xref: /dflybsd-src/sys/kern/kern_resource.c (revision 9b5a99654f820c32a9fb8fa9bae3c3b12fe27a0b)
1 /*-
2  * Copyright (c) 1982, 1986, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_resource.c	8.5 (Berkeley) 1/21/94
39  * $FreeBSD: src/sys/kern/kern_resource.c,v 1.55.2.5 2001/11/03 01:41:08 ps Exp $
40  * $DragonFly: src/sys/kern/kern_resource.c,v 1.32 2007/05/03 23:04:31 dillon Exp $
41  */
42 
43 #include "opt_compat.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/file.h>
49 #include <sys/kern_syscall.h>
50 #include <sys/kernel.h>
51 #include <sys/resourcevar.h>
52 #include <sys/malloc.h>
53 #include <sys/proc.h>
54 #include <sys/time.h>
55 #include <sys/lockf.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_param.h>
59 #include <sys/lock.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_map.h>
62 
63 #include <sys/thread2.h>
64 
65 static int donice (struct proc *chgp, int n);
66 
67 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
68 #define	UIHASH(uid)	(&uihashtbl[(uid) & uihash])
69 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
70 static u_long uihash;		/* size of hash table - 1 */
71 
72 static struct uidinfo	*uicreate (uid_t uid);
73 static struct uidinfo	*uilookup (uid_t uid);
74 
75 /*
76  * Resource controls and accounting.
77  */
78 
79 struct getpriority_info {
80 	int low;
81 	int who;
82 };
83 
84 static int getpriority_callback(struct proc *p, void *data);
85 
86 int
87 sys_getpriority(struct getpriority_args *uap)
88 {
89 	struct getpriority_info info;
90 	struct proc *curp = curproc;
91 	struct proc *p;
92 	int low = PRIO_MAX + 1;
93 
94 	switch (uap->which) {
95 	case PRIO_PROCESS:
96 		if (uap->who == 0)
97 			p = curp;
98 		else
99 			p = pfind(uap->who);
100 		if (p == 0)
101 			break;
102 		if (!PRISON_CHECK(curp->p_ucred, p->p_ucred))
103 			break;
104 		low = p->p_nice;
105 		break;
106 
107 	case PRIO_PGRP:
108 	{
109 		struct pgrp *pg;
110 
111 		if (uap->who == 0)
112 			pg = curp->p_pgrp;
113 		else if ((pg = pgfind(uap->who)) == NULL)
114 			break;
115 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
116 			if ((PRISON_CHECK(curp->p_ucred, p->p_ucred) && p->p_nice < low))
117 				low = p->p_nice;
118 		}
119 		break;
120 	}
121 	case PRIO_USER:
122 		if (uap->who == 0)
123 			uap->who = curp->p_ucred->cr_uid;
124 		info.low = low;
125 		info.who = uap->who;
126 		allproc_scan(getpriority_callback, &info);
127 		low = info.low;
128 		break;
129 
130 	default:
131 		return (EINVAL);
132 	}
133 	if (low == PRIO_MAX + 1)
134 		return (ESRCH);
135 	uap->sysmsg_result = low;
136 	return (0);
137 }
138 
139 /*
140  * Figure out the current lowest nice priority for processes owned
141  * by the specified user.
142  */
143 static
144 int
145 getpriority_callback(struct proc *p, void *data)
146 {
147 	struct getpriority_info *info = data;
148 
149 	if (PRISON_CHECK(curproc->p_ucred, p->p_ucred) &&
150 	    p->p_ucred->cr_uid == info->who &&
151 	    p->p_nice < info->low) {
152 		info->low = p->p_nice;
153 	}
154 	return(0);
155 }
156 
157 struct setpriority_info {
158 	int prio;
159 	int who;
160 	int error;
161 	int found;
162 };
163 
164 static int setpriority_callback(struct proc *p, void *data);
165 
166 int
167 sys_setpriority(struct setpriority_args *uap)
168 {
169 	struct setpriority_info info;
170 	struct proc *curp = curproc;
171 	struct proc *p;
172 	int found = 0, error = 0;
173 
174 	switch (uap->which) {
175 	case PRIO_PROCESS:
176 		if (uap->who == 0)
177 			p = curp;
178 		else
179 			p = pfind(uap->who);
180 		if (p == 0)
181 			break;
182 		if (!PRISON_CHECK(curp->p_ucred, p->p_ucred))
183 			break;
184 		error = donice(p, uap->prio);
185 		found++;
186 		break;
187 
188 	case PRIO_PGRP:
189 	{
190 		struct pgrp *pg;
191 
192 		if (uap->who == 0)
193 			pg = curp->p_pgrp;
194 		else if ((pg = pgfind(uap->who)) == NULL)
195 			break;
196 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
197 			if (PRISON_CHECK(curp->p_ucred, p->p_ucred)) {
198 				error = donice(p, uap->prio);
199 				found++;
200 			}
201 		}
202 		break;
203 	}
204 	case PRIO_USER:
205 		if (uap->who == 0)
206 			uap->who = curp->p_ucred->cr_uid;
207 		info.prio = uap->prio;
208 		info.who = uap->who;
209 		info.error = 0;
210 		info.found = 0;
211 		allproc_scan(setpriority_callback, &info);
212 		error = info.error;
213 		found = info.found;
214 		break;
215 
216 	default:
217 		return (EINVAL);
218 	}
219 	if (found == 0)
220 		return (ESRCH);
221 	return (error);
222 }
223 
224 static
225 int
226 setpriority_callback(struct proc *p, void *data)
227 {
228 	struct setpriority_info *info = data;
229 	int error;
230 
231 	if (p->p_ucred->cr_uid == info->who &&
232 	    PRISON_CHECK(curproc->p_ucred, p->p_ucred)) {
233 		error = donice(p, info->prio);
234 		if (error)
235 			info->error = error;
236 		++info->found;
237 	}
238 	return(0);
239 }
240 
241 static int
242 donice(struct proc *chgp, int n)
243 {
244 	struct proc *curp = curproc;
245 	struct ucred *cr = curp->p_ucred;
246 	struct lwp *lp;
247 
248 	if (cr->cr_uid && cr->cr_ruid &&
249 	    cr->cr_uid != chgp->p_ucred->cr_uid &&
250 	    cr->cr_ruid != chgp->p_ucred->cr_uid)
251 		return (EPERM);
252 	if (n > PRIO_MAX)
253 		n = PRIO_MAX;
254 	if (n < PRIO_MIN)
255 		n = PRIO_MIN;
256 	if (n < chgp->p_nice && suser_cred(cr, 0))
257 		return (EACCES);
258 	chgp->p_nice = n;
259 	FOREACH_LWP_IN_PROC(lp, chgp)
260 		chgp->p_usched->resetpriority(lp);
261 	return (0);
262 }
263 
264 int
265 sys_lwp_rtprio(struct lwp_rtprio_args *uap)
266 {
267 	struct proc *p = curproc;
268 	struct lwp *lp;
269 	struct rtprio rtp;
270 	struct ucred *cr = p->p_ucred;
271 	int error;
272 
273 	error = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
274 	if (error)
275 		return error;
276 
277 	if (uap->pid < 0) {
278 		return EINVAL;
279 	} else if (uap->pid == 0) {
280 		/* curproc already loaded on p */
281 	} else {
282 		p = pfind(uap->pid);
283 	}
284 
285 	if (p == 0) {
286 		return ESRCH;
287 	}
288 
289 	if (uap->tid < -1) {
290 		return EINVAL;
291 	} else if (uap->tid == -1) {
292 		/*
293 		 * sadly, tid can be 0 so we can't use 0 here
294 		 * like sys_rtprio()
295 		 */
296 		lp = curthread->td_lwp;
297 	} else {
298 		FOREACH_LWP_IN_PROC(lp, p) {
299 			if (lp->lwp_tid == uap->tid) {
300 				break;
301 			}
302 		}
303 		if (!lp) {
304 			return ESRCH;
305 		}
306 	}
307 
308 	switch (uap->function) {
309 	case RTP_LOOKUP:
310 		return (copyout(&lp->lwp_rtprio, uap->rtp,
311 				sizeof(struct rtprio)));
312 	case RTP_SET:
313 		if (cr->cr_uid && cr->cr_ruid &&
314 		    cr->cr_uid != p->p_ucred->cr_uid &&
315 		    cr->cr_ruid != p->p_ucred->cr_uid) {
316 			return EPERM;
317 		}
318 		/* disallow setting rtprio in most cases if not superuser */
319 		if (suser_cred(cr, 0)) {
320 			/* can't set someone else's */
321 			if (uap->pid) { /* XXX */
322 				return EPERM;
323 			}
324 			/* can't set realtime priority */
325 /*
326  * Realtime priority has to be restricted for reasons which should be
327  * obvious. However, for idle priority, there is a potential for
328  * system deadlock if an idleprio process gains a lock on a resource
329  * that other processes need (and the idleprio process can't run
330  * due to a CPU-bound normal process). Fix me! XXX
331  */
332  			if (RTP_PRIO_IS_REALTIME(rtp.type)) {
333 				return EPERM;
334 			}
335 		}
336 		switch (rtp.type) {
337 #ifdef RTP_PRIO_FIFO
338 		case RTP_PRIO_FIFO:
339 #endif
340 		case RTP_PRIO_REALTIME:
341 		case RTP_PRIO_NORMAL:
342 		case RTP_PRIO_IDLE:
343 			if (rtp.prio > RTP_PRIO_MAX)
344 				return EINVAL;
345 			lp->lwp_rtprio = rtp;
346 			return 0;
347 		default:
348 			return EINVAL;
349 		}
350 	default:
351 		return EINVAL;
352 	}
353 	panic("can't get here");
354 }
355 
356 /*
357  * Set realtime priority
358  */
359 /* ARGSUSED */
360 int
361 sys_rtprio(struct rtprio_args *uap)
362 {
363 	struct proc *curp = curproc;
364 	struct proc *p;
365 	struct lwp *lp;
366 	struct ucred *cr = curp->p_ucred;
367 	struct rtprio rtp;
368 	int error;
369 
370 	error = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
371 	if (error)
372 		return (error);
373 
374 	if (uap->pid == 0)
375 		p = curp;
376 	else
377 		p = pfind(uap->pid);
378 
379 	if (p == 0)
380 		return (ESRCH);
381 
382 	/* XXX lwp */
383 	lp = FIRST_LWP_IN_PROC(p);
384 	switch (uap->function) {
385 	case RTP_LOOKUP:
386 		return (copyout(&lp->lwp_rtprio, uap->rtp, sizeof(struct rtprio)));
387 	case RTP_SET:
388 		if (cr->cr_uid && cr->cr_ruid &&
389 		    cr->cr_uid != p->p_ucred->cr_uid &&
390 		    cr->cr_ruid != p->p_ucred->cr_uid)
391 		        return (EPERM);
392 		/* disallow setting rtprio in most cases if not superuser */
393 		if (suser_cred(cr, 0)) {
394 			/* can't set someone else's */
395 			if (uap->pid)
396 				return (EPERM);
397 			/* can't set realtime priority */
398 /*
399  * Realtime priority has to be restricted for reasons which should be
400  * obvious. However, for idle priority, there is a potential for
401  * system deadlock if an idleprio process gains a lock on a resource
402  * that other processes need (and the idleprio process can't run
403  * due to a CPU-bound normal process). Fix me! XXX
404  */
405  			if (RTP_PRIO_IS_REALTIME(rtp.type))
406 				return (EPERM);
407 		}
408 		switch (rtp.type) {
409 #ifdef RTP_PRIO_FIFO
410 		case RTP_PRIO_FIFO:
411 #endif
412 		case RTP_PRIO_REALTIME:
413 		case RTP_PRIO_NORMAL:
414 		case RTP_PRIO_IDLE:
415 			if (rtp.prio > RTP_PRIO_MAX)
416 				return (EINVAL);
417 			lp->lwp_rtprio = rtp;
418 			return (0);
419 		default:
420 			return (EINVAL);
421 		}
422 
423 	default:
424 		return (EINVAL);
425 	}
426 }
427 
428 int
429 sys_setrlimit(struct __setrlimit_args *uap)
430 {
431 	struct rlimit alim;
432 	int error;
433 
434 	error = copyin(uap->rlp, &alim, sizeof(alim));
435 	if (error)
436 		return (error);
437 
438 	error = kern_setrlimit(uap->which, &alim);
439 
440 	return (error);
441 }
442 
443 int
444 sys_getrlimit(struct __getrlimit_args *uap)
445 {
446 	struct rlimit lim;
447 	int error;
448 
449 	error = kern_getrlimit(uap->which, &lim);
450 
451 	if (error == 0)
452 		error = copyout(&lim, uap->rlp, sizeof(*uap->rlp));
453 	return error;
454 }
455 
456 /*
457  * Transform the running time and tick information in lwp lp's thread into user,
458  * system, and interrupt time usage.
459  *
460  * Since we are limited to statclock tick granularity this is a statisical
461  * calculation which will be correct over the long haul, but should not be
462  * expected to measure fine grained deltas.
463  */
464 void
465 calcru(struct lwp *lp, struct timeval *up, struct timeval *sp)
466 {
467 	struct thread *td = lp->lwp_thread;
468 
469 	/*
470 	 * Calculate at the statclock level.  YYY if the thread is owned by
471 	 * another cpu we need to forward the request to the other cpu, or
472 	 * have a token to interlock the information.
473 	 */
474 	crit_enter();
475 	up->tv_sec = td->td_uticks / 1000000;
476 	up->tv_usec = td->td_uticks % 1000000;
477 	sp->tv_sec = td->td_sticks / 1000000;
478 	sp->tv_usec = td->td_sticks % 1000000;
479 	crit_exit();
480 }
481 
482 /*
483  * Aggregate resource statistics of all lwps of a process.
484  *
485  * proc.p_ru keeps track of all statistics directly related to a proc.  This
486  * consists of RSS usage and nswap information and aggregate numbers for all
487  * former lwps of this proc.
488  *
489  * proc.p_cru is the sum of all stats of reaped children.
490  *
491  * lwp.lwp_ru contains the stats directly related to one specific lwp, meaning
492  * packet, scheduler switch or page fault counts, etc.  This information gets
493  * added to lwp.lwp_proc.p_ru when the lwp exits.
494  */
495 void
496 calcru_proc(struct proc *p, struct rusage *ru)
497 {
498 	struct timeval upt, spt;
499 	long *rip1, *rip2;
500 	struct lwp *lp;
501 
502 	*ru = p->p_ru;
503 
504 	FOREACH_LWP_IN_PROC(lp, p) {
505 		calcru(lp, &upt, &spt);
506 		timevaladd(&ru->ru_utime, &upt);
507 		timevaladd(&ru->ru_stime, &spt);
508 		for (rip1 = &ru->ru_first, rip2 = &lp->lwp_ru.ru_first;
509 		     rip1 <= &ru->ru_last;
510 		     rip1++, rip2++)
511 			*rip1 += *rip2;
512 	}
513 }
514 
515 
516 /* ARGSUSED */
517 int
518 sys_getrusage(struct getrusage_args *uap)
519 {
520 	struct rusage ru;
521 	struct rusage *rup;
522 
523 	switch (uap->who) {
524 
525 	case RUSAGE_SELF:
526 		rup = &ru;
527 		calcru_proc(curproc, rup);
528 		break;
529 
530 	case RUSAGE_CHILDREN:
531 		rup = &curproc->p_cru;
532 		break;
533 
534 	default:
535 		return (EINVAL);
536 	}
537 	return (copyout((caddr_t)rup, (caddr_t)uap->rusage,
538 	    sizeof (struct rusage)));
539 }
540 
541 void
542 ruadd(struct rusage *ru, struct rusage *ru2)
543 {
544 	long *ip, *ip2;
545 	int i;
546 
547 	timevaladd(&ru->ru_utime, &ru2->ru_utime);
548 	timevaladd(&ru->ru_stime, &ru2->ru_stime);
549 	if (ru->ru_maxrss < ru2->ru_maxrss)
550 		ru->ru_maxrss = ru2->ru_maxrss;
551 	ip = &ru->ru_first; ip2 = &ru2->ru_first;
552 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
553 		*ip++ += *ip2++;
554 }
555 
556 /*
557  * Find the uidinfo structure for a uid.  This structure is used to
558  * track the total resource consumption (process count, socket buffer
559  * size, etc.) for the uid and impose limits.
560  */
561 void
562 uihashinit(void)
563 {
564 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
565 }
566 
567 static struct uidinfo *
568 uilookup(uid_t uid)
569 {
570 	struct	uihashhead *uipp;
571 	struct	uidinfo *uip;
572 
573 	uipp = UIHASH(uid);
574 	LIST_FOREACH(uip, uipp, ui_hash) {
575 		if (uip->ui_uid == uid)
576 			break;
577 	}
578 	return (uip);
579 }
580 
581 static struct uidinfo *
582 uicreate(uid_t uid)
583 {
584 	struct	uidinfo *uip, *norace;
585 
586 	/*
587 	 * Allocate space and check for a race
588 	 */
589 	MALLOC(uip, struct uidinfo *, sizeof(*uip), M_UIDINFO, M_WAITOK);
590 	norace = uilookup(uid);
591 	if (norace != NULL) {
592 		FREE(uip, M_UIDINFO);
593 		return (norace);
594 	}
595 
596 	/*
597 	 * Initialize structure and enter it into the hash table
598 	 */
599 	LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
600 	uip->ui_uid = uid;
601 	uip->ui_proccnt = 0;
602 	uip->ui_sbsize = 0;
603 	uip->ui_ref = 0;
604 	uip->ui_posixlocks = 0;
605 	varsymset_init(&uip->ui_varsymset, NULL);
606 	return (uip);
607 }
608 
609 struct uidinfo *
610 uifind(uid_t uid)
611 {
612 	struct	uidinfo *uip;
613 
614 	uip = uilookup(uid);
615 	if (uip == NULL)
616 		uip = uicreate(uid);
617 	uip->ui_ref++;
618 	return (uip);
619 }
620 
621 static __inline void
622 uifree(struct uidinfo *uip)
623 {
624 	if (uip->ui_sbsize != 0)
625 		/* XXX no %qd in kernel.  Truncate. */
626 		kprintf("freeing uidinfo: uid = %d, sbsize = %ld\n",
627 		    uip->ui_uid, (long)uip->ui_sbsize);
628 	if (uip->ui_proccnt != 0)
629 		kprintf("freeing uidinfo: uid = %d, proccnt = %ld\n",
630 		    uip->ui_uid, uip->ui_proccnt);
631 	LIST_REMOVE(uip, ui_hash);
632 	varsymset_clean(&uip->ui_varsymset);
633 	FREE(uip, M_UIDINFO);
634 }
635 
636 void
637 uihold(struct uidinfo *uip)
638 {
639 	++uip->ui_ref;
640 	KKASSERT(uip->ui_ref > 0);
641 }
642 
643 void
644 uidrop(struct uidinfo *uip)
645 {
646 	KKASSERT(uip->ui_ref > 0);
647 	if (--uip->ui_ref == 0)
648 		uifree(uip);
649 }
650 
651 void
652 uireplace(struct uidinfo **puip, struct uidinfo *nuip)
653 {
654 	uidrop(*puip);
655 	*puip = nuip;
656 }
657 
658 /*
659  * Change the count associated with number of processes
660  * a given user is using.  When 'max' is 0, don't enforce a limit
661  */
662 int
663 chgproccnt(struct uidinfo *uip, int diff, int max)
664 {
665 	/* don't allow them to exceed max, but allow subtraction */
666 	if (diff > 0 && uip->ui_proccnt + diff > max && max != 0)
667 		return (0);
668 	uip->ui_proccnt += diff;
669 	if (uip->ui_proccnt < 0)
670 		kprintf("negative proccnt for uid = %d\n", uip->ui_uid);
671 	return (1);
672 }
673 
674 /*
675  * Change the total socket buffer size a user has used.
676  */
677 int
678 chgsbsize(struct uidinfo *uip, u_long *hiwat, u_long to, rlim_t max)
679 {
680 	rlim_t new;
681 
682 	crit_enter();
683 	new = uip->ui_sbsize + to - *hiwat;
684 	/* don't allow them to exceed max, but allow subtraction */
685 	if (to > *hiwat && new > max) {
686 		crit_exit();
687 		return (0);
688 	}
689 	uip->ui_sbsize = new;
690 	*hiwat = to;
691 	if (uip->ui_sbsize < 0)
692 		kprintf("negative sbsize for uid = %d\n", uip->ui_uid);
693 	crit_exit();
694 	return (1);
695 }
696 
697