1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94
37 */
38
39 #include <sys/cdefs.h>
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mutex.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/refcount.h>
52 #include <sys/racct.h>
53 #include <sys/resourcevar.h>
54 #include <sys/rwlock.h>
55 #include <sys/sched.h>
56 #include <sys/sx.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/time.h>
61 #include <sys/umtxvar.h>
62
63 #include <vm/vm.h>
64 #include <vm/vm_param.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_extern.h>
68
69 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
70 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
71 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
72 static struct rwlock uihashtbl_lock;
73 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
74 static u_long uihash; /* size of hash table - 1 */
75
76 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
77 struct timeval *up, struct timeval *sp);
78 static int donice(struct thread *td, struct proc *chgp, int n);
79 static struct uidinfo *uilookup(uid_t uid);
80 static void ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td);
81
82 /*
83 * Resource controls and accounting.
84 */
85 #ifndef _SYS_SYSPROTO_H_
86 struct getpriority_args {
87 int which;
88 int who;
89 };
90 #endif
91 int
sys_getpriority(struct thread * td,struct getpriority_args * uap)92 sys_getpriority(struct thread *td, struct getpriority_args *uap)
93 {
94
95 return (kern_getpriority(td, uap->which, uap->who));
96 }
97
98 int
kern_getpriority(struct thread * td,int which,int who)99 kern_getpriority(struct thread *td, int which, int who)
100 {
101 struct proc *p;
102 struct pgrp *pg;
103 int error, low;
104
105 error = 0;
106 low = PRIO_MAX + 1;
107 switch (which) {
108 case PRIO_PROCESS:
109 if (who == 0)
110 low = td->td_proc->p_nice;
111 else {
112 p = pfind(who);
113 if (p == NULL)
114 break;
115 if (p_cansee(td, p) == 0)
116 low = p->p_nice;
117 PROC_UNLOCK(p);
118 }
119 break;
120
121 case PRIO_PGRP:
122 sx_slock(&proctree_lock);
123 if (who == 0) {
124 pg = td->td_proc->p_pgrp;
125 PGRP_LOCK(pg);
126 } else {
127 pg = pgfind(who);
128 if (pg == NULL) {
129 sx_sunlock(&proctree_lock);
130 break;
131 }
132 }
133 sx_sunlock(&proctree_lock);
134 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
135 PROC_LOCK(p);
136 if (p->p_state == PRS_NORMAL &&
137 p_cansee(td, p) == 0) {
138 if (p->p_nice < low)
139 low = p->p_nice;
140 }
141 PROC_UNLOCK(p);
142 }
143 PGRP_UNLOCK(pg);
144 break;
145
146 case PRIO_USER:
147 if (who == 0)
148 who = td->td_ucred->cr_uid;
149 sx_slock(&allproc_lock);
150 FOREACH_PROC_IN_SYSTEM(p) {
151 PROC_LOCK(p);
152 if (p->p_state == PRS_NORMAL &&
153 p_cansee(td, p) == 0 &&
154 p->p_ucred->cr_uid == who) {
155 if (p->p_nice < low)
156 low = p->p_nice;
157 }
158 PROC_UNLOCK(p);
159 }
160 sx_sunlock(&allproc_lock);
161 break;
162
163 default:
164 error = EINVAL;
165 break;
166 }
167 if (low == PRIO_MAX + 1 && error == 0)
168 error = ESRCH;
169 td->td_retval[0] = low;
170 return (error);
171 }
172
173 #ifndef _SYS_SYSPROTO_H_
174 struct setpriority_args {
175 int which;
176 int who;
177 int prio;
178 };
179 #endif
180 int
sys_setpriority(struct thread * td,struct setpriority_args * uap)181 sys_setpriority(struct thread *td, struct setpriority_args *uap)
182 {
183
184 return (kern_setpriority(td, uap->which, uap->who, uap->prio));
185 }
186
187 int
kern_setpriority(struct thread * td,int which,int who,int prio)188 kern_setpriority(struct thread *td, int which, int who, int prio)
189 {
190 struct proc *curp, *p;
191 struct pgrp *pg;
192 int found = 0, error = 0;
193
194 curp = td->td_proc;
195 switch (which) {
196 case PRIO_PROCESS:
197 if (who == 0) {
198 PROC_LOCK(curp);
199 error = donice(td, curp, prio);
200 PROC_UNLOCK(curp);
201 } else {
202 p = pfind(who);
203 if (p == NULL)
204 break;
205 error = p_cansee(td, p);
206 if (error == 0)
207 error = donice(td, p, prio);
208 PROC_UNLOCK(p);
209 }
210 found++;
211 break;
212
213 case PRIO_PGRP:
214 sx_slock(&proctree_lock);
215 if (who == 0) {
216 pg = curp->p_pgrp;
217 PGRP_LOCK(pg);
218 } else {
219 pg = pgfind(who);
220 if (pg == NULL) {
221 sx_sunlock(&proctree_lock);
222 break;
223 }
224 }
225 sx_sunlock(&proctree_lock);
226 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
227 PROC_LOCK(p);
228 if (p->p_state == PRS_NORMAL &&
229 p_cansee(td, p) == 0) {
230 error = donice(td, p, prio);
231 found++;
232 }
233 PROC_UNLOCK(p);
234 }
235 PGRP_UNLOCK(pg);
236 break;
237
238 case PRIO_USER:
239 if (who == 0)
240 who = td->td_ucred->cr_uid;
241 sx_slock(&allproc_lock);
242 FOREACH_PROC_IN_SYSTEM(p) {
243 PROC_LOCK(p);
244 if (p->p_state == PRS_NORMAL &&
245 p->p_ucred->cr_uid == who &&
246 p_cansee(td, p) == 0) {
247 error = donice(td, p, prio);
248 found++;
249 }
250 PROC_UNLOCK(p);
251 }
252 sx_sunlock(&allproc_lock);
253 break;
254
255 default:
256 error = EINVAL;
257 break;
258 }
259 if (found == 0 && error == 0)
260 error = ESRCH;
261 return (error);
262 }
263
264 /*
265 * Set "nice" for a (whole) process.
266 */
267 static int
donice(struct thread * td,struct proc * p,int n)268 donice(struct thread *td, struct proc *p, int n)
269 {
270 int error;
271
272 PROC_LOCK_ASSERT(p, MA_OWNED);
273 if ((error = p_cansched(td, p)))
274 return (error);
275 if (n > PRIO_MAX)
276 n = PRIO_MAX;
277 if (n < PRIO_MIN)
278 n = PRIO_MIN;
279 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
280 return (EACCES);
281 sched_nice(p, n);
282 return (0);
283 }
284
285 static int unprivileged_idprio;
286 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
287 &unprivileged_idprio, 0,
288 "Allow non-root users to set an idle priority (deprecated)");
289
290 /*
291 * Set realtime priority for LWP.
292 */
293 #ifndef _SYS_SYSPROTO_H_
294 struct rtprio_thread_args {
295 int function;
296 lwpid_t lwpid;
297 struct rtprio *rtp;
298 };
299 #endif
300 int
sys_rtprio_thread(struct thread * td,struct rtprio_thread_args * uap)301 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
302 {
303 struct proc *p;
304 struct rtprio rtp;
305 struct thread *td1;
306 int cierror, error;
307
308 /* Perform copyin before acquiring locks if needed. */
309 if (uap->function == RTP_SET)
310 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
311 else
312 cierror = 0;
313
314 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
315 p = td->td_proc;
316 td1 = td;
317 PROC_LOCK(p);
318 } else {
319 td1 = tdfind(uap->lwpid, -1);
320 if (td1 == NULL)
321 return (ESRCH);
322 p = td1->td_proc;
323 }
324
325 switch (uap->function) {
326 case RTP_LOOKUP:
327 if ((error = p_cansee(td, p)))
328 break;
329 pri_to_rtp(td1, &rtp);
330 PROC_UNLOCK(p);
331 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
332 case RTP_SET:
333 if ((error = p_cansched(td, p)) || (error = cierror))
334 break;
335
336 /* Disallow setting rtprio in most cases if not superuser. */
337
338 /*
339 * Realtime priority has to be restricted for reasons which
340 * should be obvious. However, for idleprio processes, there is
341 * a potential for system deadlock if an idleprio process gains
342 * a lock on a resource that other processes need (and the
343 * idleprio process can't run due to a CPU-bound normal
344 * process). Fix me! XXX
345 *
346 * This problem is not only related to idleprio process.
347 * A user level program can obtain a file lock and hold it
348 * indefinitely. Additionally, without idleprio processes it is
349 * still conceivable that a program with low priority will never
350 * get to run. In short, allowing this feature might make it
351 * easier to lock a resource indefinitely, but it is not the
352 * only thing that makes it possible.
353 */
354 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
355 (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
356 break;
357 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
358 unprivileged_idprio == 0 &&
359 (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
360 break;
361 error = rtp_to_pri(&rtp, td1);
362 break;
363 default:
364 error = EINVAL;
365 break;
366 }
367 PROC_UNLOCK(p);
368 return (error);
369 }
370
371 /*
372 * Set realtime priority.
373 */
374 #ifndef _SYS_SYSPROTO_H_
375 struct rtprio_args {
376 int function;
377 pid_t pid;
378 struct rtprio *rtp;
379 };
380 #endif
381 int
sys_rtprio(struct thread * td,struct rtprio_args * uap)382 sys_rtprio(struct thread *td, struct rtprio_args *uap)
383 {
384 struct proc *p;
385 struct thread *tdp;
386 struct rtprio rtp;
387 int cierror, error;
388
389 /* Perform copyin before acquiring locks if needed. */
390 if (uap->function == RTP_SET)
391 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
392 else
393 cierror = 0;
394
395 if (uap->pid == 0) {
396 p = td->td_proc;
397 PROC_LOCK(p);
398 } else {
399 p = pfind(uap->pid);
400 if (p == NULL)
401 return (ESRCH);
402 }
403
404 switch (uap->function) {
405 case RTP_LOOKUP:
406 if ((error = p_cansee(td, p)))
407 break;
408 /*
409 * Return OUR priority if no pid specified,
410 * or if one is, report the highest priority
411 * in the process. There isn't much more you can do as
412 * there is only room to return a single priority.
413 * Note: specifying our own pid is not the same
414 * as leaving it zero.
415 */
416 if (uap->pid == 0) {
417 pri_to_rtp(td, &rtp);
418 } else {
419 struct rtprio rtp2;
420
421 rtp.type = RTP_PRIO_IDLE;
422 rtp.prio = RTP_PRIO_MAX;
423 FOREACH_THREAD_IN_PROC(p, tdp) {
424 pri_to_rtp(tdp, &rtp2);
425 if (rtp2.type < rtp.type ||
426 (rtp2.type == rtp.type &&
427 rtp2.prio < rtp.prio)) {
428 rtp.type = rtp2.type;
429 rtp.prio = rtp2.prio;
430 }
431 }
432 }
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 /*
440 * Disallow setting rtprio in most cases if not superuser.
441 * See the comment in sys_rtprio_thread about idprio
442 * threads holding a lock.
443 */
444 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
445 (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
446 break;
447 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
448 unprivileged_idprio == 0 &&
449 (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
450 break;
451
452 /*
453 * If we are setting our own priority, set just our
454 * thread but if we are doing another process,
455 * do all the threads on that process. If we
456 * specify our own pid we do the latter.
457 */
458 if (uap->pid == 0) {
459 error = rtp_to_pri(&rtp, td);
460 } else {
461 FOREACH_THREAD_IN_PROC(p, td) {
462 if ((error = rtp_to_pri(&rtp, td)) != 0)
463 break;
464 }
465 }
466 break;
467 default:
468 error = EINVAL;
469 break;
470 }
471 PROC_UNLOCK(p);
472 return (error);
473 }
474
475 int
rtp_to_pri(struct rtprio * rtp,struct thread * td)476 rtp_to_pri(struct rtprio *rtp, struct thread *td)
477 {
478 u_char newpri, oldclass, oldpri;
479
480 switch (RTP_PRIO_BASE(rtp->type)) {
481 case RTP_PRIO_REALTIME:
482 if (rtp->prio > RTP_PRIO_MAX)
483 return (EINVAL);
484 newpri = PRI_MIN_REALTIME + rtp->prio;
485 break;
486 case RTP_PRIO_NORMAL:
487 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
488 return (EINVAL);
489 newpri = PRI_MIN_TIMESHARE + rtp->prio;
490 break;
491 case RTP_PRIO_IDLE:
492 if (rtp->prio > RTP_PRIO_MAX)
493 return (EINVAL);
494 newpri = PRI_MIN_IDLE + rtp->prio;
495 break;
496 default:
497 return (EINVAL);
498 }
499
500 thread_lock(td);
501 oldclass = td->td_pri_class;
502 sched_class(td, rtp->type); /* XXX fix */
503 oldpri = td->td_user_pri;
504 sched_user_prio(td, newpri);
505 if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
506 td->td_pri_class != RTP_PRIO_NORMAL))
507 sched_prio(td, td->td_user_pri);
508 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
509 critical_enter();
510 thread_unlock(td);
511 umtx_pi_adjust(td, oldpri);
512 critical_exit();
513 } else
514 thread_unlock(td);
515 return (0);
516 }
517
518 void
pri_to_rtp(struct thread * td,struct rtprio * rtp)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
osetrlimit(struct thread * td,struct osetrlimit_args * uap)548 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
549 {
550 struct orlimit olim;
551 struct rlimit lim;
552 int error;
553
554 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
555 return (error);
556 lim.rlim_cur = olim.rlim_cur;
557 lim.rlim_max = olim.rlim_max;
558 error = kern_setrlimit(td, uap->which, &lim);
559 return (error);
560 }
561
562 #ifndef _SYS_SYSPROTO_H_
563 struct ogetrlimit_args {
564 u_int which;
565 struct orlimit *rlp;
566 };
567 #endif
568 int
ogetrlimit(struct thread * td,struct ogetrlimit_args * uap)569 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
570 {
571 struct orlimit olim;
572 struct rlimit rl;
573 int error;
574
575 if (uap->which >= RLIM_NLIMITS)
576 return (EINVAL);
577 lim_rlimit(td, uap->which, &rl);
578
579 /*
580 * XXX would be more correct to convert only RLIM_INFINITY to the
581 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
582 * values. Most 64->32 and 32->16 conversions, including not
583 * unimportant ones of uids are even more broken than what we
584 * do here (they blindly truncate). We don't do this correctly
585 * here since we have little experience with EOVERFLOW yet.
586 * Elsewhere, getuid() can't fail...
587 */
588 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
589 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
590 error = copyout(&olim, uap->rlp, sizeof(olim));
591 return (error);
592 }
593 #endif /* COMPAT_43 */
594
595 #ifndef _SYS_SYSPROTO_H_
596 struct setrlimit_args {
597 u_int which;
598 struct rlimit *rlp;
599 };
600 #endif
601 int
sys_setrlimit(struct thread * td,struct setrlimit_args * uap)602 sys_setrlimit(struct thread *td, struct setrlimit_args *uap)
603 {
604 struct rlimit alim;
605 int error;
606
607 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
608 return (error);
609 error = kern_setrlimit(td, uap->which, &alim);
610 return (error);
611 }
612
613 static void
lim_cb(void * arg)614 lim_cb(void *arg)
615 {
616 struct rlimit rlim;
617 struct thread *td;
618 struct proc *p;
619
620 p = arg;
621 PROC_LOCK_ASSERT(p, MA_OWNED);
622 /*
623 * Check if the process exceeds its cpu resource allocation. If
624 * it reaches the max, arrange to kill the process in ast().
625 */
626 if (p->p_cpulimit == RLIM_INFINITY)
627 return;
628 PROC_STATLOCK(p);
629 FOREACH_THREAD_IN_PROC(p, td) {
630 ruxagg(p, td);
631 }
632 PROC_STATUNLOCK(p);
633 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
634 lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
635 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
636 killproc(p, "exceeded maximum CPU limit");
637 } else {
638 if (p->p_cpulimit < rlim.rlim_max)
639 p->p_cpulimit += 5;
640 kern_psignal(p, SIGXCPU);
641 }
642 }
643 if ((p->p_flag & P_WEXIT) == 0)
644 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
645 lim_cb, p, C_PREL(1));
646 }
647
648 int
kern_setrlimit(struct thread * td,u_int which,struct rlimit * limp)649 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
650 {
651
652 return (kern_proc_setrlimit(td, td->td_proc, which, limp));
653 }
654
655 int
kern_proc_setrlimit(struct thread * td,struct proc * p,u_int which,struct rlimit * limp)656 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
657 struct rlimit *limp)
658 {
659 struct plimit *newlim, *oldlim, *oldlim_td;
660 struct rlimit *alimp;
661 struct rlimit oldssiz;
662 int error;
663
664 if (which >= RLIM_NLIMITS)
665 return (EINVAL);
666
667 /*
668 * Preserve historical bugs by treating negative limits as unsigned.
669 */
670 if (limp->rlim_cur < 0)
671 limp->rlim_cur = RLIM_INFINITY;
672 if (limp->rlim_max < 0)
673 limp->rlim_max = RLIM_INFINITY;
674
675 oldssiz.rlim_cur = 0;
676 newlim = lim_alloc();
677 PROC_LOCK(p);
678 oldlim = p->p_limit;
679 alimp = &oldlim->pl_rlimit[which];
680 if (limp->rlim_cur > alimp->rlim_max ||
681 limp->rlim_max > alimp->rlim_max)
682 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
683 PROC_UNLOCK(p);
684 lim_free(newlim);
685 return (error);
686 }
687 if (limp->rlim_cur > limp->rlim_max)
688 limp->rlim_cur = limp->rlim_max;
689 lim_copy(newlim, oldlim);
690 alimp = &newlim->pl_rlimit[which];
691
692 switch (which) {
693 case RLIMIT_CPU:
694 if (limp->rlim_cur != RLIM_INFINITY &&
695 p->p_cpulimit == RLIM_INFINITY)
696 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
697 lim_cb, p, C_PREL(1));
698 p->p_cpulimit = limp->rlim_cur;
699 break;
700 case RLIMIT_DATA:
701 if (limp->rlim_cur > maxdsiz)
702 limp->rlim_cur = maxdsiz;
703 if (limp->rlim_max > maxdsiz)
704 limp->rlim_max = maxdsiz;
705 break;
706
707 case RLIMIT_STACK:
708 if (limp->rlim_cur > maxssiz)
709 limp->rlim_cur = maxssiz;
710 if (limp->rlim_max > maxssiz)
711 limp->rlim_max = maxssiz;
712 oldssiz = *alimp;
713 if (p->p_sysent->sv_fixlimit != NULL)
714 p->p_sysent->sv_fixlimit(&oldssiz,
715 RLIMIT_STACK);
716 break;
717
718 case RLIMIT_NOFILE:
719 if (limp->rlim_cur > maxfilesperproc)
720 limp->rlim_cur = maxfilesperproc;
721 if (limp->rlim_max > maxfilesperproc)
722 limp->rlim_max = maxfilesperproc;
723 break;
724
725 case RLIMIT_NPROC:
726 if (limp->rlim_cur > maxprocperuid)
727 limp->rlim_cur = maxprocperuid;
728 if (limp->rlim_max > maxprocperuid)
729 limp->rlim_max = maxprocperuid;
730 if (limp->rlim_cur < 1)
731 limp->rlim_cur = 1;
732 if (limp->rlim_max < 1)
733 limp->rlim_max = 1;
734 break;
735 }
736 if (p->p_sysent->sv_fixlimit != NULL)
737 p->p_sysent->sv_fixlimit(limp, which);
738 *alimp = *limp;
739 p->p_limit = newlim;
740 PROC_UPDATE_COW(p);
741 oldlim_td = NULL;
742 if (td == curthread && PROC_COW_CHANGECOUNT(td, p) == 1) {
743 oldlim_td = lim_cowsync();
744 thread_cow_synced(td);
745 }
746 PROC_UNLOCK(p);
747 if (oldlim_td != NULL) {
748 MPASS(oldlim_td == oldlim);
749 lim_freen(oldlim, 2);
750 } else {
751 lim_free(oldlim);
752 }
753
754 if (which == RLIMIT_STACK &&
755 /*
756 * Skip calls from exec_new_vmspace(), done when stack is
757 * not mapped yet.
758 */
759 (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
760 /*
761 * Stack is allocated to the max at exec time with only
762 * "rlim_cur" bytes accessible. If stack limit is going
763 * up make more accessible, if going down make inaccessible.
764 */
765 if (limp->rlim_cur != oldssiz.rlim_cur) {
766 vm_offset_t addr;
767 vm_size_t size;
768 vm_prot_t prot;
769
770 if (limp->rlim_cur > oldssiz.rlim_cur) {
771 prot = p->p_sysent->sv_stackprot;
772 size = limp->rlim_cur - oldssiz.rlim_cur;
773 addr = round_page(p->p_vmspace->vm_stacktop) -
774 limp->rlim_cur;
775 } else {
776 prot = VM_PROT_NONE;
777 size = oldssiz.rlim_cur - limp->rlim_cur;
778 addr = round_page(p->p_vmspace->vm_stacktop) -
779 oldssiz.rlim_cur;
780 }
781 addr = trunc_page(addr);
782 size = round_page(size);
783 (void)vm_map_protect(&p->p_vmspace->vm_map,
784 addr, addr + size, prot, 0,
785 VM_MAP_PROTECT_SET_PROT);
786 }
787 }
788
789 return (0);
790 }
791
792 #ifndef _SYS_SYSPROTO_H_
793 struct getrlimit_args {
794 u_int which;
795 struct rlimit *rlp;
796 };
797 #endif
798 /* ARGSUSED */
799 int
sys_getrlimit(struct thread * td,struct getrlimit_args * uap)800 sys_getrlimit(struct thread *td, struct getrlimit_args *uap)
801 {
802 struct rlimit rlim;
803 int error;
804
805 if (uap->which >= RLIM_NLIMITS)
806 return (EINVAL);
807 lim_rlimit(td, uap->which, &rlim);
808 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
809 return (error);
810 }
811
812 static int
getrlimitusage_one(struct proc * p,u_int which,int flags,rlim_t * res)813 getrlimitusage_one(struct proc *p, u_int which, int flags, rlim_t *res)
814 {
815 struct thread *td;
816 struct uidinfo *ui;
817 struct vmspace *vm;
818 uid_t uid;
819 int error;
820
821 error = 0;
822 PROC_LOCK(p);
823 uid = (flags & GETRLIMITUSAGE_EUID) == 0 ? p->p_ucred->cr_ruid :
824 p->p_ucred->cr_uid;
825 PROC_UNLOCK(p);
826
827 ui = uifind(uid);
828 vm = vmspace_acquire_ref(p);
829
830 switch (which) {
831 case RLIMIT_CPU:
832 PROC_LOCK(p);
833 PROC_STATLOCK(p);
834 FOREACH_THREAD_IN_PROC(p, td)
835 ruxagg(p, td);
836 *res = p->p_rux.rux_runtime;
837 PROC_STATUNLOCK(p);
838 PROC_UNLOCK(p);
839 *res /= cpu_tickrate();
840 break;
841 case RLIMIT_FSIZE:
842 error = ENXIO;
843 break;
844 case RLIMIT_DATA:
845 if (vm == NULL)
846 error = ENXIO;
847 else
848 *res = vm->vm_dsize * PAGE_SIZE;
849 break;
850 case RLIMIT_STACK:
851 if (vm == NULL)
852 error = ENXIO;
853 else
854 *res = vm->vm_ssize * PAGE_SIZE;
855 break;
856 case RLIMIT_CORE:
857 error = ENXIO;
858 break;
859 case RLIMIT_RSS:
860 if (vm == NULL)
861 error = ENXIO;
862 else
863 *res = vmspace_resident_count(vm) * PAGE_SIZE;
864 break;
865 case RLIMIT_MEMLOCK:
866 if (vm == NULL)
867 error = ENXIO;
868 else
869 *res = pmap_wired_count(vmspace_pmap(vm)) * PAGE_SIZE;
870 break;
871 case RLIMIT_NPROC:
872 *res = ui->ui_proccnt;
873 break;
874 case RLIMIT_NOFILE:
875 *res = proc_nfiles(p);
876 break;
877 case RLIMIT_SBSIZE:
878 *res = ui->ui_sbsize;
879 break;
880 case RLIMIT_VMEM:
881 if (vm == NULL)
882 error = ENXIO;
883 else
884 *res = vm->vm_map.size;
885 break;
886 case RLIMIT_NPTS:
887 *res = ui->ui_ptscnt;
888 break;
889 case RLIMIT_SWAP:
890 *res = ui->ui_vmsize;
891 break;
892 case RLIMIT_KQUEUES:
893 *res = ui->ui_kqcnt;
894 break;
895 case RLIMIT_UMTXP:
896 *res = ui->ui_umtxcnt;
897 break;
898 case RLIMIT_PIPEBUF:
899 *res = ui->ui_pipecnt;
900 break;
901 default:
902 error = EINVAL;
903 break;
904 }
905
906 vmspace_free(vm);
907 uifree(ui);
908 return (error);
909 }
910
911 int
sys_getrlimitusage(struct thread * td,struct getrlimitusage_args * uap)912 sys_getrlimitusage(struct thread *td, struct getrlimitusage_args *uap)
913 {
914 rlim_t res;
915 int error;
916
917 if ((uap->flags & ~(GETRLIMITUSAGE_EUID)) != 0)
918 return (EINVAL);
919 error = getrlimitusage_one(curproc, uap->which, uap->flags, &res);
920 if (error == 0)
921 error = copyout(&res, uap->res, sizeof(res));
922 return (error);
923 }
924
925 /*
926 * Transform the running time and tick information for children of proc p
927 * into user and system time usage.
928 */
929 void
calccru(struct proc * p,struct timeval * up,struct timeval * sp)930 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
931 {
932
933 PROC_LOCK_ASSERT(p, MA_OWNED);
934 calcru1(p, &p->p_crux, up, sp);
935 }
936
937 /*
938 * Transform the running time and tick information in proc p into user
939 * and system time usage. If appropriate, include the current time slice
940 * on this CPU.
941 */
942 void
calcru(struct proc * p,struct timeval * up,struct timeval * sp)943 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
944 {
945 struct thread *td;
946 uint64_t runtime, u;
947
948 PROC_LOCK_ASSERT(p, MA_OWNED);
949 PROC_STATLOCK_ASSERT(p, MA_OWNED);
950 /*
951 * If we are getting stats for the current process, then add in the
952 * stats that this thread has accumulated in its current time slice.
953 * We reset the thread and CPU state as if we had performed a context
954 * switch right here.
955 */
956 td = curthread;
957 if (td->td_proc == p) {
958 u = cpu_ticks();
959 runtime = u - PCPU_GET(switchtime);
960 td->td_runtime += runtime;
961 td->td_incruntime += runtime;
962 PCPU_SET(switchtime, u);
963 }
964 /* Make sure the per-thread stats are current. */
965 FOREACH_THREAD_IN_PROC(p, td) {
966 if (td->td_incruntime == 0)
967 continue;
968 ruxagg(p, td);
969 }
970 calcru1(p, &p->p_rux, up, sp);
971 }
972
973 /* Collect resource usage for a single thread. */
974 void
rufetchtd(struct thread * td,struct rusage * ru)975 rufetchtd(struct thread *td, struct rusage *ru)
976 {
977 struct proc *p;
978 uint64_t runtime, u;
979
980 p = td->td_proc;
981 PROC_STATLOCK_ASSERT(p, MA_OWNED);
982 THREAD_LOCK_ASSERT(td, MA_OWNED);
983 /*
984 * If we are getting stats for the current thread, then add in the
985 * stats that this thread has accumulated in its current time slice.
986 * We reset the thread and CPU state as if we had performed a context
987 * switch right here.
988 */
989 if (td == curthread) {
990 u = cpu_ticks();
991 runtime = u - PCPU_GET(switchtime);
992 td->td_runtime += runtime;
993 td->td_incruntime += runtime;
994 PCPU_SET(switchtime, u);
995 }
996 ruxagg_locked(p, td);
997 *ru = td->td_ru;
998 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
999 }
1000
1001 static uint64_t
mul64_by_fraction(uint64_t a,uint64_t b,uint64_t c)1002 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
1003 {
1004 uint64_t acc, bh, bl;
1005 int i, s, sa, sb;
1006
1007 /*
1008 * Calculate (a * b) / c accurately enough without overflowing. c
1009 * must be nonzero, and its top bit must be 0. a or b must be
1010 * <= c, and the implementation is tuned for b <= c.
1011 *
1012 * The comments about times are for use in calcru1() with units of
1013 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
1014 *
1015 * Let n be the number of top zero bits in c. Each iteration
1016 * either returns, or reduces b by right shifting it by at least n.
1017 * The number of iterations is at most 1 + 64 / n, and the error is
1018 * at most the number of iterations.
1019 *
1020 * It is very unusual to need even 2 iterations. Previous
1021 * implementations overflowed essentially by returning early in the
1022 * first iteration, with n = 38 giving overflow at 105+ hours and
1023 * n = 32 giving overlow at at 388+ days despite a more careful
1024 * calculation. 388 days is a reasonable uptime, and the calculation
1025 * needs to work for the uptime times the number of CPUs since 'a'
1026 * is per-process.
1027 */
1028 if (a >= (uint64_t)1 << 63)
1029 return (0); /* Unsupported arg -- can't happen. */
1030 acc = 0;
1031 for (i = 0; i < 128; i++) {
1032 sa = flsll(a);
1033 sb = flsll(b);
1034 if (sa + sb <= 64)
1035 /* Up to 105 hours on first iteration. */
1036 return (acc + (a * b) / c);
1037 if (a >= c) {
1038 /*
1039 * This reduction is based on a = q * c + r, with the
1040 * remainder r < c. 'a' may be large to start, and
1041 * moving bits from b into 'a' at the end of the loop
1042 * sets the top bit of 'a', so the reduction makes
1043 * significant progress.
1044 */
1045 acc += (a / c) * b;
1046 a %= c;
1047 sa = flsll(a);
1048 if (sa + sb <= 64)
1049 /* Up to 388 days on first iteration. */
1050 return (acc + (a * b) / c);
1051 }
1052
1053 /*
1054 * This step writes a * b as a * ((bh << s) + bl) =
1055 * a * (bh << s) + a * bl = (a << s) * bh + a * bl. The 2
1056 * additive terms are handled separately. Splitting in
1057 * this way is linear except for rounding errors.
1058 *
1059 * s = 64 - sa is the maximum such that a << s fits in 64
1060 * bits. Since a < c and c has at least 1 zero top bit,
1061 * sa < 64 and s > 0. Thus this step makes progress by
1062 * reducing b (it increases 'a', but taking remainders on
1063 * the next iteration completes the reduction).
1064 *
1065 * Finally, the choice for s is just what is needed to keep
1066 * a * bl from overflowing, so we don't need complications
1067 * like a recursive call mul64_by_fraction(a, bl, c) to
1068 * handle the second additive term.
1069 */
1070 s = 64 - sa;
1071 bh = b >> s;
1072 bl = b - (bh << s);
1073 acc += (a * bl) / c;
1074 a <<= s;
1075 b = bh;
1076 }
1077 return (0); /* Algorithm failure -- can't happen. */
1078 }
1079
1080 static void
calcru1(struct proc * p,struct rusage_ext * ruxp,struct timeval * up,struct timeval * sp)1081 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
1082 struct timeval *sp)
1083 {
1084 /* {user, system, interrupt, total} {ticks, usec}: */
1085 uint64_t ut, uu, st, su, it, tt, tu;
1086
1087 ut = ruxp->rux_uticks;
1088 st = ruxp->rux_sticks;
1089 it = ruxp->rux_iticks;
1090 tt = ut + st + it;
1091 if (tt == 0) {
1092 /* Avoid divide by zero */
1093 st = 1;
1094 tt = 1;
1095 }
1096 tu = cputick2usec(ruxp->rux_runtime);
1097 if ((int64_t)tu < 0) {
1098 /* XXX: this should be an assert /phk */
1099 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
1100 (intmax_t)tu, p->p_pid, p->p_comm);
1101 tu = ruxp->rux_tu;
1102 }
1103
1104 /* Subdivide tu. Avoid overflow in the multiplications. */
1105 if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
1106 /* Up to 76 hours when stathz is 128. */
1107 uu = (tu * ut) / tt;
1108 su = (tu * st) / tt;
1109 } else {
1110 uu = mul64_by_fraction(tu, ut, tt);
1111 su = mul64_by_fraction(tu, st, tt);
1112 }
1113
1114 if (tu >= ruxp->rux_tu) {
1115 /*
1116 * The normal case, time increased.
1117 * Enforce monotonicity of bucketed numbers.
1118 */
1119 if (uu < ruxp->rux_uu)
1120 uu = ruxp->rux_uu;
1121 if (su < ruxp->rux_su)
1122 su = ruxp->rux_su;
1123 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1124 /*
1125 * When we calibrate the cputicker, it is not uncommon to
1126 * see the presumably fixed frequency increase slightly over
1127 * time as a result of thermal stabilization and NTP
1128 * discipline (of the reference clock). We therefore ignore
1129 * a bit of backwards slop because we expect to catch up
1130 * shortly. We use a 3 microsecond limit to catch low
1131 * counts and a 1% limit for high counts.
1132 */
1133 uu = ruxp->rux_uu;
1134 su = ruxp->rux_su;
1135 tu = ruxp->rux_tu;
1136 } else if (vm_guest == VM_GUEST_NO) { /* tu < ruxp->rux_tu */
1137 /*
1138 * What happened here was likely that a laptop, which ran at
1139 * a reduced clock frequency at boot, kicked into high gear.
1140 * The wisdom of spamming this message in that case is
1141 * dubious, but it might also be indicative of something
1142 * serious, so lets keep it and hope laptops can be made
1143 * more truthful about their CPU speed via ACPI.
1144 */
1145 printf("calcru: runtime went backwards from %ju usec "
1146 "to %ju usec for pid %d (%s)\n",
1147 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1148 p->p_pid, p->p_comm);
1149 }
1150
1151 ruxp->rux_uu = uu;
1152 ruxp->rux_su = su;
1153 ruxp->rux_tu = tu;
1154
1155 up->tv_sec = uu / 1000000;
1156 up->tv_usec = uu % 1000000;
1157 sp->tv_sec = su / 1000000;
1158 sp->tv_usec = su % 1000000;
1159 }
1160
1161 #ifndef _SYS_SYSPROTO_H_
1162 struct getrusage_args {
1163 int who;
1164 struct rusage *rusage;
1165 };
1166 #endif
1167 int
sys_getrusage(struct thread * td,struct getrusage_args * uap)1168 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1169 {
1170 struct rusage ru;
1171 int error;
1172
1173 error = kern_getrusage(td, uap->who, &ru);
1174 if (error == 0)
1175 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1176 return (error);
1177 }
1178
1179 int
kern_getrusage(struct thread * td,int who,struct rusage * rup)1180 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1181 {
1182 struct proc *p;
1183 int error;
1184
1185 error = 0;
1186 p = td->td_proc;
1187 PROC_LOCK(p);
1188 switch (who) {
1189 case RUSAGE_SELF:
1190 rufetchcalc(p, rup, &rup->ru_utime,
1191 &rup->ru_stime);
1192 break;
1193
1194 case RUSAGE_CHILDREN:
1195 *rup = p->p_stats->p_cru;
1196 calccru(p, &rup->ru_utime, &rup->ru_stime);
1197 break;
1198
1199 case RUSAGE_THREAD:
1200 PROC_STATLOCK(p);
1201 thread_lock(td);
1202 rufetchtd(td, rup);
1203 thread_unlock(td);
1204 PROC_STATUNLOCK(p);
1205 break;
1206
1207 default:
1208 error = EINVAL;
1209 }
1210 PROC_UNLOCK(p);
1211 return (error);
1212 }
1213
1214 void
rucollect(struct rusage * ru,struct rusage * ru2)1215 rucollect(struct rusage *ru, struct rusage *ru2)
1216 {
1217 long *ip, *ip2;
1218 int i;
1219
1220 if (ru->ru_maxrss < ru2->ru_maxrss)
1221 ru->ru_maxrss = ru2->ru_maxrss;
1222 ip = &ru->ru_first;
1223 ip2 = &ru2->ru_first;
1224 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1225 *ip++ += *ip2++;
1226 }
1227
1228 void
ruadd(struct rusage * ru,struct rusage_ext * rux,struct rusage * ru2,struct rusage_ext * rux2)1229 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1230 struct rusage_ext *rux2)
1231 {
1232
1233 rux->rux_runtime += rux2->rux_runtime;
1234 rux->rux_uticks += rux2->rux_uticks;
1235 rux->rux_sticks += rux2->rux_sticks;
1236 rux->rux_iticks += rux2->rux_iticks;
1237 rux->rux_uu += rux2->rux_uu;
1238 rux->rux_su += rux2->rux_su;
1239 rux->rux_tu += rux2->rux_tu;
1240 rucollect(ru, ru2);
1241 }
1242
1243 /*
1244 * Aggregate tick counts into the proc's rusage_ext.
1245 */
1246 static void
ruxagg_ext_locked(struct rusage_ext * rux,struct thread * td)1247 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1248 {
1249
1250 rux->rux_runtime += td->td_incruntime;
1251 rux->rux_uticks += td->td_uticks;
1252 rux->rux_sticks += td->td_sticks;
1253 rux->rux_iticks += td->td_iticks;
1254 }
1255
1256 void
ruxagg_locked(struct proc * p,struct thread * td)1257 ruxagg_locked(struct proc *p, struct thread *td)
1258 {
1259 THREAD_LOCK_ASSERT(td, MA_OWNED);
1260 PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1261
1262 ruxagg_ext_locked(&p->p_rux, td);
1263 ruxagg_ext_locked(&td->td_rux, td);
1264 td->td_incruntime = 0;
1265 td->td_uticks = 0;
1266 td->td_iticks = 0;
1267 td->td_sticks = 0;
1268 }
1269
1270 void
ruxagg(struct proc * p,struct thread * td)1271 ruxagg(struct proc *p, struct thread *td)
1272 {
1273
1274 thread_lock(td);
1275 ruxagg_locked(p, td);
1276 thread_unlock(td);
1277 }
1278
1279 /*
1280 * Update the rusage_ext structure and fetch a valid aggregate rusage
1281 * for proc p if storage for one is supplied.
1282 */
1283 void
rufetch(struct proc * p,struct rusage * ru)1284 rufetch(struct proc *p, struct rusage *ru)
1285 {
1286 struct thread *td;
1287
1288 PROC_STATLOCK_ASSERT(p, MA_OWNED);
1289
1290 *ru = p->p_ru;
1291 if (p->p_numthreads > 0) {
1292 FOREACH_THREAD_IN_PROC(p, td) {
1293 ruxagg(p, td);
1294 rucollect(ru, &td->td_ru);
1295 }
1296 }
1297 }
1298
1299 /*
1300 * Atomically perform a rufetch and a calcru together.
1301 * Consumers, can safely assume the calcru is executed only once
1302 * rufetch is completed.
1303 */
1304 void
rufetchcalc(struct proc * p,struct rusage * ru,struct timeval * up,struct timeval * sp)1305 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1306 struct timeval *sp)
1307 {
1308
1309 PROC_STATLOCK(p);
1310 rufetch(p, ru);
1311 calcru(p, up, sp);
1312 PROC_STATUNLOCK(p);
1313 }
1314
1315 /*
1316 * Allocate a new resource limits structure and initialize its
1317 * reference count and mutex pointer.
1318 */
1319 struct plimit *
lim_alloc(void)1320 lim_alloc(void)
1321 {
1322 struct plimit *limp;
1323
1324 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1325 refcount_init(&limp->pl_refcnt, 1);
1326 return (limp);
1327 }
1328
1329 struct plimit *
lim_hold(struct plimit * limp)1330 lim_hold(struct plimit *limp)
1331 {
1332
1333 refcount_acquire(&limp->pl_refcnt);
1334 return (limp);
1335 }
1336
1337 struct plimit *
lim_cowsync(void)1338 lim_cowsync(void)
1339 {
1340 struct thread *td;
1341 struct proc *p;
1342 struct plimit *oldlimit;
1343
1344 td = curthread;
1345 p = td->td_proc;
1346 PROC_LOCK_ASSERT(p, MA_OWNED);
1347
1348 if (td->td_limit == p->p_limit)
1349 return (NULL);
1350
1351 oldlimit = td->td_limit;
1352 td->td_limit = lim_hold(p->p_limit);
1353
1354 return (oldlimit);
1355 }
1356
1357 void
lim_fork(struct proc * p1,struct proc * p2)1358 lim_fork(struct proc *p1, struct proc *p2)
1359 {
1360
1361 PROC_LOCK_ASSERT(p1, MA_OWNED);
1362 PROC_LOCK_ASSERT(p2, MA_OWNED);
1363
1364 p2->p_limit = lim_hold(p1->p_limit);
1365 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1366 if (p1->p_cpulimit != RLIM_INFINITY)
1367 callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1368 lim_cb, p2, C_PREL(1));
1369 }
1370
1371 void
lim_free(struct plimit * limp)1372 lim_free(struct plimit *limp)
1373 {
1374
1375 if (refcount_release(&limp->pl_refcnt))
1376 free((void *)limp, M_PLIMIT);
1377 }
1378
1379 void
lim_freen(struct plimit * limp,int n)1380 lim_freen(struct plimit *limp, int n)
1381 {
1382
1383 if (refcount_releasen(&limp->pl_refcnt, n))
1384 free((void *)limp, M_PLIMIT);
1385 }
1386
1387 void
limbatch_add(struct limbatch * lb,struct thread * td)1388 limbatch_add(struct limbatch *lb, struct thread *td)
1389 {
1390 struct plimit *limp;
1391
1392 MPASS(td->td_limit != NULL);
1393 limp = td->td_limit;
1394
1395 if (lb->limp != limp) {
1396 if (lb->count != 0) {
1397 lim_freen(lb->limp, lb->count);
1398 lb->count = 0;
1399 }
1400 lb->limp = limp;
1401 }
1402
1403 lb->count++;
1404 }
1405
1406 void
limbatch_final(struct limbatch * lb)1407 limbatch_final(struct limbatch *lb)
1408 {
1409
1410 MPASS(lb->count != 0);
1411 lim_freen(lb->limp, lb->count);
1412 }
1413
1414 /*
1415 * Make a copy of the plimit structure.
1416 * We share these structures copy-on-write after fork.
1417 */
1418 void
lim_copy(struct plimit * dst,struct plimit * src)1419 lim_copy(struct plimit *dst, struct plimit *src)
1420 {
1421
1422 KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1423 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1424 }
1425
1426 /*
1427 * Return the hard limit for a particular system resource. The
1428 * which parameter specifies the index into the rlimit array.
1429 */
1430 rlim_t
lim_max(struct thread * td,int which)1431 lim_max(struct thread *td, int which)
1432 {
1433 struct rlimit rl;
1434
1435 lim_rlimit(td, which, &rl);
1436 return (rl.rlim_max);
1437 }
1438
1439 rlim_t
lim_max_proc(struct proc * p,int which)1440 lim_max_proc(struct proc *p, int which)
1441 {
1442 struct rlimit rl;
1443
1444 lim_rlimit_proc(p, which, &rl);
1445 return (rl.rlim_max);
1446 }
1447
1448 /*
1449 * Return the current (soft) limit for a particular system resource.
1450 * The which parameter which specifies the index into the rlimit array
1451 */
rlim_t(lim_cur)1452 rlim_t
1453 (lim_cur)(struct thread *td, int which)
1454 {
1455 struct rlimit rl;
1456
1457 lim_rlimit(td, which, &rl);
1458 return (rl.rlim_cur);
1459 }
1460
1461 rlim_t
lim_cur_proc(struct proc * p,int which)1462 lim_cur_proc(struct proc *p, int which)
1463 {
1464 struct rlimit rl;
1465
1466 lim_rlimit_proc(p, which, &rl);
1467 return (rl.rlim_cur);
1468 }
1469
1470 /*
1471 * Return a copy of the entire rlimit structure for the system limit
1472 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1473 */
1474 void
lim_rlimit(struct thread * td,int which,struct rlimit * rlp)1475 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1476 {
1477 struct proc *p = td->td_proc;
1478
1479 MPASS(td == curthread);
1480 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1481 ("request for invalid resource limit"));
1482 *rlp = td->td_limit->pl_rlimit[which];
1483 if (p->p_sysent->sv_fixlimit != NULL)
1484 p->p_sysent->sv_fixlimit(rlp, which);
1485 }
1486
1487 void
lim_rlimit_proc(struct proc * p,int which,struct rlimit * rlp)1488 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1489 {
1490
1491 PROC_LOCK_ASSERT(p, MA_OWNED);
1492 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1493 ("request for invalid resource limit"));
1494 *rlp = p->p_limit->pl_rlimit[which];
1495 if (p->p_sysent->sv_fixlimit != NULL)
1496 p->p_sysent->sv_fixlimit(rlp, which);
1497 }
1498
1499 void
uihashinit(void)1500 uihashinit(void)
1501 {
1502
1503 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1504 rw_init(&uihashtbl_lock, "uidinfo hash");
1505 }
1506
1507 /*
1508 * Look up a uidinfo struct for the parameter uid.
1509 * uihashtbl_lock must be locked.
1510 * Increase refcount on uidinfo struct returned.
1511 */
1512 static struct uidinfo *
uilookup(uid_t uid)1513 uilookup(uid_t uid)
1514 {
1515 struct uihashhead *uipp;
1516 struct uidinfo *uip;
1517
1518 rw_assert(&uihashtbl_lock, RA_LOCKED);
1519 uipp = UIHASH(uid);
1520 LIST_FOREACH(uip, uipp, ui_hash)
1521 if (uip->ui_uid == uid) {
1522 uihold(uip);
1523 break;
1524 }
1525
1526 return (uip);
1527 }
1528
1529 /*
1530 * Find or allocate a struct uidinfo for a particular uid.
1531 * Returns with uidinfo struct referenced.
1532 * uifree() should be called on a struct uidinfo when released.
1533 */
1534 struct uidinfo *
uifind(uid_t uid)1535 uifind(uid_t uid)
1536 {
1537 struct uidinfo *new_uip, *uip;
1538 struct ucred *cred;
1539
1540 cred = curthread->td_ucred;
1541 if (cred->cr_uidinfo->ui_uid == uid) {
1542 uip = cred->cr_uidinfo;
1543 uihold(uip);
1544 return (uip);
1545 } else if (cred->cr_ruidinfo->ui_uid == uid) {
1546 uip = cred->cr_ruidinfo;
1547 uihold(uip);
1548 return (uip);
1549 }
1550
1551 rw_rlock(&uihashtbl_lock);
1552 uip = uilookup(uid);
1553 rw_runlock(&uihashtbl_lock);
1554 if (uip != NULL)
1555 return (uip);
1556
1557 new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1558 racct_create(&new_uip->ui_racct);
1559 refcount_init(&new_uip->ui_ref, 1);
1560 new_uip->ui_uid = uid;
1561
1562 rw_wlock(&uihashtbl_lock);
1563 /*
1564 * There's a chance someone created our uidinfo while we
1565 * were in malloc and not holding the lock, so we have to
1566 * make sure we don't insert a duplicate uidinfo.
1567 */
1568 if ((uip = uilookup(uid)) == NULL) {
1569 LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1570 rw_wunlock(&uihashtbl_lock);
1571 uip = new_uip;
1572 } else {
1573 rw_wunlock(&uihashtbl_lock);
1574 racct_destroy(&new_uip->ui_racct);
1575 free(new_uip, M_UIDINFO);
1576 }
1577 return (uip);
1578 }
1579
1580 /*
1581 * Place another refcount on a uidinfo struct.
1582 */
1583 void
uihold(struct uidinfo * uip)1584 uihold(struct uidinfo *uip)
1585 {
1586
1587 refcount_acquire(&uip->ui_ref);
1588 }
1589
1590 /*-
1591 * Since uidinfo structs have a long lifetime, we use an
1592 * opportunistic refcounting scheme to avoid locking the lookup hash
1593 * for each release.
1594 *
1595 * If the refcount hits 0, we need to free the structure,
1596 * which means we need to lock the hash.
1597 * Optimal case:
1598 * After locking the struct and lowering the refcount, if we find
1599 * that we don't need to free, simply unlock and return.
1600 * Suboptimal case:
1601 * If refcount lowering results in need to free, bump the count
1602 * back up, lose the lock and acquire the locks in the proper
1603 * order to try again.
1604 */
1605 void
uifree(struct uidinfo * uip)1606 uifree(struct uidinfo *uip)
1607 {
1608
1609 if (refcount_release_if_not_last(&uip->ui_ref))
1610 return;
1611
1612 rw_wlock(&uihashtbl_lock);
1613 if (refcount_release(&uip->ui_ref) == 0) {
1614 rw_wunlock(&uihashtbl_lock);
1615 return;
1616 }
1617
1618 racct_destroy(&uip->ui_racct);
1619 LIST_REMOVE(uip, ui_hash);
1620 rw_wunlock(&uihashtbl_lock);
1621
1622 if (uip->ui_sbsize != 0)
1623 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1624 uip->ui_uid, uip->ui_sbsize);
1625 if (uip->ui_proccnt != 0)
1626 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1627 uip->ui_uid, uip->ui_proccnt);
1628 if (uip->ui_vmsize != 0)
1629 printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1630 uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1631 if (uip->ui_ptscnt != 0)
1632 printf("freeing uidinfo: uid = %d, ptscnt = %ld\n",
1633 uip->ui_uid, uip->ui_ptscnt);
1634 if (uip->ui_kqcnt != 0)
1635 printf("freeing uidinfo: uid = %d, kqcnt = %ld\n",
1636 uip->ui_uid, uip->ui_kqcnt);
1637 if (uip->ui_umtxcnt != 0)
1638 printf("freeing uidinfo: uid = %d, umtxcnt = %ld\n",
1639 uip->ui_uid, uip->ui_umtxcnt);
1640 if (uip->ui_pipecnt != 0)
1641 printf("freeing uidinfo: uid = %d, pipecnt = %ld\n",
1642 uip->ui_uid, uip->ui_pipecnt);
1643 free(uip, M_UIDINFO);
1644 }
1645
1646 #ifdef RACCT
1647 void
ui_racct_foreach(void (* callback)(struct racct * racct,void * arg2,void * arg3),void (* pre)(void),void (* post)(void),void * arg2,void * arg3)1648 ui_racct_foreach(void (*callback)(struct racct *racct,
1649 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1650 void *arg2, void *arg3)
1651 {
1652 struct uidinfo *uip;
1653 struct uihashhead *uih;
1654
1655 rw_rlock(&uihashtbl_lock);
1656 if (pre != NULL)
1657 (pre)();
1658 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1659 LIST_FOREACH(uip, uih, ui_hash) {
1660 (callback)(uip->ui_racct, arg2, arg3);
1661 }
1662 }
1663 if (post != NULL)
1664 (post)();
1665 rw_runlock(&uihashtbl_lock);
1666 }
1667 #endif
1668
1669 static inline int
chglimit(struct uidinfo * uip,long * limit,int diff,rlim_t max,const char * name)1670 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1671 {
1672 long new;
1673
1674 /* Don't allow them to exceed max, but allow subtraction. */
1675 new = atomic_fetchadd_long(limit, (long)diff) + diff;
1676 if (diff > 0 && max != 0) {
1677 if (new < 0 || new > max) {
1678 atomic_subtract_long(limit, (long)diff);
1679 return (0);
1680 }
1681 } else if (new < 0)
1682 printf("negative %s for uid = %d\n", name, uip->ui_uid);
1683 return (1);
1684 }
1685
1686 /*
1687 * Change the count associated with number of processes
1688 * a given user is using. When 'max' is 0, don't enforce a limit
1689 */
1690 int
chgproccnt(struct uidinfo * uip,int diff,rlim_t max)1691 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1692 {
1693
1694 return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1695 }
1696
1697 /*
1698 * Change the total socket buffer size a user has used.
1699 */
1700 int
chgsbsize(struct uidinfo * uip,u_int * hiwat,u_int to,rlim_t max)1701 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1702 {
1703 int diff, rv;
1704
1705 diff = to - *hiwat;
1706 if (diff > 0 && max == 0) {
1707 rv = 0;
1708 } else {
1709 rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1710 if (rv != 0)
1711 *hiwat = to;
1712 }
1713 return (rv);
1714 }
1715
1716 /*
1717 * Change the count associated with number of pseudo-terminals
1718 * a given user is using. When 'max' is 0, don't enforce a limit
1719 */
1720 int
chgptscnt(struct uidinfo * uip,int diff,rlim_t max)1721 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1722 {
1723
1724 return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1725 }
1726
1727 int
chgkqcnt(struct uidinfo * uip,int diff,rlim_t max)1728 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1729 {
1730
1731 return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1732 }
1733
1734 int
chgumtxcnt(struct uidinfo * uip,int diff,rlim_t max)1735 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1736 {
1737
1738 return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));
1739 }
1740
1741 int
chgpipecnt(struct uidinfo * uip,int diff,rlim_t max)1742 chgpipecnt(struct uidinfo *uip, int diff, rlim_t max)
1743 {
1744
1745 return (chglimit(uip, &uip->ui_pipecnt, diff, max, "pipecnt"));
1746 }
1747
1748 static int
sysctl_kern_proc_rlimit_usage(SYSCTL_HANDLER_ARGS)1749 sysctl_kern_proc_rlimit_usage(SYSCTL_HANDLER_ARGS)
1750 {
1751 rlim_t resval[RLIM_NLIMITS];
1752 struct proc *p;
1753 size_t len;
1754 int error, *name, i;
1755
1756 name = (int *)arg1;
1757 if ((u_int)arg2 != 1 && (u_int)arg2 != 2)
1758 return (EINVAL);
1759 if (req->newptr != NULL)
1760 return (EINVAL);
1761
1762 error = pget((pid_t)name[0], PGET_WANTREAD, &p);
1763 if (error != 0)
1764 return (error);
1765
1766 if ((u_int)arg2 == 1) {
1767 len = sizeof(resval);
1768 memset(resval, 0, sizeof(resval));
1769 for (i = 0; i < RLIM_NLIMITS; i++) {
1770 error = getrlimitusage_one(p, (unsigned)i, 0,
1771 &resval[i]);
1772 if (error == ENXIO) {
1773 resval[i] = -1;
1774 error = 0;
1775 } else if (error != 0) {
1776 break;
1777 }
1778 }
1779 } else {
1780 len = sizeof(resval[0]);
1781 error = getrlimitusage_one(p, (unsigned)name[1], 0,
1782 &resval[0]);
1783 if (error == ENXIO) {
1784 resval[0] = -1;
1785 error = 0;
1786 }
1787 }
1788 if (error == 0)
1789 error = SYSCTL_OUT(req, resval, len);
1790 PRELE(p);
1791 return (error);
1792 }
1793 static SYSCTL_NODE(_kern_proc, KERN_PROC_RLIMIT_USAGE, rlimit_usage,
1794 CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE,
1795 sysctl_kern_proc_rlimit_usage,
1796 "Process limited resources usage info");
1797