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/6.1/sys/kern/kern_resource.c 158179 2006-04-30 16:44:43Z 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 PROC_LOCK(p);
233 if (p->p_ucred->cr_uid == uap->who &&
234 !p_cansee(td, p)) {
235 error = donice(td, p, uap->prio);
236 found++;
237 }
238 PROC_UNLOCK(p);
239 }
240 sx_sunlock(&allproc_lock);
241 break;
242
243 default:
244 error = EINVAL;
245 break;
246 }
247 if (found == 0 && error == 0)
248 error = ESRCH;
249 return (error);
250 }
251
252 /*
253 * Set "nice" for a (whole) process.
254 */
255 static int
256 donice(struct thread *td, struct proc *p, int n)
257 {
258 int error;
259
260 PROC_LOCK_ASSERT(p, MA_OWNED);
261 if ((error = p_cansched(td, p)))
262 return (error);
263 if (n > PRIO_MAX)
264 n = PRIO_MAX;
265 if (n < PRIO_MIN)
266 n = PRIO_MIN;
267 if (n < p->p_nice && suser(td) != 0)
268 return (EACCES);
269 mtx_lock_spin(&sched_lock);
270 sched_nice(p, n);
271 mtx_unlock_spin(&sched_lock);
272 return (0);
273 }
274
275 /*
276 * Set realtime priority.
277 *
278 * MPSAFE
279 */
280 #ifndef _SYS_SYSPROTO_H_
281 struct rtprio_args {
282 int function;
283 pid_t pid;
284 struct rtprio *rtp;
285 };
286 #endif
287
288 int
289 rtprio(td, uap)
290 struct thread *td; /* curthread */
291 register struct rtprio_args *uap;
292 {
293 struct proc *curp;
294 struct proc *p;
295 struct ksegrp *kg;
296 struct rtprio rtp;
297 int cierror, error;
298
299 /* Perform copyin before acquiring locks if needed. */
300 if (uap->function == RTP_SET)
301 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
302 else
303 cierror = 0;
304
305 curp = td->td_proc;
306 if (uap->pid == 0) {
307 p = curp;
308 PROC_LOCK(p);
309 } else {
310 p = pfind(uap->pid);
311 if (p == NULL)
312 return (ESRCH);
313 }
314
315 switch (uap->function) {
316 case RTP_LOOKUP:
317 if ((error = p_cansee(td, p)))
318 break;
319 mtx_lock_spin(&sched_lock);
320 /*
321 * Return OUR priority if no pid specified,
322 * or if one is, report the highest priority
323 * in the process. There isn't much more you can do as
324 * there is only room to return a single priority.
325 * XXXKSE: maybe need a new interface to report
326 * priorities of multiple system scope threads.
327 * Note: specifying our own pid is not the same
328 * as leaving it zero.
329 */
330 if (uap->pid == 0) {
331 pri_to_rtp(td->td_ksegrp, &rtp);
332 } else {
333 struct rtprio rtp2;
334
335 rtp.type = RTP_PRIO_IDLE;
336 rtp.prio = RTP_PRIO_MAX;
337 FOREACH_KSEGRP_IN_PROC(p, kg) {
338 pri_to_rtp(kg, &rtp2);
339 if (rtp2.type < rtp.type ||
340 (rtp2.type == rtp.type &&
341 rtp2.prio < rtp.prio)) {
342 rtp.type = rtp2.type;
343 rtp.prio = rtp2.prio;
344 }
345 }
346 }
347 mtx_unlock_spin(&sched_lock);
348 PROC_UNLOCK(p);
349 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
350 case RTP_SET:
351 if ((error = p_cansched(td, p)) || (error = cierror))
352 break;
353
354 /* Disallow setting rtprio in most cases if not superuser. */
355 if (suser(td) != 0) {
356 /* can't set someone else's */
357 if (uap->pid) {
358 error = EPERM;
359 break;
360 }
361 /* can't set realtime priority */
362 /*
363 * Realtime priority has to be restricted for reasons which should be
364 * obvious. However, for idle priority, there is a potential for
365 * system deadlock if an idleprio process gains a lock on a resource
366 * that other processes need (and the idleprio process can't run
367 * due to a CPU-bound normal process). Fix me! XXX
368 */
369 #if 0
370 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
371 #else
372 if (rtp.type != RTP_PRIO_NORMAL) {
373 #endif
374 error = EPERM;
375 break;
376 }
377 }
378
379 /*
380 * If we are setting our own priority, set just our
381 * KSEGRP but if we are doing another process,
382 * do all the groups on that process. If we
383 * specify our own pid we do the latter.
384 */
385 mtx_lock_spin(&sched_lock);
386 if (uap->pid == 0) {
387 error = rtp_to_pri(&rtp, td->td_ksegrp);
388 } else {
389 FOREACH_KSEGRP_IN_PROC(p, kg) {
390 if ((error = rtp_to_pri(&rtp, kg)) != 0) {
391 break;
392 }
393 }
394 }
395 mtx_unlock_spin(&sched_lock);
396 break;
397 default:
398 error = EINVAL;
399 break;
400 }
401 PROC_UNLOCK(p);
402 return (error);
403 }
404
405 int
406 rtp_to_pri(struct rtprio *rtp, struct ksegrp *kg)
407 {
408
409 mtx_assert(&sched_lock, MA_OWNED);
410 if (rtp->prio > RTP_PRIO_MAX)
411 return (EINVAL);
412 switch (RTP_PRIO_BASE(rtp->type)) {
413 case RTP_PRIO_REALTIME:
414 kg->kg_user_pri = PRI_MIN_REALTIME + rtp->prio;
415 break;
416 case RTP_PRIO_NORMAL:
417 kg->kg_user_pri = PRI_MIN_TIMESHARE + rtp->prio;
418 break;
419 case RTP_PRIO_IDLE:
420 kg->kg_user_pri = PRI_MIN_IDLE + rtp->prio;
421 break;
422 default:
423 return (EINVAL);
424 }
425 sched_class(kg, rtp->type);
426 if (curthread->td_ksegrp == kg) {
427 sched_prio(curthread, kg->kg_user_pri); /* XXX dubious */
428 }
429 return (0);
430 }
431
432 void
433 pri_to_rtp(struct ksegrp *kg, struct rtprio *rtp)
434 {
435
436 mtx_assert(&sched_lock, MA_OWNED);
437 switch (PRI_BASE(kg->kg_pri_class)) {
438 case PRI_REALTIME:
439 rtp->prio = kg->kg_user_pri - PRI_MIN_REALTIME;
440 break;
441 case PRI_TIMESHARE:
442 rtp->prio = kg->kg_user_pri - PRI_MIN_TIMESHARE;
443 break;
444 case PRI_IDLE:
445 rtp->prio = kg->kg_user_pri - PRI_MIN_IDLE;
446 break;
447 default:
448 break;
449 }
450 rtp->type = kg->kg_pri_class;
451 }
452
453 #if defined(COMPAT_43)
454 #ifndef _SYS_SYSPROTO_H_
455 struct osetrlimit_args {
456 u_int which;
457 struct orlimit *rlp;
458 };
459 #endif
460 /*
461 * MPSAFE
462 */
463 int
464 osetrlimit(td, uap)
465 struct thread *td;
466 register struct osetrlimit_args *uap;
467 {
468 struct orlimit olim;
469 struct rlimit lim;
470 int error;
471
472 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
473 return (error);
474 lim.rlim_cur = olim.rlim_cur;
475 lim.rlim_max = olim.rlim_max;
476 error = kern_setrlimit(td, uap->which, &lim);
477 return (error);
478 }
479
480 #ifndef _SYS_SYSPROTO_H_
481 struct ogetrlimit_args {
482 u_int which;
483 struct orlimit *rlp;
484 };
485 #endif
486 /*
487 * MPSAFE
488 */
489 int
490 ogetrlimit(td, uap)
491 struct thread *td;
492 register struct ogetrlimit_args *uap;
493 {
494 struct orlimit olim;
495 struct rlimit rl;
496 struct proc *p;
497 int error;
498
499 if (uap->which >= RLIM_NLIMITS)
500 return (EINVAL);
501 p = td->td_proc;
502 PROC_LOCK(p);
503 lim_rlimit(p, uap->which, &rl);
504 PROC_UNLOCK(p);
505
506 /*
507 * XXX would be more correct to convert only RLIM_INFINITY to the
508 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
509 * values. Most 64->32 and 32->16 conversions, including not
510 * unimportant ones of uids are even more broken than what we
511 * do here (they blindly truncate). We don't do this correctly
512 * here since we have little experience with EOVERFLOW yet.
513 * Elsewhere, getuid() can't fail...
514 */
515 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
516 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
517 error = copyout(&olim, uap->rlp, sizeof(olim));
518 return (error);
519 }
520 #endif /* COMPAT_43 */
521
522 #ifndef _SYS_SYSPROTO_H_
523 struct __setrlimit_args {
524 u_int which;
525 struct rlimit *rlp;
526 };
527 #endif
528 /*
529 * MPSAFE
530 */
531 int
532 setrlimit(td, uap)
533 struct thread *td;
534 register struct __setrlimit_args *uap;
535 {
536 struct rlimit alim;
537 int error;
538
539 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
540 return (error);
541 error = kern_setrlimit(td, uap->which, &alim);
542 return (error);
543 }
544
545 int
546 kern_setrlimit(td, which, limp)
547 struct thread *td;
548 u_int which;
549 struct rlimit *limp;
550 {
551 struct plimit *newlim, *oldlim;
552 struct proc *p;
553 register struct rlimit *alimp;
554 rlim_t oldssiz;
555 int error;
556
557 if (which >= RLIM_NLIMITS)
558 return (EINVAL);
559
560 /*
561 * Preserve historical bugs by treating negative limits as unsigned.
562 */
563 if (limp->rlim_cur < 0)
564 limp->rlim_cur = RLIM_INFINITY;
565 if (limp->rlim_max < 0)
566 limp->rlim_max = RLIM_INFINITY;
567
568 oldssiz = 0;
569 p = td->td_proc;
570 newlim = lim_alloc();
571 PROC_LOCK(p);
572 oldlim = p->p_limit;
573 alimp = &oldlim->pl_rlimit[which];
574 if (limp->rlim_cur > alimp->rlim_max ||
575 limp->rlim_max > alimp->rlim_max)
576 if ((error = suser_cred(td->td_ucred, SUSER_ALLOWJAIL))) {
577 PROC_UNLOCK(p);
578 lim_free(newlim);
579 return (error);
580 }
581 if (limp->rlim_cur > limp->rlim_max)
582 limp->rlim_cur = limp->rlim_max;
583 lim_copy(newlim, oldlim);
584 alimp = &newlim->pl_rlimit[which];
585
586 switch (which) {
587
588 case RLIMIT_CPU:
589 mtx_lock_spin(&sched_lock);
590 p->p_cpulimit = limp->rlim_cur;
591 mtx_unlock_spin(&sched_lock);
592 break;
593 case RLIMIT_DATA:
594 if (limp->rlim_cur > maxdsiz)
595 limp->rlim_cur = maxdsiz;
596 if (limp->rlim_max > maxdsiz)
597 limp->rlim_max = maxdsiz;
598 break;
599
600 case RLIMIT_STACK:
601 if (limp->rlim_cur > maxssiz)
602 limp->rlim_cur = maxssiz;
603 if (limp->rlim_max > maxssiz)
604 limp->rlim_max = maxssiz;
605 oldssiz = alimp->rlim_cur;
606 break;
607
608 case RLIMIT_NOFILE:
609 if (limp->rlim_cur > maxfilesperproc)
610 limp->rlim_cur = maxfilesperproc;
611 if (limp->rlim_max > maxfilesperproc)
612 limp->rlim_max = maxfilesperproc;
613 break;
614
615 case RLIMIT_NPROC:
616 if (limp->rlim_cur > maxprocperuid)
617 limp->rlim_cur = maxprocperuid;
618 if (limp->rlim_max > maxprocperuid)
619 limp->rlim_max = maxprocperuid;
620 if (limp->rlim_cur < 1)
621 limp->rlim_cur = 1;
622 if (limp->rlim_max < 1)
623 limp->rlim_max = 1;
624 break;
625 }
626 *alimp = *limp;
627 p->p_limit = newlim;
628 PROC_UNLOCK(p);
629 lim_free(oldlim);
630
631 if (which == RLIMIT_STACK) {
632 /*
633 * Stack is allocated to the max at exec time with only
634 * "rlim_cur" bytes accessible. If stack limit is going
635 * up make more accessible, if going down make inaccessible.
636 */
637 if (limp->rlim_cur != oldssiz) {
638 vm_offset_t addr;
639 vm_size_t size;
640 vm_prot_t prot;
641
642 if (limp->rlim_cur > oldssiz) {
643 prot = p->p_sysent->sv_stackprot;
644 size = limp->rlim_cur - oldssiz;
645 addr = p->p_sysent->sv_usrstack -
646 limp->rlim_cur;
647 } else {
648 prot = VM_PROT_NONE;
649 size = oldssiz - limp->rlim_cur;
650 addr = p->p_sysent->sv_usrstack - oldssiz;
651 }
652 addr = trunc_page(addr);
653 size = round_page(size);
654 (void)vm_map_protect(&p->p_vmspace->vm_map,
655 addr, addr + size, prot, FALSE);
656 }
657 }
658
659 if (td->td_proc->p_sysent->sv_fixlimits != NULL) {
660 struct image_params imgp;
661
662 imgp.proc = td->td_proc;
663 td->td_proc->p_sysent->sv_fixlimits(&imgp);
664 }
665 return (0);
666 }
667
668 #ifndef _SYS_SYSPROTO_H_
669 struct __getrlimit_args {
670 u_int which;
671 struct rlimit *rlp;
672 };
673 #endif
674 /*
675 * MPSAFE
676 */
677 /* ARGSUSED */
678 int
679 getrlimit(td, uap)
680 struct thread *td;
681 register struct __getrlimit_args *uap;
682 {
683 struct rlimit rlim;
684 struct proc *p;
685 int error;
686
687 if (uap->which >= RLIM_NLIMITS)
688 return (EINVAL);
689 p = td->td_proc;
690 PROC_LOCK(p);
691 lim_rlimit(p, uap->which, &rlim);
692 PROC_UNLOCK(p);
693 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
694 return (error);
695 }
696
697 /*
698 * Transform the running time and tick information in proc p into user,
699 * system, and interrupt time usage.
700 */
701 void
702 calcru(p, up, sp)
703 struct proc *p;
704 struct timeval *up;
705 struct timeval *sp;
706 {
707 struct bintime bt;
708 struct rusage_ext rux;
709 struct thread *td;
710 int bt_valid;
711
712 PROC_LOCK_ASSERT(p, MA_OWNED);
713 mtx_assert(&sched_lock, MA_NOTOWNED);
714 bt_valid = 0;
715 mtx_lock_spin(&sched_lock);
716 rux = p->p_rux;
717 FOREACH_THREAD_IN_PROC(p, td) {
718 if (TD_IS_RUNNING(td)) {
719 /*
720 * Adjust for the current time slice. This is
721 * actually fairly important since the error here is
722 * on the order of a time quantum which is much
723 * greater than the precision of binuptime().
724 */
725 KASSERT(td->td_oncpu != NOCPU,
726 ("%s: running thread has no CPU", __func__));
727 if (!bt_valid) {
728 binuptime(&bt);
729 bt_valid = 1;
730 }
731 bintime_add(&rux.rux_runtime, &bt);
732 bintime_sub(&rux.rux_runtime,
733 &pcpu_find(td->td_oncpu)->pc_switchtime);
734 }
735 }
736 mtx_unlock_spin(&sched_lock);
737 calcru1(p, &rux, up, sp);
738 p->p_rux.rux_uu = rux.rux_uu;
739 p->p_rux.rux_su = rux.rux_su;
740 p->p_rux.rux_iu = rux.rux_iu;
741 }
742
743 void
744 calccru(p, up, sp)
745 struct proc *p;
746 struct timeval *up;
747 struct timeval *sp;
748 {
749
750 PROC_LOCK_ASSERT(p, MA_OWNED);
751 calcru1(p, &p->p_crux, up, sp);
752 }
753
754 static void
755 calcru1(p, ruxp, up, sp)
756 struct proc *p;
757 struct rusage_ext *ruxp;
758 struct timeval *up;
759 struct timeval *sp;
760 {
761 struct timeval tv;
762 /* {user, system, interrupt, total} {ticks, usec}; previous tu: */
763 u_int64_t ut, uu, st, su, it, iu, tt, tu, ptu;
764
765 ut = ruxp->rux_uticks;
766 st = ruxp->rux_sticks;
767 it = ruxp->rux_iticks;
768 tt = ut + st + it;
769 if (tt == 0) {
770 st = 1;
771 tt = 1;
772 }
773 bintime2timeval(&ruxp->rux_runtime, &tv);
774 tu = (u_int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
775 ptu = ruxp->rux_uu + ruxp->rux_su + ruxp->rux_iu;
776 if (tu < ptu) {
777 printf(
778 "calcru: runtime went backwards from %ju usec to %ju usec for pid %d (%s)\n",
779 (uintmax_t)ptu, (uintmax_t)tu, p->p_pid, p->p_comm);
780 tu = ptu;
781 }
782 if ((int64_t)tu < 0) {
783 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
784 (intmax_t)tu, p->p_pid, p->p_comm);
785 tu = ptu;
786 }
787
788 /* Subdivide tu. */
789 uu = (tu * ut) / tt;
790 su = (tu * st) / tt;
791 iu = tu - uu - su;
792
793 /* Enforce monotonicity. */
794 if (uu < ruxp->rux_uu || su < ruxp->rux_su || iu < ruxp->rux_iu) {
795 if (uu < ruxp->rux_uu)
796 uu = ruxp->rux_uu;
797 else if (uu + ruxp->rux_su + ruxp->rux_iu > tu)
798 uu = tu - ruxp->rux_su - ruxp->rux_iu;
799 if (st == 0)
800 su = ruxp->rux_su;
801 else {
802 su = ((tu - uu) * st) / (st + it);
803 if (su < ruxp->rux_su)
804 su = ruxp->rux_su;
805 else if (uu + su + ruxp->rux_iu > tu)
806 su = tu - uu - ruxp->rux_iu;
807 }
808 KASSERT(uu + su + ruxp->rux_iu <= tu,
809 ("calcru: monotonisation botch 1"));
810 iu = tu - uu - su;
811 KASSERT(iu >= ruxp->rux_iu,
812 ("calcru: monotonisation botch 2"));
813 }
814 ruxp->rux_uu = uu;
815 ruxp->rux_su = su;
816 ruxp->rux_iu = iu;
817
818 up->tv_sec = uu / 1000000;
819 up->tv_usec = uu % 1000000;
820 sp->tv_sec = su / 1000000;
821 sp->tv_usec = su % 1000000;
822 }
823
824 #ifndef _SYS_SYSPROTO_H_
825 struct getrusage_args {
826 int who;
827 struct rusage *rusage;
828 };
829 #endif
830 /*
831 * MPSAFE
832 */
833 int
834 getrusage(td, uap)
835 register struct thread *td;
836 register struct getrusage_args *uap;
837 {
838 struct rusage ru;
839 int error;
840
841 error = kern_getrusage(td, uap->who, &ru);
842 if (error == 0)
843 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
844 return (error);
845 }
846
847 int
848 kern_getrusage(td, who, rup)
849 struct thread *td;
850 int who;
851 struct rusage *rup;
852 {
853 struct proc *p;
854
855 p = td->td_proc;
856 PROC_LOCK(p);
857 switch (who) {
858
859 case RUSAGE_SELF:
860 *rup = p->p_stats->p_ru;
861 calcru(p, &rup->ru_utime, &rup->ru_stime);
862 break;
863
864 case RUSAGE_CHILDREN:
865 *rup = p->p_stats->p_cru;
866 calccru(p, &rup->ru_utime, &rup->ru_stime);
867 break;
868
869 default:
870 PROC_UNLOCK(p);
871 return (EINVAL);
872 }
873 PROC_UNLOCK(p);
874 return (0);
875 }
876
877 void
878 ruadd(ru, rux, ru2, rux2)
879 struct rusage *ru;
880 struct rusage_ext *rux;
881 struct rusage *ru2;
882 struct rusage_ext *rux2;
883 {
884 register long *ip, *ip2;
885 register int i;
886
887 bintime_add(&rux->rux_runtime, &rux2->rux_runtime);
888 rux->rux_uticks += rux2->rux_uticks;
889 rux->rux_sticks += rux2->rux_sticks;
890 rux->rux_iticks += rux2->rux_iticks;
891 rux->rux_uu += rux2->rux_uu;
892 rux->rux_su += rux2->rux_su;
893 rux->rux_iu += rux2->rux_iu;
894 if (ru->ru_maxrss < ru2->ru_maxrss)
895 ru->ru_maxrss = ru2->ru_maxrss;
896 ip = &ru->ru_first;
897 ip2 = &ru2->ru_first;
898 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
899 *ip++ += *ip2++;
900 }
901
902 /*
903 * Allocate a new resource limits structure and initialize its
904 * reference count and mutex pointer.
905 */
906 struct plimit *
907 lim_alloc()
908 {
909 struct plimit *limp;
910
911 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
912 limp->pl_refcnt = 1;
913 limp->pl_mtx = mtx_pool_alloc(mtxpool_sleep);
914 return (limp);
915 }
916
917 struct plimit *
918 lim_hold(limp)
919 struct plimit *limp;
920 {
921
922 LIM_LOCK(limp);
923 limp->pl_refcnt++;
924 LIM_UNLOCK(limp);
925 return (limp);
926 }
927
928 void
929 lim_free(limp)
930 struct plimit *limp;
931 {
932
933 LIM_LOCK(limp);
934 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
935 if (--limp->pl_refcnt == 0) {
936 LIM_UNLOCK(limp);
937 free((void *)limp, M_PLIMIT);
938 return;
939 }
940 LIM_UNLOCK(limp);
941 }
942
943 /*
944 * Make a copy of the plimit structure.
945 * We share these structures copy-on-write after fork.
946 */
947 void
948 lim_copy(dst, src)
949 struct plimit *dst, *src;
950 {
951
952 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
953 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
954 }
955
956 /*
957 * Return the hard limit for a particular system resource. The
958 * which parameter specifies the index into the rlimit array.
959 */
960 rlim_t
961 lim_max(struct proc *p, int which)
962 {
963 struct rlimit rl;
964
965 lim_rlimit(p, which, &rl);
966 return (rl.rlim_max);
967 }
968
969 /*
970 * Return the current (soft) limit for a particular system resource.
971 * The which parameter which specifies the index into the rlimit array
972 */
973 rlim_t
974 lim_cur(struct proc *p, int which)
975 {
976 struct rlimit rl;
977
978 lim_rlimit(p, which, &rl);
979 return (rl.rlim_cur);
980 }
981
982 /*
983 * Return a copy of the entire rlimit structure for the system limit
984 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
985 */
986 void
987 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
988 {
989
990 PROC_LOCK_ASSERT(p, MA_OWNED);
991 KASSERT(which >= 0 && which < RLIM_NLIMITS,
992 ("request for invalid resource limit"));
993 *rlp = p->p_limit->pl_rlimit[which];
994 }
995
996 /*
997 * Find the uidinfo structure for a uid. This structure is used to
998 * track the total resource consumption (process count, socket buffer
999 * size, etc.) for the uid and impose limits.
1000 */
1001 void
1002 uihashinit()
1003 {
1004
1005 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1006 mtx_init(&uihashtbl_mtx, "uidinfo hash", NULL, MTX_DEF);
1007 }
1008
1009 /*
1010 * Look up a uidinfo struct for the parameter uid.
1011 * uihashtbl_mtx must be locked.
1012 */
1013 static struct uidinfo *
1014 uilookup(uid)
1015 uid_t uid;
1016 {
1017 struct uihashhead *uipp;
1018 struct uidinfo *uip;
1019
1020 mtx_assert(&uihashtbl_mtx, MA_OWNED);
1021 uipp = UIHASH(uid);
1022 LIST_FOREACH(uip, uipp, ui_hash)
1023 if (uip->ui_uid == uid)
1024 break;
1025
1026 return (uip);
1027 }
1028
1029 /*
1030 * Find or allocate a struct uidinfo for a particular uid.
1031 * Increase refcount on uidinfo struct returned.
1032 * uifree() should be called on a struct uidinfo when released.
1033 */
1034 struct uidinfo *
1035 uifind(uid)
1036 uid_t uid;
1037 {
1038 struct uidinfo *old_uip, *uip;
1039
1040 mtx_lock(&uihashtbl_mtx);
1041 uip = uilookup(uid);
1042 if (uip == NULL) {
1043 mtx_unlock(&uihashtbl_mtx);
1044 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1045 mtx_lock(&uihashtbl_mtx);
1046 /*
1047 * There's a chance someone created our uidinfo while we
1048 * were in malloc and not holding the lock, so we have to
1049 * make sure we don't insert a duplicate uidinfo.
1050 */
1051 if ((old_uip = uilookup(uid)) != NULL) {
1052 /* Someone else beat us to it. */
1053 free(uip, M_UIDINFO);
1054 uip = old_uip;
1055 } else {
1056 uip->ui_mtxp = mtx_pool_alloc(mtxpool_sleep);
1057 uip->ui_uid = uid;
1058 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1059 }
1060 }
1061 uihold(uip);
1062 mtx_unlock(&uihashtbl_mtx);
1063 return (uip);
1064 }
1065
1066 /*
1067 * Place another refcount on a uidinfo struct.
1068 */
1069 void
1070 uihold(uip)
1071 struct uidinfo *uip;
1072 {
1073
1074 UIDINFO_LOCK(uip);
1075 uip->ui_ref++;
1076 UIDINFO_UNLOCK(uip);
1077 }
1078
1079 /*-
1080 * Since uidinfo structs have a long lifetime, we use an
1081 * opportunistic refcounting scheme to avoid locking the lookup hash
1082 * for each release.
1083 *
1084 * If the refcount hits 0, we need to free the structure,
1085 * which means we need to lock the hash.
1086 * Optimal case:
1087 * After locking the struct and lowering the refcount, if we find
1088 * that we don't need to free, simply unlock and return.
1089 * Suboptimal case:
1090 * If refcount lowering results in need to free, bump the count
1091 * back up, loose the lock and aquire the locks in the proper
1092 * order to try again.
1093 */
1094 void
1095 uifree(uip)
1096 struct uidinfo *uip;
1097 {
1098
1099 /* Prepare for optimal case. */
1100 UIDINFO_LOCK(uip);
1101
1102 if (--uip->ui_ref != 0) {
1103 UIDINFO_UNLOCK(uip);
1104 return;
1105 }
1106
1107 /* Prepare for suboptimal case. */
1108 uip->ui_ref++;
1109 UIDINFO_UNLOCK(uip);
1110 mtx_lock(&uihashtbl_mtx);
1111 UIDINFO_LOCK(uip);
1112
1113 /*
1114 * We must subtract one from the count again because we backed out
1115 * our initial subtraction before dropping the lock.
1116 * Since another thread may have added a reference after we dropped the
1117 * initial lock we have to test for zero again.
1118 */
1119 if (--uip->ui_ref == 0) {
1120 LIST_REMOVE(uip, ui_hash);
1121 mtx_unlock(&uihashtbl_mtx);
1122 if (uip->ui_sbsize != 0)
1123 printf("freeing uidinfo: uid = %d, sbsize = %jd\n",
1124 uip->ui_uid, (intmax_t)uip->ui_sbsize);
1125 if (uip->ui_proccnt != 0)
1126 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1127 uip->ui_uid, uip->ui_proccnt);
1128 UIDINFO_UNLOCK(uip);
1129 FREE(uip, M_UIDINFO);
1130 return;
1131 }
1132
1133 mtx_unlock(&uihashtbl_mtx);
1134 UIDINFO_UNLOCK(uip);
1135 }
1136
1137 /*
1138 * Change the count associated with number of processes
1139 * a given user is using. When 'max' is 0, don't enforce a limit
1140 */
1141 int
1142 chgproccnt(uip, diff, max)
1143 struct uidinfo *uip;
1144 int diff;
1145 int max;
1146 {
1147
1148 UIDINFO_LOCK(uip);
1149 /* Don't allow them to exceed max, but allow subtraction. */
1150 if (diff > 0 && uip->ui_proccnt + diff > max && max != 0) {
1151 UIDINFO_UNLOCK(uip);
1152 return (0);
1153 }
1154 uip->ui_proccnt += diff;
1155 if (uip->ui_proccnt < 0)
1156 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1157 UIDINFO_UNLOCK(uip);
1158 return (1);
1159 }
1160
1161 /*
1162 * Change the total socket buffer size a user has used.
1163 */
1164 int
1165 chgsbsize(uip, hiwat, to, max)
1166 struct uidinfo *uip;
1167 u_int *hiwat;
1168 u_int to;
1169 rlim_t max;
1170 {
1171 rlim_t new;
1172
1173 UIDINFO_LOCK(uip);
1174 new = uip->ui_sbsize + to - *hiwat;
1175 /* Don't allow them to exceed max, but allow subtraction. */
1176 if (to > *hiwat && new > max) {
1177 UIDINFO_UNLOCK(uip);
1178 return (0);
1179 }
1180 uip->ui_sbsize = new;
1181 UIDINFO_UNLOCK(uip);
1182 *hiwat = to;
1183 if (new < 0)
1184 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1185 return (1);
1186 }
Cache object: 86f7c6994664153b91b14c99b5f0a204
|