xref: /dragonfly/sys/kern/kern_resource.c (revision 6bd457ed) 
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 1.22 2005/06/25 20:03:28 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 int
80 getpriority(struct getpriority_args *uap)
81 {
82 	struct proc *curp = curproc;
83 	struct proc *p;
84 	int low = PRIO_MAX + 1;
85 
86 	switch (uap->which) {
87 	case PRIO_PROCESS:
88 		if (uap->who == 0)
89 			p = curp;
90 		else
91 			p = pfind(uap->who);
92 		if (p == 0)
93 			break;
94 		if (!PRISON_CHECK(curp->p_ucred, p->p_ucred))
95 			break;
96 		low = p->p_nice;
97 		break;
98 
99 	case PRIO_PGRP:
100 	{
101 		struct pgrp *pg;
102 
103 		if (uap->who == 0)
104 			pg = curp->p_pgrp;
105 		else if ((pg = pgfind(uap->who)) == NULL)
106 			break;
107 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
108 			if ((PRISON_CHECK(curp->p_ucred, p->p_ucred) && p->p_nice < low))
109 				low = p->p_nice;
110 		}
111 		break;
112 	}
113 	case PRIO_USER:
114 		if (uap->who == 0)
115 			uap->who = curp->p_ucred->cr_uid;
116 		FOREACH_PROC_IN_SYSTEM(p)
117 			if (PRISON_CHECK(curp->p_ucred, p->p_ucred) &&
118 			    p->p_ucred->cr_uid == uap->who &&
119 			    p->p_nice < low)
120 				low = p->p_nice;
121 		break;
122 
123 	default:
124 		return (EINVAL);
125 	}
126 	if (low == PRIO_MAX + 1)
127 		return (ESRCH);
128 	uap->sysmsg_result = low;
129 	return (0);
130 }
131 
132 /* ARGSUSED */
133 int
134 setpriority(struct setpriority_args *uap)
135 {
136 	struct proc *curp = curproc;
137 	struct proc *p;
138 	int found = 0, error = 0;
139 
140 	switch (uap->which) {
141 
142 	case PRIO_PROCESS:
143 		if (uap->who == 0)
144 			p = curp;
145 		else
146 			p = pfind(uap->who);
147 		if (p == 0)
148 			break;
149 		if (!PRISON_CHECK(curp->p_ucred, p->p_ucred))
150 			break;
151 		error = donice(p, uap->prio);
152 		found++;
153 		break;
154 
155 	case PRIO_PGRP:
156 	{
157 		struct pgrp *pg;
158 
159 		if (uap->who == 0)
160 			pg = curp->p_pgrp;
161 		else if ((pg = pgfind(uap->who)) == NULL)
162 			break;
163 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
164 			if (PRISON_CHECK(curp->p_ucred, p->p_ucred)) {
165 				error = donice(p, uap->prio);
166 				found++;
167 			}
168 		}
169 		break;
170 	}
171 	case PRIO_USER:
172 		if (uap->who == 0)
173 			uap->who = curp->p_ucred->cr_uid;
174 		FOREACH_PROC_IN_SYSTEM(p)
175 			if (p->p_ucred->cr_uid == uap->who &&
176 			    PRISON_CHECK(curp->p_ucred, p->p_ucred)) {
177 				error = donice(p, uap->prio);
178 				found++;
179 			}
180 		break;
181 
182 	default:
183 		return (EINVAL);
184 	}
185 	if (found == 0)
186 		return (ESRCH);
187 	return (error);
188 }
189 
190 static int
191 donice(struct proc *chgp, int n)
192 {
193 	struct proc *curp = curproc;
194 	struct ucred *cr = curp->p_ucred;
195 
196 	if (cr->cr_uid && cr->cr_ruid &&
197 	    cr->cr_uid != chgp->p_ucred->cr_uid &&
198 	    cr->cr_ruid != chgp->p_ucred->cr_uid)
199 		return (EPERM);
200 	if (n > PRIO_MAX)
201 		n = PRIO_MAX;
202 	if (n < PRIO_MIN)
203 		n = PRIO_MIN;
204 	if (n < chgp->p_nice && suser_cred(cr, 0))
205 		return (EACCES);
206 	chgp->p_nice = n;
207 	chgp->p_usched->resetpriority(chgp);
208 	return (0);
209 }
210 
211 /*
212  * Set realtime priority
213  */
214 /* ARGSUSED */
215 int
216 rtprio(struct rtprio_args *uap)
217 {
218 	struct proc *curp = curproc;
219 	struct proc *p;
220 	struct ucred *cr = curp->p_ucred;
221 	struct rtprio rtp;
222 	int error;
223 
224 	error = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
225 	if (error)
226 		return (error);
227 
228 	if (uap->pid == 0)
229 		p = curp;
230 	else
231 		p = pfind(uap->pid);
232 
233 	if (p == 0)
234 		return (ESRCH);
235 
236 	switch (uap->function) {
237 	case RTP_LOOKUP:
238 		return (copyout(&p->p_rtprio, uap->rtp, sizeof(struct rtprio)));
239 	case RTP_SET:
240 		if (cr->cr_uid && cr->cr_ruid &&
241 		    cr->cr_uid != p->p_ucred->cr_uid &&
242 		    cr->cr_ruid != p->p_ucred->cr_uid)
243 		        return (EPERM);
244 		/* disallow setting rtprio in most cases if not superuser */
245 		if (suser_cred(cr, 0)) {
246 			/* can't set someone else's */
247 			if (uap->pid)
248 				return (EPERM);
249 			/* can't set realtime priority */
250 /*
251  * Realtime priority has to be restricted for reasons which should be
252  * obvious. However, for idle priority, there is a potential for
253  * system deadlock if an idleprio process gains a lock on a resource
254  * that other processes need (and the idleprio process can't run
255  * due to a CPU-bound normal process). Fix me! XXX
256  */
257  			if (RTP_PRIO_IS_REALTIME(rtp.type))
258 				return (EPERM);
259 		}
260 		switch (rtp.type) {
261 #ifdef RTP_PRIO_FIFO
262 		case RTP_PRIO_FIFO:
263 #endif
264 		case RTP_PRIO_REALTIME:
265 		case RTP_PRIO_NORMAL:
266 		case RTP_PRIO_IDLE:
267 			if (rtp.prio > RTP_PRIO_MAX)
268 				return (EINVAL);
269 			p->p_rtprio = rtp;
270 			return (0);
271 		default:
272 			return (EINVAL);
273 		}
274 
275 	default:
276 		return (EINVAL);
277 	}
278 }
279 
280 int
281 setrlimit(struct __setrlimit_args *uap)
282 {
283 	struct rlimit alim;
284 	int error;
285 
286 	error = copyin(uap->rlp, &alim, sizeof(alim));
287 	if (error)
288 		return (error);
289 
290 	error = kern_setrlimit(uap->which, &alim);
291 
292 	return (error);
293 }
294 
295 int
296 kern_setrlimit(u_int which, struct rlimit *limp)
297 {
298 	struct proc *p = curproc;
299 	struct rlimit *alimp;
300 	int error;
301 
302 	if (which >= RLIM_NLIMITS)
303 		return (EINVAL);
304 	alimp = &p->p_rlimit[which];
305 
306 	/*
307 	 * Preserve historical bugs by treating negative limits as unsigned.
308 	 */
309 	if (limp->rlim_cur < 0)
310 		limp->rlim_cur = RLIM_INFINITY;
311 	if (limp->rlim_max < 0)
312 		limp->rlim_max = RLIM_INFINITY;
313 
314 	if (limp->rlim_cur > alimp->rlim_max ||
315 	    limp->rlim_max > alimp->rlim_max)
316 		if ((error = suser_cred(p->p_ucred, PRISON_ROOT)))
317 			return (error);
318 	if (limp->rlim_cur > limp->rlim_max)
319 		limp->rlim_cur = limp->rlim_max;
320 	if (p->p_limit->p_refcnt > 1 &&
321 	    (p->p_limit->p_lflags & PL_SHAREMOD) == 0) {
322 		p->p_limit->p_refcnt--;
323 		p->p_limit = limcopy(p->p_limit);
324 		alimp = &p->p_rlimit[which];
325 	}
326 
327 	switch (which) {
328 
329 	case RLIMIT_CPU:
330 		if (limp->rlim_cur > RLIM_INFINITY / (rlim_t)1000000)
331 			p->p_limit->p_cpulimit = RLIM_INFINITY;
332 		else
333 			p->p_limit->p_cpulimit =
334 			    (rlim_t)1000000 * limp->rlim_cur;
335 		break;
336 	case RLIMIT_DATA:
337 		if (limp->rlim_cur > maxdsiz)
338 			limp->rlim_cur = maxdsiz;
339 		if (limp->rlim_max > maxdsiz)
340 			limp->rlim_max = maxdsiz;
341 		break;
342 
343 	case RLIMIT_STACK:
344 		if (limp->rlim_cur > maxssiz)
345 			limp->rlim_cur = maxssiz;
346 		if (limp->rlim_max > maxssiz)
347 			limp->rlim_max = maxssiz;
348 		/*
349 		 * Stack is allocated to the max at exec time with only
350 		 * "rlim_cur" bytes accessible.  If stack limit is going
351 		 * up make more accessible, if going down make inaccessible.
352 		 */
353 		if (limp->rlim_cur != alimp->rlim_cur) {
354 			vm_offset_t addr;
355 			vm_size_t size;
356 			vm_prot_t prot;
357 
358 			if (limp->rlim_cur > alimp->rlim_cur) {
359 				prot = VM_PROT_ALL;
360 				size = limp->rlim_cur - alimp->rlim_cur;
361 				addr = USRSTACK - limp->rlim_cur;
362 			} else {
363 				prot = VM_PROT_NONE;
364 				size = alimp->rlim_cur - limp->rlim_cur;
365 				addr = USRSTACK - alimp->rlim_cur;
366 			}
367 			addr = trunc_page(addr);
368 			size = round_page(size);
369 			(void) vm_map_protect(&p->p_vmspace->vm_map,
370 					      addr, addr+size, prot, FALSE);
371 		}
372 		break;
373 
374 	case RLIMIT_NOFILE:
375 		if (limp->rlim_cur > maxfilesperproc)
376 			limp->rlim_cur = maxfilesperproc;
377 		if (limp->rlim_max > maxfilesperproc)
378 			limp->rlim_max = maxfilesperproc;
379 		break;
380 
381 	case RLIMIT_NPROC:
382 		if (limp->rlim_cur > maxprocperuid)
383 			limp->rlim_cur = maxprocperuid;
384 		if (limp->rlim_max > maxprocperuid)
385 			limp->rlim_max = maxprocperuid;
386 		if (limp->rlim_cur < 1)
387 			limp->rlim_cur = 1;
388 		if (limp->rlim_max < 1)
389 			limp->rlim_max = 1;
390 		break;
391 	case RLIMIT_POSIXLOCKS:
392 		if (limp->rlim_cur > maxposixlocksperuid)
393 			limp->rlim_cur = maxposixlocksperuid;
394 		if (limp->rlim_max > maxposixlocksperuid)
395 			limp->rlim_max = maxposixlocksperuid;
396 		break;
397 	}
398 	*alimp = *limp;
399 	return (0);
400 }
401 
402 /*
403  * The rlimit indexed by which is returned in the second argument.
404  */
405 int
406 kern_getrlimit(u_int which, struct rlimit *limp)
407 {
408 	struct thread *td = curthread;
409 	struct proc *p = td->td_proc;
410 
411 	if (which >= RLIM_NLIMITS)
412 		return (EINVAL);
413 
414 	*limp = p->p_rlimit[which];
415 
416 	return (0);
417 }
418 
419 int
420 getrlimit(struct __getrlimit_args *uap)
421 {
422 	struct rlimit lim;
423 	int error;
424 
425 	error = kern_getrlimit(uap->which, &lim);
426 
427 	if (error == 0)
428 		error = copyout(&lim, uap->rlp, sizeof(*uap->rlp));
429 	return error;
430 }
431 
432 /*
433  * Transform the running time and tick information in proc p into user,
434  * system, and interrupt time usage.
435  *
436  * Since we are limited to statclock tick granularity this is a statisical
437  * calculation which will be correct over the long haul, but should not be
438  * expected to measure fine grained deltas.
439  */
440 void
441 calcru(struct proc *p, struct timeval *up, struct timeval *sp,
442 	struct timeval *ip)
443 {
444 	struct thread *td = p->p_thread;
445 
446 	/*
447 	 * Calculate at the statclock level.  YYY if the thread is owned by
448 	 * another cpu we need to forward the request to the other cpu, or
449 	 * have a token to interlock the information.
450 	 */
451 	crit_enter();
452 	up->tv_sec = td->td_uticks / 1000000;
453 	up->tv_usec = td->td_uticks % 1000000;
454 	sp->tv_sec = td->td_sticks / 1000000;
455 	sp->tv_usec = td->td_sticks % 1000000;
456 	if (ip != NULL) {
457 		ip->tv_sec = td->td_iticks / 1000000;
458 		ip->tv_usec = td->td_iticks % 1000000;
459 	}
460 	crit_exit();
461 }
462 
463 /* ARGSUSED */
464 int
465 getrusage(struct getrusage_args *uap)
466 {
467 	struct proc *p = curproc;
468 	struct rusage *rup;
469 
470 	switch (uap->who) {
471 
472 	case RUSAGE_SELF:
473 		rup = &p->p_stats->p_ru;
474 		calcru(p, &rup->ru_utime, &rup->ru_stime, NULL);
475 		break;
476 
477 	case RUSAGE_CHILDREN:
478 		rup = &p->p_stats->p_cru;
479 		break;
480 
481 	default:
482 		return (EINVAL);
483 	}
484 	return (copyout((caddr_t)rup, (caddr_t)uap->rusage,
485 	    sizeof (struct rusage)));
486 }
487 
488 void
489 ruadd(struct rusage *ru, struct rusage *ru2)
490 {
491 	long *ip, *ip2;
492 	int i;
493 
494 	timevaladd(&ru->ru_utime, &ru2->ru_utime);
495 	timevaladd(&ru->ru_stime, &ru2->ru_stime);
496 	if (ru->ru_maxrss < ru2->ru_maxrss)
497 		ru->ru_maxrss = ru2->ru_maxrss;
498 	ip = &ru->ru_first; ip2 = &ru2->ru_first;
499 	for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
500 		*ip++ += *ip2++;
501 }
502 
503 /*
504  * Make a copy of the plimit structure.
505  * We share these structures copy-on-write after fork,
506  * and copy when a limit is changed.
507  */
508 struct plimit *
509 limcopy(struct plimit *lim)
510 {
511 	struct plimit *copy;
512 
513 	MALLOC(copy, struct plimit *, sizeof(struct plimit),
514 	    M_SUBPROC, M_WAITOK);
515 	bcopy(lim->pl_rlimit, copy->pl_rlimit, sizeof(struct plimit));
516 	copy->p_lflags = 0;
517 	copy->p_refcnt = 1;
518 	return (copy);
519 }
520 
521 /*
522  * Find the uidinfo structure for a uid.  This structure is used to
523  * track the total resource consumption (process count, socket buffer
524  * size, etc.) for the uid and impose limits.
525  */
526 void
527 uihashinit(void)
528 {
529 	uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
530 }
531 
532 static struct uidinfo *
533 uilookup(uid_t uid)
534 {
535 	struct	uihashhead *uipp;
536 	struct	uidinfo *uip;
537 
538 	uipp = UIHASH(uid);
539 	LIST_FOREACH(uip, uipp, ui_hash) {
540 		if (uip->ui_uid == uid)
541 			break;
542 	}
543 	return (uip);
544 }
545 
546 static struct uidinfo *
547 uicreate(uid_t uid)
548 {
549 	struct	uidinfo *uip, *norace;
550 
551 	/*
552 	 * Allocate space and check for a race
553 	 */
554 	MALLOC(uip, struct uidinfo *, sizeof(*uip), M_UIDINFO, M_WAITOK);
555 	norace = uilookup(uid);
556 	if (norace != NULL) {
557 		FREE(uip, M_UIDINFO);
558 		return (norace);
559 	}
560 
561 	/*
562 	 * Initialize structure and enter it into the hash table
563 	 */
564 	LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
565 	uip->ui_uid = uid;
566 	uip->ui_proccnt = 0;
567 	uip->ui_sbsize = 0;
568 	uip->ui_ref = 0;
569 	uip->ui_posixlocks = 0;
570 	varsymset_init(&uip->ui_varsymset, NULL);
571 	return (uip);
572 }
573 
574 struct uidinfo *
575 uifind(uid_t uid)
576 {
577 	struct	uidinfo *uip;
578 
579 	uip = uilookup(uid);
580 	if (uip == NULL)
581 		uip = uicreate(uid);
582 	uip->ui_ref++;
583 	return (uip);
584 }
585 
586 static __inline void
587 uifree(struct uidinfo *uip)
588 {
589 	if (uip->ui_sbsize != 0)
590 		/* XXX no %qd in kernel.  Truncate. */
591 		printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
592 		    uip->ui_uid, (long)uip->ui_sbsize);
593 	if (uip->ui_proccnt != 0)
594 		printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
595 		    uip->ui_uid, uip->ui_proccnt);
596 	LIST_REMOVE(uip, ui_hash);
597 	varsymset_clean(&uip->ui_varsymset);
598 	FREE(uip, M_UIDINFO);
599 }
600 
601 void
602 uihold(struct uidinfo *uip)
603 {
604 	++uip->ui_ref;
605 	KKASSERT(uip->ui_ref > 0);
606 }
607 
608 void
609 uidrop(struct uidinfo *uip)
610 {
611 	KKASSERT(uip->ui_ref > 0);
612 	if (--uip->ui_ref == 0)
613 		uifree(uip);
614 }
615 
616 void
617 uireplace(struct uidinfo **puip, struct uidinfo *nuip)
618 {
619 	uidrop(*puip);
620 	*puip = nuip;
621 }
622 
623 /*
624  * Change the count associated with number of processes
625  * a given user is using.  When 'max' is 0, don't enforce a limit
626  */
627 int
628 chgproccnt(struct uidinfo *uip, int diff, int max)
629 {
630 	/* don't allow them to exceed max, but allow subtraction */
631 	if (diff > 0 && uip->ui_proccnt + diff > max && max != 0)
632 		return (0);
633 	uip->ui_proccnt += diff;
634 	if (uip->ui_proccnt < 0)
635 		printf("negative proccnt for uid = %d\n", uip->ui_uid);
636 	return (1);
637 }
638 
639 /*
640  * Change the total socket buffer size a user has used.
641  */
642 int
643 chgsbsize(struct uidinfo *uip, u_long *hiwat, u_long to, rlim_t max)
644 {
645 	rlim_t new;
646 
647 	crit_enter();
648 	new = uip->ui_sbsize + to - *hiwat;
649 	/* don't allow them to exceed max, but allow subtraction */
650 	if (to > *hiwat && new > max) {
651 		crit_exit();
652 		return (0);
653 	}
654 	uip->ui_sbsize = new;
655 	*hiwat = to;
656 	if (uip->ui_sbsize < 0)
657 		printf("negative sbsize for uid = %d\n", uip->ui_uid);
658 	crit_exit();
659 	return (1);
660 }
661 
662