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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/8.1/sys/kern/kern_resource.c 208580 2010-05-26 19:26:28Z kib $");
39
40 #include "opt_compat.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/sysproto.h>
45 #include <sys/file.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mutex.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/refcount.h>
53 #include <sys/resourcevar.h>
54 #include <sys/rwlock.h>
55 #include <sys/sched.h>
56 #include <sys/sx.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/time.h>
60 #include <sys/umtx.h>
61
62 #include <vm/vm.h>
63 #include <vm/vm_param.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_map.h>
66
67
68 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
69 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
70 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
71 static struct rwlock uihashtbl_lock;
72 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
73 static u_long uihash; /* size of hash table - 1 */
74
75 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
76 struct timeval *up, struct timeval *sp);
77 static int donice(struct thread *td, struct proc *chgp, int n);
78 static struct uidinfo *uilookup(uid_t uid);
79 static void ruxagg_locked(struct rusage_ext *rux, struct thread *td);
80
81 /*
82 * Resource controls and accounting.
83 */
84 #ifndef _SYS_SYSPROTO_H_
85 struct getpriority_args {
86 int which;
87 int who;
88 };
89 #endif
90 int
91 getpriority(td, uap)
92 struct thread *td;
93 register struct getpriority_args *uap;
94 {
95 struct proc *p;
96 struct pgrp *pg;
97 int error, low;
98
99 error = 0;
100 low = PRIO_MAX + 1;
101 switch (uap->which) {
102
103 case PRIO_PROCESS:
104 if (uap->who == 0)
105 low = td->td_proc->p_nice;
106 else {
107 p = pfind(uap->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 (uap->who == 0) {
119 pg = td->td_proc->p_pgrp;
120 PGRP_LOCK(pg);
121 } else {
122 pg = pgfind(uap->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_cansee(td, p) == 0) {
132 if (p->p_nice < low)
133 low = p->p_nice;
134 }
135 PROC_UNLOCK(p);
136 }
137 PGRP_UNLOCK(pg);
138 break;
139
140 case PRIO_USER:
141 if (uap->who == 0)
142 uap->who = td->td_ucred->cr_uid;
143 sx_slock(&allproc_lock);
144 FOREACH_PROC_IN_SYSTEM(p) {
145 /* Do not bother to check PRS_NEW processes */
146 if (p->p_state == PRS_NEW)
147 continue;
148 PROC_LOCK(p);
149 if (p_cansee(td, p) == 0 &&
150 p->p_ucred->cr_uid == uap->who) {
151 if (p->p_nice < low)
152 low = p->p_nice;
153 }
154 PROC_UNLOCK(p);
155 }
156 sx_sunlock(&allproc_lock);
157 break;
158
159 default:
160 error = EINVAL;
161 break;
162 }
163 if (low == PRIO_MAX + 1 && error == 0)
164 error = ESRCH;
165 td->td_retval[0] = low;
166 return (error);
167 }
168
169 #ifndef _SYS_SYSPROTO_H_
170 struct setpriority_args {
171 int which;
172 int who;
173 int prio;
174 };
175 #endif
176 int
177 setpriority(td, uap)
178 struct thread *td;
179 struct setpriority_args *uap;
180 {
181 struct proc *curp, *p;
182 struct pgrp *pg;
183 int found = 0, error = 0;
184
185 curp = td->td_proc;
186 switch (uap->which) {
187 case PRIO_PROCESS:
188 if (uap->who == 0) {
189 PROC_LOCK(curp);
190 error = donice(td, curp, uap->prio);
191 PROC_UNLOCK(curp);
192 } else {
193 p = pfind(uap->who);
194 if (p == NULL)
195 break;
196 error = p_cansee(td, p);
197 if (error == 0)
198 error = donice(td, p, uap->prio);
199 PROC_UNLOCK(p);
200 }
201 found++;
202 break;
203
204 case PRIO_PGRP:
205 sx_slock(&proctree_lock);
206 if (uap->who == 0) {
207 pg = curp->p_pgrp;
208 PGRP_LOCK(pg);
209 } else {
210 pg = pgfind(uap->who);
211 if (pg == NULL) {
212 sx_sunlock(&proctree_lock);
213 break;
214 }
215 }
216 sx_sunlock(&proctree_lock);
217 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
218 PROC_LOCK(p);
219 if (p_cansee(td, p) == 0) {
220 error = donice(td, p, uap->prio);
221 found++;
222 }
223 PROC_UNLOCK(p);
224 }
225 PGRP_UNLOCK(pg);
226 break;
227
228 case PRIO_USER:
229 if (uap->who == 0)
230 uap->who = td->td_ucred->cr_uid;
231 sx_slock(&allproc_lock);
232 FOREACH_PROC_IN_SYSTEM(p) {
233 PROC_LOCK(p);
234 if (p->p_ucred->cr_uid == uap->who &&
235 p_cansee(td, p) == 0) {
236 error = donice(td, p, uap->prio);
237 found++;
238 }
239 PROC_UNLOCK(p);
240 }
241 sx_sunlock(&allproc_lock);
242 break;
243
244 default:
245 error = EINVAL;
246 break;
247 }
248 if (found == 0 && error == 0)
249 error = ESRCH;
250 return (error);
251 }
252
253 /*
254 * Set "nice" for a (whole) process.
255 */
256 static int
257 donice(struct thread *td, struct proc *p, int n)
258 {
259 int error;
260
261 PROC_LOCK_ASSERT(p, MA_OWNED);
262 if ((error = p_cansched(td, p)))
263 return (error);
264 if (n > PRIO_MAX)
265 n = PRIO_MAX;
266 if (n < PRIO_MIN)
267 n = PRIO_MIN;
268 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
269 return (EACCES);
270 sched_nice(p, n);
271 return (0);
272 }
273
274 /*
275 * Set realtime priority for LWP.
276 */
277 #ifndef _SYS_SYSPROTO_H_
278 struct rtprio_thread_args {
279 int function;
280 lwpid_t lwpid;
281 struct rtprio *rtp;
282 };
283 #endif
284 int
285 rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
286 {
287 struct proc *p;
288 struct rtprio rtp;
289 struct thread *td1;
290 int cierror, error;
291
292 /* Perform copyin before acquiring locks if needed. */
293 if (uap->function == RTP_SET)
294 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
295 else
296 cierror = 0;
297
298 /*
299 * Though lwpid is unique, only current process is supported
300 * since there is no efficient way to look up a LWP yet.
301 */
302 p = td->td_proc;
303 PROC_LOCK(p);
304
305 switch (uap->function) {
306 case RTP_LOOKUP:
307 if ((error = p_cansee(td, p)))
308 break;
309 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
310 td1 = td;
311 else
312 td1 = thread_find(p, uap->lwpid);
313 if (td1 != NULL)
314 pri_to_rtp(td1, &rtp);
315 else
316 error = ESRCH;
317 PROC_UNLOCK(p);
318 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
319 case RTP_SET:
320 if ((error = p_cansched(td, p)) || (error = cierror))
321 break;
322
323 /* Disallow setting rtprio in most cases if not superuser. */
324 /*
325 * Realtime priority has to be restricted for reasons which should be
326 * obvious. However, for idle priority, there is a potential for
327 * system deadlock if an idleprio process gains a lock on a resource
328 * that other processes need (and the idleprio process can't run
329 * due to a CPU-bound normal process). Fix me! XXX
330 */
331 #if 0
332 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
333 #else
334 if (rtp.type != RTP_PRIO_NORMAL) {
335 #endif
336 error = priv_check(td, PRIV_SCHED_RTPRIO);
337 if (error)
338 break;
339 }
340
341 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
342 td1 = td;
343 else
344 td1 = thread_find(p, uap->lwpid);
345 if (td1 != NULL)
346 error = rtp_to_pri(&rtp, td1);
347 else
348 error = ESRCH;
349 break;
350 default:
351 error = EINVAL;
352 break;
353 }
354 PROC_UNLOCK(p);
355 return (error);
356 }
357
358 /*
359 * Set realtime priority.
360 */
361 #ifndef _SYS_SYSPROTO_H_
362 struct rtprio_args {
363 int function;
364 pid_t pid;
365 struct rtprio *rtp;
366 };
367 #endif
368 int
369 rtprio(td, uap)
370 struct thread *td; /* curthread */
371 register struct rtprio_args *uap;
372 {
373 struct proc *p;
374 struct thread *tdp;
375 struct rtprio rtp;
376 int cierror, error;
377
378 /* Perform copyin before acquiring locks if needed. */
379 if (uap->function == RTP_SET)
380 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
381 else
382 cierror = 0;
383
384 if (uap->pid == 0) {
385 p = td->td_proc;
386 PROC_LOCK(p);
387 } else {
388 p = pfind(uap->pid);
389 if (p == NULL)
390 return (ESRCH);
391 }
392
393 switch (uap->function) {
394 case RTP_LOOKUP:
395 if ((error = p_cansee(td, p)))
396 break;
397 /*
398 * Return OUR priority if no pid specified,
399 * or if one is, report the highest priority
400 * in the process. There isn't much more you can do as
401 * there is only room to return a single priority.
402 * Note: specifying our own pid is not the same
403 * as leaving it zero.
404 */
405 if (uap->pid == 0) {
406 pri_to_rtp(td, &rtp);
407 } else {
408 struct rtprio rtp2;
409
410 rtp.type = RTP_PRIO_IDLE;
411 rtp.prio = RTP_PRIO_MAX;
412 FOREACH_THREAD_IN_PROC(p, tdp) {
413 pri_to_rtp(tdp, &rtp2);
414 if (rtp2.type < rtp.type ||
415 (rtp2.type == rtp.type &&
416 rtp2.prio < rtp.prio)) {
417 rtp.type = rtp2.type;
418 rtp.prio = rtp2.prio;
419 }
420 }
421 }
422 PROC_UNLOCK(p);
423 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
424 case RTP_SET:
425 if ((error = p_cansched(td, p)) || (error = cierror))
426 break;
427
428 /* Disallow setting rtprio in most cases if not superuser. */
429 /*
430 * Realtime priority has to be restricted for reasons which should be
431 * obvious. However, for idle priority, there is a potential for
432 * system deadlock if an idleprio process gains a lock on a resource
433 * that other processes need (and the idleprio process can't run
434 * due to a CPU-bound normal process). Fix me! XXX
435 */
436 #if 0
437 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
438 #else
439 if (rtp.type != RTP_PRIO_NORMAL) {
440 #endif
441 error = priv_check(td, PRIV_SCHED_RTPRIO);
442 if (error)
443 break;
444 }
445
446 /*
447 * If we are setting our own priority, set just our
448 * thread but if we are doing another process,
449 * do all the threads on that process. If we
450 * specify our own pid we do the latter.
451 */
452 if (uap->pid == 0) {
453 error = rtp_to_pri(&rtp, td);
454 } else {
455 FOREACH_THREAD_IN_PROC(p, td) {
456 if ((error = rtp_to_pri(&rtp, td)) != 0)
457 break;
458 }
459 }
460 break;
461 default:
462 error = EINVAL;
463 break;
464 }
465 PROC_UNLOCK(p);
466 return (error);
467 }
468
469 int
470 rtp_to_pri(struct rtprio *rtp, struct thread *td)
471 {
472 u_char newpri;
473 u_char oldpri;
474
475 thread_lock(td);
476 switch (RTP_PRIO_BASE(rtp->type)) {
477 case RTP_PRIO_REALTIME:
478 if (rtp->prio > RTP_PRIO_MAX) {
479 thread_unlock(td);
480 return (EINVAL);
481 }
482 newpri = PRI_MIN_REALTIME + rtp->prio;
483 break;
484 case RTP_PRIO_NORMAL:
485 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) {
486 thread_unlock(td);
487 return (EINVAL);
488 }
489 newpri = PRI_MIN_TIMESHARE + rtp->prio;
490 break;
491 case RTP_PRIO_IDLE:
492 newpri = PRI_MIN_IDLE + rtp->prio;
493 break;
494 default:
495 thread_unlock(td);
496 return (EINVAL);
497 }
498 sched_class(td, rtp->type); /* XXX fix */
499 oldpri = td->td_user_pri;
500 sched_user_prio(td, newpri);
501 if (curthread == td)
502 sched_prio(curthread, td->td_user_pri); /* XXX dubious */
503 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
504 thread_unlock(td);
505 umtx_pi_adjust(td, oldpri);
506 } else
507 thread_unlock(td);
508 return (0);
509 }
510
511 void
512 pri_to_rtp(struct thread *td, struct rtprio *rtp)
513 {
514
515 thread_lock(td);
516 switch (PRI_BASE(td->td_pri_class)) {
517 case PRI_REALTIME:
518 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
519 break;
520 case PRI_TIMESHARE:
521 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
522 break;
523 case PRI_IDLE:
524 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
525 break;
526 default:
527 break;
528 }
529 rtp->type = td->td_pri_class;
530 thread_unlock(td);
531 }
532
533 #if defined(COMPAT_43)
534 #ifndef _SYS_SYSPROTO_H_
535 struct osetrlimit_args {
536 u_int which;
537 struct orlimit *rlp;
538 };
539 #endif
540 int
541 osetrlimit(td, uap)
542 struct thread *td;
543 register 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(td, uap)
565 struct thread *td;
566 register struct ogetrlimit_args *uap;
567 {
568 struct orlimit olim;
569 struct rlimit rl;
570 struct proc *p;
571 int error;
572
573 if (uap->which >= RLIM_NLIMITS)
574 return (EINVAL);
575 p = td->td_proc;
576 PROC_LOCK(p);
577 lim_rlimit(p, uap->which, &rl);
578 PROC_UNLOCK(p);
579
580 /*
581 * XXX would be more correct to convert only RLIM_INFINITY to the
582 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
583 * values. Most 64->32 and 32->16 conversions, including not
584 * unimportant ones of uids are even more broken than what we
585 * do here (they blindly truncate). We don't do this correctly
586 * here since we have little experience with EOVERFLOW yet.
587 * Elsewhere, getuid() can't fail...
588 */
589 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
590 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
591 error = copyout(&olim, uap->rlp, sizeof(olim));
592 return (error);
593 }
594 #endif /* COMPAT_43 */
595
596 #ifndef _SYS_SYSPROTO_H_
597 struct __setrlimit_args {
598 u_int which;
599 struct rlimit *rlp;
600 };
601 #endif
602 int
603 setrlimit(td, uap)
604 struct thread *td;
605 register struct __setrlimit_args *uap;
606 {
607 struct rlimit alim;
608 int error;
609
610 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
611 return (error);
612 error = kern_setrlimit(td, uap->which, &alim);
613 return (error);
614 }
615
616 static void
617 lim_cb(void *arg)
618 {
619 struct rlimit rlim;
620 struct thread *td;
621 struct proc *p;
622
623 p = arg;
624 PROC_LOCK_ASSERT(p, MA_OWNED);
625 /*
626 * Check if the process exceeds its cpu resource allocation. If
627 * it reaches the max, arrange to kill the process in ast().
628 */
629 if (p->p_cpulimit == RLIM_INFINITY)
630 return;
631 PROC_SLOCK(p);
632 FOREACH_THREAD_IN_PROC(p, td) {
633 ruxagg(p, td);
634 }
635 PROC_SUNLOCK(p);
636 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
637 lim_rlimit(p, RLIMIT_CPU, &rlim);
638 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
639 killproc(p, "exceeded maximum CPU limit");
640 } else {
641 if (p->p_cpulimit < rlim.rlim_max)
642 p->p_cpulimit += 5;
643 psignal(p, SIGXCPU);
644 }
645 }
646 if ((p->p_flag & P_WEXIT) == 0)
647 callout_reset(&p->p_limco, hz, lim_cb, p);
648 }
649
650 int
651 kern_setrlimit(td, which, limp)
652 struct thread *td;
653 u_int which;
654 struct rlimit *limp;
655 {
656 struct plimit *newlim, *oldlim;
657 struct proc *p;
658 register struct rlimit *alimp;
659 struct rlimit oldssiz;
660 int error;
661
662 if (which >= RLIM_NLIMITS)
663 return (EINVAL);
664
665 /*
666 * Preserve historical bugs by treating negative limits as unsigned.
667 */
668 if (limp->rlim_cur < 0)
669 limp->rlim_cur = RLIM_INFINITY;
670 if (limp->rlim_max < 0)
671 limp->rlim_max = RLIM_INFINITY;
672
673 oldssiz.rlim_cur = 0;
674 p = td->td_proc;
675 newlim = lim_alloc();
676 PROC_LOCK(p);
677 oldlim = p->p_limit;
678 alimp = &oldlim->pl_rlimit[which];
679 if (limp->rlim_cur > alimp->rlim_max ||
680 limp->rlim_max > alimp->rlim_max)
681 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
682 PROC_UNLOCK(p);
683 lim_free(newlim);
684 return (error);
685 }
686 if (limp->rlim_cur > limp->rlim_max)
687 limp->rlim_cur = limp->rlim_max;
688 lim_copy(newlim, oldlim);
689 alimp = &newlim->pl_rlimit[which];
690
691 switch (which) {
692
693 case RLIMIT_CPU:
694 if (limp->rlim_cur != RLIM_INFINITY &&
695 p->p_cpulimit == RLIM_INFINITY)
696 callout_reset(&p->p_limco, hz, lim_cb, p);
697 p->p_cpulimit = limp->rlim_cur;
698 break;
699 case RLIMIT_DATA:
700 if (limp->rlim_cur > maxdsiz)
701 limp->rlim_cur = maxdsiz;
702 if (limp->rlim_max > maxdsiz)
703 limp->rlim_max = maxdsiz;
704 break;
705
706 case RLIMIT_STACK:
707 if (limp->rlim_cur > maxssiz)
708 limp->rlim_cur = maxssiz;
709 if (limp->rlim_max > maxssiz)
710 limp->rlim_max = maxssiz;
711 oldssiz = *alimp;
712 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
713 td->td_proc->p_sysent->sv_fixlimit(&oldssiz,
714 RLIMIT_STACK);
715 break;
716
717 case RLIMIT_NOFILE:
718 if (limp->rlim_cur > maxfilesperproc)
719 limp->rlim_cur = maxfilesperproc;
720 if (limp->rlim_max > maxfilesperproc)
721 limp->rlim_max = maxfilesperproc;
722 break;
723
724 case RLIMIT_NPROC:
725 if (limp->rlim_cur > maxprocperuid)
726 limp->rlim_cur = maxprocperuid;
727 if (limp->rlim_max > maxprocperuid)
728 limp->rlim_max = maxprocperuid;
729 if (limp->rlim_cur < 1)
730 limp->rlim_cur = 1;
731 if (limp->rlim_max < 1)
732 limp->rlim_max = 1;
733 break;
734 }
735 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
736 td->td_proc->p_sysent->sv_fixlimit(limp, which);
737 *alimp = *limp;
738 p->p_limit = newlim;
739 PROC_UNLOCK(p);
740 lim_free(oldlim);
741
742 if (which == RLIMIT_STACK) {
743 /*
744 * Stack is allocated to the max at exec time with only
745 * "rlim_cur" bytes accessible. If stack limit is going
746 * up make more accessible, if going down make inaccessible.
747 */
748 if (limp->rlim_cur != oldssiz.rlim_cur) {
749 vm_offset_t addr;
750 vm_size_t size;
751 vm_prot_t prot;
752
753 if (limp->rlim_cur > oldssiz.rlim_cur) {
754 prot = p->p_sysent->sv_stackprot;
755 size = limp->rlim_cur - oldssiz.rlim_cur;
756 addr = p->p_sysent->sv_usrstack -
757 limp->rlim_cur;
758 } else {
759 prot = VM_PROT_NONE;
760 size = oldssiz.rlim_cur - limp->rlim_cur;
761 addr = p->p_sysent->sv_usrstack -
762 oldssiz.rlim_cur;
763 }
764 addr = trunc_page(addr);
765 size = round_page(size);
766 (void)vm_map_protect(&p->p_vmspace->vm_map,
767 addr, addr + size, prot, FALSE);
768 }
769 }
770
771 return (0);
772 }
773
774 #ifndef _SYS_SYSPROTO_H_
775 struct __getrlimit_args {
776 u_int which;
777 struct rlimit *rlp;
778 };
779 #endif
780 /* ARGSUSED */
781 int
782 getrlimit(td, uap)
783 struct thread *td;
784 register struct __getrlimit_args *uap;
785 {
786 struct rlimit rlim;
787 struct proc *p;
788 int error;
789
790 if (uap->which >= RLIM_NLIMITS)
791 return (EINVAL);
792 p = td->td_proc;
793 PROC_LOCK(p);
794 lim_rlimit(p, uap->which, &rlim);
795 PROC_UNLOCK(p);
796 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
797 return (error);
798 }
799
800 /*
801 * Transform the running time and tick information for children of proc p
802 * into user and system time usage.
803 */
804 void
805 calccru(p, up, sp)
806 struct proc *p;
807 struct timeval *up;
808 struct timeval *sp;
809 {
810
811 PROC_LOCK_ASSERT(p, MA_OWNED);
812 calcru1(p, &p->p_crux, up, sp);
813 }
814
815 /*
816 * Transform the running time and tick information in proc p into user
817 * and system time usage. If appropriate, include the current time slice
818 * on this CPU.
819 */
820 void
821 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
822 {
823 struct thread *td;
824 uint64_t u;
825
826 PROC_LOCK_ASSERT(p, MA_OWNED);
827 PROC_SLOCK_ASSERT(p, MA_OWNED);
828 /*
829 * If we are getting stats for the current process, then add in the
830 * stats that this thread has accumulated in its current time slice.
831 * We reset the thread and CPU state as if we had performed a context
832 * switch right here.
833 */
834 td = curthread;
835 if (td->td_proc == p) {
836 u = cpu_ticks();
837 p->p_rux.rux_runtime += u - PCPU_GET(switchtime);
838 PCPU_SET(switchtime, u);
839 }
840 /* Make sure the per-thread stats are current. */
841 FOREACH_THREAD_IN_PROC(p, td) {
842 if (td->td_incruntime == 0)
843 continue;
844 ruxagg(p, td);
845 }
846 calcru1(p, &p->p_rux, up, sp);
847 }
848
849 static void
850 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
851 struct timeval *sp)
852 {
853 /* {user, system, interrupt, total} {ticks, usec}: */
854 u_int64_t ut, uu, st, su, it, tt, tu;
855
856 ut = ruxp->rux_uticks;
857 st = ruxp->rux_sticks;
858 it = ruxp->rux_iticks;
859 tt = ut + st + it;
860 if (tt == 0) {
861 /* Avoid divide by zero */
862 st = 1;
863 tt = 1;
864 }
865 tu = cputick2usec(ruxp->rux_runtime);
866 if ((int64_t)tu < 0) {
867 /* XXX: this should be an assert /phk */
868 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
869 (intmax_t)tu, p->p_pid, p->p_comm);
870 tu = ruxp->rux_tu;
871 }
872
873 if (tu >= ruxp->rux_tu) {
874 /*
875 * The normal case, time increased.
876 * Enforce monotonicity of bucketed numbers.
877 */
878 uu = (tu * ut) / tt;
879 if (uu < ruxp->rux_uu)
880 uu = ruxp->rux_uu;
881 su = (tu * st) / tt;
882 if (su < ruxp->rux_su)
883 su = ruxp->rux_su;
884 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
885 /*
886 * When we calibrate the cputicker, it is not uncommon to
887 * see the presumably fixed frequency increase slightly over
888 * time as a result of thermal stabilization and NTP
889 * discipline (of the reference clock). We therefore ignore
890 * a bit of backwards slop because we expect to catch up
891 * shortly. We use a 3 microsecond limit to catch low
892 * counts and a 1% limit for high counts.
893 */
894 uu = ruxp->rux_uu;
895 su = ruxp->rux_su;
896 tu = ruxp->rux_tu;
897 } else { /* tu < ruxp->rux_tu */
898 /*
899 * What happened here was likely that a laptop, which ran at
900 * a reduced clock frequency at boot, kicked into high gear.
901 * The wisdom of spamming this message in that case is
902 * dubious, but it might also be indicative of something
903 * serious, so lets keep it and hope laptops can be made
904 * more truthful about their CPU speed via ACPI.
905 */
906 printf("calcru: runtime went backwards from %ju usec "
907 "to %ju usec for pid %d (%s)\n",
908 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
909 p->p_pid, p->p_comm);
910 uu = (tu * ut) / tt;
911 su = (tu * st) / tt;
912 }
913
914 ruxp->rux_uu = uu;
915 ruxp->rux_su = su;
916 ruxp->rux_tu = tu;
917
918 up->tv_sec = uu / 1000000;
919 up->tv_usec = uu % 1000000;
920 sp->tv_sec = su / 1000000;
921 sp->tv_usec = su % 1000000;
922 }
923
924 #ifndef _SYS_SYSPROTO_H_
925 struct getrusage_args {
926 int who;
927 struct rusage *rusage;
928 };
929 #endif
930 int
931 getrusage(td, uap)
932 register struct thread *td;
933 register struct getrusage_args *uap;
934 {
935 struct rusage ru;
936 int error;
937
938 error = kern_getrusage(td, uap->who, &ru);
939 if (error == 0)
940 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
941 return (error);
942 }
943
944 int
945 kern_getrusage(struct thread *td, int who, struct rusage *rup)
946 {
947 struct proc *p;
948 int error;
949
950 error = 0;
951 p = td->td_proc;
952 PROC_LOCK(p);
953 switch (who) {
954 case RUSAGE_SELF:
955 rufetchcalc(p, rup, &rup->ru_utime,
956 &rup->ru_stime);
957 break;
958
959 case RUSAGE_CHILDREN:
960 *rup = p->p_stats->p_cru;
961 calccru(p, &rup->ru_utime, &rup->ru_stime);
962 break;
963
964 case RUSAGE_THREAD:
965 PROC_SLOCK(p);
966 ruxagg(p, td);
967 PROC_SUNLOCK(p);
968 thread_lock(td);
969 *rup = td->td_ru;
970 calcru1(p, &td->td_rux, &rup->ru_utime, &rup->ru_stime);
971 thread_unlock(td);
972 break;
973
974 default:
975 error = EINVAL;
976 }
977 PROC_UNLOCK(p);
978 return (error);
979 }
980
981 void
982 rucollect(struct rusage *ru, struct rusage *ru2)
983 {
984 long *ip, *ip2;
985 int i;
986
987 if (ru->ru_maxrss < ru2->ru_maxrss)
988 ru->ru_maxrss = ru2->ru_maxrss;
989 ip = &ru->ru_first;
990 ip2 = &ru2->ru_first;
991 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
992 *ip++ += *ip2++;
993 }
994
995 void
996 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
997 struct rusage_ext *rux2)
998 {
999
1000 rux->rux_runtime += rux2->rux_runtime;
1001 rux->rux_uticks += rux2->rux_uticks;
1002 rux->rux_sticks += rux2->rux_sticks;
1003 rux->rux_iticks += rux2->rux_iticks;
1004 rux->rux_uu += rux2->rux_uu;
1005 rux->rux_su += rux2->rux_su;
1006 rux->rux_tu += rux2->rux_tu;
1007 rucollect(ru, ru2);
1008 }
1009
1010 /*
1011 * Aggregate tick counts into the proc's rusage_ext.
1012 */
1013 static void
1014 ruxagg_locked(struct rusage_ext *rux, struct thread *td)
1015 {
1016
1017 THREAD_LOCK_ASSERT(td, MA_OWNED);
1018 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED);
1019 rux->rux_runtime += td->td_incruntime;
1020 rux->rux_uticks += td->td_uticks;
1021 rux->rux_sticks += td->td_sticks;
1022 rux->rux_iticks += td->td_iticks;
1023 }
1024
1025 void
1026 ruxagg(struct proc *p, struct thread *td)
1027 {
1028
1029 thread_lock(td);
1030 ruxagg_locked(&p->p_rux, td);
1031 ruxagg_locked(&td->td_rux, td);
1032 td->td_incruntime = 0;
1033 td->td_uticks = 0;
1034 td->td_iticks = 0;
1035 td->td_sticks = 0;
1036 thread_unlock(td);
1037 }
1038
1039 /*
1040 * Update the rusage_ext structure and fetch a valid aggregate rusage
1041 * for proc p if storage for one is supplied.
1042 */
1043 void
1044 rufetch(struct proc *p, struct rusage *ru)
1045 {
1046 struct thread *td;
1047
1048 PROC_SLOCK_ASSERT(p, MA_OWNED);
1049
1050 *ru = p->p_ru;
1051 if (p->p_numthreads > 0) {
1052 FOREACH_THREAD_IN_PROC(p, td) {
1053 ruxagg(p, td);
1054 rucollect(ru, &td->td_ru);
1055 }
1056 }
1057 }
1058
1059 /*
1060 * Atomically perform a rufetch and a calcru together.
1061 * Consumers, can safely assume the calcru is executed only once
1062 * rufetch is completed.
1063 */
1064 void
1065 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1066 struct timeval *sp)
1067 {
1068
1069 PROC_SLOCK(p);
1070 rufetch(p, ru);
1071 calcru(p, up, sp);
1072 PROC_SUNLOCK(p);
1073 }
1074
1075 /*
1076 * Allocate a new resource limits structure and initialize its
1077 * reference count and mutex pointer.
1078 */
1079 struct plimit *
1080 lim_alloc()
1081 {
1082 struct plimit *limp;
1083
1084 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1085 refcount_init(&limp->pl_refcnt, 1);
1086 return (limp);
1087 }
1088
1089 struct plimit *
1090 lim_hold(limp)
1091 struct plimit *limp;
1092 {
1093
1094 refcount_acquire(&limp->pl_refcnt);
1095 return (limp);
1096 }
1097
1098 void
1099 lim_fork(struct proc *p1, struct proc *p2)
1100 {
1101 p2->p_limit = lim_hold(p1->p_limit);
1102 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1103 if (p1->p_cpulimit != RLIM_INFINITY)
1104 callout_reset(&p2->p_limco, hz, lim_cb, p2);
1105 }
1106
1107 void
1108 lim_free(limp)
1109 struct plimit *limp;
1110 {
1111
1112 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
1113 if (refcount_release(&limp->pl_refcnt))
1114 free((void *)limp, M_PLIMIT);
1115 }
1116
1117 /*
1118 * Make a copy of the plimit structure.
1119 * We share these structures copy-on-write after fork.
1120 */
1121 void
1122 lim_copy(dst, src)
1123 struct plimit *dst, *src;
1124 {
1125
1126 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
1127 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1128 }
1129
1130 /*
1131 * Return the hard limit for a particular system resource. The
1132 * which parameter specifies the index into the rlimit array.
1133 */
1134 rlim_t
1135 lim_max(struct proc *p, int which)
1136 {
1137 struct rlimit rl;
1138
1139 lim_rlimit(p, which, &rl);
1140 return (rl.rlim_max);
1141 }
1142
1143 /*
1144 * Return the current (soft) limit for a particular system resource.
1145 * The which parameter which specifies the index into the rlimit array
1146 */
1147 rlim_t
1148 lim_cur(struct proc *p, int which)
1149 {
1150 struct rlimit rl;
1151
1152 lim_rlimit(p, which, &rl);
1153 return (rl.rlim_cur);
1154 }
1155
1156 /*
1157 * Return a copy of the entire rlimit structure for the system limit
1158 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1159 */
1160 void
1161 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
1162 {
1163
1164 PROC_LOCK_ASSERT(p, MA_OWNED);
1165 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1166 ("request for invalid resource limit"));
1167 *rlp = p->p_limit->pl_rlimit[which];
1168 if (p->p_sysent->sv_fixlimit != NULL)
1169 p->p_sysent->sv_fixlimit(rlp, which);
1170 }
1171
1172 /*
1173 * Find the uidinfo structure for a uid. This structure is used to
1174 * track the total resource consumption (process count, socket buffer
1175 * size, etc.) for the uid and impose limits.
1176 */
1177 void
1178 uihashinit()
1179 {
1180
1181 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1182 rw_init(&uihashtbl_lock, "uidinfo hash");
1183 }
1184
1185 /*
1186 * Look up a uidinfo struct for the parameter uid.
1187 * uihashtbl_lock must be locked.
1188 */
1189 static struct uidinfo *
1190 uilookup(uid)
1191 uid_t uid;
1192 {
1193 struct uihashhead *uipp;
1194 struct uidinfo *uip;
1195
1196 rw_assert(&uihashtbl_lock, RA_LOCKED);
1197 uipp = UIHASH(uid);
1198 LIST_FOREACH(uip, uipp, ui_hash)
1199 if (uip->ui_uid == uid)
1200 break;
1201
1202 return (uip);
1203 }
1204
1205 /*
1206 * Find or allocate a struct uidinfo for a particular uid.
1207 * Increase refcount on uidinfo struct returned.
1208 * uifree() should be called on a struct uidinfo when released.
1209 */
1210 struct uidinfo *
1211 uifind(uid)
1212 uid_t uid;
1213 {
1214 struct uidinfo *old_uip, *uip;
1215
1216 rw_rlock(&uihashtbl_lock);
1217 uip = uilookup(uid);
1218 if (uip == NULL) {
1219 rw_runlock(&uihashtbl_lock);
1220 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1221 rw_wlock(&uihashtbl_lock);
1222 /*
1223 * There's a chance someone created our uidinfo while we
1224 * were in malloc and not holding the lock, so we have to
1225 * make sure we don't insert a duplicate uidinfo.
1226 */
1227 if ((old_uip = uilookup(uid)) != NULL) {
1228 /* Someone else beat us to it. */
1229 free(uip, M_UIDINFO);
1230 uip = old_uip;
1231 } else {
1232 refcount_init(&uip->ui_ref, 0);
1233 uip->ui_uid = uid;
1234 mtx_init(&uip->ui_vmsize_mtx, "ui_vmsize", NULL,
1235 MTX_DEF);
1236 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1237 }
1238 }
1239 uihold(uip);
1240 rw_unlock(&uihashtbl_lock);
1241 return (uip);
1242 }
1243
1244 /*
1245 * Place another refcount on a uidinfo struct.
1246 */
1247 void
1248 uihold(uip)
1249 struct uidinfo *uip;
1250 {
1251
1252 refcount_acquire(&uip->ui_ref);
1253 }
1254
1255 /*-
1256 * Since uidinfo structs have a long lifetime, we use an
1257 * opportunistic refcounting scheme to avoid locking the lookup hash
1258 * for each release.
1259 *
1260 * If the refcount hits 0, we need to free the structure,
1261 * which means we need to lock the hash.
1262 * Optimal case:
1263 * After locking the struct and lowering the refcount, if we find
1264 * that we don't need to free, simply unlock and return.
1265 * Suboptimal case:
1266 * If refcount lowering results in need to free, bump the count
1267 * back up, lose the lock and acquire the locks in the proper
1268 * order to try again.
1269 */
1270 void
1271 uifree(uip)
1272 struct uidinfo *uip;
1273 {
1274 int old;
1275
1276 /* Prepare for optimal case. */
1277 old = uip->ui_ref;
1278 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1))
1279 return;
1280
1281 /* Prepare for suboptimal case. */
1282 rw_wlock(&uihashtbl_lock);
1283 if (refcount_release(&uip->ui_ref)) {
1284 LIST_REMOVE(uip, ui_hash);
1285 rw_wunlock(&uihashtbl_lock);
1286 if (uip->ui_sbsize != 0)
1287 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1288 uip->ui_uid, uip->ui_sbsize);
1289 if (uip->ui_proccnt != 0)
1290 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1291 uip->ui_uid, uip->ui_proccnt);
1292 if (uip->ui_vmsize != 0)
1293 printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1294 uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1295 mtx_destroy(&uip->ui_vmsize_mtx);
1296 free(uip, M_UIDINFO);
1297 return;
1298 }
1299 /*
1300 * Someone added a reference between atomic_cmpset_int() and
1301 * rw_wlock(&uihashtbl_lock).
1302 */
1303 rw_wunlock(&uihashtbl_lock);
1304 }
1305
1306 /*
1307 * Change the count associated with number of processes
1308 * a given user is using. When 'max' is 0, don't enforce a limit
1309 */
1310 int
1311 chgproccnt(uip, diff, max)
1312 struct uidinfo *uip;
1313 int diff;
1314 rlim_t max;
1315 {
1316
1317 /* Don't allow them to exceed max, but allow subtraction. */
1318 if (diff > 0 && max != 0) {
1319 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) {
1320 atomic_subtract_long(&uip->ui_proccnt, (long)diff);
1321 return (0);
1322 }
1323 } else {
1324 atomic_add_long(&uip->ui_proccnt, (long)diff);
1325 if (uip->ui_proccnt < 0)
1326 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1327 }
1328 return (1);
1329 }
1330
1331 /*
1332 * Change the total socket buffer size a user has used.
1333 */
1334 int
1335 chgsbsize(uip, hiwat, to, max)
1336 struct uidinfo *uip;
1337 u_int *hiwat;
1338 u_int to;
1339 rlim_t max;
1340 {
1341 int diff;
1342
1343 diff = to - *hiwat;
1344 if (diff > 0) {
1345 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) {
1346 atomic_subtract_long(&uip->ui_sbsize, (long)diff);
1347 return (0);
1348 }
1349 } else {
1350 atomic_add_long(&uip->ui_sbsize, (long)diff);
1351 if (uip->ui_sbsize < 0)
1352 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1353 }
1354 *hiwat = to;
1355 return (1);
1356 }
1357
1358 /*
1359 * Change the count associated with number of pseudo-terminals
1360 * a given user is using. When 'max' is 0, don't enforce a limit
1361 */
1362 int
1363 chgptscnt(uip, diff, max)
1364 struct uidinfo *uip;
1365 int diff;
1366 rlim_t max;
1367 {
1368
1369 /* Don't allow them to exceed max, but allow subtraction. */
1370 if (diff > 0 && max != 0) {
1371 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) {
1372 atomic_subtract_long(&uip->ui_ptscnt, (long)diff);
1373 return (0);
1374 }
1375 } else {
1376 atomic_add_long(&uip->ui_ptscnt, (long)diff);
1377 if (uip->ui_ptscnt < 0)
1378 printf("negative ptscnt for uid = %d\n", uip->ui_uid);
1379 }
1380 return (1);
1381 }
Cache object: ba3e2ceb858fc109a888c385d0decbfb
|