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("$FreeBSD$");
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/sched.h>
55 #include <sys/sx.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysent.h>
58 #include <sys/time.h>
59 #include <sys/umtx.h>
60
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65
66
67 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
68 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
69 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
70 static struct mtx uihashtbl_mtx;
71 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
72 static u_long uihash; /* size of hash table - 1 */
73
74 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
75 struct timeval *up, struct timeval *sp);
76 static int donice(struct thread *td, struct proc *chgp, int n);
77 static struct uidinfo *uilookup(uid_t uid);
78
79 /*
80 * Resource controls and accounting.
81 */
82 #ifndef _SYS_SYSPROTO_H_
83 struct getpriority_args {
84 int which;
85 int who;
86 };
87 #endif
88 int
89 getpriority(td, uap)
90 struct thread *td;
91 register struct getpriority_args *uap;
92 {
93 struct proc *p;
94 struct pgrp *pg;
95 int error, low;
96
97 error = 0;
98 low = PRIO_MAX + 1;
99 switch (uap->which) {
100
101 case PRIO_PROCESS:
102 if (uap->who == 0)
103 low = td->td_proc->p_nice;
104 else {
105 p = pfind(uap->who);
106 if (p == NULL)
107 break;
108 if (p_cansee(td, p) == 0)
109 low = p->p_nice;
110 PROC_UNLOCK(p);
111 }
112 break;
113
114 case PRIO_PGRP:
115 sx_slock(&proctree_lock);
116 if (uap->who == 0) {
117 pg = td->td_proc->p_pgrp;
118 PGRP_LOCK(pg);
119 } else {
120 pg = pgfind(uap->who);
121 if (pg == NULL) {
122 sx_sunlock(&proctree_lock);
123 break;
124 }
125 }
126 sx_sunlock(&proctree_lock);
127 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
128 PROC_LOCK(p);
129 if (!p_cansee(td, p)) {
130 if (p->p_nice < low)
131 low = p->p_nice;
132 }
133 PROC_UNLOCK(p);
134 }
135 PGRP_UNLOCK(pg);
136 break;
137
138 case PRIO_USER:
139 if (uap->who == 0)
140 uap->who = td->td_ucred->cr_uid;
141 sx_slock(&allproc_lock);
142 FOREACH_PROC_IN_SYSTEM(p) {
143 /* Do not bother to check PRS_NEW processes */
144 if (p->p_state == PRS_NEW)
145 continue;
146 PROC_LOCK(p);
147 if (!p_cansee(td, p) &&
148 p->p_ucred->cr_uid == uap->who) {
149 if (p->p_nice < low)
150 low = p->p_nice;
151 }
152 PROC_UNLOCK(p);
153 }
154 sx_sunlock(&allproc_lock);
155 break;
156
157 default:
158 error = EINVAL;
159 break;
160 }
161 if (low == PRIO_MAX + 1 && error == 0)
162 error = ESRCH;
163 td->td_retval[0] = low;
164 return (error);
165 }
166
167 #ifndef _SYS_SYSPROTO_H_
168 struct setpriority_args {
169 int which;
170 int who;
171 int prio;
172 };
173 #endif
174 int
175 setpriority(td, uap)
176 struct thread *td;
177 struct setpriority_args *uap;
178 {
179 struct proc *curp, *p;
180 struct pgrp *pg;
181 int found = 0, error = 0;
182
183 curp = td->td_proc;
184 switch (uap->which) {
185 case PRIO_PROCESS:
186 if (uap->who == 0) {
187 PROC_LOCK(curp);
188 error = donice(td, curp, uap->prio);
189 PROC_UNLOCK(curp);
190 } else {
191 p = pfind(uap->who);
192 if (p == 0)
193 break;
194 if (p_cansee(td, p) == 0)
195 error = donice(td, p, uap->prio);
196 PROC_UNLOCK(p);
197 }
198 found++;
199 break;
200
201 case PRIO_PGRP:
202 sx_slock(&proctree_lock);
203 if (uap->who == 0) {
204 pg = curp->p_pgrp;
205 PGRP_LOCK(pg);
206 } else {
207 pg = pgfind(uap->who);
208 if (pg == NULL) {
209 sx_sunlock(&proctree_lock);
210 break;
211 }
212 }
213 sx_sunlock(&proctree_lock);
214 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
215 PROC_LOCK(p);
216 if (!p_cansee(td, p)) {
217 error = donice(td, p, uap->prio);
218 found++;
219 }
220 PROC_UNLOCK(p);
221 }
222 PGRP_UNLOCK(pg);
223 break;
224
225 case PRIO_USER:
226 if (uap->who == 0)
227 uap->who = td->td_ucred->cr_uid;
228 sx_slock(&allproc_lock);
229 FOREACH_PROC_IN_SYSTEM(p) {
230 PROC_LOCK(p);
231 if (p->p_ucred->cr_uid == uap->who &&
232 !p_cansee(td, p)) {
233 error = donice(td, p, uap->prio);
234 found++;
235 }
236 PROC_UNLOCK(p);
237 }
238 sx_sunlock(&allproc_lock);
239 break;
240
241 default:
242 error = EINVAL;
243 break;
244 }
245 if (found == 0 && error == 0)
246 error = ESRCH;
247 return (error);
248 }
249
250 /*
251 * Set "nice" for a (whole) process.
252 */
253 static int
254 donice(struct thread *td, struct proc *p, int n)
255 {
256 int error;
257
258 PROC_LOCK_ASSERT(p, MA_OWNED);
259 if ((error = p_cansched(td, p)))
260 return (error);
261 if (n > PRIO_MAX)
262 n = PRIO_MAX;
263 if (n < PRIO_MIN)
264 n = PRIO_MIN;
265 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
266 return (EACCES);
267 PROC_SLOCK(p);
268 sched_nice(p, n);
269 PROC_SUNLOCK(p);
270 return (0);
271 }
272
273 /*
274 * Set realtime priority for LWP.
275 */
276 #ifndef _SYS_SYSPROTO_H_
277 struct rtprio_thread_args {
278 int function;
279 lwpid_t lwpid;
280 struct rtprio *rtp;
281 };
282 #endif
283 int
284 rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
285 {
286 struct proc *curp;
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 curp = td->td_proc;
299 /*
300 * Though lwpid is unique, only current process is supported
301 * since there is no efficient way to look up a LWP yet.
302 */
303 p = curp;
304 PROC_LOCK(p);
305
306 switch (uap->function) {
307 case RTP_LOOKUP:
308 if ((error = p_cansee(td, p)))
309 break;
310 PROC_SLOCK(p);
311 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
312 td1 = td;
313 else
314 td1 = thread_find(p, uap->lwpid);
315 if (td1 != NULL)
316 pri_to_rtp(td1, &rtp);
317 else
318 error = ESRCH;
319 PROC_SUNLOCK(p);
320 PROC_UNLOCK(p);
321 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
322 case RTP_SET:
323 if ((error = p_cansched(td, p)) || (error = cierror))
324 break;
325
326 /* Disallow setting rtprio in most cases if not superuser. */
327 /*
328 * Realtime priority has to be restricted for reasons which should be
329 * obvious. However, for idle priority, there is a potential for
330 * system deadlock if an idleprio process gains a lock on a resource
331 * that other processes need (and the idleprio process can't run
332 * due to a CPU-bound normal process). Fix me! XXX
333 */
334 #if 0
335 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
336 #else
337 if (rtp.type != RTP_PRIO_NORMAL) {
338 #endif
339 error = priv_check(td, PRIV_SCHED_RTPRIO);
340 if (error)
341 break;
342 }
343
344 PROC_SLOCK(p);
345 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
346 td1 = td;
347 else
348 td1 = thread_find(p, uap->lwpid);
349 if (td1 != NULL)
350 error = rtp_to_pri(&rtp, td1);
351 else
352 error = ESRCH;
353 PROC_SUNLOCK(p);
354 break;
355 default:
356 error = EINVAL;
357 break;
358 }
359 PROC_UNLOCK(p);
360 return (error);
361 }
362
363 /*
364 * Set realtime priority.
365 */
366 #ifndef _SYS_SYSPROTO_H_
367 struct rtprio_args {
368 int function;
369 pid_t pid;
370 struct rtprio *rtp;
371 };
372 #endif
373 int
374 rtprio(td, uap)
375 struct thread *td; /* curthread */
376 register struct rtprio_args *uap;
377 {
378 struct proc *curp;
379 struct proc *p;
380 struct thread *tdp;
381 struct rtprio rtp;
382 int cierror, error;
383
384 /* Perform copyin before acquiring locks if needed. */
385 if (uap->function == RTP_SET)
386 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
387 else
388 cierror = 0;
389
390 curp = td->td_proc;
391 if (uap->pid == 0) {
392 p = curp;
393 PROC_LOCK(p);
394 } else {
395 p = pfind(uap->pid);
396 if (p == NULL)
397 return (ESRCH);
398 }
399
400 switch (uap->function) {
401 case RTP_LOOKUP:
402 if ((error = p_cansee(td, p)))
403 break;
404 PROC_SLOCK(p);
405 /*
406 * Return OUR priority if no pid specified,
407 * or if one is, report the highest priority
408 * in the process. There isn't much more you can do as
409 * there is only room to return a single priority.
410 * XXXKSE: maybe need a new interface to report
411 * priorities of multiple system scope threads.
412 * Note: specifying our own pid is not the same
413 * as leaving it zero.
414 */
415 if (uap->pid == 0) {
416 pri_to_rtp(td, &rtp);
417 } else {
418 struct rtprio rtp2;
419
420 rtp.type = RTP_PRIO_IDLE;
421 rtp.prio = RTP_PRIO_MAX;
422 FOREACH_THREAD_IN_PROC(p, tdp) {
423 pri_to_rtp(tdp, &rtp2);
424 if (rtp2.type < rtp.type ||
425 (rtp2.type == rtp.type &&
426 rtp2.prio < rtp.prio)) {
427 rtp.type = rtp2.type;
428 rtp.prio = rtp2.prio;
429 }
430 }
431 }
432 PROC_SUNLOCK(p);
433 PROC_UNLOCK(p);
434 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
435 case RTP_SET:
436 if ((error = p_cansched(td, p)) || (error = cierror))
437 break;
438
439 /* Disallow setting rtprio in most cases if not superuser. */
440 /*
441 * Realtime priority has to be restricted for reasons which should be
442 * obvious. However, for idle priority, there is a potential for
443 * system deadlock if an idleprio process gains a lock on a resource
444 * that other processes need (and the idleprio process can't run
445 * due to a CPU-bound normal process). Fix me! XXX
446 */
447 #if 0
448 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
449 #else
450 if (rtp.type != RTP_PRIO_NORMAL) {
451 #endif
452 error = priv_check(td, PRIV_SCHED_RTPRIO);
453 if (error)
454 break;
455 }
456
457 /*
458 * If we are setting our own priority, set just our
459 * thread but if we are doing another process,
460 * do all the threads on that process. If we
461 * specify our own pid we do the latter.
462 */
463 PROC_SLOCK(p);
464 if (uap->pid == 0) {
465 error = rtp_to_pri(&rtp, td);
466 } else {
467 FOREACH_THREAD_IN_PROC(p, td) {
468 if ((error = rtp_to_pri(&rtp, td)) != 0)
469 break;
470 }
471 }
472 PROC_SUNLOCK(p);
473 break;
474 default:
475 error = EINVAL;
476 break;
477 }
478 PROC_UNLOCK(p);
479 return (error);
480 }
481
482 int
483 rtp_to_pri(struct rtprio *rtp, struct thread *td)
484 {
485 u_char newpri;
486 u_char oldpri;
487
488 if (rtp->prio > RTP_PRIO_MAX)
489 return (EINVAL);
490 thread_lock(td);
491 switch (RTP_PRIO_BASE(rtp->type)) {
492 case RTP_PRIO_REALTIME:
493 newpri = PRI_MIN_REALTIME + rtp->prio;
494 break;
495 case RTP_PRIO_NORMAL:
496 newpri = PRI_MIN_TIMESHARE + rtp->prio;
497 break;
498 case RTP_PRIO_IDLE:
499 newpri = PRI_MIN_IDLE + rtp->prio;
500 break;
501 default:
502 thread_unlock(td);
503 return (EINVAL);
504 }
505 sched_class(td, rtp->type); /* XXX fix */
506 oldpri = td->td_user_pri;
507 sched_user_prio(td, newpri);
508 if (curthread == td)
509 sched_prio(curthread, td->td_user_pri); /* XXX dubious */
510 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
511 thread_unlock(td);
512 umtx_pi_adjust(td, oldpri);
513 } else
514 thread_unlock(td);
515 return (0);
516 }
517
518 void
519 pri_to_rtp(struct thread *td, struct rtprio *rtp)
520 {
521
522 thread_lock(td);
523 switch (PRI_BASE(td->td_pri_class)) {
524 case PRI_REALTIME:
525 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
526 break;
527 case PRI_TIMESHARE:
528 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
529 break;
530 case PRI_IDLE:
531 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
532 break;
533 default:
534 break;
535 }
536 rtp->type = td->td_pri_class;
537 thread_unlock(td);
538 }
539
540 #if defined(COMPAT_43)
541 #ifndef _SYS_SYSPROTO_H_
542 struct osetrlimit_args {
543 u_int which;
544 struct orlimit *rlp;
545 };
546 #endif
547 int
548 osetrlimit(td, uap)
549 struct thread *td;
550 register struct osetrlimit_args *uap;
551 {
552 struct orlimit olim;
553 struct rlimit lim;
554 int error;
555
556 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
557 return (error);
558 lim.rlim_cur = olim.rlim_cur;
559 lim.rlim_max = olim.rlim_max;
560 error = kern_setrlimit(td, uap->which, &lim);
561 return (error);
562 }
563
564 #ifndef _SYS_SYSPROTO_H_
565 struct ogetrlimit_args {
566 u_int which;
567 struct orlimit *rlp;
568 };
569 #endif
570 int
571 ogetrlimit(td, uap)
572 struct thread *td;
573 register struct ogetrlimit_args *uap;
574 {
575 struct orlimit olim;
576 struct rlimit rl;
577 struct proc *p;
578 int error;
579
580 if (uap->which >= RLIM_NLIMITS)
581 return (EINVAL);
582 p = td->td_proc;
583 PROC_LOCK(p);
584 lim_rlimit(p, uap->which, &rl);
585 PROC_UNLOCK(p);
586
587 /*
588 * XXX would be more correct to convert only RLIM_INFINITY to the
589 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
590 * values. Most 64->32 and 32->16 conversions, including not
591 * unimportant ones of uids are even more broken than what we
592 * do here (they blindly truncate). We don't do this correctly
593 * here since we have little experience with EOVERFLOW yet.
594 * Elsewhere, getuid() can't fail...
595 */
596 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
597 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
598 error = copyout(&olim, uap->rlp, sizeof(olim));
599 return (error);
600 }
601 #endif /* COMPAT_43 */
602
603 #ifndef _SYS_SYSPROTO_H_
604 struct __setrlimit_args {
605 u_int which;
606 struct rlimit *rlp;
607 };
608 #endif
609 int
610 setrlimit(td, uap)
611 struct thread *td;
612 register struct __setrlimit_args *uap;
613 {
614 struct rlimit alim;
615 int error;
616
617 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
618 return (error);
619 error = kern_setrlimit(td, uap->which, &alim);
620 return (error);
621 }
622
623 static void
624 lim_cb(void *arg)
625 {
626 struct rlimit rlim;
627 struct thread *td;
628 struct proc *p;
629
630 p = arg;
631 PROC_LOCK_ASSERT(p, MA_OWNED);
632 /*
633 * Check if the process exceeds its cpu resource allocation. If
634 * it reaches the max, arrange to kill the process in ast().
635 */
636 if (p->p_cpulimit == RLIM_INFINITY)
637 return;
638 PROC_SLOCK(p);
639 FOREACH_THREAD_IN_PROC(p, td) {
640 thread_lock(td);
641 ruxagg(&p->p_rux, td);
642 thread_unlock(td);
643 }
644 PROC_SUNLOCK(p);
645 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
646 lim_rlimit(p, RLIMIT_CPU, &rlim);
647 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
648 killproc(p, "exceeded maximum CPU limit");
649 } else {
650 if (p->p_cpulimit < rlim.rlim_max)
651 p->p_cpulimit += 5;
652 psignal(p, SIGXCPU);
653 }
654 }
655 callout_reset(&p->p_limco, hz, lim_cb, p);
656 }
657
658 int
659 kern_setrlimit(td, which, limp)
660 struct thread *td;
661 u_int which;
662 struct rlimit *limp;
663 {
664 struct plimit *newlim, *oldlim;
665 struct proc *p;
666 register struct rlimit *alimp;
667 struct rlimit oldssiz;
668 int error;
669
670 if (which >= RLIM_NLIMITS)
671 return (EINVAL);
672
673 /*
674 * Preserve historical bugs by treating negative limits as unsigned.
675 */
676 if (limp->rlim_cur < 0)
677 limp->rlim_cur = RLIM_INFINITY;
678 if (limp->rlim_max < 0)
679 limp->rlim_max = RLIM_INFINITY;
680
681 oldssiz.rlim_cur = 0;
682 p = td->td_proc;
683 newlim = lim_alloc();
684 PROC_LOCK(p);
685 oldlim = p->p_limit;
686 alimp = &oldlim->pl_rlimit[which];
687 if (limp->rlim_cur > alimp->rlim_max ||
688 limp->rlim_max > alimp->rlim_max)
689 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
690 PROC_UNLOCK(p);
691 lim_free(newlim);
692 return (error);
693 }
694 if (limp->rlim_cur > limp->rlim_max)
695 limp->rlim_cur = limp->rlim_max;
696 lim_copy(newlim, oldlim);
697 alimp = &newlim->pl_rlimit[which];
698
699 switch (which) {
700
701 case RLIMIT_CPU:
702 if (limp->rlim_cur != RLIM_INFINITY &&
703 p->p_cpulimit == RLIM_INFINITY)
704 callout_reset(&p->p_limco, hz, lim_cb, p);
705 PROC_SLOCK(p);
706 p->p_cpulimit = limp->rlim_cur;
707 PROC_SUNLOCK(p);
708 break;
709 case RLIMIT_DATA:
710 if (limp->rlim_cur > maxdsiz)
711 limp->rlim_cur = maxdsiz;
712 if (limp->rlim_max > maxdsiz)
713 limp->rlim_max = maxdsiz;
714 break;
715
716 case RLIMIT_STACK:
717 if (limp->rlim_cur > maxssiz)
718 limp->rlim_cur = maxssiz;
719 if (limp->rlim_max > maxssiz)
720 limp->rlim_max = maxssiz;
721 oldssiz = *alimp;
722 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
723 td->td_proc->p_sysent->sv_fixlimit(&oldssiz,
724 RLIMIT_STACK);
725 break;
726
727 case RLIMIT_NOFILE:
728 if (limp->rlim_cur > maxfilesperproc)
729 limp->rlim_cur = maxfilesperproc;
730 if (limp->rlim_max > maxfilesperproc)
731 limp->rlim_max = maxfilesperproc;
732 break;
733
734 case RLIMIT_NPROC:
735 if (limp->rlim_cur > maxprocperuid)
736 limp->rlim_cur = maxprocperuid;
737 if (limp->rlim_max > maxprocperuid)
738 limp->rlim_max = maxprocperuid;
739 if (limp->rlim_cur < 1)
740 limp->rlim_cur = 1;
741 if (limp->rlim_max < 1)
742 limp->rlim_max = 1;
743 break;
744 }
745 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
746 td->td_proc->p_sysent->sv_fixlimit(limp, which);
747 *alimp = *limp;
748 p->p_limit = newlim;
749 PROC_UNLOCK(p);
750 lim_free(oldlim);
751
752 if (which == RLIMIT_STACK) {
753 /*
754 * Stack is allocated to the max at exec time with only
755 * "rlim_cur" bytes accessible. If stack limit is going
756 * up make more accessible, if going down make inaccessible.
757 */
758 if (limp->rlim_cur != oldssiz.rlim_cur) {
759 vm_offset_t addr;
760 vm_size_t size;
761 vm_prot_t prot;
762
763 if (limp->rlim_cur > oldssiz.rlim_cur) {
764 prot = p->p_sysent->sv_stackprot;
765 size = limp->rlim_cur - oldssiz.rlim_cur;
766 addr = p->p_sysent->sv_usrstack -
767 limp->rlim_cur;
768 } else {
769 prot = VM_PROT_NONE;
770 size = oldssiz.rlim_cur - limp->rlim_cur;
771 addr = p->p_sysent->sv_usrstack -
772 oldssiz.rlim_cur;
773 }
774 addr = trunc_page(addr);
775 size = round_page(size);
776 (void)vm_map_protect(&p->p_vmspace->vm_map,
777 addr, addr + size, prot, FALSE);
778 }
779 }
780
781 return (0);
782 }
783
784 #ifndef _SYS_SYSPROTO_H_
785 struct __getrlimit_args {
786 u_int which;
787 struct rlimit *rlp;
788 };
789 #endif
790 /* ARGSUSED */
791 int
792 getrlimit(td, uap)
793 struct thread *td;
794 register struct __getrlimit_args *uap;
795 {
796 struct rlimit rlim;
797 struct proc *p;
798 int error;
799
800 if (uap->which >= RLIM_NLIMITS)
801 return (EINVAL);
802 p = td->td_proc;
803 PROC_LOCK(p);
804 lim_rlimit(p, uap->which, &rlim);
805 PROC_UNLOCK(p);
806 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
807 return (error);
808 }
809
810 /*
811 * Transform the running time and tick information for children of proc p
812 * into user and system time usage.
813 */
814 void
815 calccru(p, up, sp)
816 struct proc *p;
817 struct timeval *up;
818 struct timeval *sp;
819 {
820
821 PROC_LOCK_ASSERT(p, MA_OWNED);
822 calcru1(p, &p->p_crux, up, sp);
823 }
824
825 /*
826 * Transform the running time and tick information in proc p into user
827 * and system time usage. If appropriate, include the current time slice
828 * on this CPU.
829 */
830 void
831 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
832 {
833 struct thread *td;
834 uint64_t u;
835
836 PROC_LOCK_ASSERT(p, MA_OWNED);
837 PROC_SLOCK_ASSERT(p, MA_OWNED);
838 /*
839 * If we are getting stats for the current process, then add in the
840 * stats that this thread has accumulated in its current time slice.
841 * We reset the thread and CPU state as if we had performed a context
842 * switch right here.
843 */
844 td = curthread;
845 if (td->td_proc == p) {
846 u = cpu_ticks();
847 p->p_rux.rux_runtime += u - PCPU_GET(switchtime);
848 PCPU_SET(switchtime, u);
849 }
850 /* Make sure the per-thread stats are current. */
851 FOREACH_THREAD_IN_PROC(p, td) {
852 if (td->td_incruntime == 0)
853 continue;
854 thread_lock(td);
855 ruxagg(&p->p_rux, td);
856 thread_unlock(td);
857 }
858 calcru1(p, &p->p_rux, up, sp);
859 }
860
861 static void
862 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
863 struct timeval *sp)
864 {
865 /* {user, system, interrupt, total} {ticks, usec}: */
866 u_int64_t ut, uu, st, su, it, tt, tu;
867
868 ut = ruxp->rux_uticks;
869 st = ruxp->rux_sticks;
870 it = ruxp->rux_iticks;
871 tt = ut + st + it;
872 if (tt == 0) {
873 /* Avoid divide by zero */
874 st = 1;
875 tt = 1;
876 }
877 tu = cputick2usec(ruxp->rux_runtime);
878 if ((int64_t)tu < 0) {
879 /* XXX: this should be an assert /phk */
880 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
881 (intmax_t)tu, p->p_pid, p->p_comm);
882 tu = ruxp->rux_tu;
883 }
884
885 if (tu >= ruxp->rux_tu) {
886 /*
887 * The normal case, time increased.
888 * Enforce monotonicity of bucketed numbers.
889 */
890 uu = (tu * ut) / tt;
891 if (uu < ruxp->rux_uu)
892 uu = ruxp->rux_uu;
893 su = (tu * st) / tt;
894 if (su < ruxp->rux_su)
895 su = ruxp->rux_su;
896 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
897 /*
898 * When we calibrate the cputicker, it is not uncommon to
899 * see the presumably fixed frequency increase slightly over
900 * time as a result of thermal stabilization and NTP
901 * discipline (of the reference clock). We therefore ignore
902 * a bit of backwards slop because we expect to catch up
903 * shortly. We use a 3 microsecond limit to catch low
904 * counts and a 1% limit for high counts.
905 */
906 uu = ruxp->rux_uu;
907 su = ruxp->rux_su;
908 tu = ruxp->rux_tu;
909 } else { /* tu < ruxp->rux_tu */
910 /*
911 * What happene here was likely that a laptop, which ran at
912 * a reduced clock frequency at boot, kicked into high gear.
913 * The wisdom of spamming this message in that case is
914 * dubious, but it might also be indicative of something
915 * serious, so lets keep it and hope laptops can be made
916 * more truthful about their CPU speed via ACPI.
917 */
918 printf("calcru: runtime went backwards from %ju usec "
919 "to %ju usec for pid %d (%s)\n",
920 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
921 p->p_pid, p->p_comm);
922 uu = (tu * ut) / tt;
923 su = (tu * st) / tt;
924 }
925
926 ruxp->rux_uu = uu;
927 ruxp->rux_su = su;
928 ruxp->rux_tu = tu;
929
930 up->tv_sec = uu / 1000000;
931 up->tv_usec = uu % 1000000;
932 sp->tv_sec = su / 1000000;
933 sp->tv_usec = su % 1000000;
934 }
935
936 #ifndef _SYS_SYSPROTO_H_
937 struct getrusage_args {
938 int who;
939 struct rusage *rusage;
940 };
941 #endif
942 int
943 getrusage(td, uap)
944 register struct thread *td;
945 register struct getrusage_args *uap;
946 {
947 struct rusage ru;
948 int error;
949
950 error = kern_getrusage(td, uap->who, &ru);
951 if (error == 0)
952 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
953 return (error);
954 }
955
956 int
957 kern_getrusage(td, who, rup)
958 struct thread *td;
959 int who;
960 struct rusage *rup;
961 {
962 struct proc *p;
963
964 p = td->td_proc;
965 PROC_LOCK(p);
966 switch (who) {
967
968 case RUSAGE_SELF:
969 rufetchcalc(p, rup, &rup->ru_utime,
970 &rup->ru_stime);
971 break;
972
973 case RUSAGE_CHILDREN:
974 *rup = p->p_stats->p_cru;
975 calccru(p, &rup->ru_utime, &rup->ru_stime);
976 break;
977
978 default:
979 PROC_UNLOCK(p);
980 return (EINVAL);
981 }
982 PROC_UNLOCK(p);
983 return (0);
984 }
985
986 void
987 rucollect(struct rusage *ru, struct rusage *ru2)
988 {
989 long *ip, *ip2;
990 int i;
991
992 if (ru->ru_maxrss < ru2->ru_maxrss)
993 ru->ru_maxrss = ru2->ru_maxrss;
994 ip = &ru->ru_first;
995 ip2 = &ru2->ru_first;
996 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
997 *ip++ += *ip2++;
998 }
999
1000 void
1001 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1002 struct rusage_ext *rux2)
1003 {
1004
1005 rux->rux_runtime += rux2->rux_runtime;
1006 rux->rux_uticks += rux2->rux_uticks;
1007 rux->rux_sticks += rux2->rux_sticks;
1008 rux->rux_iticks += rux2->rux_iticks;
1009 rux->rux_uu += rux2->rux_uu;
1010 rux->rux_su += rux2->rux_su;
1011 rux->rux_tu += rux2->rux_tu;
1012 rucollect(ru, ru2);
1013 }
1014
1015 /*
1016 * Aggregate tick counts into the proc's rusage_ext.
1017 */
1018 void
1019 ruxagg(struct rusage_ext *rux, struct thread *td)
1020 {
1021
1022 THREAD_LOCK_ASSERT(td, MA_OWNED);
1023 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED);
1024 rux->rux_runtime += td->td_incruntime;
1025 rux->rux_uticks += td->td_uticks;
1026 rux->rux_sticks += td->td_sticks;
1027 rux->rux_iticks += td->td_iticks;
1028 td->td_incruntime = 0;
1029 td->td_uticks = 0;
1030 td->td_iticks = 0;
1031 td->td_sticks = 0;
1032 }
1033
1034 /*
1035 * Update the rusage_ext structure and fetch a valid aggregate rusage
1036 * for proc p if storage for one is supplied.
1037 */
1038 void
1039 rufetch(struct proc *p, struct rusage *ru)
1040 {
1041 struct thread *td;
1042
1043 PROC_SLOCK_ASSERT(p, MA_OWNED);
1044
1045 *ru = p->p_ru;
1046 if (p->p_numthreads > 0) {
1047 FOREACH_THREAD_IN_PROC(p, td) {
1048 thread_lock(td);
1049 ruxagg(&p->p_rux, td);
1050 thread_unlock(td);
1051 rucollect(ru, &td->td_ru);
1052 }
1053 }
1054 }
1055
1056 /*
1057 * Atomically perform a rufetch and a calcru together.
1058 * Consumers, can safely assume the calcru is executed only once
1059 * rufetch is completed.
1060 */
1061 void
1062 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1063 struct timeval *sp)
1064 {
1065
1066 PROC_SLOCK(p);
1067 rufetch(p, ru);
1068 calcru(p, up, sp);
1069 PROC_SUNLOCK(p);
1070 }
1071
1072 /*
1073 * Allocate a new resource limits structure and initialize its
1074 * reference count and mutex pointer.
1075 */
1076 struct plimit *
1077 lim_alloc()
1078 {
1079 struct plimit *limp;
1080
1081 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1082 refcount_init(&limp->pl_refcnt, 1);
1083 return (limp);
1084 }
1085
1086 struct plimit *
1087 lim_hold(limp)
1088 struct plimit *limp;
1089 {
1090
1091 refcount_acquire(&limp->pl_refcnt);
1092 return (limp);
1093 }
1094
1095 void
1096 lim_fork(struct proc *p1, struct proc *p2)
1097 {
1098 p2->p_limit = lim_hold(p1->p_limit);
1099 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1100 if (p1->p_cpulimit != RLIM_INFINITY)
1101 callout_reset(&p2->p_limco, hz, lim_cb, p2);
1102 }
1103
1104 void
1105 lim_free(limp)
1106 struct plimit *limp;
1107 {
1108
1109 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
1110 if (refcount_release(&limp->pl_refcnt))
1111 free((void *)limp, M_PLIMIT);
1112 }
1113
1114 /*
1115 * Make a copy of the plimit structure.
1116 * We share these structures copy-on-write after fork.
1117 */
1118 void
1119 lim_copy(dst, src)
1120 struct plimit *dst, *src;
1121 {
1122
1123 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
1124 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1125 }
1126
1127 /*
1128 * Return the hard limit for a particular system resource. The
1129 * which parameter specifies the index into the rlimit array.
1130 */
1131 rlim_t
1132 lim_max(struct proc *p, int which)
1133 {
1134 struct rlimit rl;
1135
1136 lim_rlimit(p, which, &rl);
1137 return (rl.rlim_max);
1138 }
1139
1140 /*
1141 * Return the current (soft) limit for a particular system resource.
1142 * The which parameter which specifies the index into the rlimit array
1143 */
1144 rlim_t
1145 lim_cur(struct proc *p, int which)
1146 {
1147 struct rlimit rl;
1148
1149 lim_rlimit(p, which, &rl);
1150 return (rl.rlim_cur);
1151 }
1152
1153 /*
1154 * Return a copy of the entire rlimit structure for the system limit
1155 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1156 */
1157 void
1158 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
1159 {
1160
1161 PROC_LOCK_ASSERT(p, MA_OWNED);
1162 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1163 ("request for invalid resource limit"));
1164 *rlp = p->p_limit->pl_rlimit[which];
1165 if (p->p_sysent->sv_fixlimit != NULL)
1166 p->p_sysent->sv_fixlimit(rlp, which);
1167 }
1168
1169 /*
1170 * Find the uidinfo structure for a uid. This structure is used to
1171 * track the total resource consumption (process count, socket buffer
1172 * size, etc.) for the uid and impose limits.
1173 */
1174 void
1175 uihashinit()
1176 {
1177
1178 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1179 mtx_init(&uihashtbl_mtx, "uidinfo hash", NULL, MTX_DEF);
1180 }
1181
1182 /*
1183 * Look up a uidinfo struct for the parameter uid.
1184 * uihashtbl_mtx must be locked.
1185 */
1186 static struct uidinfo *
1187 uilookup(uid)
1188 uid_t uid;
1189 {
1190 struct uihashhead *uipp;
1191 struct uidinfo *uip;
1192
1193 mtx_assert(&uihashtbl_mtx, MA_OWNED);
1194 uipp = UIHASH(uid);
1195 LIST_FOREACH(uip, uipp, ui_hash)
1196 if (uip->ui_uid == uid)
1197 break;
1198
1199 return (uip);
1200 }
1201
1202 /*
1203 * Find or allocate a struct uidinfo for a particular uid.
1204 * Increase refcount on uidinfo struct returned.
1205 * uifree() should be called on a struct uidinfo when released.
1206 */
1207 struct uidinfo *
1208 uifind(uid)
1209 uid_t uid;
1210 {
1211 struct uidinfo *old_uip, *uip;
1212
1213 mtx_lock(&uihashtbl_mtx);
1214 uip = uilookup(uid);
1215 if (uip == NULL) {
1216 mtx_unlock(&uihashtbl_mtx);
1217 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1218 mtx_lock(&uihashtbl_mtx);
1219 /*
1220 * There's a chance someone created our uidinfo while we
1221 * were in malloc and not holding the lock, so we have to
1222 * make sure we don't insert a duplicate uidinfo.
1223 */
1224 if ((old_uip = uilookup(uid)) != NULL) {
1225 /* Someone else beat us to it. */
1226 free(uip, M_UIDINFO);
1227 uip = old_uip;
1228 } else {
1229 uip->ui_mtxp = mtx_pool_alloc(mtxpool_sleep);
1230 uip->ui_uid = uid;
1231 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1232 }
1233 }
1234 uihold(uip);
1235 mtx_unlock(&uihashtbl_mtx);
1236 return (uip);
1237 }
1238
1239 /*
1240 * Place another refcount on a uidinfo struct.
1241 */
1242 void
1243 uihold(uip)
1244 struct uidinfo *uip;
1245 {
1246
1247 UIDINFO_LOCK(uip);
1248 uip->ui_ref++;
1249 UIDINFO_UNLOCK(uip);
1250 }
1251
1252 /*-
1253 * Since uidinfo structs have a long lifetime, we use an
1254 * opportunistic refcounting scheme to avoid locking the lookup hash
1255 * for each release.
1256 *
1257 * If the refcount hits 0, we need to free the structure,
1258 * which means we need to lock the hash.
1259 * Optimal case:
1260 * After locking the struct and lowering the refcount, if we find
1261 * that we don't need to free, simply unlock and return.
1262 * Suboptimal case:
1263 * If refcount lowering results in need to free, bump the count
1264 * back up, lose the lock and acquire the locks in the proper
1265 * order to try again.
1266 */
1267 void
1268 uifree(uip)
1269 struct uidinfo *uip;
1270 {
1271
1272 /* Prepare for optimal case. */
1273 UIDINFO_LOCK(uip);
1274
1275 if (--uip->ui_ref != 0) {
1276 UIDINFO_UNLOCK(uip);
1277 return;
1278 }
1279
1280 /* Prepare for suboptimal case. */
1281 uip->ui_ref++;
1282 UIDINFO_UNLOCK(uip);
1283 mtx_lock(&uihashtbl_mtx);
1284 UIDINFO_LOCK(uip);
1285
1286 /*
1287 * We must subtract one from the count again because we backed out
1288 * our initial subtraction before dropping the lock.
1289 * Since another thread may have added a reference after we dropped the
1290 * initial lock we have to test for zero again.
1291 */
1292 if (--uip->ui_ref == 0) {
1293 LIST_REMOVE(uip, ui_hash);
1294 mtx_unlock(&uihashtbl_mtx);
1295 if (uip->ui_sbsize != 0)
1296 printf("freeing uidinfo: uid = %d, sbsize = %jd\n",
1297 uip->ui_uid, (intmax_t)uip->ui_sbsize);
1298 if (uip->ui_proccnt != 0)
1299 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1300 uip->ui_uid, uip->ui_proccnt);
1301 UIDINFO_UNLOCK(uip);
1302 FREE(uip, M_UIDINFO);
1303 return;
1304 }
1305
1306 mtx_unlock(&uihashtbl_mtx);
1307 UIDINFO_UNLOCK(uip);
1308 }
1309
1310 /*
1311 * Change the count associated with number of processes
1312 * a given user is using. When 'max' is 0, don't enforce a limit
1313 */
1314 int
1315 chgproccnt(uip, diff, max)
1316 struct uidinfo *uip;
1317 int diff;
1318 int max;
1319 {
1320
1321 UIDINFO_LOCK(uip);
1322 /* Don't allow them to exceed max, but allow subtraction. */
1323 if (diff > 0 && uip->ui_proccnt + diff > max && max != 0) {
1324 UIDINFO_UNLOCK(uip);
1325 return (0);
1326 }
1327 uip->ui_proccnt += diff;
1328 if (uip->ui_proccnt < 0)
1329 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1330 UIDINFO_UNLOCK(uip);
1331 return (1);
1332 }
1333
1334 /*
1335 * Change the total socket buffer size a user has used.
1336 */
1337 int
1338 chgsbsize(uip, hiwat, to, max)
1339 struct uidinfo *uip;
1340 u_int *hiwat;
1341 u_int to;
1342 rlim_t max;
1343 {
1344 rlim_t new;
1345
1346 UIDINFO_LOCK(uip);
1347 new = uip->ui_sbsize + to - *hiwat;
1348 /* Don't allow them to exceed max, but allow subtraction. */
1349 if (to > *hiwat && new > max) {
1350 UIDINFO_UNLOCK(uip);
1351 return (0);
1352 }
1353 uip->ui_sbsize = new;
1354 UIDINFO_UNLOCK(uip);
1355 *hiwat = to;
1356 if (new < 0)
1357 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1358 return (1);
1359 }
Cache object: 6fe704452d38706dfa712eae6258f46c
|