xref: /netbsd-src/sys/kern/kern_resource.c (revision 4724848cf0da353df257f730694b7882798e5daf)
1 /*	$NetBSD: kern_resource.c,v 1.189 2022/04/09 23:38:33 riastradh Exp $	*/
2 
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	8.8 (Berkeley) 2/14/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: kern_resource.c,v 1.189 2022/04/09 23:38:33 riastradh Exp $");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/file.h>
46 #include <sys/resourcevar.h>
47 #include <sys/kmem.h>
48 #include <sys/namei.h>
49 #include <sys/pool.h>
50 #include <sys/proc.h>
51 #include <sys/sysctl.h>
52 #include <sys/timevar.h>
53 #include <sys/kauth.h>
54 #include <sys/atomic.h>
55 #include <sys/mount.h>
56 #include <sys/syscallargs.h>
57 #include <sys/atomic.h>
58 
59 #include <uvm/uvm_extern.h>
60 
61 /*
62  * Maximum process data and stack limits.
63  * They are variables so they are patchable.
64  */
65 rlim_t			maxdmap = MAXDSIZ;
66 rlim_t			maxsmap = MAXSSIZ;
67 
68 static pool_cache_t	plimit_cache	__read_mostly;
69 static pool_cache_t	pstats_cache	__read_mostly;
70 
71 static kauth_listener_t	resource_listener;
72 static struct sysctllog	*proc_sysctllog;
73 
74 static int	donice(struct lwp *, struct proc *, int);
75 static void	sysctl_proc_setup(void);
76 
77 static int
78 resource_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
79     void *arg0, void *arg1, void *arg2, void *arg3)
80 {
81 	struct proc *p;
82 	int result;
83 
84 	result = KAUTH_RESULT_DEFER;
85 	p = arg0;
86 
87 	switch (action) {
88 	case KAUTH_PROCESS_NICE:
89 		if (kauth_cred_geteuid(cred) != kauth_cred_geteuid(p->p_cred) &&
90 		    kauth_cred_getuid(cred) != kauth_cred_geteuid(p->p_cred)) {
91 			break;
92 		}
93 
94 		if ((u_long)arg1 >= p->p_nice)
95 			result = KAUTH_RESULT_ALLOW;
96 
97 		break;
98 
99 	case KAUTH_PROCESS_RLIMIT: {
100 		enum kauth_process_req req;
101 
102 		req = (enum kauth_process_req)(uintptr_t)arg1;
103 
104 		switch (req) {
105 		case KAUTH_REQ_PROCESS_RLIMIT_GET:
106 			result = KAUTH_RESULT_ALLOW;
107 			break;
108 
109 		case KAUTH_REQ_PROCESS_RLIMIT_SET: {
110 			struct rlimit *new_rlimit;
111 			u_long which;
112 
113 			if ((p != curlwp->l_proc) &&
114 			    (proc_uidmatch(cred, p->p_cred) != 0))
115 				break;
116 
117 			new_rlimit = arg2;
118 			which = (u_long)arg3;
119 
120 			if (new_rlimit->rlim_max <= p->p_rlimit[which].rlim_max)
121 				result = KAUTH_RESULT_ALLOW;
122 
123 			break;
124 			}
125 
126 		default:
127 			break;
128 		}
129 
130 		break;
131 	}
132 
133 	default:
134 		break;
135 	}
136 
137 	return result;
138 }
139 
140 void
141 resource_init(void)
142 {
143 
144 	plimit_cache = pool_cache_init(sizeof(struct plimit), 0, 0, 0,
145 	    "plimitpl", NULL, IPL_NONE, NULL, NULL, NULL);
146 	pstats_cache = pool_cache_init(sizeof(struct pstats), 0, 0, 0,
147 	    "pstatspl", NULL, IPL_NONE, NULL, NULL, NULL);
148 
149 	resource_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
150 	    resource_listener_cb, NULL);
151 
152 	sysctl_proc_setup();
153 }
154 
155 /*
156  * Resource controls and accounting.
157  */
158 
159 int
160 sys_getpriority(struct lwp *l, const struct sys_getpriority_args *uap,
161     register_t *retval)
162 {
163 	/* {
164 		syscallarg(int) which;
165 		syscallarg(id_t) who;
166 	} */
167 	struct proc *curp = l->l_proc, *p;
168 	id_t who = SCARG(uap, who);
169 	int low = NZERO + PRIO_MAX + 1;
170 
171 	mutex_enter(&proc_lock);
172 	switch (SCARG(uap, which)) {
173 	case PRIO_PROCESS:
174 		p = who ? proc_find(who) : curp;
175 		if (p != NULL)
176 			low = p->p_nice;
177 		break;
178 
179 	case PRIO_PGRP: {
180 		struct pgrp *pg;
181 
182 		if (who == 0)
183 			pg = curp->p_pgrp;
184 		else if ((pg = pgrp_find(who)) == NULL)
185 			break;
186 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
187 			if (p->p_nice < low)
188 				low = p->p_nice;
189 		}
190 		break;
191 	}
192 
193 	case PRIO_USER:
194 		if (who == 0)
195 			who = (int)kauth_cred_geteuid(l->l_cred);
196 		PROCLIST_FOREACH(p, &allproc) {
197 			mutex_enter(p->p_lock);
198 			if (kauth_cred_geteuid(p->p_cred) ==
199 			    (uid_t)who && p->p_nice < low)
200 				low = p->p_nice;
201 			mutex_exit(p->p_lock);
202 		}
203 		break;
204 
205 	default:
206 		mutex_exit(&proc_lock);
207 		return EINVAL;
208 	}
209 	mutex_exit(&proc_lock);
210 
211 	if (low == NZERO + PRIO_MAX + 1) {
212 		return ESRCH;
213 	}
214 	*retval = low - NZERO;
215 	return 0;
216 }
217 
218 int
219 sys_setpriority(struct lwp *l, const struct sys_setpriority_args *uap,
220     register_t *retval)
221 {
222 	/* {
223 		syscallarg(int) which;
224 		syscallarg(id_t) who;
225 		syscallarg(int) prio;
226 	} */
227 	struct proc *curp = l->l_proc, *p;
228 	id_t who = SCARG(uap, who);
229 	int found = 0, error = 0;
230 
231 	mutex_enter(&proc_lock);
232 	switch (SCARG(uap, which)) {
233 	case PRIO_PROCESS:
234 		p = who ? proc_find(who) : curp;
235 		if (p != NULL) {
236 			mutex_enter(p->p_lock);
237 			found++;
238 			error = donice(l, p, SCARG(uap, prio));
239 			mutex_exit(p->p_lock);
240 		}
241 		break;
242 
243 	case PRIO_PGRP: {
244 		struct pgrp *pg;
245 
246 		if (who == 0)
247 			pg = curp->p_pgrp;
248 		else if ((pg = pgrp_find(who)) == NULL)
249 			break;
250 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
251 			mutex_enter(p->p_lock);
252 			found++;
253 			error = donice(l, p, SCARG(uap, prio));
254 			mutex_exit(p->p_lock);
255 			if (error)
256 				break;
257 		}
258 		break;
259 	}
260 
261 	case PRIO_USER:
262 		if (who == 0)
263 			who = (int)kauth_cred_geteuid(l->l_cred);
264 		PROCLIST_FOREACH(p, &allproc) {
265 			mutex_enter(p->p_lock);
266 			if (kauth_cred_geteuid(p->p_cred) ==
267 			    (uid_t)SCARG(uap, who)) {
268 				found++;
269 				error = donice(l, p, SCARG(uap, prio));
270 			}
271 			mutex_exit(p->p_lock);
272 			if (error)
273 				break;
274 		}
275 		break;
276 
277 	default:
278 		mutex_exit(&proc_lock);
279 		return EINVAL;
280 	}
281 	mutex_exit(&proc_lock);
282 
283 	return (found == 0) ? ESRCH : error;
284 }
285 
286 /*
287  * Renice a process.
288  *
289  * Call with the target process' credentials locked.
290  */
291 static int
292 donice(struct lwp *l, struct proc *chgp, int n)
293 {
294 	kauth_cred_t cred = l->l_cred;
295 
296 	KASSERT(mutex_owned(chgp->p_lock));
297 
298 	if (kauth_cred_geteuid(cred) && kauth_cred_getuid(cred) &&
299 	    kauth_cred_geteuid(cred) != kauth_cred_geteuid(chgp->p_cred) &&
300 	    kauth_cred_getuid(cred) != kauth_cred_geteuid(chgp->p_cred))
301 		return EPERM;
302 
303 	if (n > PRIO_MAX) {
304 		n = PRIO_MAX;
305 	}
306 	if (n < PRIO_MIN) {
307 		n = PRIO_MIN;
308 	}
309 	n += NZERO;
310 
311 	if (kauth_authorize_process(cred, KAUTH_PROCESS_NICE, chgp,
312 	    KAUTH_ARG(n), NULL, NULL)) {
313 		return EACCES;
314 	}
315 
316 	sched_nice(chgp, n);
317 	return 0;
318 }
319 
320 int
321 sys_setrlimit(struct lwp *l, const struct sys_setrlimit_args *uap,
322     register_t *retval)
323 {
324 	/* {
325 		syscallarg(int) which;
326 		syscallarg(const struct rlimit *) rlp;
327 	} */
328 	int error, which = SCARG(uap, which);
329 	struct rlimit alim;
330 
331 	error = copyin(SCARG(uap, rlp), &alim, sizeof(struct rlimit));
332 	if (error) {
333 		return error;
334 	}
335 	return dosetrlimit(l, l->l_proc, which, &alim);
336 }
337 
338 int
339 dosetrlimit(struct lwp *l, struct proc *p, int which, struct rlimit *limp)
340 {
341 	struct rlimit *alimp;
342 	int error;
343 
344 	if ((u_int)which >= RLIM_NLIMITS)
345 		return EINVAL;
346 
347 	if (limp->rlim_cur > limp->rlim_max) {
348 		/*
349 		 * This is programming error. According to SUSv2, we should
350 		 * return error in this case.
351 		 */
352 		return EINVAL;
353 	}
354 
355 	alimp = &p->p_rlimit[which];
356 	/* if we don't change the value, no need to limcopy() */
357 	if (limp->rlim_cur == alimp->rlim_cur &&
358 	    limp->rlim_max == alimp->rlim_max)
359 		return 0;
360 
361 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
362 	    p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_SET), limp, KAUTH_ARG(which));
363 	if (error)
364 		return error;
365 
366 	lim_privatise(p);
367 	/* p->p_limit is now unchangeable */
368 	alimp = &p->p_rlimit[which];
369 
370 	switch (which) {
371 
372 	case RLIMIT_DATA:
373 		if (limp->rlim_cur > maxdmap)
374 			limp->rlim_cur = maxdmap;
375 		if (limp->rlim_max > maxdmap)
376 			limp->rlim_max = maxdmap;
377 		break;
378 
379 	case RLIMIT_STACK:
380 		if (limp->rlim_cur > maxsmap)
381 			limp->rlim_cur = maxsmap;
382 		if (limp->rlim_max > maxsmap)
383 			limp->rlim_max = maxsmap;
384 
385 		/*
386 		 * Return EINVAL if the new stack size limit is lower than
387 		 * current usage. Otherwise, the process would get SIGSEGV the
388 		 * moment it would try to access anything on its current stack.
389 		 * This conforms to SUSv2.
390 		 */
391 		if (btoc(limp->rlim_cur) < p->p_vmspace->vm_ssize ||
392 		    btoc(limp->rlim_max) < p->p_vmspace->vm_ssize) {
393 			return EINVAL;
394 		}
395 
396 		/*
397 		 * Stack is allocated to the max at exec time with
398 		 * only "rlim_cur" bytes accessible (In other words,
399 		 * allocates stack dividing two contiguous regions at
400 		 * "rlim_cur" bytes boundary).
401 		 *
402 		 * Since allocation is done in terms of page, roundup
403 		 * "rlim_cur" (otherwise, contiguous regions
404 		 * overlap).  If stack limit is going up make more
405 		 * accessible, if going down make inaccessible.
406 		 */
407 		limp->rlim_max = round_page(limp->rlim_max);
408 		limp->rlim_cur = round_page(limp->rlim_cur);
409 		if (limp->rlim_cur != alimp->rlim_cur) {
410 			vaddr_t addr;
411 			vsize_t size;
412 			vm_prot_t prot;
413 			char *base, *tmp;
414 
415 			base = p->p_vmspace->vm_minsaddr;
416 			if (limp->rlim_cur > alimp->rlim_cur) {
417 				prot = VM_PROT_READ | VM_PROT_WRITE;
418 				size = limp->rlim_cur - alimp->rlim_cur;
419 				tmp = STACK_GROW(base, alimp->rlim_cur);
420 			} else {
421 				prot = VM_PROT_NONE;
422 				size = alimp->rlim_cur - limp->rlim_cur;
423 				tmp = STACK_GROW(base, limp->rlim_cur);
424 			}
425 			addr = (vaddr_t)STACK_ALLOC(tmp, size);
426 			(void) uvm_map_protect(&p->p_vmspace->vm_map,
427 			    addr, addr + size, prot, false);
428 		}
429 		break;
430 
431 	case RLIMIT_NOFILE:
432 		if (limp->rlim_cur > maxfiles)
433 			limp->rlim_cur = maxfiles;
434 		if (limp->rlim_max > maxfiles)
435 			limp->rlim_max = maxfiles;
436 		break;
437 
438 	case RLIMIT_NPROC:
439 		if (limp->rlim_cur > maxproc)
440 			limp->rlim_cur = maxproc;
441 		if (limp->rlim_max > maxproc)
442 			limp->rlim_max = maxproc;
443 		break;
444 
445 	case RLIMIT_NTHR:
446 		if (limp->rlim_cur > maxlwp)
447 			limp->rlim_cur = maxlwp;
448 		if (limp->rlim_max > maxlwp)
449 			limp->rlim_max = maxlwp;
450 		break;
451 	}
452 
453 	mutex_enter(&p->p_limit->pl_lock);
454 	*alimp = *limp;
455 	mutex_exit(&p->p_limit->pl_lock);
456 	return 0;
457 }
458 
459 int
460 sys_getrlimit(struct lwp *l, const struct sys_getrlimit_args *uap,
461     register_t *retval)
462 {
463 	/* {
464 		syscallarg(int) which;
465 		syscallarg(struct rlimit *) rlp;
466 	} */
467 	struct proc *p = l->l_proc;
468 	int which = SCARG(uap, which);
469 	struct rlimit rl;
470 
471 	if ((u_int)which >= RLIM_NLIMITS)
472 		return EINVAL;
473 
474 	mutex_enter(p->p_lock);
475 	memcpy(&rl, &p->p_rlimit[which], sizeof(rl));
476 	mutex_exit(p->p_lock);
477 
478 	return copyout(&rl, SCARG(uap, rlp), sizeof(rl));
479 }
480 
481 /*
482  * Transform the running time and tick information in proc p into user,
483  * system, and interrupt time usage.
484  *
485  * Should be called with p->p_lock held unless called from exit1().
486  */
487 void
488 calcru(struct proc *p, struct timeval *up, struct timeval *sp,
489     struct timeval *ip, struct timeval *rp)
490 {
491 	uint64_t u, st, ut, it, tot, dt;
492 	struct lwp *l;
493 	struct bintime tm;
494 	struct timeval tv;
495 
496 	KASSERT(p->p_stat == SDEAD || mutex_owned(p->p_lock));
497 
498 	mutex_spin_enter(&p->p_stmutex);
499 	st = p->p_sticks;
500 	ut = p->p_uticks;
501 	it = p->p_iticks;
502 	mutex_spin_exit(&p->p_stmutex);
503 
504 	tm = p->p_rtime;
505 
506 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
507 		lwp_lock(l);
508 		bintime_add(&tm, &l->l_rtime);
509 		if ((l->l_pflag & LP_RUNNING) != 0 &&
510 		    (l->l_pflag & (LP_INTR | LP_TIMEINTR)) != LP_INTR) {
511 			struct bintime diff;
512 			/*
513 			 * Adjust for the current time slice.  This is
514 			 * actually fairly important since the error
515 			 * here is on the order of a time quantum,
516 			 * which is much greater than the sampling
517 			 * error.
518 			 */
519 			binuptime(&diff);
520 			membar_consumer(); /* for softint_dispatch() */
521 			bintime_sub(&diff, &l->l_stime);
522 			bintime_add(&tm, &diff);
523 		}
524 		lwp_unlock(l);
525 	}
526 
527 	tot = st + ut + it;
528 	bintime2timeval(&tm, &tv);
529 	u = (uint64_t)tv.tv_sec * 1000000ul + tv.tv_usec;
530 
531 	if (tot == 0) {
532 		/* No ticks, so can't use to share time out, split 50-50 */
533 		st = ut = u / 2;
534 	} else {
535 		st = (u * st) / tot;
536 		ut = (u * ut) / tot;
537 	}
538 
539 	/*
540 	 * Try to avoid lying to the users (too much)
541 	 *
542 	 * Of course, user/sys time are based on sampling (ie: statistics)
543 	 * so that would be impossible, but convincing the mark
544 	 * that we have used less ?time this call than we had
545 	 * last time, is beyond reasonable...  (the con fails!)
546 	 *
547 	 * Note that since actual used time cannot decrease, either
548 	 * utime or stime (or both) must be greater now than last time
549 	 * (or both the same) - if one seems to have decreased, hold
550 	 * it constant and steal the necessary bump from the other
551 	 * which must have increased.
552 	 */
553 	if (p->p_xutime > ut) {
554 		dt = p->p_xutime - ut;
555 		st -= uimin(dt, st);
556 		ut = p->p_xutime;
557 	} else if (p->p_xstime > st) {
558 		dt = p->p_xstime - st;
559 		ut -= uimin(dt, ut);
560 		st = p->p_xstime;
561 	}
562 
563 	if (sp != NULL) {
564 		p->p_xstime = st;
565 		sp->tv_sec = st / 1000000;
566 		sp->tv_usec = st % 1000000;
567 	}
568 	if (up != NULL) {
569 		p->p_xutime = ut;
570 		up->tv_sec = ut / 1000000;
571 		up->tv_usec = ut % 1000000;
572 	}
573 	if (ip != NULL) {
574 		if (it != 0)		/* it != 0 --> tot != 0 */
575 			it = (u * it) / tot;
576 		ip->tv_sec = it / 1000000;
577 		ip->tv_usec = it % 1000000;
578 	}
579 	if (rp != NULL) {
580 		*rp = tv;
581 	}
582 }
583 
584 int
585 sys___getrusage50(struct lwp *l, const struct sys___getrusage50_args *uap,
586     register_t *retval)
587 {
588 	/* {
589 		syscallarg(int) who;
590 		syscallarg(struct rusage *) rusage;
591 	} */
592 	int error;
593 	struct rusage ru;
594 	struct proc *p = l->l_proc;
595 
596 	error = getrusage1(p, SCARG(uap, who), &ru);
597 	if (error != 0)
598 		return error;
599 
600 	return copyout(&ru, SCARG(uap, rusage), sizeof(ru));
601 }
602 
603 int
604 getrusage1(struct proc *p, int who, struct rusage *ru) {
605 
606 	switch (who) {
607 	case RUSAGE_SELF:
608 		mutex_enter(p->p_lock);
609 		ruspace(p);
610 		memcpy(ru, &p->p_stats->p_ru, sizeof(*ru));
611 		calcru(p, &ru->ru_utime, &ru->ru_stime, NULL, NULL);
612 		rulwps(p, ru);
613 		mutex_exit(p->p_lock);
614 		break;
615 	case RUSAGE_CHILDREN:
616 		mutex_enter(p->p_lock);
617 		memcpy(ru, &p->p_stats->p_cru, sizeof(*ru));
618 		mutex_exit(p->p_lock);
619 		break;
620 	default:
621 		return EINVAL;
622 	}
623 
624 	return 0;
625 }
626 
627 void
628 ruspace(struct proc *p)
629 {
630 	struct vmspace *vm = p->p_vmspace;
631 	struct rusage *ru = &p->p_stats->p_ru;
632 
633 	ru->ru_ixrss = vm->vm_tsize << (PAGE_SHIFT - 10);
634 	ru->ru_idrss = vm->vm_dsize << (PAGE_SHIFT - 10);
635 	ru->ru_isrss = vm->vm_ssize << (PAGE_SHIFT - 10);
636 #ifdef __HAVE_NO_PMAP_STATS
637 	/* We don't keep track of the max so we get the current */
638 	ru->ru_maxrss = vm_resident_count(vm) << (PAGE_SHIFT - 10);
639 #else
640 	ru->ru_maxrss = vm->vm_rssmax << (PAGE_SHIFT - 10);
641 #endif
642 }
643 
644 void
645 ruadd(struct rusage *ru, struct rusage *ru2)
646 {
647 	long *ip, *ip2;
648 	int i;
649 
650 	timeradd(&ru->ru_utime, &ru2->ru_utime, &ru->ru_utime);
651 	timeradd(&ru->ru_stime, &ru2->ru_stime, &ru->ru_stime);
652 	if (ru->ru_maxrss < ru2->ru_maxrss)
653 		ru->ru_maxrss = ru2->ru_maxrss;
654 	ip = &ru->ru_first; ip2 = &ru2->ru_first;
655 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
656 		*ip++ += *ip2++;
657 }
658 
659 void
660 rulwps(proc_t *p, struct rusage *ru)
661 {
662 	lwp_t *l;
663 
664 	KASSERT(mutex_owned(p->p_lock));
665 
666 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
667 		ruadd(ru, &l->l_ru);
668 		ru->ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
669 		ru->ru_nivcsw += l->l_nivcsw;
670 	}
671 }
672 
673 /*
674  * lim_copy: make a copy of the plimit structure.
675  *
676  * We use copy-on-write after fork, and copy when a limit is changed.
677  */
678 struct plimit *
679 lim_copy(struct plimit *lim)
680 {
681 	struct plimit *newlim;
682 	char *corename;
683 	size_t alen, len;
684 
685 	newlim = pool_cache_get(plimit_cache, PR_WAITOK);
686 	mutex_init(&newlim->pl_lock, MUTEX_DEFAULT, IPL_NONE);
687 	newlim->pl_writeable = false;
688 	newlim->pl_refcnt = 1;
689 	newlim->pl_sv_limit = NULL;
690 
691 	mutex_enter(&lim->pl_lock);
692 	memcpy(newlim->pl_rlimit, lim->pl_rlimit,
693 	    sizeof(struct rlimit) * RLIM_NLIMITS);
694 
695 	/*
696 	 * Note: the common case is a use of default core name.
697 	 */
698 	alen = 0;
699 	corename = NULL;
700 	for (;;) {
701 		if (lim->pl_corename == defcorename) {
702 			newlim->pl_corename = defcorename;
703 			newlim->pl_cnlen = 0;
704 			break;
705 		}
706 		len = lim->pl_cnlen;
707 		if (len == alen) {
708 			newlim->pl_corename = corename;
709 			newlim->pl_cnlen = len;
710 			memcpy(corename, lim->pl_corename, len);
711 			corename = NULL;
712 			break;
713 		}
714 		mutex_exit(&lim->pl_lock);
715 		if (corename) {
716 			kmem_free(corename, alen);
717 		}
718 		alen = len;
719 		corename = kmem_alloc(alen, KM_SLEEP);
720 		mutex_enter(&lim->pl_lock);
721 	}
722 	mutex_exit(&lim->pl_lock);
723 
724 	if (corename) {
725 		kmem_free(corename, alen);
726 	}
727 	return newlim;
728 }
729 
730 void
731 lim_addref(struct plimit *lim)
732 {
733 	atomic_inc_uint(&lim->pl_refcnt);
734 }
735 
736 /*
737  * lim_privatise: give a process its own private plimit structure.
738  */
739 void
740 lim_privatise(proc_t *p)
741 {
742 	struct plimit *lim = p->p_limit, *newlim;
743 
744 	if (lim->pl_writeable) {
745 		return;
746 	}
747 
748 	newlim = lim_copy(lim);
749 
750 	mutex_enter(p->p_lock);
751 	if (p->p_limit->pl_writeable) {
752 		/* Other thread won the race. */
753 		mutex_exit(p->p_lock);
754 		lim_free(newlim);
755 		return;
756 	}
757 
758 	/*
759 	 * Since p->p_limit can be accessed without locked held,
760 	 * old limit structure must not be deleted yet.
761 	 */
762 	newlim->pl_sv_limit = p->p_limit;
763 	newlim->pl_writeable = true;
764 	p->p_limit = newlim;
765 	mutex_exit(p->p_lock);
766 }
767 
768 void
769 lim_setcorename(proc_t *p, char *name, size_t len)
770 {
771 	struct plimit *lim;
772 	char *oname;
773 	size_t olen;
774 
775 	lim_privatise(p);
776 	lim = p->p_limit;
777 
778 	mutex_enter(&lim->pl_lock);
779 	oname = lim->pl_corename;
780 	olen = lim->pl_cnlen;
781 	lim->pl_corename = name;
782 	lim->pl_cnlen = len;
783 	mutex_exit(&lim->pl_lock);
784 
785 	if (oname != defcorename) {
786 		kmem_free(oname, olen);
787 	}
788 }
789 
790 void
791 lim_free(struct plimit *lim)
792 {
793 	struct plimit *sv_lim;
794 
795 	do {
796 		membar_release();
797 		if (atomic_dec_uint_nv(&lim->pl_refcnt) > 0) {
798 			return;
799 		}
800 		membar_acquire();
801 		if (lim->pl_corename != defcorename) {
802 			kmem_free(lim->pl_corename, lim->pl_cnlen);
803 		}
804 		sv_lim = lim->pl_sv_limit;
805 		mutex_destroy(&lim->pl_lock);
806 		pool_cache_put(plimit_cache, lim);
807 	} while ((lim = sv_lim) != NULL);
808 }
809 
810 struct pstats *
811 pstatscopy(struct pstats *ps)
812 {
813 	struct pstats *nps;
814 	size_t len;
815 
816 	nps = pool_cache_get(pstats_cache, PR_WAITOK);
817 
818 	len = (char *)&nps->pstat_endzero - (char *)&nps->pstat_startzero;
819 	memset(&nps->pstat_startzero, 0, len);
820 
821 	len = (char *)&nps->pstat_endcopy - (char *)&nps->pstat_startcopy;
822 	memcpy(&nps->pstat_startcopy, &ps->pstat_startcopy, len);
823 
824 	return nps;
825 }
826 
827 void
828 pstatsfree(struct pstats *ps)
829 {
830 
831 	pool_cache_put(pstats_cache, ps);
832 }
833 
834 /*
835  * sysctl_proc_findproc: a routine for sysctl proc subtree helpers that
836  * need to pick a valid process by PID.
837  *
838  * => Hold a reference on the process, on success.
839  */
840 static int
841 sysctl_proc_findproc(lwp_t *l, pid_t pid, proc_t **p2)
842 {
843 	proc_t *p;
844 	int error;
845 
846 	if (pid == PROC_CURPROC) {
847 		p = l->l_proc;
848 	} else {
849 		mutex_enter(&proc_lock);
850 		p = proc_find(pid);
851 		if (p == NULL) {
852 			mutex_exit(&proc_lock);
853 			return ESRCH;
854 		}
855 	}
856 	error = rw_tryenter(&p->p_reflock, RW_READER) ? 0 : EBUSY;
857 	if (pid != PROC_CURPROC) {
858 		mutex_exit(&proc_lock);
859 	}
860 	*p2 = p;
861 	return error;
862 }
863 
864 /*
865  * sysctl_proc_paxflags: helper routine to get process's paxctl flags
866  */
867 static int
868 sysctl_proc_paxflags(SYSCTLFN_ARGS)
869 {
870 	struct proc *p;
871 	struct sysctlnode node;
872 	int paxflags;
873 	int error;
874 
875 	/* First, validate the request. */
876 	if (namelen != 0 || name[-1] != PROC_PID_PAXFLAGS)
877 		return EINVAL;
878 
879 	/* Find the process.  Hold a reference (p_reflock), if found. */
880 	error = sysctl_proc_findproc(l, (pid_t)name[-2], &p);
881 	if (error)
882 		return error;
883 
884 	/* XXX-elad */
885 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_CANSEE, p,
886 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
887 	if (error) {
888 		rw_exit(&p->p_reflock);
889 		return error;
890 	}
891 
892 	/* Retrieve the limits. */
893 	node = *rnode;
894 	paxflags = p->p_pax;
895 	node.sysctl_data = &paxflags;
896 
897 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
898 
899 	/* If attempting to write new value, it's an error */
900 	if (error == 0 && newp != NULL)
901 		error = EACCES;
902 
903 	rw_exit(&p->p_reflock);
904 	return error;
905 }
906 
907 /*
908  * sysctl_proc_corename: helper routine to get or set the core file name
909  * for a process specified by PID.
910  */
911 static int
912 sysctl_proc_corename(SYSCTLFN_ARGS)
913 {
914 	struct proc *p;
915 	struct plimit *lim;
916 	char *cnbuf, *cname;
917 	struct sysctlnode node;
918 	size_t len;
919 	int error;
920 
921 	/* First, validate the request. */
922 	if (namelen != 0 || name[-1] != PROC_PID_CORENAME)
923 		return EINVAL;
924 
925 	/* Find the process.  Hold a reference (p_reflock), if found. */
926 	error = sysctl_proc_findproc(l, (pid_t)name[-2], &p);
927 	if (error)
928 		return error;
929 
930 	/* XXX-elad */
931 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_CANSEE, p,
932 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
933 	if (error) {
934 		rw_exit(&p->p_reflock);
935 		return error;
936 	}
937 
938 	cnbuf = PNBUF_GET();
939 
940 	if (oldp) {
941 		/* Get case: copy the core name into the buffer. */
942 		error = kauth_authorize_process(l->l_cred,
943 		    KAUTH_PROCESS_CORENAME, p,
944 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CORENAME_GET), NULL, NULL);
945 		if (error) {
946 			goto done;
947 		}
948 		lim = p->p_limit;
949 		mutex_enter(&lim->pl_lock);
950 		strlcpy(cnbuf, lim->pl_corename, MAXPATHLEN);
951 		mutex_exit(&lim->pl_lock);
952 	}
953 
954 	node = *rnode;
955 	node.sysctl_data = cnbuf;
956 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
957 
958 	/* Return if error, or if caller is only getting the core name. */
959 	if (error || newp == NULL) {
960 		goto done;
961 	}
962 
963 	/*
964 	 * Set case.  Check permission and then validate new core name.
965 	 * It must be either "core", "/core", or end in ".core".
966 	 */
967 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_CORENAME,
968 	    p, KAUTH_ARG(KAUTH_REQ_PROCESS_CORENAME_SET), cnbuf, NULL);
969 	if (error) {
970 		goto done;
971 	}
972 	len = strlen(cnbuf);
973 	if ((len < 4 || strcmp(cnbuf + len - 4, "core") != 0) ||
974 	    (len > 4 && cnbuf[len - 5] != '/' && cnbuf[len - 5] != '.')) {
975 		error = EINVAL;
976 		goto done;
977 	}
978 
979 	/* Allocate, copy and set the new core name for plimit structure. */
980 	cname = kmem_alloc(++len, KM_NOSLEEP);
981 	if (cname == NULL) {
982 		error = ENOMEM;
983 		goto done;
984 	}
985 	memcpy(cname, cnbuf, len);
986 	lim_setcorename(p, cname, len);
987 done:
988 	rw_exit(&p->p_reflock);
989 	PNBUF_PUT(cnbuf);
990 	return error;
991 }
992 
993 /*
994  * sysctl_proc_stop: helper routine for checking/setting the stop flags.
995  */
996 static int
997 sysctl_proc_stop(SYSCTLFN_ARGS)
998 {
999 	struct proc *p;
1000 	int isset, flag, error = 0;
1001 	struct sysctlnode node;
1002 
1003 	if (namelen != 0)
1004 		return EINVAL;
1005 
1006 	/* Find the process.  Hold a reference (p_reflock), if found. */
1007 	error = sysctl_proc_findproc(l, (pid_t)name[-2], &p);
1008 	if (error)
1009 		return error;
1010 
1011 	/* XXX-elad */
1012 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_CANSEE, p,
1013 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
1014 	if (error) {
1015 		goto out;
1016 	}
1017 
1018 	/* Determine the flag. */
1019 	switch (rnode->sysctl_num) {
1020 	case PROC_PID_STOPFORK:
1021 		flag = PS_STOPFORK;
1022 		break;
1023 	case PROC_PID_STOPEXEC:
1024 		flag = PS_STOPEXEC;
1025 		break;
1026 	case PROC_PID_STOPEXIT:
1027 		flag = PS_STOPEXIT;
1028 		break;
1029 	default:
1030 		error = EINVAL;
1031 		goto out;
1032 	}
1033 	isset = (p->p_flag & flag) ? 1 : 0;
1034 	node = *rnode;
1035 	node.sysctl_data = &isset;
1036 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1037 
1038 	/* Return if error, or if callers is only getting the flag. */
1039 	if (error || newp == NULL) {
1040 		goto out;
1041 	}
1042 
1043 	/* Check if caller can set the flags. */
1044 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_STOPFLAG,
1045 	    p, KAUTH_ARG(flag), NULL, NULL);
1046 	if (error) {
1047 		goto out;
1048 	}
1049 	mutex_enter(p->p_lock);
1050 	if (isset) {
1051 		p->p_sflag |= flag;
1052 	} else {
1053 		p->p_sflag &= ~flag;
1054 	}
1055 	mutex_exit(p->p_lock);
1056 out:
1057 	rw_exit(&p->p_reflock);
1058 	return error;
1059 }
1060 
1061 /*
1062  * sysctl_proc_plimit: helper routine to get/set rlimits of a process.
1063  */
1064 static int
1065 sysctl_proc_plimit(SYSCTLFN_ARGS)
1066 {
1067 	struct proc *p;
1068 	u_int limitno;
1069 	int which, error = 0;
1070         struct rlimit alim;
1071 	struct sysctlnode node;
1072 
1073 	if (namelen != 0)
1074 		return EINVAL;
1075 
1076 	which = name[-1];
1077 	if (which != PROC_PID_LIMIT_TYPE_SOFT &&
1078 	    which != PROC_PID_LIMIT_TYPE_HARD)
1079 		return EINVAL;
1080 
1081 	limitno = name[-2] - 1;
1082 	if (limitno >= RLIM_NLIMITS)
1083 		return EINVAL;
1084 
1085 	if (name[-3] != PROC_PID_LIMIT)
1086 		return EINVAL;
1087 
1088 	/* Find the process.  Hold a reference (p_reflock), if found. */
1089 	error = sysctl_proc_findproc(l, (pid_t)name[-4], &p);
1090 	if (error)
1091 		return error;
1092 
1093 	/* XXX-elad */
1094 	error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_CANSEE, p,
1095 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
1096 	if (error)
1097 		goto out;
1098 
1099 	/* Check if caller can retrieve the limits. */
1100 	if (newp == NULL) {
1101 		error = kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
1102 		    p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_GET), &alim,
1103 		    KAUTH_ARG(which));
1104 		if (error)
1105 			goto out;
1106 	}
1107 
1108 	/* Retrieve the limits. */
1109 	node = *rnode;
1110 	memcpy(&alim, &p->p_rlimit[limitno], sizeof(alim));
1111 	if (which == PROC_PID_LIMIT_TYPE_HARD) {
1112 		node.sysctl_data = &alim.rlim_max;
1113 	} else {
1114 		node.sysctl_data = &alim.rlim_cur;
1115 	}
1116 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1117 
1118 	/* Return if error, or if we are only retrieving the limits. */
1119 	if (error || newp == NULL) {
1120 		goto out;
1121 	}
1122 	error = dosetrlimit(l, p, limitno, &alim);
1123 out:
1124 	rw_exit(&p->p_reflock);
1125 	return error;
1126 }
1127 
1128 /*
1129  * Setup sysctl nodes.
1130  */
1131 static void
1132 sysctl_proc_setup(void)
1133 {
1134 
1135 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1136 		       CTLFLAG_PERMANENT|CTLFLAG_ANYNUMBER,
1137 		       CTLTYPE_NODE, "curproc",
1138 		       SYSCTL_DESCR("Per-process settings"),
1139 		       NULL, 0, NULL, 0,
1140 		       CTL_PROC, PROC_CURPROC, CTL_EOL);
1141 
1142 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1143 		       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
1144 		       CTLTYPE_INT, "paxflags",
1145 		       SYSCTL_DESCR("Process PAX control flags"),
1146 		       sysctl_proc_paxflags, 0, NULL, 0,
1147 		       CTL_PROC, PROC_CURPROC, PROC_PID_PAXFLAGS, CTL_EOL);
1148 
1149 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1150 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE,
1151 		       CTLTYPE_STRING, "corename",
1152 		       SYSCTL_DESCR("Core file name"),
1153 		       sysctl_proc_corename, 0, NULL, MAXPATHLEN,
1154 		       CTL_PROC, PROC_CURPROC, PROC_PID_CORENAME, CTL_EOL);
1155 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1156 		       CTLFLAG_PERMANENT,
1157 		       CTLTYPE_NODE, "rlimit",
1158 		       SYSCTL_DESCR("Process limits"),
1159 		       NULL, 0, NULL, 0,
1160 		       CTL_PROC, PROC_CURPROC, PROC_PID_LIMIT, CTL_EOL);
1161 
1162 #define create_proc_plimit(s, n) do {					\
1163 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,			\
1164 		       CTLFLAG_PERMANENT,				\
1165 		       CTLTYPE_NODE, s,					\
1166 		       SYSCTL_DESCR("Process " s " limits"),		\
1167 		       NULL, 0, NULL, 0,				\
1168 		       CTL_PROC, PROC_CURPROC, PROC_PID_LIMIT, n,	\
1169 		       CTL_EOL);					\
1170 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,			\
1171 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE, \
1172 		       CTLTYPE_QUAD, "soft",				\
1173 		       SYSCTL_DESCR("Process soft " s " limit"),	\
1174 		       sysctl_proc_plimit, 0, NULL, 0,			\
1175 		       CTL_PROC, PROC_CURPROC, PROC_PID_LIMIT, n,	\
1176 		       PROC_PID_LIMIT_TYPE_SOFT, CTL_EOL);		\
1177 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,			\
1178 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE, \
1179 		       CTLTYPE_QUAD, "hard",				\
1180 		       SYSCTL_DESCR("Process hard " s " limit"),	\
1181 		       sysctl_proc_plimit, 0, NULL, 0,			\
1182 		       CTL_PROC, PROC_CURPROC, PROC_PID_LIMIT, n,	\
1183 		       PROC_PID_LIMIT_TYPE_HARD, CTL_EOL);		\
1184 	} while (0/*CONSTCOND*/)
1185 
1186 	create_proc_plimit("cputime",		PROC_PID_LIMIT_CPU);
1187 	create_proc_plimit("filesize",		PROC_PID_LIMIT_FSIZE);
1188 	create_proc_plimit("datasize",		PROC_PID_LIMIT_DATA);
1189 	create_proc_plimit("stacksize",		PROC_PID_LIMIT_STACK);
1190 	create_proc_plimit("coredumpsize",	PROC_PID_LIMIT_CORE);
1191 	create_proc_plimit("memoryuse",		PROC_PID_LIMIT_RSS);
1192 	create_proc_plimit("memorylocked",	PROC_PID_LIMIT_MEMLOCK);
1193 	create_proc_plimit("maxproc",		PROC_PID_LIMIT_NPROC);
1194 	create_proc_plimit("descriptors",	PROC_PID_LIMIT_NOFILE);
1195 	create_proc_plimit("sbsize",		PROC_PID_LIMIT_SBSIZE);
1196 	create_proc_plimit("vmemoryuse",	PROC_PID_LIMIT_AS);
1197 	create_proc_plimit("maxlwp",		PROC_PID_LIMIT_NTHR);
1198 
1199 #undef create_proc_plimit
1200 
1201 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1202 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE,
1203 		       CTLTYPE_INT, "stopfork",
1204 		       SYSCTL_DESCR("Stop process at fork(2)"),
1205 		       sysctl_proc_stop, 0, NULL, 0,
1206 		       CTL_PROC, PROC_CURPROC, PROC_PID_STOPFORK, CTL_EOL);
1207 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1208 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE,
1209 		       CTLTYPE_INT, "stopexec",
1210 		       SYSCTL_DESCR("Stop process at execve(2)"),
1211 		       sysctl_proc_stop, 0, NULL, 0,
1212 		       CTL_PROC, PROC_CURPROC, PROC_PID_STOPEXEC, CTL_EOL);
1213 	sysctl_createv(&proc_sysctllog, 0, NULL, NULL,
1214 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE|CTLFLAG_ANYWRITE,
1215 		       CTLTYPE_INT, "stopexit",
1216 		       SYSCTL_DESCR("Stop process before completing exit"),
1217 		       sysctl_proc_stop, 0, NULL, 0,
1218 		       CTL_PROC, PROC_CURPROC, PROC_PID_STOPEXIT, CTL_EOL);
1219 }
1220