xref: /openbsd-src/sys/kern/kern_resource.c (revision f5a6fea6afbf41ba68c0a53f5a4f2277e47251f3) 
1 /*	$OpenBSD: kern_resource.c,v 1.83 2024/05/22 09:20:22 claudio Exp $	*/
2 /*	$NetBSD: kern_resource.c,v 1.38 1996/10/23 07:19:38 matthias Exp $	*/
3 
4 /*-
5  * Copyright (c) 1982, 1986, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_resource.c	8.5 (Berkeley) 1/21/94
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/file.h>
44 #include <sys/resourcevar.h>
45 #include <sys/pool.h>
46 #include <sys/proc.h>
47 #include <sys/ktrace.h>
48 #include <sys/sched.h>
49 #include <sys/signalvar.h>
50 
51 #include <sys/mount.h>
52 #include <sys/syscallargs.h>
53 
54 #include <uvm/uvm_extern.h>
55 #include <uvm/uvm.h>
56 
57 /* Resource usage check interval in msec */
58 #define RUCHECK_INTERVAL	1000
59 
60 /* SIGXCPU interval in seconds of process runtime */
61 #define SIGXCPU_INTERVAL	5
62 
63 struct plimit	*lim_copy(struct plimit *);
64 struct plimit	*lim_write_begin(void);
65 void		 lim_write_commit(struct plimit *);
66 
67 void	tuagg_sub(struct tusage *, struct proc *, const struct timespec *);
68 
69 /*
70  * Patchable maximum data and stack limits.
71  */
72 rlim_t maxdmap = MAXDSIZ;
73 rlim_t maxsmap = MAXSSIZ;
74 
75 /*
76  * Serializes resource limit updates.
77  * This lock has to be held together with ps_mtx when updating
78  * the process' ps_limit.
79  */
80 struct rwlock rlimit_lock = RWLOCK_INITIALIZER("rlimitlk");
81 
82 /*
83  * Resource controls and accounting.
84  */
85 
86 int
87 sys_getpriority(struct proc *curp, void *v, register_t *retval)
88 {
89 	struct sys_getpriority_args /* {
90 		syscallarg(int) which;
91 		syscallarg(id_t) who;
92 	} */ *uap = v;
93 	struct process *pr;
94 	int low = NZERO + PRIO_MAX + 1;
95 
96 	switch (SCARG(uap, which)) {
97 
98 	case PRIO_PROCESS:
99 		if (SCARG(uap, who) == 0)
100 			pr = curp->p_p;
101 		else
102 			pr = prfind(SCARG(uap, who));
103 		if (pr == NULL)
104 			break;
105 		if (pr->ps_nice < low)
106 			low = pr->ps_nice;
107 		break;
108 
109 	case PRIO_PGRP: {
110 		struct pgrp *pg;
111 
112 		if (SCARG(uap, who) == 0)
113 			pg = curp->p_p->ps_pgrp;
114 		else if ((pg = pgfind(SCARG(uap, who))) == NULL)
115 			break;
116 		LIST_FOREACH(pr, &pg->pg_members, ps_pglist)
117 			if (pr->ps_nice < low)
118 				low = pr->ps_nice;
119 		break;
120 	}
121 
122 	case PRIO_USER:
123 		if (SCARG(uap, who) == 0)
124 			SCARG(uap, who) = curp->p_ucred->cr_uid;
125 		LIST_FOREACH(pr, &allprocess, ps_list)
126 			if (pr->ps_ucred->cr_uid == SCARG(uap, who) &&
127 			    pr->ps_nice < low)
128 				low = pr->ps_nice;
129 		break;
130 
131 	default:
132 		return (EINVAL);
133 	}
134 	if (low == NZERO + PRIO_MAX + 1)
135 		return (ESRCH);
136 	*retval = low - NZERO;
137 	return (0);
138 }
139 
140 int
141 sys_setpriority(struct proc *curp, void *v, register_t *retval)
142 {
143 	struct sys_setpriority_args /* {
144 		syscallarg(int) which;
145 		syscallarg(id_t) who;
146 		syscallarg(int) prio;
147 	} */ *uap = v;
148 	struct process *pr;
149 	int found = 0, error = 0;
150 
151 	switch (SCARG(uap, which)) {
152 
153 	case PRIO_PROCESS:
154 		if (SCARG(uap, who) == 0)
155 			pr = curp->p_p;
156 		else
157 			pr = prfind(SCARG(uap, who));
158 		if (pr == NULL)
159 			break;
160 		error = donice(curp, pr, SCARG(uap, prio));
161 		found = 1;
162 		break;
163 
164 	case PRIO_PGRP: {
165 		struct pgrp *pg;
166 
167 		if (SCARG(uap, who) == 0)
168 			pg = curp->p_p->ps_pgrp;
169 		else if ((pg = pgfind(SCARG(uap, who))) == NULL)
170 			break;
171 		LIST_FOREACH(pr, &pg->pg_members, ps_pglist) {
172 			error = donice(curp, pr, SCARG(uap, prio));
173 			found = 1;
174 		}
175 		break;
176 	}
177 
178 	case PRIO_USER:
179 		if (SCARG(uap, who) == 0)
180 			SCARG(uap, who) = curp->p_ucred->cr_uid;
181 		LIST_FOREACH(pr, &allprocess, ps_list)
182 			if (pr->ps_ucred->cr_uid == SCARG(uap, who)) {
183 				error = donice(curp, pr, SCARG(uap, prio));
184 				found = 1;
185 			}
186 		break;
187 
188 	default:
189 		return (EINVAL);
190 	}
191 	if (!found)
192 		return (ESRCH);
193 	return (error);
194 }
195 
196 int
197 donice(struct proc *curp, struct process *chgpr, int n)
198 {
199 	struct ucred *ucred = curp->p_ucred;
200 	struct proc *p;
201 	int s;
202 
203 	if (ucred->cr_uid != 0 && ucred->cr_ruid != 0 &&
204 	    ucred->cr_uid != chgpr->ps_ucred->cr_uid &&
205 	    ucred->cr_ruid != chgpr->ps_ucred->cr_uid)
206 		return (EPERM);
207 	if (n > PRIO_MAX)
208 		n = PRIO_MAX;
209 	if (n < PRIO_MIN)
210 		n = PRIO_MIN;
211 	n += NZERO;
212 	if (n < chgpr->ps_nice && suser(curp))
213 		return (EACCES);
214 	chgpr->ps_nice = n;
215 	mtx_enter(&chgpr->ps_mtx);
216 	SCHED_LOCK(s);
217 	TAILQ_FOREACH(p, &chgpr->ps_threads, p_thr_link) {
218 		setpriority(p, p->p_estcpu, n);
219 	}
220 	SCHED_UNLOCK(s);
221 	mtx_leave(&chgpr->ps_mtx);
222 	return (0);
223 }
224 
225 int
226 sys_setrlimit(struct proc *p, void *v, register_t *retval)
227 {
228 	struct sys_setrlimit_args /* {
229 		syscallarg(int) which;
230 		syscallarg(const struct rlimit *) rlp;
231 	} */ *uap = v;
232 	struct rlimit alim;
233 	int error;
234 
235 	error = copyin((caddr_t)SCARG(uap, rlp), (caddr_t)&alim,
236 		       sizeof (struct rlimit));
237 	if (error)
238 		return (error);
239 #ifdef KTRACE
240 	if (KTRPOINT(p, KTR_STRUCT))
241 		ktrrlimit(p, &alim);
242 #endif
243 	return (dosetrlimit(p, SCARG(uap, which), &alim));
244 }
245 
246 int
247 dosetrlimit(struct proc *p, u_int which, struct rlimit *limp)
248 {
249 	struct rlimit *alimp;
250 	struct plimit *limit;
251 	rlim_t maxlim;
252 	int error;
253 
254 	if (which >= RLIM_NLIMITS || limp->rlim_cur > limp->rlim_max)
255 		return (EINVAL);
256 
257 	rw_enter_write(&rlimit_lock);
258 
259 	alimp = &p->p_p->ps_limit->pl_rlimit[which];
260 	if (limp->rlim_max > alimp->rlim_max) {
261 		if ((error = suser(p)) != 0) {
262 			rw_exit_write(&rlimit_lock);
263 			return (error);
264 		}
265 	}
266 
267 	/* Get exclusive write access to the limit structure. */
268 	limit = lim_write_begin();
269 	alimp = &limit->pl_rlimit[which];
270 
271 	switch (which) {
272 	case RLIMIT_DATA:
273 		maxlim = maxdmap;
274 		break;
275 	case RLIMIT_STACK:
276 		maxlim = maxsmap;
277 		break;
278 	case RLIMIT_NOFILE:
279 		maxlim = maxfiles;
280 		break;
281 	case RLIMIT_NPROC:
282 		maxlim = maxprocess;
283 		break;
284 	default:
285 		maxlim = RLIM_INFINITY;
286 		break;
287 	}
288 
289 	if (limp->rlim_max > maxlim)
290 		limp->rlim_max = maxlim;
291 	if (limp->rlim_cur > limp->rlim_max)
292 		limp->rlim_cur = limp->rlim_max;
293 
294 	if (which == RLIMIT_CPU && limp->rlim_cur != RLIM_INFINITY &&
295 	    alimp->rlim_cur == RLIM_INFINITY)
296 		timeout_add_msec(&p->p_p->ps_rucheck_to, RUCHECK_INTERVAL);
297 
298 	if (which == RLIMIT_STACK) {
299 		/*
300 		 * Stack is allocated to the max at exec time with only
301 		 * "rlim_cur" bytes accessible.  If stack limit is going
302 		 * up make more accessible, if going down make inaccessible.
303 		 */
304 		if (limp->rlim_cur != alimp->rlim_cur) {
305 			vaddr_t addr;
306 			vsize_t size;
307 			vm_prot_t prot;
308 			struct vmspace *vm = p->p_vmspace;
309 
310 			if (limp->rlim_cur > alimp->rlim_cur) {
311 				prot = PROT_READ | PROT_WRITE;
312 				size = limp->rlim_cur - alimp->rlim_cur;
313 #ifdef MACHINE_STACK_GROWS_UP
314 				addr = (vaddr_t)vm->vm_maxsaddr +
315 				    alimp->rlim_cur;
316 #else
317 				addr = (vaddr_t)vm->vm_minsaddr -
318 				    limp->rlim_cur;
319 #endif
320 			} else {
321 				prot = PROT_NONE;
322 				size = alimp->rlim_cur - limp->rlim_cur;
323 #ifdef MACHINE_STACK_GROWS_UP
324 				addr = (vaddr_t)vm->vm_maxsaddr +
325 				    limp->rlim_cur;
326 #else
327 				addr = (vaddr_t)vm->vm_minsaddr -
328 				    alimp->rlim_cur;
329 #endif
330 			}
331 			addr = trunc_page(addr);
332 			size = round_page(size);
333 			KERNEL_LOCK();
334 			(void) uvm_map_protect(&vm->vm_map, addr,
335 			    addr+size, prot, UVM_ET_STACK, FALSE, FALSE);
336 			KERNEL_UNLOCK();
337 		}
338 	}
339 
340 	*alimp = *limp;
341 
342 	lim_write_commit(limit);
343 	rw_exit_write(&rlimit_lock);
344 
345 	return (0);
346 }
347 
348 int
349 sys_getrlimit(struct proc *p, void *v, register_t *retval)
350 {
351 	struct sys_getrlimit_args /* {
352 		syscallarg(int) which;
353 		syscallarg(struct rlimit *) rlp;
354 	} */ *uap = v;
355 	struct plimit *limit;
356 	struct rlimit alimp;
357 	int error;
358 
359 	if (SCARG(uap, which) < 0 || SCARG(uap, which) >= RLIM_NLIMITS)
360 		return (EINVAL);
361 	limit = lim_read_enter();
362 	alimp = limit->pl_rlimit[SCARG(uap, which)];
363 	lim_read_leave(limit);
364 	error = copyout(&alimp, SCARG(uap, rlp), sizeof(struct rlimit));
365 #ifdef KTRACE
366 	if (error == 0 && KTRPOINT(p, KTR_STRUCT))
367 		ktrrlimit(p, &alimp);
368 #endif
369 	return (error);
370 }
371 
372 void
373 tuagg_sub(struct tusage *tup, struct proc *p, const struct timespec *ts)
374 {
375 	if (ts != NULL)
376 		timespecadd(&tup->tu_runtime, ts, &tup->tu_runtime);
377 	tup->tu_uticks += p->p_uticks;
378 	tup->tu_sticks += p->p_sticks;
379 	tup->tu_iticks += p->p_iticks;
380 }
381 
382 /*
383  * Aggregate a single thread's immediate time counts into the running
384  * totals for the thread and process
385  */
386 void
387 tuagg_locked(struct process *pr, struct proc *p, const struct timespec *ts)
388 {
389 	tuagg_sub(&pr->ps_tu, p, ts);
390 	tuagg_sub(&p->p_tu, p, ts);
391 	p->p_uticks = 0;
392 	p->p_sticks = 0;
393 	p->p_iticks = 0;
394 }
395 
396 void
397 tuagg(struct process *pr, struct proc *p)
398 {
399 	int s;
400 
401 	SCHED_LOCK(s);
402 	tuagg_locked(pr, p, NULL);
403 	SCHED_UNLOCK(s);
404 }
405 
406 /*
407  * Transform the running time and tick information in a struct tusage
408  * into user, system, and interrupt time usage.
409  */
410 void
411 calctsru(struct tusage *tup, struct timespec *up, struct timespec *sp,
412     struct timespec *ip)
413 {
414 	u_quad_t st, ut, it;
415 
416 	st = tup->tu_sticks;
417 	ut = tup->tu_uticks;
418 	it = tup->tu_iticks;
419 
420 	if (st + ut + it == 0) {
421 		timespecclear(up);
422 		timespecclear(sp);
423 		if (ip != NULL)
424 			timespecclear(ip);
425 		return;
426 	}
427 
428 	st = st * 1000000000 / stathz;
429 	sp->tv_sec = st / 1000000000;
430 	sp->tv_nsec = st % 1000000000;
431 	ut = ut * 1000000000 / stathz;
432 	up->tv_sec = ut / 1000000000;
433 	up->tv_nsec = ut % 1000000000;
434 	if (ip != NULL) {
435 		it = it * 1000000000 / stathz;
436 		ip->tv_sec = it / 1000000000;
437 		ip->tv_nsec = it % 1000000000;
438 	}
439 }
440 
441 void
442 calcru(struct tusage *tup, struct timeval *up, struct timeval *sp,
443     struct timeval *ip)
444 {
445 	struct timespec u, s, i;
446 
447 	calctsru(tup, &u, &s, ip != NULL ? &i : NULL);
448 	TIMESPEC_TO_TIMEVAL(up, &u);
449 	TIMESPEC_TO_TIMEVAL(sp, &s);
450 	if (ip != NULL)
451 		TIMESPEC_TO_TIMEVAL(ip, &i);
452 }
453 
454 int
455 sys_getrusage(struct proc *p, void *v, register_t *retval)
456 {
457 	struct sys_getrusage_args /* {
458 		syscallarg(int) who;
459 		syscallarg(struct rusage *) rusage;
460 	} */ *uap = v;
461 	struct rusage ru;
462 	int error;
463 
464 	error = dogetrusage(p, SCARG(uap, who), &ru);
465 	if (error == 0) {
466 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
467 #ifdef KTRACE
468 		if (error == 0 && KTRPOINT(p, KTR_STRUCT))
469 			ktrrusage(p, &ru);
470 #endif
471 	}
472 	return (error);
473 }
474 
475 int
476 dogetrusage(struct proc *p, int who, struct rusage *rup)
477 {
478 	struct process *pr = p->p_p;
479 	struct proc *q;
480 
481 	KERNEL_ASSERT_LOCKED();
482 
483 	switch (who) {
484 	case RUSAGE_SELF:
485 		/* start with the sum of dead threads, if any */
486 		if (pr->ps_ru != NULL)
487 			*rup = *pr->ps_ru;
488 		else
489 			memset(rup, 0, sizeof(*rup));
490 
491 		/* add on all living threads */
492 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
493 			ruadd(rup, &q->p_ru);
494 			tuagg(pr, q);
495 		}
496 
497 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
498 		break;
499 
500 	case RUSAGE_THREAD:
501 		*rup = p->p_ru;
502 		calcru(&p->p_tu, &rup->ru_utime, &rup->ru_stime, NULL);
503 		break;
504 
505 	case RUSAGE_CHILDREN:
506 		*rup = pr->ps_cru;
507 		break;
508 
509 	default:
510 		return (EINVAL);
511 	}
512 	return (0);
513 }
514 
515 void
516 ruadd(struct rusage *ru, struct rusage *ru2)
517 {
518 	long *ip, *ip2;
519 	int i;
520 
521 	timeradd(&ru->ru_utime, &ru2->ru_utime, &ru->ru_utime);
522 	timeradd(&ru->ru_stime, &ru2->ru_stime, &ru->ru_stime);
523 	if (ru->ru_maxrss < ru2->ru_maxrss)
524 		ru->ru_maxrss = ru2->ru_maxrss;
525 	ip = &ru->ru_first; ip2 = &ru2->ru_first;
526 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
527 		*ip++ += *ip2++;
528 }
529 
530 /*
531  * Check if the process exceeds its cpu resource allocation.
532  * If over max, kill it.
533  */
534 void
535 rucheck(void *arg)
536 {
537 	struct rlimit rlim;
538 	struct process *pr = arg;
539 	time_t runtime;
540 	int s;
541 
542 	KERNEL_ASSERT_LOCKED();
543 
544 	SCHED_LOCK(s);
545 	runtime = pr->ps_tu.tu_runtime.tv_sec;
546 	SCHED_UNLOCK(s);
547 
548 	mtx_enter(&pr->ps_mtx);
549 	rlim = pr->ps_limit->pl_rlimit[RLIMIT_CPU];
550 	mtx_leave(&pr->ps_mtx);
551 
552 	if ((rlim_t)runtime >= rlim.rlim_cur) {
553 		if ((rlim_t)runtime >= rlim.rlim_max) {
554 			prsignal(pr, SIGKILL);
555 		} else if (runtime >= pr->ps_nextxcpu) {
556 			prsignal(pr, SIGXCPU);
557 			pr->ps_nextxcpu = runtime + SIGXCPU_INTERVAL;
558 		}
559 	}
560 
561 	timeout_add_msec(&pr->ps_rucheck_to, RUCHECK_INTERVAL);
562 }
563 
564 struct pool plimit_pool;
565 
566 void
567 lim_startup(struct plimit *limit0)
568 {
569 	rlim_t lim;
570 	int i;
571 
572 	pool_init(&plimit_pool, sizeof(struct plimit), 0, IPL_MPFLOOR,
573 	    PR_WAITOK, "plimitpl", NULL);
574 
575 	for (i = 0; i < nitems(limit0->pl_rlimit); i++)
576 		limit0->pl_rlimit[i].rlim_cur =
577 		    limit0->pl_rlimit[i].rlim_max = RLIM_INFINITY;
578 	limit0->pl_rlimit[RLIMIT_NOFILE].rlim_cur = NOFILE;
579 	limit0->pl_rlimit[RLIMIT_NOFILE].rlim_max = MIN(NOFILE_MAX,
580 	    (maxfiles - NOFILE > NOFILE) ? maxfiles - NOFILE : NOFILE);
581 	limit0->pl_rlimit[RLIMIT_NPROC].rlim_cur = MAXUPRC;
582 	lim = ptoa(uvmexp.free);
583 	limit0->pl_rlimit[RLIMIT_RSS].rlim_max = lim;
584 	lim = ptoa(64*1024);		/* Default to very low */
585 	limit0->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
586 	limit0->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
587 	refcnt_init(&limit0->pl_refcnt);
588 }
589 
590 /*
591  * Make a copy of the plimit structure.
592  * We share these structures copy-on-write after fork,
593  * and copy when a limit is changed.
594  */
595 struct plimit *
596 lim_copy(struct plimit *lim)
597 {
598 	struct plimit *newlim;
599 
600 	newlim = pool_get(&plimit_pool, PR_WAITOK);
601 	memcpy(newlim->pl_rlimit, lim->pl_rlimit,
602 	    sizeof(struct rlimit) * RLIM_NLIMITS);
603 	refcnt_init(&newlim->pl_refcnt);
604 	return (newlim);
605 }
606 
607 void
608 lim_free(struct plimit *lim)
609 {
610 	if (refcnt_rele(&lim->pl_refcnt) == 0)
611 		return;
612 	pool_put(&plimit_pool, lim);
613 }
614 
615 void
616 lim_fork(struct process *parent, struct process *child)
617 {
618 	struct plimit *limit;
619 
620 	mtx_enter(&parent->ps_mtx);
621 	limit = parent->ps_limit;
622 	refcnt_take(&limit->pl_refcnt);
623 	mtx_leave(&parent->ps_mtx);
624 
625 	child->ps_limit = limit;
626 
627 	if (limit->pl_rlimit[RLIMIT_CPU].rlim_cur != RLIM_INFINITY)
628 		timeout_add_msec(&child->ps_rucheck_to, RUCHECK_INTERVAL);
629 }
630 
631 /*
632  * Return an exclusive write reference to the process' resource limit structure.
633  * The caller has to release the structure by calling lim_write_commit().
634  *
635  * This invalidates any plimit read reference held by the calling thread.
636  */
637 struct plimit *
638 lim_write_begin(void)
639 {
640 	struct plimit *limit;
641 	struct proc *p = curproc;
642 
643 	rw_assert_wrlock(&rlimit_lock);
644 
645 	if (p->p_limit != NULL)
646 		lim_free(p->p_limit);
647 	p->p_limit = NULL;
648 
649 	/*
650 	 * It is safe to access ps_limit here without holding ps_mtx
651 	 * because rlimit_lock excludes other writers.
652 	 */
653 
654 	limit = p->p_p->ps_limit;
655 	if (P_HASSIBLING(p) || refcnt_shared(&limit->pl_refcnt))
656 		limit = lim_copy(limit);
657 
658 	return (limit);
659 }
660 
661 /*
662  * Finish exclusive write access to the plimit structure.
663  * This makes the structure visible to other threads in the process.
664  */
665 void
666 lim_write_commit(struct plimit *limit)
667 {
668 	struct plimit *olimit;
669 	struct proc *p = curproc;
670 
671 	rw_assert_wrlock(&rlimit_lock);
672 
673 	if (limit != p->p_p->ps_limit) {
674 		mtx_enter(&p->p_p->ps_mtx);
675 		olimit = p->p_p->ps_limit;
676 		p->p_p->ps_limit = limit;
677 		mtx_leave(&p->p_p->ps_mtx);
678 
679 		lim_free(olimit);
680 	}
681 }
682 
683 /*
684  * Begin read access to the process' resource limit structure.
685  * The access has to be finished by calling lim_read_leave().
686  *
687  * Sections denoted by lim_read_enter() and lim_read_leave() cannot nest.
688  */
689 struct plimit *
690 lim_read_enter(void)
691 {
692 	struct plimit *limit;
693 	struct proc *p = curproc;
694 	struct process *pr = p->p_p;
695 
696 	/*
697 	 * This thread might not observe the latest value of ps_limit
698 	 * if another thread updated the limits very recently on another CPU.
699 	 * However, the anomaly should disappear quickly, especially if
700 	 * there is any synchronization activity between the threads (or
701 	 * the CPUs).
702 	 */
703 
704 	limit = p->p_limit;
705 	if (limit != pr->ps_limit) {
706 		mtx_enter(&pr->ps_mtx);
707 		limit = pr->ps_limit;
708 		refcnt_take(&limit->pl_refcnt);
709 		mtx_leave(&pr->ps_mtx);
710 		if (p->p_limit != NULL)
711 			lim_free(p->p_limit);
712 		p->p_limit = limit;
713 	}
714 	KASSERT(limit != NULL);
715 	return (limit);
716 }
717 
718 /*
719  * Get the value of the resource limit in given process.
720  */
721 rlim_t
722 lim_cur_proc(struct proc *p, int which)
723 {
724 	struct process *pr = p->p_p;
725 	rlim_t val;
726 
727 	KASSERT(which >= 0 && which < RLIM_NLIMITS);
728 
729 	mtx_enter(&pr->ps_mtx);
730 	val = pr->ps_limit->pl_rlimit[which].rlim_cur;
731 	mtx_leave(&pr->ps_mtx);
732 	return (val);
733 }
734