xref: /freebsd/sys/kern/kern_resource.c (revision fe75646a0234a261c0013bf1840fdac4acaf0cec)
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 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/sysproto.h>
40 #include <sys/file.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/refcount.h>
48 #include <sys/racct.h>
49 #include <sys/resourcevar.h>
50 #include <sys/rwlock.h>
51 #include <sys/sched.h>
52 #include <sys/sx.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysent.h>
56 #include <sys/time.h>
57 #include <sys/umtxvar.h>
58 
59 #include <vm/vm.h>
60 #include <vm/vm_param.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_map.h>
63 
64 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
65 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
66 #define	UIHASH(uid)	(&uihashtbl[(uid) & uihash])
67 static struct rwlock uihashtbl_lock;
68 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
69 static u_long uihash;		/* size of hash table - 1 */
70 
71 static void	calcru1(struct proc *p, struct rusage_ext *ruxp,
72 		    struct timeval *up, struct timeval *sp);
73 static int	donice(struct thread *td, struct proc *chgp, int n);
74 static struct uidinfo *uilookup(uid_t uid);
75 static void	ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td);
76 
77 /*
78  * Resource controls and accounting.
79  */
80 #ifndef _SYS_SYSPROTO_H_
81 struct getpriority_args {
82 	int	which;
83 	int	who;
84 };
85 #endif
86 int
87 sys_getpriority(struct thread *td, struct getpriority_args *uap)
88 {
89 
90 	return (kern_getpriority(td, uap->which, uap->who));
91 }
92 
93 int
94 kern_getpriority(struct thread *td, int which, int who)
95 {
96 	struct proc *p;
97 	struct pgrp *pg;
98 	int error, low;
99 
100 	error = 0;
101 	low = PRIO_MAX + 1;
102 	switch (which) {
103 	case PRIO_PROCESS:
104 		if (who == 0)
105 			low = td->td_proc->p_nice;
106 		else {
107 			p = pfind(who);
108 			if (p == NULL)
109 				break;
110 			if (p_cansee(td, p) == 0)
111 				low = p->p_nice;
112 			PROC_UNLOCK(p);
113 		}
114 		break;
115 
116 	case PRIO_PGRP:
117 		sx_slock(&proctree_lock);
118 		if (who == 0) {
119 			pg = td->td_proc->p_pgrp;
120 			PGRP_LOCK(pg);
121 		} else {
122 			pg = pgfind(who);
123 			if (pg == NULL) {
124 				sx_sunlock(&proctree_lock);
125 				break;
126 			}
127 		}
128 		sx_sunlock(&proctree_lock);
129 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
130 			PROC_LOCK(p);
131 			if (p->p_state == PRS_NORMAL &&
132 			    p_cansee(td, p) == 0) {
133 				if (p->p_nice < low)
134 					low = p->p_nice;
135 			}
136 			PROC_UNLOCK(p);
137 		}
138 		PGRP_UNLOCK(pg);
139 		break;
140 
141 	case PRIO_USER:
142 		if (who == 0)
143 			who = td->td_ucred->cr_uid;
144 		sx_slock(&allproc_lock);
145 		FOREACH_PROC_IN_SYSTEM(p) {
146 			PROC_LOCK(p);
147 			if (p->p_state == PRS_NORMAL &&
148 			    p_cansee(td, p) == 0 &&
149 			    p->p_ucred->cr_uid == who) {
150 				if (p->p_nice < low)
151 					low = p->p_nice;
152 			}
153 			PROC_UNLOCK(p);
154 		}
155 		sx_sunlock(&allproc_lock);
156 		break;
157 
158 	default:
159 		error = EINVAL;
160 		break;
161 	}
162 	if (low == PRIO_MAX + 1 && error == 0)
163 		error = ESRCH;
164 	td->td_retval[0] = low;
165 	return (error);
166 }
167 
168 #ifndef _SYS_SYSPROTO_H_
169 struct setpriority_args {
170 	int	which;
171 	int	who;
172 	int	prio;
173 };
174 #endif
175 int
176 sys_setpriority(struct thread *td, struct setpriority_args *uap)
177 {
178 
179 	return (kern_setpriority(td, uap->which, uap->who, uap->prio));
180 }
181 
182 int
183 kern_setpriority(struct thread *td, int which, int who, int prio)
184 {
185 	struct proc *curp, *p;
186 	struct pgrp *pg;
187 	int found = 0, error = 0;
188 
189 	curp = td->td_proc;
190 	switch (which) {
191 	case PRIO_PROCESS:
192 		if (who == 0) {
193 			PROC_LOCK(curp);
194 			error = donice(td, curp, prio);
195 			PROC_UNLOCK(curp);
196 		} else {
197 			p = pfind(who);
198 			if (p == NULL)
199 				break;
200 			error = p_cansee(td, p);
201 			if (error == 0)
202 				error = donice(td, p, prio);
203 			PROC_UNLOCK(p);
204 		}
205 		found++;
206 		break;
207 
208 	case PRIO_PGRP:
209 		sx_slock(&proctree_lock);
210 		if (who == 0) {
211 			pg = curp->p_pgrp;
212 			PGRP_LOCK(pg);
213 		} else {
214 			pg = pgfind(who);
215 			if (pg == NULL) {
216 				sx_sunlock(&proctree_lock);
217 				break;
218 			}
219 		}
220 		sx_sunlock(&proctree_lock);
221 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
222 			PROC_LOCK(p);
223 			if (p->p_state == PRS_NORMAL &&
224 			    p_cansee(td, p) == 0) {
225 				error = donice(td, p, prio);
226 				found++;
227 			}
228 			PROC_UNLOCK(p);
229 		}
230 		PGRP_UNLOCK(pg);
231 		break;
232 
233 	case PRIO_USER:
234 		if (who == 0)
235 			who = td->td_ucred->cr_uid;
236 		sx_slock(&allproc_lock);
237 		FOREACH_PROC_IN_SYSTEM(p) {
238 			PROC_LOCK(p);
239 			if (p->p_state == PRS_NORMAL &&
240 			    p->p_ucred->cr_uid == who &&
241 			    p_cansee(td, p) == 0) {
242 				error = donice(td, p, prio);
243 				found++;
244 			}
245 			PROC_UNLOCK(p);
246 		}
247 		sx_sunlock(&allproc_lock);
248 		break;
249 
250 	default:
251 		error = EINVAL;
252 		break;
253 	}
254 	if (found == 0 && error == 0)
255 		error = ESRCH;
256 	return (error);
257 }
258 
259 /*
260  * Set "nice" for a (whole) process.
261  */
262 static int
263 donice(struct thread *td, struct proc *p, int n)
264 {
265 	int error;
266 
267 	PROC_LOCK_ASSERT(p, MA_OWNED);
268 	if ((error = p_cansched(td, p)))
269 		return (error);
270 	if (n > PRIO_MAX)
271 		n = PRIO_MAX;
272 	if (n < PRIO_MIN)
273 		n = PRIO_MIN;
274 	if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
275 		return (EACCES);
276 	sched_nice(p, n);
277 	return (0);
278 }
279 
280 static int unprivileged_idprio;
281 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
282     &unprivileged_idprio, 0,
283     "Allow non-root users to set an idle priority (deprecated)");
284 
285 /*
286  * Set realtime priority for LWP.
287  */
288 #ifndef _SYS_SYSPROTO_H_
289 struct rtprio_thread_args {
290 	int		function;
291 	lwpid_t		lwpid;
292 	struct rtprio	*rtp;
293 };
294 #endif
295 int
296 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
297 {
298 	struct proc *p;
299 	struct rtprio rtp;
300 	struct thread *td1;
301 	int cierror, error;
302 
303 	/* Perform copyin before acquiring locks if needed. */
304 	if (uap->function == RTP_SET)
305 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
306 	else
307 		cierror = 0;
308 
309 	if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
310 		p = td->td_proc;
311 		td1 = td;
312 		PROC_LOCK(p);
313 	} else {
314 		td1 = tdfind(uap->lwpid, -1);
315 		if (td1 == NULL)
316 			return (ESRCH);
317 		p = td1->td_proc;
318 	}
319 
320 	switch (uap->function) {
321 	case RTP_LOOKUP:
322 		if ((error = p_cansee(td, p)))
323 			break;
324 		pri_to_rtp(td1, &rtp);
325 		PROC_UNLOCK(p);
326 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
327 	case RTP_SET:
328 		if ((error = p_cansched(td, p)) || (error = cierror))
329 			break;
330 
331 		/* Disallow setting rtprio in most cases if not superuser. */
332 
333 		/*
334 		 * Realtime priority has to be restricted for reasons which
335 		 * should be obvious.  However, for idleprio processes, there is
336 		 * a potential for system deadlock if an idleprio process gains
337 		 * a lock on a resource that other processes need (and the
338 		 * idleprio process can't run due to a CPU-bound normal
339 		 * process).  Fix me!  XXX
340 		 *
341 		 * This problem is not only related to idleprio process.
342 		 * A user level program can obtain a file lock and hold it
343 		 * indefinitely.  Additionally, without idleprio processes it is
344 		 * still conceivable that a program with low priority will never
345 		 * get to run.  In short, allowing this feature might make it
346 		 * easier to lock a resource indefinitely, but it is not the
347 		 * only thing that makes it possible.
348 		 */
349 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
350 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
351 			break;
352 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
353 		    unprivileged_idprio == 0 &&
354 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
355 			break;
356 		error = rtp_to_pri(&rtp, td1);
357 		break;
358 	default:
359 		error = EINVAL;
360 		break;
361 	}
362 	PROC_UNLOCK(p);
363 	return (error);
364 }
365 
366 /*
367  * Set realtime priority.
368  */
369 #ifndef _SYS_SYSPROTO_H_
370 struct rtprio_args {
371 	int		function;
372 	pid_t		pid;
373 	struct rtprio	*rtp;
374 };
375 #endif
376 int
377 sys_rtprio(struct thread *td, struct rtprio_args *uap)
378 {
379 	struct proc *p;
380 	struct thread *tdp;
381 	struct rtprio rtp;
382 	int cierror, error;
383 
384 	/* Perform copyin before acquiring locks if needed. */
385 	if (uap->function == RTP_SET)
386 		cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
387 	else
388 		cierror = 0;
389 
390 	if (uap->pid == 0) {
391 		p = td->td_proc;
392 		PROC_LOCK(p);
393 	} else {
394 		p = pfind(uap->pid);
395 		if (p == NULL)
396 			return (ESRCH);
397 	}
398 
399 	switch (uap->function) {
400 	case RTP_LOOKUP:
401 		if ((error = p_cansee(td, p)))
402 			break;
403 		/*
404 		 * Return OUR priority if no pid specified,
405 		 * or if one is, report the highest priority
406 		 * in the process.  There isn't much more you can do as
407 		 * there is only room to return a single priority.
408 		 * Note: specifying our own pid is not the same
409 		 * as leaving it zero.
410 		 */
411 		if (uap->pid == 0) {
412 			pri_to_rtp(td, &rtp);
413 		} else {
414 			struct rtprio rtp2;
415 
416 			rtp.type = RTP_PRIO_IDLE;
417 			rtp.prio = RTP_PRIO_MAX;
418 			FOREACH_THREAD_IN_PROC(p, tdp) {
419 				pri_to_rtp(tdp, &rtp2);
420 				if (rtp2.type <  rtp.type ||
421 				    (rtp2.type == rtp.type &&
422 				    rtp2.prio < rtp.prio)) {
423 					rtp.type = rtp2.type;
424 					rtp.prio = rtp2.prio;
425 				}
426 			}
427 		}
428 		PROC_UNLOCK(p);
429 		return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
430 	case RTP_SET:
431 		if ((error = p_cansched(td, p)) || (error = cierror))
432 			break;
433 
434 		/*
435 		 * Disallow setting rtprio in most cases if not superuser.
436 		 * See the comment in sys_rtprio_thread about idprio
437 		 * threads holding a lock.
438 		 */
439 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
440 		    (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
441 			break;
442 		if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
443 		    unprivileged_idprio == 0 &&
444 		    (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
445 			break;
446 
447 		/*
448 		 * If we are setting our own priority, set just our
449 		 * thread but if we are doing another process,
450 		 * do all the threads on that process. If we
451 		 * specify our own pid we do the latter.
452 		 */
453 		if (uap->pid == 0) {
454 			error = rtp_to_pri(&rtp, td);
455 		} else {
456 			FOREACH_THREAD_IN_PROC(p, td) {
457 				if ((error = rtp_to_pri(&rtp, td)) != 0)
458 					break;
459 			}
460 		}
461 		break;
462 	default:
463 		error = EINVAL;
464 		break;
465 	}
466 	PROC_UNLOCK(p);
467 	return (error);
468 }
469 
470 int
471 rtp_to_pri(struct rtprio *rtp, struct thread *td)
472 {
473 	u_char  newpri, oldclass, oldpri;
474 
475 	switch (RTP_PRIO_BASE(rtp->type)) {
476 	case RTP_PRIO_REALTIME:
477 		if (rtp->prio > RTP_PRIO_MAX)
478 			return (EINVAL);
479 		newpri = PRI_MIN_REALTIME + rtp->prio;
480 		break;
481 	case RTP_PRIO_NORMAL:
482 		if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
483 			return (EINVAL);
484 		newpri = PRI_MIN_TIMESHARE + rtp->prio;
485 		break;
486 	case RTP_PRIO_IDLE:
487 		if (rtp->prio > RTP_PRIO_MAX)
488 			return (EINVAL);
489 		newpri = PRI_MIN_IDLE + rtp->prio;
490 		break;
491 	default:
492 		return (EINVAL);
493 	}
494 
495 	thread_lock(td);
496 	oldclass = td->td_pri_class;
497 	sched_class(td, rtp->type);	/* XXX fix */
498 	oldpri = td->td_user_pri;
499 	sched_user_prio(td, newpri);
500 	if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
501 	    td->td_pri_class != RTP_PRIO_NORMAL))
502 		sched_prio(td, td->td_user_pri);
503 	if (TD_ON_UPILOCK(td) && oldpri != newpri) {
504 		critical_enter();
505 		thread_unlock(td);
506 		umtx_pi_adjust(td, oldpri);
507 		critical_exit();
508 	} else
509 		thread_unlock(td);
510 	return (0);
511 }
512 
513 void
514 pri_to_rtp(struct thread *td, struct rtprio *rtp)
515 {
516 
517 	thread_lock(td);
518 	switch (PRI_BASE(td->td_pri_class)) {
519 	case PRI_REALTIME:
520 		rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
521 		break;
522 	case PRI_TIMESHARE:
523 		rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
524 		break;
525 	case PRI_IDLE:
526 		rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
527 		break;
528 	default:
529 		break;
530 	}
531 	rtp->type = td->td_pri_class;
532 	thread_unlock(td);
533 }
534 
535 #if defined(COMPAT_43)
536 #ifndef _SYS_SYSPROTO_H_
537 struct osetrlimit_args {
538 	u_int	which;
539 	struct	orlimit *rlp;
540 };
541 #endif
542 int
543 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
544 {
545 	struct orlimit olim;
546 	struct rlimit lim;
547 	int error;
548 
549 	if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
550 		return (error);
551 	lim.rlim_cur = olim.rlim_cur;
552 	lim.rlim_max = olim.rlim_max;
553 	error = kern_setrlimit(td, uap->which, &lim);
554 	return (error);
555 }
556 
557 #ifndef _SYS_SYSPROTO_H_
558 struct ogetrlimit_args {
559 	u_int	which;
560 	struct	orlimit *rlp;
561 };
562 #endif
563 int
564 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
565 {
566 	struct orlimit olim;
567 	struct rlimit rl;
568 	int error;
569 
570 	if (uap->which >= RLIM_NLIMITS)
571 		return (EINVAL);
572 	lim_rlimit(td, uap->which, &rl);
573 
574 	/*
575 	 * XXX would be more correct to convert only RLIM_INFINITY to the
576 	 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
577 	 * values.  Most 64->32 and 32->16 conversions, including not
578 	 * unimportant ones of uids are even more broken than what we
579 	 * do here (they blindly truncate).  We don't do this correctly
580 	 * here since we have little experience with EOVERFLOW yet.
581 	 * Elsewhere, getuid() can't fail...
582 	 */
583 	olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
584 	olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
585 	error = copyout(&olim, uap->rlp, sizeof(olim));
586 	return (error);
587 }
588 #endif /* COMPAT_43 */
589 
590 #ifndef _SYS_SYSPROTO_H_
591 struct setrlimit_args {
592 	u_int	which;
593 	struct	rlimit *rlp;
594 };
595 #endif
596 int
597 sys_setrlimit(struct thread *td, struct setrlimit_args *uap)
598 {
599 	struct rlimit alim;
600 	int error;
601 
602 	if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
603 		return (error);
604 	error = kern_setrlimit(td, uap->which, &alim);
605 	return (error);
606 }
607 
608 static void
609 lim_cb(void *arg)
610 {
611 	struct rlimit rlim;
612 	struct thread *td;
613 	struct proc *p;
614 
615 	p = arg;
616 	PROC_LOCK_ASSERT(p, MA_OWNED);
617 	/*
618 	 * Check if the process exceeds its cpu resource allocation.  If
619 	 * it reaches the max, arrange to kill the process in ast().
620 	 */
621 	if (p->p_cpulimit == RLIM_INFINITY)
622 		return;
623 	PROC_STATLOCK(p);
624 	FOREACH_THREAD_IN_PROC(p, td) {
625 		ruxagg(p, td);
626 	}
627 	PROC_STATUNLOCK(p);
628 	if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
629 		lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
630 		if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
631 			killproc(p, "exceeded maximum CPU limit");
632 		} else {
633 			if (p->p_cpulimit < rlim.rlim_max)
634 				p->p_cpulimit += 5;
635 			kern_psignal(p, SIGXCPU);
636 		}
637 	}
638 	if ((p->p_flag & P_WEXIT) == 0)
639 		callout_reset_sbt(&p->p_limco, SBT_1S, 0,
640 		    lim_cb, p, C_PREL(1));
641 }
642 
643 int
644 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
645 {
646 
647 	return (kern_proc_setrlimit(td, td->td_proc, which, limp));
648 }
649 
650 int
651 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
652     struct rlimit *limp)
653 {
654 	struct plimit *newlim, *oldlim, *oldlim_td;
655 	struct rlimit *alimp;
656 	struct rlimit oldssiz;
657 	int error;
658 
659 	if (which >= RLIM_NLIMITS)
660 		return (EINVAL);
661 
662 	/*
663 	 * Preserve historical bugs by treating negative limits as unsigned.
664 	 */
665 	if (limp->rlim_cur < 0)
666 		limp->rlim_cur = RLIM_INFINITY;
667 	if (limp->rlim_max < 0)
668 		limp->rlim_max = RLIM_INFINITY;
669 
670 	oldssiz.rlim_cur = 0;
671 	newlim = lim_alloc();
672 	PROC_LOCK(p);
673 	oldlim = p->p_limit;
674 	alimp = &oldlim->pl_rlimit[which];
675 	if (limp->rlim_cur > alimp->rlim_max ||
676 	    limp->rlim_max > alimp->rlim_max)
677 		if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
678 			PROC_UNLOCK(p);
679 			lim_free(newlim);
680 			return (error);
681 		}
682 	if (limp->rlim_cur > limp->rlim_max)
683 		limp->rlim_cur = limp->rlim_max;
684 	lim_copy(newlim, oldlim);
685 	alimp = &newlim->pl_rlimit[which];
686 
687 	switch (which) {
688 	case RLIMIT_CPU:
689 		if (limp->rlim_cur != RLIM_INFINITY &&
690 		    p->p_cpulimit == RLIM_INFINITY)
691 			callout_reset_sbt(&p->p_limco, SBT_1S, 0,
692 			    lim_cb, p, C_PREL(1));
693 		p->p_cpulimit = limp->rlim_cur;
694 		break;
695 	case RLIMIT_DATA:
696 		if (limp->rlim_cur > maxdsiz)
697 			limp->rlim_cur = maxdsiz;
698 		if (limp->rlim_max > maxdsiz)
699 			limp->rlim_max = maxdsiz;
700 		break;
701 
702 	case RLIMIT_STACK:
703 		if (limp->rlim_cur > maxssiz)
704 			limp->rlim_cur = maxssiz;
705 		if (limp->rlim_max > maxssiz)
706 			limp->rlim_max = maxssiz;
707 		oldssiz = *alimp;
708 		if (p->p_sysent->sv_fixlimit != NULL)
709 			p->p_sysent->sv_fixlimit(&oldssiz,
710 			    RLIMIT_STACK);
711 		break;
712 
713 	case RLIMIT_NOFILE:
714 		if (limp->rlim_cur > maxfilesperproc)
715 			limp->rlim_cur = maxfilesperproc;
716 		if (limp->rlim_max > maxfilesperproc)
717 			limp->rlim_max = maxfilesperproc;
718 		break;
719 
720 	case RLIMIT_NPROC:
721 		if (limp->rlim_cur > maxprocperuid)
722 			limp->rlim_cur = maxprocperuid;
723 		if (limp->rlim_max > maxprocperuid)
724 			limp->rlim_max = maxprocperuid;
725 		if (limp->rlim_cur < 1)
726 			limp->rlim_cur = 1;
727 		if (limp->rlim_max < 1)
728 			limp->rlim_max = 1;
729 		break;
730 	}
731 	if (p->p_sysent->sv_fixlimit != NULL)
732 		p->p_sysent->sv_fixlimit(limp, which);
733 	*alimp = *limp;
734 	p->p_limit = newlim;
735 	PROC_UPDATE_COW(p);
736 	oldlim_td = NULL;
737 	if (td == curthread && PROC_COW_CHANGECOUNT(td, p) == 1) {
738 		oldlim_td = lim_cowsync();
739 		thread_cow_synced(td);
740 	}
741 	PROC_UNLOCK(p);
742 	if (oldlim_td != NULL) {
743 		MPASS(oldlim_td == oldlim);
744 		lim_freen(oldlim, 2);
745 	} else {
746 		lim_free(oldlim);
747 	}
748 
749 	if (which == RLIMIT_STACK &&
750 	    /*
751 	     * Skip calls from exec_new_vmspace(), done when stack is
752 	     * not mapped yet.
753 	     */
754 	    (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
755 		/*
756 		 * Stack is allocated to the max at exec time with only
757 		 * "rlim_cur" bytes accessible.  If stack limit is going
758 		 * up make more accessible, if going down make inaccessible.
759 		 */
760 		if (limp->rlim_cur != oldssiz.rlim_cur) {
761 			vm_offset_t addr;
762 			vm_size_t size;
763 			vm_prot_t prot;
764 
765 			if (limp->rlim_cur > oldssiz.rlim_cur) {
766 				prot = p->p_sysent->sv_stackprot;
767 				size = limp->rlim_cur - oldssiz.rlim_cur;
768 				addr = round_page(p->p_vmspace->vm_stacktop) -
769 				    limp->rlim_cur;
770 			} else {
771 				prot = VM_PROT_NONE;
772 				size = oldssiz.rlim_cur - limp->rlim_cur;
773 				addr = round_page(p->p_vmspace->vm_stacktop) -
774 				    oldssiz.rlim_cur;
775 			}
776 			addr = trunc_page(addr);
777 			size = round_page(size);
778 			(void)vm_map_protect(&p->p_vmspace->vm_map,
779 			    addr, addr + size, prot, 0,
780 			    VM_MAP_PROTECT_SET_PROT);
781 		}
782 	}
783 
784 	return (0);
785 }
786 
787 #ifndef _SYS_SYSPROTO_H_
788 struct getrlimit_args {
789 	u_int	which;
790 	struct	rlimit *rlp;
791 };
792 #endif
793 /* ARGSUSED */
794 int
795 sys_getrlimit(struct thread *td, struct getrlimit_args *uap)
796 {
797 	struct rlimit rlim;
798 	int error;
799 
800 	if (uap->which >= RLIM_NLIMITS)
801 		return (EINVAL);
802 	lim_rlimit(td, uap->which, &rlim);
803 	error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
804 	return (error);
805 }
806 
807 /*
808  * Transform the running time and tick information for children of proc p
809  * into user and system time usage.
810  */
811 void
812 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
813 {
814 
815 	PROC_LOCK_ASSERT(p, MA_OWNED);
816 	calcru1(p, &p->p_crux, up, sp);
817 }
818 
819 /*
820  * Transform the running time and tick information in proc p into user
821  * and system time usage.  If appropriate, include the current time slice
822  * on this CPU.
823  */
824 void
825 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
826 {
827 	struct thread *td;
828 	uint64_t runtime, u;
829 
830 	PROC_LOCK_ASSERT(p, MA_OWNED);
831 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
832 	/*
833 	 * If we are getting stats for the current process, then add in the
834 	 * stats that this thread has accumulated in its current time slice.
835 	 * We reset the thread and CPU state as if we had performed a context
836 	 * switch right here.
837 	 */
838 	td = curthread;
839 	if (td->td_proc == p) {
840 		u = cpu_ticks();
841 		runtime = u - PCPU_GET(switchtime);
842 		td->td_runtime += runtime;
843 		td->td_incruntime += runtime;
844 		PCPU_SET(switchtime, u);
845 	}
846 	/* Make sure the per-thread stats are current. */
847 	FOREACH_THREAD_IN_PROC(p, td) {
848 		if (td->td_incruntime == 0)
849 			continue;
850 		ruxagg(p, td);
851 	}
852 	calcru1(p, &p->p_rux, up, sp);
853 }
854 
855 /* Collect resource usage for a single thread. */
856 void
857 rufetchtd(struct thread *td, struct rusage *ru)
858 {
859 	struct proc *p;
860 	uint64_t runtime, u;
861 
862 	p = td->td_proc;
863 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
864 	THREAD_LOCK_ASSERT(td, MA_OWNED);
865 	/*
866 	 * If we are getting stats for the current thread, then add in the
867 	 * stats that this thread has accumulated in its current time slice.
868 	 * We reset the thread and CPU state as if we had performed a context
869 	 * switch right here.
870 	 */
871 	if (td == curthread) {
872 		u = cpu_ticks();
873 		runtime = u - PCPU_GET(switchtime);
874 		td->td_runtime += runtime;
875 		td->td_incruntime += runtime;
876 		PCPU_SET(switchtime, u);
877 	}
878 	ruxagg_locked(p, td);
879 	*ru = td->td_ru;
880 	calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
881 }
882 
883 static uint64_t
884 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
885 {
886 	uint64_t acc, bh, bl;
887 	int i, s, sa, sb;
888 
889 	/*
890 	 * Calculate (a * b) / c accurately enough without overflowing.  c
891 	 * must be nonzero, and its top bit must be 0.  a or b must be
892 	 * <= c, and the implementation is tuned for b <= c.
893 	 *
894 	 * The comments about times are for use in calcru1() with units of
895 	 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
896 	 *
897 	 * Let n be the number of top zero bits in c.  Each iteration
898 	 * either returns, or reduces b by right shifting it by at least n.
899 	 * The number of iterations is at most 1 + 64 / n, and the error is
900 	 * at most the number of iterations.
901 	 *
902 	 * It is very unusual to need even 2 iterations.  Previous
903 	 * implementations overflowed essentially by returning early in the
904 	 * first iteration, with n = 38 giving overflow at 105+ hours and
905 	 * n = 32 giving overlow at at 388+ days despite a more careful
906 	 * calculation.  388 days is a reasonable uptime, and the calculation
907 	 * needs to work for the uptime times the number of CPUs since 'a'
908 	 * is per-process.
909 	 */
910 	if (a >= (uint64_t)1 << 63)
911 		return (0);		/* Unsupported arg -- can't happen. */
912 	acc = 0;
913 	for (i = 0; i < 128; i++) {
914 		sa = flsll(a);
915 		sb = flsll(b);
916 		if (sa + sb <= 64)
917 			/* Up to 105 hours on first iteration. */
918 			return (acc + (a * b) / c);
919 		if (a >= c) {
920 			/*
921 			 * This reduction is based on a = q * c + r, with the
922 			 * remainder r < c.  'a' may be large to start, and
923 			 * moving bits from b into 'a' at the end of the loop
924 			 * sets the top bit of 'a', so the reduction makes
925 			 * significant progress.
926 			 */
927 			acc += (a / c) * b;
928 			a %= c;
929 			sa = flsll(a);
930 			if (sa + sb <= 64)
931 				/* Up to 388 days on first iteration. */
932 				return (acc + (a * b) / c);
933 		}
934 
935 		/*
936 		 * This step writes a * b as a * ((bh << s) + bl) =
937 		 * a * (bh << s) + a * bl = (a << s) * bh + a * bl.  The 2
938 		 * additive terms are handled separately.  Splitting in
939 		 * this way is linear except for rounding errors.
940 		 *
941 		 * s = 64 - sa is the maximum such that a << s fits in 64
942 		 * bits.  Since a < c and c has at least 1 zero top bit,
943 		 * sa < 64 and s > 0.  Thus this step makes progress by
944 		 * reducing b (it increases 'a', but taking remainders on
945 		 * the next iteration completes the reduction).
946 		 *
947 		 * Finally, the choice for s is just what is needed to keep
948 		 * a * bl from overflowing, so we don't need complications
949 		 * like a recursive call mul64_by_fraction(a, bl, c) to
950 		 * handle the second additive term.
951 		 */
952 		s = 64 - sa;
953 		bh = b >> s;
954 		bl = b - (bh << s);
955 		acc += (a * bl) / c;
956 		a <<= s;
957 		b = bh;
958 	}
959 	return (0);		/* Algorithm failure -- can't happen. */
960 }
961 
962 static void
963 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
964     struct timeval *sp)
965 {
966 	/* {user, system, interrupt, total} {ticks, usec}: */
967 	uint64_t ut, uu, st, su, it, tt, tu;
968 
969 	ut = ruxp->rux_uticks;
970 	st = ruxp->rux_sticks;
971 	it = ruxp->rux_iticks;
972 	tt = ut + st + it;
973 	if (tt == 0) {
974 		/* Avoid divide by zero */
975 		st = 1;
976 		tt = 1;
977 	}
978 	tu = cputick2usec(ruxp->rux_runtime);
979 	if ((int64_t)tu < 0) {
980 		/* XXX: this should be an assert /phk */
981 		printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
982 		    (intmax_t)tu, p->p_pid, p->p_comm);
983 		tu = ruxp->rux_tu;
984 	}
985 
986 	/* Subdivide tu.  Avoid overflow in the multiplications. */
987 	if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
988 		/* Up to 76 hours when stathz is 128. */
989 		uu = (tu * ut) / tt;
990 		su = (tu * st) / tt;
991 	} else {
992 		uu = mul64_by_fraction(tu, ut, tt);
993 		su = mul64_by_fraction(tu, st, tt);
994 	}
995 
996 	if (tu >= ruxp->rux_tu) {
997 		/*
998 		 * The normal case, time increased.
999 		 * Enforce monotonicity of bucketed numbers.
1000 		 */
1001 		if (uu < ruxp->rux_uu)
1002 			uu = ruxp->rux_uu;
1003 		if (su < ruxp->rux_su)
1004 			su = ruxp->rux_su;
1005 	} else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1006 		/*
1007 		 * When we calibrate the cputicker, it is not uncommon to
1008 		 * see the presumably fixed frequency increase slightly over
1009 		 * time as a result of thermal stabilization and NTP
1010 		 * discipline (of the reference clock).  We therefore ignore
1011 		 * a bit of backwards slop because we  expect to catch up
1012 		 * shortly.  We use a 3 microsecond limit to catch low
1013 		 * counts and a 1% limit for high counts.
1014 		 */
1015 		uu = ruxp->rux_uu;
1016 		su = ruxp->rux_su;
1017 		tu = ruxp->rux_tu;
1018 	} else if (vm_guest == VM_GUEST_NO) {  /* tu < ruxp->rux_tu */
1019 		/*
1020 		 * What happened here was likely that a laptop, which ran at
1021 		 * a reduced clock frequency at boot, kicked into high gear.
1022 		 * The wisdom of spamming this message in that case is
1023 		 * dubious, but it might also be indicative of something
1024 		 * serious, so lets keep it and hope laptops can be made
1025 		 * more truthful about their CPU speed via ACPI.
1026 		 */
1027 		printf("calcru: runtime went backwards from %ju usec "
1028 		    "to %ju usec for pid %d (%s)\n",
1029 		    (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1030 		    p->p_pid, p->p_comm);
1031 	}
1032 
1033 	ruxp->rux_uu = uu;
1034 	ruxp->rux_su = su;
1035 	ruxp->rux_tu = tu;
1036 
1037 	up->tv_sec = uu / 1000000;
1038 	up->tv_usec = uu % 1000000;
1039 	sp->tv_sec = su / 1000000;
1040 	sp->tv_usec = su % 1000000;
1041 }
1042 
1043 #ifndef _SYS_SYSPROTO_H_
1044 struct getrusage_args {
1045 	int	who;
1046 	struct	rusage *rusage;
1047 };
1048 #endif
1049 int
1050 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1051 {
1052 	struct rusage ru;
1053 	int error;
1054 
1055 	error = kern_getrusage(td, uap->who, &ru);
1056 	if (error == 0)
1057 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1058 	return (error);
1059 }
1060 
1061 int
1062 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1063 {
1064 	struct proc *p;
1065 	int error;
1066 
1067 	error = 0;
1068 	p = td->td_proc;
1069 	PROC_LOCK(p);
1070 	switch (who) {
1071 	case RUSAGE_SELF:
1072 		rufetchcalc(p, rup, &rup->ru_utime,
1073 		    &rup->ru_stime);
1074 		break;
1075 
1076 	case RUSAGE_CHILDREN:
1077 		*rup = p->p_stats->p_cru;
1078 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1079 		break;
1080 
1081 	case RUSAGE_THREAD:
1082 		PROC_STATLOCK(p);
1083 		thread_lock(td);
1084 		rufetchtd(td, rup);
1085 		thread_unlock(td);
1086 		PROC_STATUNLOCK(p);
1087 		break;
1088 
1089 	default:
1090 		error = EINVAL;
1091 	}
1092 	PROC_UNLOCK(p);
1093 	return (error);
1094 }
1095 
1096 void
1097 rucollect(struct rusage *ru, struct rusage *ru2)
1098 {
1099 	long *ip, *ip2;
1100 	int i;
1101 
1102 	if (ru->ru_maxrss < ru2->ru_maxrss)
1103 		ru->ru_maxrss = ru2->ru_maxrss;
1104 	ip = &ru->ru_first;
1105 	ip2 = &ru2->ru_first;
1106 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1107 		*ip++ += *ip2++;
1108 }
1109 
1110 void
1111 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1112     struct rusage_ext *rux2)
1113 {
1114 
1115 	rux->rux_runtime += rux2->rux_runtime;
1116 	rux->rux_uticks += rux2->rux_uticks;
1117 	rux->rux_sticks += rux2->rux_sticks;
1118 	rux->rux_iticks += rux2->rux_iticks;
1119 	rux->rux_uu += rux2->rux_uu;
1120 	rux->rux_su += rux2->rux_su;
1121 	rux->rux_tu += rux2->rux_tu;
1122 	rucollect(ru, ru2);
1123 }
1124 
1125 /*
1126  * Aggregate tick counts into the proc's rusage_ext.
1127  */
1128 static void
1129 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1130 {
1131 
1132 	rux->rux_runtime += td->td_incruntime;
1133 	rux->rux_uticks += td->td_uticks;
1134 	rux->rux_sticks += td->td_sticks;
1135 	rux->rux_iticks += td->td_iticks;
1136 }
1137 
1138 void
1139 ruxagg_locked(struct proc *p, struct thread *td)
1140 {
1141 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1142 	PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1143 
1144 	ruxagg_ext_locked(&p->p_rux, td);
1145 	ruxagg_ext_locked(&td->td_rux, td);
1146 	td->td_incruntime = 0;
1147 	td->td_uticks = 0;
1148 	td->td_iticks = 0;
1149 	td->td_sticks = 0;
1150 }
1151 
1152 void
1153 ruxagg(struct proc *p, struct thread *td)
1154 {
1155 
1156 	thread_lock(td);
1157 	ruxagg_locked(p, td);
1158 	thread_unlock(td);
1159 }
1160 
1161 /*
1162  * Update the rusage_ext structure and fetch a valid aggregate rusage
1163  * for proc p if storage for one is supplied.
1164  */
1165 void
1166 rufetch(struct proc *p, struct rusage *ru)
1167 {
1168 	struct thread *td;
1169 
1170 	PROC_STATLOCK_ASSERT(p, MA_OWNED);
1171 
1172 	*ru = p->p_ru;
1173 	if (p->p_numthreads > 0)  {
1174 		FOREACH_THREAD_IN_PROC(p, td) {
1175 			ruxagg(p, td);
1176 			rucollect(ru, &td->td_ru);
1177 		}
1178 	}
1179 }
1180 
1181 /*
1182  * Atomically perform a rufetch and a calcru together.
1183  * Consumers, can safely assume the calcru is executed only once
1184  * rufetch is completed.
1185  */
1186 void
1187 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1188     struct timeval *sp)
1189 {
1190 
1191 	PROC_STATLOCK(p);
1192 	rufetch(p, ru);
1193 	calcru(p, up, sp);
1194 	PROC_STATUNLOCK(p);
1195 }
1196 
1197 /*
1198  * Allocate a new resource limits structure and initialize its
1199  * reference count and mutex pointer.
1200  */
1201 struct plimit *
1202 lim_alloc(void)
1203 {
1204 	struct plimit *limp;
1205 
1206 	limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1207 	refcount_init(&limp->pl_refcnt, 1);
1208 	return (limp);
1209 }
1210 
1211 struct plimit *
1212 lim_hold(struct plimit *limp)
1213 {
1214 
1215 	refcount_acquire(&limp->pl_refcnt);
1216 	return (limp);
1217 }
1218 
1219 struct plimit *
1220 lim_cowsync(void)
1221 {
1222 	struct thread *td;
1223 	struct proc *p;
1224 	struct plimit *oldlimit;
1225 
1226 	td = curthread;
1227 	p = td->td_proc;
1228 	PROC_LOCK_ASSERT(p, MA_OWNED);
1229 
1230 	if (td->td_limit == p->p_limit)
1231 		return (NULL);
1232 
1233 	oldlimit = td->td_limit;
1234 	td->td_limit = lim_hold(p->p_limit);
1235 
1236 	return (oldlimit);
1237 }
1238 
1239 void
1240 lim_fork(struct proc *p1, struct proc *p2)
1241 {
1242 
1243 	PROC_LOCK_ASSERT(p1, MA_OWNED);
1244 	PROC_LOCK_ASSERT(p2, MA_OWNED);
1245 
1246 	p2->p_limit = lim_hold(p1->p_limit);
1247 	callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1248 	if (p1->p_cpulimit != RLIM_INFINITY)
1249 		callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1250 		    lim_cb, p2, C_PREL(1));
1251 }
1252 
1253 void
1254 lim_free(struct plimit *limp)
1255 {
1256 
1257 	if (refcount_release(&limp->pl_refcnt))
1258 		free((void *)limp, M_PLIMIT);
1259 }
1260 
1261 void
1262 lim_freen(struct plimit *limp, int n)
1263 {
1264 
1265 	if (refcount_releasen(&limp->pl_refcnt, n))
1266 		free((void *)limp, M_PLIMIT);
1267 }
1268 
1269 void
1270 limbatch_add(struct limbatch *lb, struct thread *td)
1271 {
1272 	struct plimit *limp;
1273 
1274 	MPASS(td->td_limit != NULL);
1275 	limp = td->td_limit;
1276 
1277 	if (lb->limp != limp) {
1278 		if (lb->count != 0) {
1279 			lim_freen(lb->limp, lb->count);
1280 			lb->count = 0;
1281 		}
1282 		lb->limp = limp;
1283 	}
1284 
1285 	lb->count++;
1286 }
1287 
1288 void
1289 limbatch_final(struct limbatch *lb)
1290 {
1291 
1292 	MPASS(lb->count != 0);
1293 	lim_freen(lb->limp, lb->count);
1294 }
1295 
1296 /*
1297  * Make a copy of the plimit structure.
1298  * We share these structures copy-on-write after fork.
1299  */
1300 void
1301 lim_copy(struct plimit *dst, struct plimit *src)
1302 {
1303 
1304 	KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1305 	bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1306 }
1307 
1308 /*
1309  * Return the hard limit for a particular system resource.  The
1310  * which parameter specifies the index into the rlimit array.
1311  */
1312 rlim_t
1313 lim_max(struct thread *td, int which)
1314 {
1315 	struct rlimit rl;
1316 
1317 	lim_rlimit(td, which, &rl);
1318 	return (rl.rlim_max);
1319 }
1320 
1321 rlim_t
1322 lim_max_proc(struct proc *p, int which)
1323 {
1324 	struct rlimit rl;
1325 
1326 	lim_rlimit_proc(p, which, &rl);
1327 	return (rl.rlim_max);
1328 }
1329 
1330 /*
1331  * Return the current (soft) limit for a particular system resource.
1332  * The which parameter which specifies the index into the rlimit array
1333  */
1334 rlim_t
1335 (lim_cur)(struct thread *td, int which)
1336 {
1337 	struct rlimit rl;
1338 
1339 	lim_rlimit(td, which, &rl);
1340 	return (rl.rlim_cur);
1341 }
1342 
1343 rlim_t
1344 lim_cur_proc(struct proc *p, int which)
1345 {
1346 	struct rlimit rl;
1347 
1348 	lim_rlimit_proc(p, which, &rl);
1349 	return (rl.rlim_cur);
1350 }
1351 
1352 /*
1353  * Return a copy of the entire rlimit structure for the system limit
1354  * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1355  */
1356 void
1357 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1358 {
1359 	struct proc *p = td->td_proc;
1360 
1361 	MPASS(td == curthread);
1362 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1363 	    ("request for invalid resource limit"));
1364 	*rlp = td->td_limit->pl_rlimit[which];
1365 	if (p->p_sysent->sv_fixlimit != NULL)
1366 		p->p_sysent->sv_fixlimit(rlp, which);
1367 }
1368 
1369 void
1370 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1371 {
1372 
1373 	PROC_LOCK_ASSERT(p, MA_OWNED);
1374 	KASSERT(which >= 0 && which < RLIM_NLIMITS,
1375 	    ("request for invalid resource limit"));
1376 	*rlp = p->p_limit->pl_rlimit[which];
1377 	if (p->p_sysent->sv_fixlimit != NULL)
1378 		p->p_sysent->sv_fixlimit(rlp, which);
1379 }
1380 
1381 void
1382 uihashinit(void)
1383 {
1384 
1385 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1386 	rw_init(&uihashtbl_lock, "uidinfo hash");
1387 }
1388 
1389 /*
1390  * Look up a uidinfo struct for the parameter uid.
1391  * uihashtbl_lock must be locked.
1392  * Increase refcount on uidinfo struct returned.
1393  */
1394 static struct uidinfo *
1395 uilookup(uid_t uid)
1396 {
1397 	struct uihashhead *uipp;
1398 	struct uidinfo *uip;
1399 
1400 	rw_assert(&uihashtbl_lock, RA_LOCKED);
1401 	uipp = UIHASH(uid);
1402 	LIST_FOREACH(uip, uipp, ui_hash)
1403 		if (uip->ui_uid == uid) {
1404 			uihold(uip);
1405 			break;
1406 		}
1407 
1408 	return (uip);
1409 }
1410 
1411 /*
1412  * Find or allocate a struct uidinfo for a particular uid.
1413  * Returns with uidinfo struct referenced.
1414  * uifree() should be called on a struct uidinfo when released.
1415  */
1416 struct uidinfo *
1417 uifind(uid_t uid)
1418 {
1419 	struct uidinfo *new_uip, *uip;
1420 	struct ucred *cred;
1421 
1422 	cred = curthread->td_ucred;
1423 	if (cred->cr_uidinfo->ui_uid == uid) {
1424 		uip = cred->cr_uidinfo;
1425 		uihold(uip);
1426 		return (uip);
1427 	} else if (cred->cr_ruidinfo->ui_uid == uid) {
1428 		uip = cred->cr_ruidinfo;
1429 		uihold(uip);
1430 		return (uip);
1431 	}
1432 
1433 	rw_rlock(&uihashtbl_lock);
1434 	uip = uilookup(uid);
1435 	rw_runlock(&uihashtbl_lock);
1436 	if (uip != NULL)
1437 		return (uip);
1438 
1439 	new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1440 	racct_create(&new_uip->ui_racct);
1441 	refcount_init(&new_uip->ui_ref, 1);
1442 	new_uip->ui_uid = uid;
1443 
1444 	rw_wlock(&uihashtbl_lock);
1445 	/*
1446 	 * There's a chance someone created our uidinfo while we
1447 	 * were in malloc and not holding the lock, so we have to
1448 	 * make sure we don't insert a duplicate uidinfo.
1449 	 */
1450 	if ((uip = uilookup(uid)) == NULL) {
1451 		LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1452 		rw_wunlock(&uihashtbl_lock);
1453 		uip = new_uip;
1454 	} else {
1455 		rw_wunlock(&uihashtbl_lock);
1456 		racct_destroy(&new_uip->ui_racct);
1457 		free(new_uip, M_UIDINFO);
1458 	}
1459 	return (uip);
1460 }
1461 
1462 /*
1463  * Place another refcount on a uidinfo struct.
1464  */
1465 void
1466 uihold(struct uidinfo *uip)
1467 {
1468 
1469 	refcount_acquire(&uip->ui_ref);
1470 }
1471 
1472 /*-
1473  * Since uidinfo structs have a long lifetime, we use an
1474  * opportunistic refcounting scheme to avoid locking the lookup hash
1475  * for each release.
1476  *
1477  * If the refcount hits 0, we need to free the structure,
1478  * which means we need to lock the hash.
1479  * Optimal case:
1480  *   After locking the struct and lowering the refcount, if we find
1481  *   that we don't need to free, simply unlock and return.
1482  * Suboptimal case:
1483  *   If refcount lowering results in need to free, bump the count
1484  *   back up, lose the lock and acquire the locks in the proper
1485  *   order to try again.
1486  */
1487 void
1488 uifree(struct uidinfo *uip)
1489 {
1490 
1491 	if (refcount_release_if_not_last(&uip->ui_ref))
1492 		return;
1493 
1494 	rw_wlock(&uihashtbl_lock);
1495 	if (refcount_release(&uip->ui_ref) == 0) {
1496 		rw_wunlock(&uihashtbl_lock);
1497 		return;
1498 	}
1499 
1500 	racct_destroy(&uip->ui_racct);
1501 	LIST_REMOVE(uip, ui_hash);
1502 	rw_wunlock(&uihashtbl_lock);
1503 
1504 	if (uip->ui_sbsize != 0)
1505 		printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1506 		    uip->ui_uid, uip->ui_sbsize);
1507 	if (uip->ui_proccnt != 0)
1508 		printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1509 		    uip->ui_uid, uip->ui_proccnt);
1510 	if (uip->ui_vmsize != 0)
1511 		printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1512 		    uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1513 	free(uip, M_UIDINFO);
1514 }
1515 
1516 #ifdef RACCT
1517 void
1518 ui_racct_foreach(void (*callback)(struct racct *racct,
1519     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1520     void *arg2, void *arg3)
1521 {
1522 	struct uidinfo *uip;
1523 	struct uihashhead *uih;
1524 
1525 	rw_rlock(&uihashtbl_lock);
1526 	if (pre != NULL)
1527 		(pre)();
1528 	for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1529 		LIST_FOREACH(uip, uih, ui_hash) {
1530 			(callback)(uip->ui_racct, arg2, arg3);
1531 		}
1532 	}
1533 	if (post != NULL)
1534 		(post)();
1535 	rw_runlock(&uihashtbl_lock);
1536 }
1537 #endif
1538 
1539 static inline int
1540 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1541 {
1542 	long new;
1543 
1544 	/* Don't allow them to exceed max, but allow subtraction. */
1545 	new = atomic_fetchadd_long(limit, (long)diff) + diff;
1546 	if (diff > 0 && max != 0) {
1547 		if (new < 0 || new > max) {
1548 			atomic_subtract_long(limit, (long)diff);
1549 			return (0);
1550 		}
1551 	} else if (new < 0)
1552 		printf("negative %s for uid = %d\n", name, uip->ui_uid);
1553 	return (1);
1554 }
1555 
1556 /*
1557  * Change the count associated with number of processes
1558  * a given user is using.  When 'max' is 0, don't enforce a limit
1559  */
1560 int
1561 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1562 {
1563 
1564 	return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1565 }
1566 
1567 /*
1568  * Change the total socket buffer size a user has used.
1569  */
1570 int
1571 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1572 {
1573 	int diff, rv;
1574 
1575 	diff = to - *hiwat;
1576 	if (diff > 0 && max == 0) {
1577 		rv = 0;
1578 	} else {
1579 		rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1580 		if (rv != 0)
1581 			*hiwat = to;
1582 	}
1583 	return (rv);
1584 }
1585 
1586 /*
1587  * Change the count associated with number of pseudo-terminals
1588  * a given user is using.  When 'max' is 0, don't enforce a limit
1589  */
1590 int
1591 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1592 {
1593 
1594 	return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1595 }
1596 
1597 int
1598 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1599 {
1600 
1601 	return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1602 }
1603 
1604 int
1605 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1606 {
1607 
1608 	return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));
1609 }
1610