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/6.3/sys/kern/kern_resource.c 173886 2007-11-24 19:45:58Z cvs2svn $");
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/imgact.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
54 #include <sys/sx.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysent.h>
57 #include <sys/time.h>
58
59 #include <vm/vm.h>
60 #include <vm/vm_param.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_map.h>
63
64
65 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
66 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
67 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
68 static struct mtx uihashtbl_mtx;
69 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
70 static u_long uihash; /* size of hash table - 1 */
71
72 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
73 struct timeval *up, struct timeval *sp);
74 static int donice(struct thread *td, struct proc *chgp, int n);
75 static struct uidinfo *uilookup(uid_t uid);
76
77 /*
78 * Resource controls and accounting.
79 */
80
81 #ifndef _SYS_SYSPROTO_H_
82 struct getpriority_args {
83 int which;
84 int who;
85 };
86 #endif
87 /*
88 * MPSAFE
89 */
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)) {
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 LIST_FOREACH(p, &allproc, p_list) {
145 PROC_LOCK(p);
146 if (!p_cansee(td, p) &&
147 p->p_ucred->cr_uid == uap->who) {
148 if (p->p_nice < low)
149 low = p->p_nice;
150 }
151 PROC_UNLOCK(p);
152 }
153 sx_sunlock(&allproc_lock);
154 break;
155
156 default:
157 error = EINVAL;
158 break;
159 }
160 if (low == PRIO_MAX + 1 && error == 0)
161 error = ESRCH;
162 td->td_retval[0] = low;
163 return (error);
164 }
165
166 #ifndef _SYS_SYSPROTO_H_
167 struct setpriority_args {
168 int which;
169 int who;
170 int prio;
171 };
172 #endif
173 /*
174 * MPSAFE
175 */
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 == 0)
195 break;
196 if (p_cansee(td, p) == 0)
197 error = donice(td, p, uap->prio);
198 PROC_UNLOCK(p);
199 }
200 found++;
201 break;
202
203 case PRIO_PGRP:
204 sx_slock(&proctree_lock);
205 if (uap->who == 0) {
206 pg = curp->p_pgrp;
207 PGRP_LOCK(pg);
208 } else {
209 pg = pgfind(uap->who);
210 if (pg == NULL) {
211 sx_sunlock(&proctree_lock);
212 break;
213 }
214 }
215 sx_sunlock(&proctree_lock);
216 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
217 PROC_LOCK(p);
218 if (!p_cansee(td, p)) {
219 error = donice(td, p, uap->prio);
220 found++;
221 }
222 PROC_UNLOCK(p);
223 }
224 PGRP_UNLOCK(pg);
225 break;
226
227 case PRIO_USER:
228 if (uap->who == 0)
229 uap->who = td->td_ucred->cr_uid;
230 sx_slock(&allproc_lock);
231 FOREACH_PROC_IN_SYSTEM(p) {
232 /* Do not bother to check PRS_NEW processes */
233 if (p->p_state == PRS_NEW)
234 continue;
235 PROC_LOCK(p);
236 if (p->p_ucred->cr_uid == uap->who &&
237 !p_cansee(td, p)) {
238 error = donice(td, p, uap->prio);
239 found++;
240 }
241 PROC_UNLOCK(p);
242 }
243 sx_sunlock(&allproc_lock);
244 break;
245
246 default:
247 error = EINVAL;
248 break;
249 }
250 if (found == 0 && error == 0)
251 error = ESRCH;
252 return (error);
253 }
254
255 /*
256 * Set "nice" for a (whole) process.
257 */
258 static int
259 donice(struct thread *td, struct proc *p, int n)
260 {
261 int error;
262
263 PROC_LOCK_ASSERT(p, MA_OWNED);
264 if ((error = p_cansched(td, p)))
265 return (error);
266 if (n > PRIO_MAX)
267 n = PRIO_MAX;
268 if (n < PRIO_MIN)
269 n = PRIO_MIN;
270 if (n < p->p_nice && suser(td) != 0)
271 return (EACCES);
272 mtx_lock_spin(&sched_lock);
273 sched_nice(p, n);
274 mtx_unlock_spin(&sched_lock);
275 return (0);
276 }
277
278 /*
279 * Set realtime priority.
280 *
281 * MPSAFE
282 */
283 #ifndef _SYS_SYSPROTO_H_
284 struct rtprio_args {
285 int function;
286 pid_t pid;
287 struct rtprio *rtp;
288 };
289 #endif
290
291 int
292 rtprio(td, uap)
293 struct thread *td; /* curthread */
294 register struct rtprio_args *uap;
295 {
296 struct proc *curp;
297 struct proc *p;
298 struct ksegrp *kg;
299 struct rtprio rtp;
300 int cierror, error;
301
302 /* Perform copyin before acquiring locks if needed. */
303 if (uap->function == RTP_SET)
304 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
305 else
306 cierror = 0;
307
308 curp = td->td_proc;
309 if (uap->pid == 0) {
310 p = curp;
311 PROC_LOCK(p);
312 } else {
313 p = pfind(uap->pid);
314 if (p == NULL)
315 return (ESRCH);
316 }
317
318 switch (uap->function) {
319 case RTP_LOOKUP:
320 if ((error = p_cansee(td, p)))
321 break;
322 mtx_lock_spin(&sched_lock);
323 /*
324 * Return OUR priority if no pid specified,
325 * or if one is, report the highest priority
326 * in the process. There isn't much more you can do as
327 * there is only room to return a single priority.
328 * XXXKSE: maybe need a new interface to report
329 * priorities of multiple system scope threads.
330 * Note: specifying our own pid is not the same
331 * as leaving it zero.
332 */
333 if (uap->pid == 0) {
334 pri_to_rtp(td->td_ksegrp, &rtp);
335 } else {
336 struct rtprio rtp2;
337
338 rtp.type = RTP_PRIO_IDLE;
339 rtp.prio = RTP_PRIO_MAX;
340 FOREACH_KSEGRP_IN_PROC(p, kg) {
341 pri_to_rtp(kg, &rtp2);
342 if (rtp2.type < rtp.type ||
343 (rtp2.type == rtp.type &&
344 rtp2.prio < rtp.prio)) {
345 rtp.type = rtp2.type;
346 rtp.prio = rtp2.prio;
347 }
348 }
349 }
350 mtx_unlock_spin(&sched_lock);
351 PROC_UNLOCK(p);
352 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
353 case RTP_SET:
354 if ((error = p_cansched(td, p)) || (error = cierror))
355 break;
356
357 /* Disallow setting rtprio in most cases if not superuser. */
358 if (suser(td) != 0) {
359 /* can't set someone else's */
360 if (uap->pid) {
361 error = EPERM;
362 break;
363 }
364 /* can't set realtime priority */
365 /*
366 * Realtime priority has to be restricted for reasons which should be
367 * obvious. However, for idle priority, there is a potential for
368 * system deadlock if an idleprio process gains a lock on a resource
369 * that other processes need (and the idleprio process can't run
370 * due to a CPU-bound normal process). Fix me! XXX
371 */
372 #if 0
373 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
374 #else
375 if (rtp.type != RTP_PRIO_NORMAL) {
376 #endif
377 error = EPERM;
378 break;
379 }
380 }
381
382 /*
383 * If we are setting our own priority, set just our
384 * KSEGRP but if we are doing another process,
385 * do all the groups on that process. If we
386 * specify our own pid we do the latter.
387 */
388 mtx_lock_spin(&sched_lock);
389 if (uap->pid == 0) {
390 error = rtp_to_pri(&rtp, td->td_ksegrp);
391 } else {
392 FOREACH_KSEGRP_IN_PROC(p, kg) {
393 if ((error = rtp_to_pri(&rtp, kg)) != 0) {
394 break;
395 }
396 }
397 }
398 mtx_unlock_spin(&sched_lock);
399 break;
400 default:
401 error = EINVAL;
402 break;
403 }
404 PROC_UNLOCK(p);
405 return (error);
406 }
407
408 int
409 rtp_to_pri(struct rtprio *rtp, struct ksegrp *kg)
410 {
411
412 mtx_assert(&sched_lock, MA_OWNED);
413 if (rtp->prio > RTP_PRIO_MAX)
414 return (EINVAL);
415 switch (RTP_PRIO_BASE(rtp->type)) {
416 case RTP_PRIO_REALTIME:
417 kg->kg_user_pri = PRI_MIN_REALTIME + rtp->prio;
418 break;
419 case RTP_PRIO_NORMAL:
420 kg->kg_user_pri = PRI_MIN_TIMESHARE + rtp->prio;
421 break;
422 case RTP_PRIO_IDLE:
423 kg->kg_user_pri = PRI_MIN_IDLE + rtp->prio;
424 break;
425 default:
426 return (EINVAL);
427 }
428 sched_class(kg, rtp->type);
429 if (curthread->td_ksegrp == kg) {
430 sched_prio(curthread, kg->kg_user_pri); /* XXX dubious */
431 }
432 return (0);
433 }
434
435 void
436 pri_to_rtp(struct ksegrp *kg, struct rtprio *rtp)
437 {
438
439 mtx_assert(&sched_lock, MA_OWNED);
440 switch (PRI_BASE(kg->kg_pri_class)) {
441 case PRI_REALTIME:
442 rtp->prio = kg->kg_user_pri - PRI_MIN_REALTIME;
443 break;
444 case PRI_TIMESHARE:
445 rtp->prio = kg->kg_user_pri - PRI_MIN_TIMESHARE;
446 break;
447 case PRI_IDLE:
448 rtp->prio = kg->kg_user_pri - PRI_MIN_IDLE;
449 break;
450 default:
451 break;
452 }
453 rtp->type = kg->kg_pri_class;
454 }
455
456 #if defined(COMPAT_43)
457 #ifndef _SYS_SYSPROTO_H_
458 struct osetrlimit_args {
459 u_int which;
460 struct orlimit *rlp;
461 };
462 #endif
463 /*
464 * MPSAFE
465 */
466 int
467 osetrlimit(td, uap)
468 struct thread *td;
469 register struct osetrlimit_args *uap;
470 {
471 struct orlimit olim;
472 struct rlimit lim;
473 int error;
474
475 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
476 return (error);
477 lim.rlim_cur = olim.rlim_cur;
478 lim.rlim_max = olim.rlim_max;
479 error = kern_setrlimit(td, uap->which, &lim);
480 return (error);
481 }
482
483 #ifndef _SYS_SYSPROTO_H_
484 struct ogetrlimit_args {
485 u_int which;
486 struct orlimit *rlp;
487 };
488 #endif
489 /*
490 * MPSAFE
491 */
492 int
493 ogetrlimit(td, uap)
494 struct thread *td;
495 register struct ogetrlimit_args *uap;
496 {
497 struct orlimit olim;
498 struct rlimit rl;
499 struct proc *p;
500 int error;
501
502 if (uap->which >= RLIM_NLIMITS)
503 return (EINVAL);
504 p = td->td_proc;
505 PROC_LOCK(p);
506 lim_rlimit(p, uap->which, &rl);
507 PROC_UNLOCK(p);
508
509 /*
510 * XXX would be more correct to convert only RLIM_INFINITY to the
511 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
512 * values. Most 64->32 and 32->16 conversions, including not
513 * unimportant ones of uids are even more broken than what we
514 * do here (they blindly truncate). We don't do this correctly
515 * here since we have little experience with EOVERFLOW yet.
516 * Elsewhere, getuid() can't fail...
517 */
518 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
519 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
520 error = copyout(&olim, uap->rlp, sizeof(olim));
521 return (error);
522 }
523 #endif /* COMPAT_43 */
524
525 #ifndef _SYS_SYSPROTO_H_
526 struct __setrlimit_args {
527 u_int which;
528 struct rlimit *rlp;
529 };
530 #endif
531 /*
532 * MPSAFE
533 */
534 int
535 setrlimit(td, uap)
536 struct thread *td;
537 register struct __setrlimit_args *uap;
538 {
539 struct rlimit alim;
540 int error;
541
542 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
543 return (error);
544 error = kern_setrlimit(td, uap->which, &alim);
545 return (error);
546 }
547
548 int
549 kern_setrlimit(td, which, limp)
550 struct thread *td;
551 u_int which;
552 struct rlimit *limp;
553 {
554 struct plimit *newlim, *oldlim;
555 struct proc *p;
556 register struct rlimit *alimp;
557 rlim_t oldssiz;
558 int error;
559
560 if (which >= RLIM_NLIMITS)
561 return (EINVAL);
562
563 /*
564 * Preserve historical bugs by treating negative limits as unsigned.
565 */
566 if (limp->rlim_cur < 0)
567 limp->rlim_cur = RLIM_INFINITY;
568 if (limp->rlim_max < 0)
569 limp->rlim_max = RLIM_INFINITY;
570
571 oldssiz = 0;
572 p = td->td_proc;
573 newlim = lim_alloc();
574 PROC_LOCK(p);
575 oldlim = p->p_limit;
576 alimp = &oldlim->pl_rlimit[which];
577 if (limp->rlim_cur > alimp->rlim_max ||
578 limp->rlim_max > alimp->rlim_max)
579 if ((error = suser_cred(td->td_ucred, SUSER_ALLOWJAIL))) {
580 PROC_UNLOCK(p);
581 lim_free(newlim);
582 return (error);
583 }
584 if (limp->rlim_cur > limp->rlim_max)
585 limp->rlim_cur = limp->rlim_max;
586 lim_copy(newlim, oldlim);
587 alimp = &newlim->pl_rlimit[which];
588
589 switch (which) {
590
591 case RLIMIT_CPU:
592 mtx_lock_spin(&sched_lock);
593 p->p_cpulimit = limp->rlim_cur;
594 mtx_unlock_spin(&sched_lock);
595 break;
596 case RLIMIT_DATA:
597 if (limp->rlim_cur > maxdsiz)
598 limp->rlim_cur = maxdsiz;
599 if (limp->rlim_max > maxdsiz)
600 limp->rlim_max = maxdsiz;
601 break;
602
603 case RLIMIT_STACK:
604 if (limp->rlim_cur > maxssiz)
605 limp->rlim_cur = maxssiz;
606 if (limp->rlim_max > maxssiz)
607 limp->rlim_max = maxssiz;
608 oldssiz = alimp->rlim_cur;
609 break;
610
611 case RLIMIT_NOFILE:
612 if (limp->rlim_cur > maxfilesperproc)
613 limp->rlim_cur = maxfilesperproc;
614 if (limp->rlim_max > maxfilesperproc)
615 limp->rlim_max = maxfilesperproc;
616 break;
617
618 case RLIMIT_NPROC:
619 if (limp->rlim_cur > maxprocperuid)
620 limp->rlim_cur = maxprocperuid;
621 if (limp->rlim_max > maxprocperuid)
622 limp->rlim_max = maxprocperuid;
623 if (limp->rlim_cur < 1)
624 limp->rlim_cur = 1;
625 if (limp->rlim_max < 1)
626 limp->rlim_max = 1;
627 break;
628 }
629 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
630 td->td_proc->p_sysent->sv_fixlimit(limp, which);
631 *alimp = *limp;
632 p->p_limit = newlim;
633 PROC_UNLOCK(p);
634 lim_free(oldlim);
635
636 if (which == RLIMIT_STACK) {
637 /*
638 * Stack is allocated to the max at exec time with only
639 * "rlim_cur" bytes accessible. If stack limit is going
640 * up make more accessible, if going down make inaccessible.
641 */
642 if (limp->rlim_cur != oldssiz) {
643 vm_offset_t addr;
644 vm_size_t size;
645 vm_prot_t prot;
646
647 if (limp->rlim_cur > oldssiz) {
648 prot = p->p_sysent->sv_stackprot;
649 size = limp->rlim_cur - oldssiz;
650 addr = p->p_sysent->sv_usrstack -
651 limp->rlim_cur;
652 } else {
653 prot = VM_PROT_NONE;
654 size = oldssiz - limp->rlim_cur;
655 addr = p->p_sysent->sv_usrstack - oldssiz;
656 }
657 addr = trunc_page(addr);
658 size = round_page(size);
659 (void)vm_map_protect(&p->p_vmspace->vm_map,
660 addr, addr + size, prot, FALSE);
661 }
662 }
663
664 return (0);
665 }
666
667 #ifndef _SYS_SYSPROTO_H_
668 struct __getrlimit_args {
669 u_int which;
670 struct rlimit *rlp;
671 };
672 #endif
673 /*
674 * MPSAFE
675 */
676 /* ARGSUSED */
677 int
678 getrlimit(td, uap)
679 struct thread *td;
680 register struct __getrlimit_args *uap;
681 {
682 struct rlimit rlim;
683 struct proc *p;
684 int error;
685
686 if (uap->which >= RLIM_NLIMITS)
687 return (EINVAL);
688 p = td->td_proc;
689 PROC_LOCK(p);
690 lim_rlimit(p, uap->which, &rlim);
691 PROC_UNLOCK(p);
692 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
693 return (error);
694 }
695
696 /*
697 * Transform the running time and tick information in proc p into user,
698 * system, and interrupt time usage.
699 */
700 void
701 calcru(p, up, sp)
702 struct proc *p;
703 struct timeval *up;
704 struct timeval *sp;
705 {
706 struct bintime bt;
707 struct rusage_ext rux;
708 struct thread *td;
709 int bt_valid;
710
711 PROC_LOCK_ASSERT(p, MA_OWNED);
712 mtx_assert(&sched_lock, MA_NOTOWNED);
713 bt_valid = 0;
714 mtx_lock_spin(&sched_lock);
715 rux = p->p_rux;
716 FOREACH_THREAD_IN_PROC(p, td) {
717 if (TD_IS_RUNNING(td)) {
718 /*
719 * Adjust for the current time slice. This is
720 * actually fairly important since the error here is
721 * on the order of a time quantum which is much
722 * greater than the precision of binuptime().
723 */
724 KASSERT(td->td_oncpu != NOCPU,
725 ("%s: running thread has no CPU", __func__));
726 if (!bt_valid) {
727 binuptime(&bt);
728 bt_valid = 1;
729 }
730 bintime_add(&rux.rux_runtime, &bt);
731 bintime_sub(&rux.rux_runtime,
732 &pcpu_find(td->td_oncpu)->pc_switchtime);
733 }
734 }
735 mtx_unlock_spin(&sched_lock);
736 calcru1(p, &rux, up, sp);
737 p->p_rux.rux_uu = rux.rux_uu;
738 p->p_rux.rux_su = rux.rux_su;
739 p->p_rux.rux_iu = rux.rux_iu;
740 }
741
742 void
743 calccru(p, up, sp)
744 struct proc *p;
745 struct timeval *up;
746 struct timeval *sp;
747 {
748
749 PROC_LOCK_ASSERT(p, MA_OWNED);
750 calcru1(p, &p->p_crux, up, sp);
751 }
752
753 static void
754 calcru1(p, ruxp, up, sp)
755 struct proc *p;
756 struct rusage_ext *ruxp;
757 struct timeval *up;
758 struct timeval *sp;
759 {
760 struct timeval tv;
761 /* {user, system, interrupt, total} {ticks, usec}; previous tu: */
762 u_int64_t ut, uu, st, su, it, iu, tt, tu, ptu;
763
764 ut = ruxp->rux_uticks;
765 st = ruxp->rux_sticks;
766 it = ruxp->rux_iticks;
767 tt = ut + st + it;
768 if (tt == 0) {
769 st = 1;
770 tt = 1;
771 }
772 bintime2timeval(&ruxp->rux_runtime, &tv);
773 tu = (u_int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
774 ptu = ruxp->rux_uu + ruxp->rux_su + ruxp->rux_iu;
775 if (tu < ptu) {
776 printf(
777 "calcru: runtime went backwards from %ju usec to %ju usec for pid %d (%s)\n",
778 (uintmax_t)ptu, (uintmax_t)tu, p->p_pid, p->p_comm);
779 tu = ptu;
780 }
781 if ((int64_t)tu < 0) {
782 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
783 (intmax_t)tu, p->p_pid, p->p_comm);
784 tu = ptu;
785 }
786
787 /* Subdivide tu. */
788 uu = (tu * ut) / tt;
789 su = (tu * st) / tt;
790 iu = tu - uu - su;
791
792 /* Enforce monotonicity. */
793 if (uu < ruxp->rux_uu || su < ruxp->rux_su || iu < ruxp->rux_iu) {
794 if (uu < ruxp->rux_uu)
795 uu = ruxp->rux_uu;
796 else if (uu + ruxp->rux_su + ruxp->rux_iu > tu)
797 uu = tu - ruxp->rux_su - ruxp->rux_iu;
798 if (st == 0)
799 su = ruxp->rux_su;
800 else {
801 su = ((tu - uu) * st) / (st + it);
802 if (su < ruxp->rux_su)
803 su = ruxp->rux_su;
804 else if (uu + su + ruxp->rux_iu > tu)
805 su = tu - uu - ruxp->rux_iu;
806 }
807 KASSERT(uu + su + ruxp->rux_iu <= tu,
808 ("calcru: monotonisation botch 1"));
809 iu = tu - uu - su;
810 KASSERT(iu >= ruxp->rux_iu,
811 ("calcru: monotonisation botch 2"));
812 }
813 ruxp->rux_uu = uu;
814 ruxp->rux_su = su;
815 ruxp->rux_iu = iu;
816
817 up->tv_sec = uu / 1000000;
818 up->tv_usec = uu % 1000000;
819 sp->tv_sec = su / 1000000;
820 sp->tv_usec = su % 1000000;
821 }
822
823 #ifndef _SYS_SYSPROTO_H_
824 struct getrusage_args {
825 int who;
826 struct rusage *rusage;
827 };
828 #endif
829 /*
830 * MPSAFE
831 */
832 int
833 getrusage(td, uap)
834 register struct thread *td;
835 register struct getrusage_args *uap;
836 {
837 struct rusage ru;
838 int error;
839
840 error = kern_getrusage(td, uap->who, &ru);
841 if (error == 0)
842 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
843 return (error);
844 }
845
846 int
847 kern_getrusage(td, who, rup)
848 struct thread *td;
849 int who;
850 struct rusage *rup;
851 {
852 struct proc *p;
853
854 p = td->td_proc;
855 PROC_LOCK(p);
856 switch (who) {
857
858 case RUSAGE_SELF:
859 *rup = p->p_stats->p_ru;
860 calcru(p, &rup->ru_utime, &rup->ru_stime);
861 break;
862
863 case RUSAGE_CHILDREN:
864 *rup = p->p_stats->p_cru;
865 calccru(p, &rup->ru_utime, &rup->ru_stime);
866 break;
867
868 default:
869 PROC_UNLOCK(p);
870 return (EINVAL);
871 }
872 PROC_UNLOCK(p);
873 return (0);
874 }
875
876 void
877 ruadd(ru, rux, ru2, rux2)
878 struct rusage *ru;
879 struct rusage_ext *rux;
880 struct rusage *ru2;
881 struct rusage_ext *rux2;
882 {
883 register long *ip, *ip2;
884 register int i;
885
886 bintime_add(&rux->rux_runtime, &rux2->rux_runtime);
887 rux->rux_uticks += rux2->rux_uticks;
888 rux->rux_sticks += rux2->rux_sticks;
889 rux->rux_iticks += rux2->rux_iticks;
890 rux->rux_uu += rux2->rux_uu;
891 rux->rux_su += rux2->rux_su;
892 rux->rux_iu += rux2->rux_iu;
893 if (ru->ru_maxrss < ru2->ru_maxrss)
894 ru->ru_maxrss = ru2->ru_maxrss;
895 ip = &ru->ru_first;
896 ip2 = &ru2->ru_first;
897 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
898 *ip++ += *ip2++;
899 }
900
901 /*
902 * Allocate a new resource limits structure and initialize its
903 * reference count and mutex pointer.
904 */
905 struct plimit *
906 lim_alloc()
907 {
908 struct plimit *limp;
909
910 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
911 limp->pl_refcnt = 1;
912 limp->pl_mtx = mtx_pool_alloc(mtxpool_sleep);
913 return (limp);
914 }
915
916 struct plimit *
917 lim_hold(limp)
918 struct plimit *limp;
919 {
920
921 LIM_LOCK(limp);
922 limp->pl_refcnt++;
923 LIM_UNLOCK(limp);
924 return (limp);
925 }
926
927 void
928 lim_free(limp)
929 struct plimit *limp;
930 {
931
932 LIM_LOCK(limp);
933 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
934 if (--limp->pl_refcnt == 0) {
935 LIM_UNLOCK(limp);
936 free((void *)limp, M_PLIMIT);
937 return;
938 }
939 LIM_UNLOCK(limp);
940 }
941
942 /*
943 * Make a copy of the plimit structure.
944 * We share these structures copy-on-write after fork.
945 */
946 void
947 lim_copy(dst, src)
948 struct plimit *dst, *src;
949 {
950
951 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
952 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
953 }
954
955 /*
956 * Return the hard limit for a particular system resource. The
957 * which parameter specifies the index into the rlimit array.
958 */
959 rlim_t
960 lim_max(struct proc *p, int which)
961 {
962 struct rlimit rl;
963
964 lim_rlimit(p, which, &rl);
965 return (rl.rlim_max);
966 }
967
968 /*
969 * Return the current (soft) limit for a particular system resource.
970 * The which parameter which specifies the index into the rlimit array
971 */
972 rlim_t
973 lim_cur(struct proc *p, int which)
974 {
975 struct rlimit rl;
976
977 lim_rlimit(p, which, &rl);
978 return (rl.rlim_cur);
979 }
980
981 /*
982 * Return a copy of the entire rlimit structure for the system limit
983 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
984 */
985 void
986 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
987 {
988
989 PROC_LOCK_ASSERT(p, MA_OWNED);
990 KASSERT(which >= 0 && which < RLIM_NLIMITS,
991 ("request for invalid resource limit"));
992 *rlp = p->p_limit->pl_rlimit[which];
993 if (p->p_sysent->sv_fixlimit != NULL)
994 p->p_sysent->sv_fixlimit(rlp, which);
995 }
996
997 /*
998 * Find the uidinfo structure for a uid. This structure is used to
999 * track the total resource consumption (process count, socket buffer
1000 * size, etc.) for the uid and impose limits.
1001 */
1002 void
1003 uihashinit()
1004 {
1005
1006 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1007 mtx_init(&uihashtbl_mtx, "uidinfo hash", NULL, MTX_DEF);
1008 }
1009
1010 /*
1011 * Look up a uidinfo struct for the parameter uid.
1012 * uihashtbl_mtx must be locked.
1013 */
1014 static struct uidinfo *
1015 uilookup(uid)
1016 uid_t uid;
1017 {
1018 struct uihashhead *uipp;
1019 struct uidinfo *uip;
1020
1021 mtx_assert(&uihashtbl_mtx, MA_OWNED);
1022 uipp = UIHASH(uid);
1023 LIST_FOREACH(uip, uipp, ui_hash)
1024 if (uip->ui_uid == uid)
1025 break;
1026
1027 return (uip);
1028 }
1029
1030 /*
1031 * Find or allocate a struct uidinfo for a particular uid.
1032 * Increase refcount on uidinfo struct returned.
1033 * uifree() should be called on a struct uidinfo when released.
1034 */
1035 struct uidinfo *
1036 uifind(uid)
1037 uid_t uid;
1038 {
1039 struct uidinfo *old_uip, *uip;
1040
1041 mtx_lock(&uihashtbl_mtx);
1042 uip = uilookup(uid);
1043 if (uip == NULL) {
1044 mtx_unlock(&uihashtbl_mtx);
1045 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1046 mtx_lock(&uihashtbl_mtx);
1047 /*
1048 * There's a chance someone created our uidinfo while we
1049 * were in malloc and not holding the lock, so we have to
1050 * make sure we don't insert a duplicate uidinfo.
1051 */
1052 if ((old_uip = uilookup(uid)) != NULL) {
1053 /* Someone else beat us to it. */
1054 free(uip, M_UIDINFO);
1055 uip = old_uip;
1056 } else {
1057 uip->ui_mtxp = mtx_pool_alloc(mtxpool_sleep);
1058 uip->ui_uid = uid;
1059 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1060 }
1061 }
1062 uihold(uip);
1063 mtx_unlock(&uihashtbl_mtx);
1064 return (uip);
1065 }
1066
1067 /*
1068 * Place another refcount on a uidinfo struct.
1069 */
1070 void
1071 uihold(uip)
1072 struct uidinfo *uip;
1073 {
1074
1075 UIDINFO_LOCK(uip);
1076 uip->ui_ref++;
1077 UIDINFO_UNLOCK(uip);
1078 }
1079
1080 /*-
1081 * Since uidinfo structs have a long lifetime, we use an
1082 * opportunistic refcounting scheme to avoid locking the lookup hash
1083 * for each release.
1084 *
1085 * If the refcount hits 0, we need to free the structure,
1086 * which means we need to lock the hash.
1087 * Optimal case:
1088 * After locking the struct and lowering the refcount, if we find
1089 * that we don't need to free, simply unlock and return.
1090 * Suboptimal case:
1091 * If refcount lowering results in need to free, bump the count
1092 * back up, loose the lock and aquire the locks in the proper
1093 * order to try again.
1094 */
1095 void
1096 uifree(uip)
1097 struct uidinfo *uip;
1098 {
1099
1100 /* Prepare for optimal case. */
1101 UIDINFO_LOCK(uip);
1102
1103 if (--uip->ui_ref != 0) {
1104 UIDINFO_UNLOCK(uip);
1105 return;
1106 }
1107
1108 /* Prepare for suboptimal case. */
1109 uip->ui_ref++;
1110 UIDINFO_UNLOCK(uip);
1111 mtx_lock(&uihashtbl_mtx);
1112 UIDINFO_LOCK(uip);
1113
1114 /*
1115 * We must subtract one from the count again because we backed out
1116 * our initial subtraction before dropping the lock.
1117 * Since another thread may have added a reference after we dropped the
1118 * initial lock we have to test for zero again.
1119 */
1120 if (--uip->ui_ref == 0) {
1121 LIST_REMOVE(uip, ui_hash);
1122 mtx_unlock(&uihashtbl_mtx);
1123 if (uip->ui_sbsize != 0)
1124 printf("freeing uidinfo: uid = %d, sbsize = %jd\n",
1125 uip->ui_uid, (intmax_t)uip->ui_sbsize);
1126 if (uip->ui_proccnt != 0)
1127 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1128 uip->ui_uid, uip->ui_proccnt);
1129 UIDINFO_UNLOCK(uip);
1130 FREE(uip, M_UIDINFO);
1131 return;
1132 }
1133
1134 mtx_unlock(&uihashtbl_mtx);
1135 UIDINFO_UNLOCK(uip);
1136 }
1137
1138 /*
1139 * Change the count associated with number of processes
1140 * a given user is using. When 'max' is 0, don't enforce a limit
1141 */
1142 int
1143 chgproccnt(uip, diff, max)
1144 struct uidinfo *uip;
1145 int diff;
1146 int max;
1147 {
1148
1149 UIDINFO_LOCK(uip);
1150 /* Don't allow them to exceed max, but allow subtraction. */
1151 if (diff > 0 && uip->ui_proccnt + diff > max && max != 0) {
1152 UIDINFO_UNLOCK(uip);
1153 return (0);
1154 }
1155 uip->ui_proccnt += diff;
1156 if (uip->ui_proccnt < 0)
1157 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1158 UIDINFO_UNLOCK(uip);
1159 return (1);
1160 }
1161
1162 /*
1163 * Change the total socket buffer size a user has used.
1164 */
1165 int
1166 chgsbsize(uip, hiwat, to, max)
1167 struct uidinfo *uip;
1168 u_int *hiwat;
1169 u_int to;
1170 rlim_t max;
1171 {
1172 rlim_t new;
1173
1174 UIDINFO_LOCK(uip);
1175 new = uip->ui_sbsize + to - *hiwat;
1176 /* Don't allow them to exceed max, but allow subtraction. */
1177 if (to > *hiwat && new > max) {
1178 UIDINFO_UNLOCK(uip);
1179 return (0);
1180 }
1181 uip->ui_sbsize = new;
1182 UIDINFO_UNLOCK(uip);
1183 *hiwat = to;
1184 if (new < 0)
1185 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1186 return (1);
1187 }
Cache object: 6d11977e50ebc09d2c54e8655471a65f
|