
1494 		printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1495 		    uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1496 	free(uip, M_UIDINFO);
1497 }
1498 
1499 #ifdef RACCT
1500 void
1501 ui_racct_foreach(void (*callback)(struct racct *racct,
1502     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1503     void *arg2, void *arg3)
1504 {
1505 	struct uidinfo *uip;
1506 	struct uihashhead *uih;
1507 
1508 	rw_rlock(&uihashtbl_lock);
1509 	if (pre != NULL)
1510 		(pre)();
1511 	for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1512 		LIST_FOREACH(uip, uih, ui_hash) {
1513 			(callback)(uip->ui_racct, arg2, arg3);
1514 		}
1515 	}
1516 	if (post != NULL)
1517 		(post)();
1518 	rw_runlock(&uihashtbl_lock);
1519 }
1520 #endif
1521 
1522 static inline int
1523 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1524 {
1525 	long new;
1526 
1527 	/* Don't allow them to exceed max, but allow subtraction. */
1528 	new = atomic_fetchadd_long(limit, (long)diff) + diff;
1529 	if (diff > 0 && max != 0) {
1530 		if (new < 0 || new > max) {
1531 			atomic_subtract_long(limit, (long)diff);
1532 			return (0);
1533 		}
1534 	} else if (new < 0)
1535 		printf("negative %s for uid = %d\n", name, uip->ui_uid);
1536 	return (1);
1537 }
1538 
1539 /*
1540  * Change the count associated with number of processes
1541  * a given user is using.  When 'max' is 0, don't enforce a limit
1542  */
1543 int
1544 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1545 {
1546 
1547 	return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1548 }
1549 
1550 /*
1551  * Change the total socket buffer size a user has used.
1552  */
1553 int
1554 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1555 {
1556 	int diff, rv;
1557 
1558 	diff = to - *hiwat;
1559 	if (diff > 0 && max == 0) {
1560 		rv = 0;
1561 	} else {
1562 		rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1563 		if (rv != 0)
1564 			*hiwat = to;
1565 	}
1566 	return (rv);
1567 }
1568 
1569 /*
1570  * Change the count associated with number of pseudo-terminals
1571  * a given user is using.  When 'max' is 0, don't enforce a limit
1572  */
1573 int
1574 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1575 {
1576 
1577 	return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1578 }
1579 
1580 int
1581 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1582 {
1583 
1584 	return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1585 }
1586 
1587 int
1588 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1589 {
1590 
1591 	return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));
1592 }
1593