1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2010 The FreeBSD Foundation
5 * All rights reserved.
6 *
7 * This software was developed by Edward Tomasz Napierala under sponsorship
8 * from the FreeBSD Foundation.
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 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * $FreeBSD$
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD$");
36
37#include "opt_sched.h"
38
39#include <sys/param.h>
40#include <sys/buf.h>
41#include <sys/systm.h>
42#include <sys/eventhandler.h>
43#include <sys/jail.h>
44#include <sys/kernel.h>
45#include <sys/kthread.h>
46#include <sys/lock.h>
47#include <sys/loginclass.h>
48#include <sys/malloc.h>
49#include <sys/mutex.h>
50#include <sys/proc.h>
51#include <sys/racct.h>
52#include <sys/resourcevar.h>
53#include <sys/sbuf.h>
54#include <sys/sched.h>
55#include <sys/sdt.h>
56#include <sys/smp.h>
57#include <sys/sx.h>
58#include <sys/sysctl.h>
59#include <sys/sysent.h>
60#include <sys/sysproto.h>
61#include <sys/umtx.h>
62#include <machine/smp.h>
63
64#ifdef RCTL
65#include <sys/rctl.h>
66#endif
67
68#ifdef RACCT
69
70FEATURE(racct, "Resource Accounting");
71
72/*
73 * Do not block processes that have their %cpu usage <= pcpu_threshold.
74 */
75static int pcpu_threshold = 1;
76#ifdef RACCT_DEFAULT_TO_DISABLED
77int racct_enable = 0;
78#else
79int racct_enable = 1;
80#endif
81
82SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW, 0, "Resource Accounting");
83SYSCTL_UINT(_kern_racct, OID_AUTO, enable, CTLFLAG_RDTUN, &racct_enable,
84    0, "Enable RACCT/RCTL");
85SYSCTL_UINT(_kern_racct, OID_AUTO, pcpu_threshold, CTLFLAG_RW, &pcpu_threshold,
86    0, "Processes with higher %cpu usage than this value can be throttled.");
87
88/*
89 * How many seconds it takes to use the scheduler %cpu calculations.  When a
90 * process starts, we compute its %cpu usage by dividing its runtime by the
91 * process wall clock time.  After RACCT_PCPU_SECS pass, we use the value
92 * provided by the scheduler.
93 */
94#define RACCT_PCPU_SECS		3
95
96struct mtx racct_lock;
97MTX_SYSINIT(racct_lock, &racct_lock, "racct lock", MTX_DEF);
98
99static uma_zone_t racct_zone;
100
101static void racct_sub_racct(struct racct *dest, const struct racct *src);
102static void racct_sub_cred_locked(struct ucred *cred, int resource,
103		uint64_t amount);
104static void racct_add_cred_locked(struct ucred *cred, int resource,
105		uint64_t amount);
106
107SDT_PROVIDER_DEFINE(racct);
108SDT_PROBE_DEFINE3(racct, , rusage, add,
109    "struct proc *", "int", "uint64_t");
110SDT_PROBE_DEFINE3(racct, , rusage, add__failure,
111    "struct proc *", "int", "uint64_t");
112SDT_PROBE_DEFINE3(racct, , rusage, add__buf,
113    "struct proc *", "const struct buf *", "int");
114SDT_PROBE_DEFINE3(racct, , rusage, add__cred,
115    "struct ucred *", "int", "uint64_t");
116SDT_PROBE_DEFINE3(racct, , rusage, add__force,
117    "struct proc *", "int", "uint64_t");
118SDT_PROBE_DEFINE3(racct, , rusage, set,
119    "struct proc *", "int", "uint64_t");
120SDT_PROBE_DEFINE3(racct, , rusage, set__failure,
121    "struct proc *", "int", "uint64_t");
122SDT_PROBE_DEFINE3(racct, , rusage, set__force,
123    "struct proc *", "int", "uint64_t");
124SDT_PROBE_DEFINE3(racct, , rusage, sub,
125    "struct proc *", "int", "uint64_t");
126SDT_PROBE_DEFINE3(racct, , rusage, sub__cred,
127    "struct ucred *", "int", "uint64_t");
128SDT_PROBE_DEFINE1(racct, , racct, create,
129    "struct racct *");
130SDT_PROBE_DEFINE1(racct, , racct, destroy,
131    "struct racct *");
132SDT_PROBE_DEFINE2(racct, , racct, join,
133    "struct racct *", "struct racct *");
134SDT_PROBE_DEFINE2(racct, , racct, join__failure,
135    "struct racct *", "struct racct *");
136SDT_PROBE_DEFINE2(racct, , racct, leave,
137    "struct racct *", "struct racct *");
138
139int racct_types[] = {
140	[RACCT_CPU] =
141		RACCT_IN_MILLIONS,
142	[RACCT_DATA] =
143		RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
144	[RACCT_STACK] =
145		RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
146	[RACCT_CORE] =
147		RACCT_DENIABLE,
148	[RACCT_RSS] =
149		RACCT_RECLAIMABLE,
150	[RACCT_MEMLOCK] =
151		RACCT_RECLAIMABLE | RACCT_DENIABLE,
152	[RACCT_NPROC] =
153		RACCT_RECLAIMABLE | RACCT_DENIABLE,
154	[RACCT_NOFILE] =
155		RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
156	[RACCT_VMEM] =
157		RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
158	[RACCT_NPTS] =
159		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
160	[RACCT_SWAP] =
161		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
162	[RACCT_NTHR] =
163		RACCT_RECLAIMABLE | RACCT_DENIABLE,
164	[RACCT_MSGQQUEUED] =
165		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
166	[RACCT_MSGQSIZE] =
167		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
168	[RACCT_NMSGQ] =
169		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
170	[RACCT_NSEM] =
171		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
172	[RACCT_NSEMOP] =
173		RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
174	[RACCT_NSHM] =
175		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
176	[RACCT_SHMSIZE] =
177		RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
178	[RACCT_WALLCLOCK] =
179		RACCT_IN_MILLIONS,
180	[RACCT_PCTCPU] =
181		RACCT_DECAYING | RACCT_DENIABLE | RACCT_IN_MILLIONS,
182	[RACCT_READBPS] =
183		RACCT_DECAYING,
184	[RACCT_WRITEBPS] =
185		RACCT_DECAYING,
186	[RACCT_READIOPS] =
187		RACCT_DECAYING,
188	[RACCT_WRITEIOPS] =
189		RACCT_DECAYING };
190
191static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE;
192
193#ifdef SCHED_4BSD
194/*
195 * Contains intermediate values for %cpu calculations to avoid using floating
196 * point in the kernel.
197 * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20)
198 * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to
199 * zero so the calculations are more straightforward.
200 */
201fixpt_t ccpu_exp[] = {
202	[0] = FSCALE * 1,
203	[1] = FSCALE * 0.95122942450071400909,
204	[2] = FSCALE * 0.90483741803595957316,
205	[3] = FSCALE * 0.86070797642505780722,
206	[4] = FSCALE * 0.81873075307798185866,
207	[5] = FSCALE * 0.77880078307140486824,
208	[6] = FSCALE * 0.74081822068171786606,
209	[7] = FSCALE * 0.70468808971871343435,
210	[8] = FSCALE * 0.67032004603563930074,
211	[9] = FSCALE * 0.63762815162177329314,
212	[10] = FSCALE * 0.60653065971263342360,
213	[11] = FSCALE * 0.57694981038048669531,
214	[12] = FSCALE * 0.54881163609402643262,
215	[13] = FSCALE * 0.52204577676101604789,
216	[14] = FSCALE * 0.49658530379140951470,
217	[15] = FSCALE * 0.47236655274101470713,
218	[16] = FSCALE * 0.44932896411722159143,
219	[17] = FSCALE * 0.42741493194872666992,
220	[18] = FSCALE * 0.40656965974059911188,
221	[19] = FSCALE * 0.38674102345450120691,
222	[20] = FSCALE * 0.36787944117144232159,
223	[21] = FSCALE * 0.34993774911115535467,
224	[22] = FSCALE * 0.33287108369807955328,
225	[23] = FSCALE * 0.31663676937905321821,
226	[24] = FSCALE * 0.30119421191220209664,
227	[25] = FSCALE * 0.28650479686019010032,
228	[26] = FSCALE * 0.27253179303401260312,
229	[27] = FSCALE * 0.25924026064589150757,
230	[28] = FSCALE * 0.24659696394160647693,
231	[29] = FSCALE * 0.23457028809379765313,
232	[30] = FSCALE * 0.22313016014842982893,
233	[31] = FSCALE * 0.21224797382674305771,
234	[32] = FSCALE * 0.20189651799465540848,
235	[33] = FSCALE * 0.19204990862075411423,
236	[34] = FSCALE * 0.18268352405273465022,
237	[35] = FSCALE * 0.17377394345044512668,
238	[36] = FSCALE * 0.16529888822158653829,
239	[37] = FSCALE * 0.15723716631362761621,
240	[38] = FSCALE * 0.14956861922263505264,
241	[39] = FSCALE * 0.14227407158651357185,
242	[40] = FSCALE * 0.13533528323661269189,
243	[41] = FSCALE * 0.12873490358780421886,
244	[42] = FSCALE * 0.12245642825298191021,
245	[43] = FSCALE * 0.11648415777349695786,
246	[44] = FSCALE * 0.11080315836233388333,
247	[45] = FSCALE * 0.10539922456186433678,
248	[46] = FSCALE * 0.10025884372280373372,
249	[47] = FSCALE * 0.09536916221554961888,
250	[48] = FSCALE * 0.09071795328941250337,
251	[49] = FSCALE * 0.08629358649937051097,
252	[50] = FSCALE * 0.08208499862389879516,
253	[51] = FSCALE * 0.07808166600115315231,
254	[52] = FSCALE * 0.07427357821433388042,
255	[53] = FSCALE * 0.07065121306042958674,
256	[54] = FSCALE * 0.06720551273974976512,
257	[55] = FSCALE * 0.06392786120670757270,
258	[56] = FSCALE * 0.06081006262521796499,
259	[57] = FSCALE * 0.05784432087483846296,
260	[58] = FSCALE * 0.05502322005640722902,
261	[59] = FSCALE * 0.05233970594843239308,
262	[60] = FSCALE * 0.04978706836786394297,
263	[61] = FSCALE * 0.04735892439114092119,
264	[62] = FSCALE * 0.04504920239355780606,
265	[63] = FSCALE * 0.04285212686704017991,
266	[64] = FSCALE * 0.04076220397836621516,
267	[65] = FSCALE * 0.03877420783172200988,
268	[66] = FSCALE * 0.03688316740124000544,
269	[67] = FSCALE * 0.03508435410084502588,
270	[68] = FSCALE * 0.03337326996032607948,
271	[69] = FSCALE * 0.03174563637806794323,
272	[70] = FSCALE * 0.03019738342231850073,
273	[71] = FSCALE * 0.02872463965423942912,
274	[72] = FSCALE * 0.02732372244729256080,
275	[73] = FSCALE * 0.02599112877875534358,
276	[74] = FSCALE * 0.02472352647033939120,
277	[75] = FSCALE * 0.02351774585600910823,
278	[76] = FSCALE * 0.02237077185616559577,
279	[77] = FSCALE * 0.02127973643837716938,
280	[78] = FSCALE * 0.02024191144580438847,
281	[79] = FSCALE * 0.01925470177538692429,
282	[80] = FSCALE * 0.01831563888873418029,
283	[81] = FSCALE * 0.01742237463949351138,
284	[82] = FSCALE * 0.01657267540176124754,
285	[83] = FSCALE * 0.01576441648485449082,
286	[84] = FSCALE * 0.01499557682047770621,
287	[85] = FSCALE * 0.01426423390899925527,
288	[86] = FSCALE * 0.01356855901220093175,
289	[87] = FSCALE * 0.01290681258047986886,
290	[88] = FSCALE * 0.01227733990306844117,
291	[89] = FSCALE * 0.01167856697039544521,
292	[90] = FSCALE * 0.01110899653824230649,
293	[91] = FSCALE * 0.01056720438385265337,
294	[92] = FSCALE * 0.01005183574463358164,
295	[93] = FSCALE * 0.00956160193054350793,
296	[94] = FSCALE * 0.00909527710169581709,
297	[95] = FSCALE * 0.00865169520312063417,
298	[96] = FSCALE * 0.00822974704902002884,
299	[97] = FSCALE * 0.00782837754922577143,
300	[98] = FSCALE * 0.00744658307092434051,
301	[99] = FSCALE * 0.00708340892905212004,
302	[100] = FSCALE * 0.00673794699908546709,
303	[101] = FSCALE * 0.00640933344625638184,
304	[102] = FSCALE * 0.00609674656551563610,
305	[103] = FSCALE * 0.00579940472684214321,
306	[104] = FSCALE * 0.00551656442076077241,
307	[105] = FSCALE * 0.00524751839918138427,
308	[106] = FSCALE * 0.00499159390691021621,
309	[107] = FSCALE * 0.00474815099941147558,
310	[108] = FSCALE * 0.00451658094261266798,
311	[109] = FSCALE * 0.00429630469075234057,
312	[110] = FSCALE * 0.00408677143846406699,
313};
314#endif
315
316#define	CCPU_EXP_MAX	110
317
318/*
319 * This function is analogical to the getpcpu() function in the ps(1) command.
320 * They should both calculate in the same way so that the racct %cpu
321 * calculations are consistent with the values showed by the ps(1) tool.
322 * The calculations are more complex in the 4BSD scheduler because of the value
323 * of the ccpu variable.  In ULE it is defined to be zero which saves us some
324 * work.
325 */
326static uint64_t
327racct_getpcpu(struct proc *p, u_int pcpu)
328{
329	u_int swtime;
330#ifdef SCHED_4BSD
331	fixpt_t pctcpu, pctcpu_next;
332#endif
333#ifdef SMP
334	struct pcpu *pc;
335	int found;
336#endif
337	fixpt_t p_pctcpu;
338	struct thread *td;
339
340	ASSERT_RACCT_ENABLED();
341
342	/*
343	 * If the process is swapped out, we count its %cpu usage as zero.
344	 * This behaviour is consistent with the userland ps(1) tool.
345	 */
346	if ((p->p_flag & P_INMEM) == 0)
347		return (0);
348	swtime = (ticks - p->p_swtick) / hz;
349
350	/*
351	 * For short-lived processes, the sched_pctcpu() returns small
352	 * values even for cpu intensive processes.  Therefore we use
353	 * our own estimate in this case.
354	 */
355	if (swtime < RACCT_PCPU_SECS)
356		return (pcpu);
357
358	p_pctcpu = 0;
359	FOREACH_THREAD_IN_PROC(p, td) {
360		if (td == PCPU_GET(idlethread))
361			continue;
362#ifdef SMP
363		found = 0;
364		STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
365			if (td == pc->pc_idlethread) {
366				found = 1;
367				break;
368			}
369		}
370		if (found)
371			continue;
372#endif
373		thread_lock(td);
374#ifdef SCHED_4BSD
375		pctcpu = sched_pctcpu(td);
376		/* Count also the yet unfinished second. */
377		pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT;
378		pctcpu_next += sched_pctcpu_delta(td);
379		p_pctcpu += max(pctcpu, pctcpu_next);
380#else
381		/*
382		 * In ULE the %cpu statistics are updated on every
383		 * sched_pctcpu() call.  So special calculations to
384		 * account for the latest (unfinished) second are
385		 * not needed.
386		 */
387		p_pctcpu += sched_pctcpu(td);
388#endif
389		thread_unlock(td);
390	}
391
392#ifdef SCHED_4BSD
393	if (swtime <= CCPU_EXP_MAX)
394		return ((100 * (uint64_t)p_pctcpu * 1000000) /
395		    (FSCALE - ccpu_exp[swtime]));
396#endif
397
398	return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE);
399}
400
401static void
402racct_add_racct(struct racct *dest, const struct racct *src)
403{
404	int i;
405
406	ASSERT_RACCT_ENABLED();
407	RACCT_LOCK_ASSERT();
408
409	/*
410	 * Update resource usage in dest.
411	 */
412	for (i = 0; i <= RACCT_MAX; i++) {
413		KASSERT(dest->r_resources[i] >= 0,
414		    ("%s: resource %d propagation meltdown: dest < 0",
415		    __func__, i));
416		KASSERT(src->r_resources[i] >= 0,
417		    ("%s: resource %d propagation meltdown: src < 0",
418		    __func__, i));
419		dest->r_resources[i] += src->r_resources[i];
420	}
421}
422
423static void
424racct_sub_racct(struct racct *dest, const struct racct *src)
425{
426	int i;
427
428	ASSERT_RACCT_ENABLED();
429	RACCT_LOCK_ASSERT();
430
431	/*
432	 * Update resource usage in dest.
433	 */
434	for (i = 0; i <= RACCT_MAX; i++) {
435		if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) {
436			KASSERT(dest->r_resources[i] >= 0,
437			    ("%s: resource %d propagation meltdown: dest < 0",
438			    __func__, i));
439			KASSERT(src->r_resources[i] >= 0,
440			    ("%s: resource %d propagation meltdown: src < 0",
441			    __func__, i));
442			KASSERT(src->r_resources[i] <= dest->r_resources[i],
443			    ("%s: resource %d propagation meltdown: src > dest",
444			    __func__, i));
445		}
446		if (RACCT_CAN_DROP(i)) {
447			dest->r_resources[i] -= src->r_resources[i];
448			if (dest->r_resources[i] < 0)
449				dest->r_resources[i] = 0;
450		}
451	}
452}
453
454void
455racct_create(struct racct **racctp)
456{
457
458	if (!racct_enable)
459		return;
460
461	SDT_PROBE1(racct, , racct, create, racctp);
462
463	KASSERT(*racctp == NULL, ("racct already allocated"));
464
465	*racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO);
466}
467
468static void
469racct_destroy_locked(struct racct **racctp)
470{
471	struct racct *racct;
472	int i;
473
474	ASSERT_RACCT_ENABLED();
475
476	SDT_PROBE1(racct, , racct, destroy, racctp);
477
478	RACCT_LOCK_ASSERT();
479	KASSERT(racctp != NULL, ("NULL racctp"));
480	KASSERT(*racctp != NULL, ("NULL racct"));
481
482	racct = *racctp;
483
484	for (i = 0; i <= RACCT_MAX; i++) {
485		if (RACCT_IS_SLOPPY(i))
486			continue;
487		if (!RACCT_IS_RECLAIMABLE(i))
488			continue;
489		KASSERT(racct->r_resources[i] == 0,
490		    ("destroying non-empty racct: "
491		    "%ju allocated for resource %d\n",
492		    racct->r_resources[i], i));
493	}
494	uma_zfree(racct_zone, racct);
495	*racctp = NULL;
496}
497
498void
499racct_destroy(struct racct **racct)
500{
501
502	if (!racct_enable)
503		return;
504
505	RACCT_LOCK();
506	racct_destroy_locked(racct);
507	RACCT_UNLOCK();
508}
509
510/*
511 * Increase consumption of 'resource' by 'amount' for 'racct',
512 * but not its parents.  Differently from other cases, 'amount' here
513 * may be less than zero.
514 */
515static void
516racct_adjust_resource(struct racct *racct, int resource,
517    int64_t amount)
518{
519
520	ASSERT_RACCT_ENABLED();
521	RACCT_LOCK_ASSERT();
522	KASSERT(racct != NULL, ("NULL racct"));
523
524	racct->r_resources[resource] += amount;
525	if (racct->r_resources[resource] < 0) {
526		KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource),
527		    ("%s: resource %d usage < 0", __func__, resource));
528		racct->r_resources[resource] = 0;
529	}
530
531	/*
532	 * There are some cases where the racct %cpu resource would grow
533	 * beyond 100% per core.  For example in racct_proc_exit() we add
534	 * the process %cpu usage to the ucred racct containers.  If too
535	 * many processes terminated in a short time span, the ucred %cpu
536	 * resource could grow too much.  Also, the 4BSD scheduler sometimes
537	 * returns for a thread more than 100% cpu usage. So we set a sane
538	 * boundary here to 100% * the maxumum number of CPUs.
539	 */
540	if ((resource == RACCT_PCTCPU) &&
541	    (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000 * (int64_t)MAXCPU))
542		racct->r_resources[RACCT_PCTCPU] = 100 * 1000000 * (int64_t)MAXCPU;
543}
544
545static int
546racct_add_locked(struct proc *p, int resource, uint64_t amount, int force)
547{
548#ifdef RCTL
549	int error;
550#endif
551
552	ASSERT_RACCT_ENABLED();
553
554	/*
555	 * We need proc lock to dereference p->p_ucred.
556	 */
557	PROC_LOCK_ASSERT(p, MA_OWNED);
558
559#ifdef RCTL
560	error = rctl_enforce(p, resource, amount);
561	if (error && !force && RACCT_IS_DENIABLE(resource)) {
562		SDT_PROBE3(racct, , rusage, add__failure, p, resource, amount);
563		return (error);
564	}
565#endif
566	racct_adjust_resource(p->p_racct, resource, amount);
567	racct_add_cred_locked(p->p_ucred, resource, amount);
568
569	return (0);
570}
571
572/*
573 * Increase allocation of 'resource' by 'amount' for process 'p'.
574 * Return 0 if it's below limits, or errno, if it's not.
575 */
576int
577racct_add(struct proc *p, int resource, uint64_t amount)
578{
579	int error;
580
581	if (!racct_enable)
582		return (0);
583
584	SDT_PROBE3(racct, , rusage, add, p, resource, amount);
585
586	RACCT_LOCK();
587	error = racct_add_locked(p, resource, amount, 0);
588	RACCT_UNLOCK();
589	return (error);
590}
591
592/*
593 * Increase allocation of 'resource' by 'amount' for process 'p'.
594 * Doesn't check for limits and never fails.
595 */
596void
597racct_add_force(struct proc *p, int resource, uint64_t amount)
598{
599
600	if (!racct_enable)
601		return;
602
603	SDT_PROBE3(racct, , rusage, add__force, p, resource, amount);
604
605	RACCT_LOCK();
606	racct_add_locked(p, resource, amount, 1);
607	RACCT_UNLOCK();
608}
609
610static void
611racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount)
612{
613	struct prison *pr;
614
615	ASSERT_RACCT_ENABLED();
616
617	racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, amount);
618	for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
619		racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
620		    amount);
621	racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, amount);
622}
623
624/*
625 * Increase allocation of 'resource' by 'amount' for credential 'cred'.
626 * Doesn't check for limits and never fails.
627 */
628void
629racct_add_cred(struct ucred *cred, int resource, uint64_t amount)
630{
631
632	if (!racct_enable)
633		return;
634
635	SDT_PROBE3(racct, , rusage, add__cred, cred, resource, amount);
636
637	RACCT_LOCK();
638	racct_add_cred_locked(cred, resource, amount);
639	RACCT_UNLOCK();
640}
641
642/*
643 * Account for disk IO resource consumption.  Checks for limits,
644 * but never fails, due to disk limits being undeniable.
645 */
646void
647racct_add_buf(struct proc *p, const struct buf *bp, int is_write)
648{
649
650	ASSERT_RACCT_ENABLED();
651	PROC_LOCK_ASSERT(p, MA_OWNED);
652
653	SDT_PROBE3(racct, , rusage, add__buf, p, bp, is_write);
654
655	RACCT_LOCK();
656	if (is_write) {
657		racct_add_locked(curproc, RACCT_WRITEBPS, bp->b_bcount, 1);
658		racct_add_locked(curproc, RACCT_WRITEIOPS, 1, 1);
659	} else {
660		racct_add_locked(curproc, RACCT_READBPS, bp->b_bcount, 1);
661		racct_add_locked(curproc, RACCT_READIOPS, 1, 1);
662	}
663	RACCT_UNLOCK();
664}
665
666static int
667racct_set_locked(struct proc *p, int resource, uint64_t amount, int force)
668{
669	int64_t old_amount, decayed_amount, diff_proc, diff_cred;
670#ifdef RCTL
671	int error;
672#endif
673
674	ASSERT_RACCT_ENABLED();
675
676	/*
677	 * We need proc lock to dereference p->p_ucred.
678	 */
679	PROC_LOCK_ASSERT(p, MA_OWNED);
680
681	old_amount = p->p_racct->r_resources[resource];
682	/*
683	 * The diffs may be negative.
684	 */
685	diff_proc = amount - old_amount;
686	if (resource == RACCT_PCTCPU) {
687		/*
688		 * Resources in per-credential racct containers may decay.
689		 * If this is the case, we need to calculate the difference
690		 * between the new amount and the proportional value of the
691		 * old amount that has decayed in the ucred racct containers.
692		 */
693		decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE;
694		diff_cred = amount - decayed_amount;
695	} else
696		diff_cred = diff_proc;
697#ifdef notyet
698	KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource),
699	    ("%s: usage of non-droppable resource %d dropping", __func__,
700	     resource));
701#endif
702#ifdef RCTL
703	if (diff_proc > 0) {
704		error = rctl_enforce(p, resource, diff_proc);
705		if (error && !force && RACCT_IS_DENIABLE(resource)) {
706			SDT_PROBE3(racct, , rusage, set__failure, p, resource,
707			    amount);
708			return (error);
709		}
710	}
711#endif
712	racct_adjust_resource(p->p_racct, resource, diff_proc);
713	if (diff_cred > 0)
714		racct_add_cred_locked(p->p_ucred, resource, diff_cred);
715	else if (diff_cred < 0)
716		racct_sub_cred_locked(p->p_ucred, resource, -diff_cred);
717
718	return (0);
719}
720
721/*
722 * Set allocation of 'resource' to 'amount' for process 'p'.
723 * Return 0 if it's below limits, or errno, if it's not.
724 *
725 * Note that decreasing the allocation always returns 0,
726 * even if it's above the limit.
727 */
728int
729racct_set(struct proc *p, int resource, uint64_t amount)
730{
731	int error;
732
733	if (!racct_enable)
734		return (0);
735
736	SDT_PROBE3(racct, , rusage, set__force, p, resource, amount);
737
738	RACCT_LOCK();
739	error = racct_set_locked(p, resource, amount, 0);
740	RACCT_UNLOCK();
741	return (error);
742}
743
744void
745racct_set_force(struct proc *p, int resource, uint64_t amount)
746{
747
748	if (!racct_enable)
749		return;
750
751	SDT_PROBE3(racct, , rusage, set, p, resource, amount);
752
753	RACCT_LOCK();
754	racct_set_locked(p, resource, amount, 1);
755	RACCT_UNLOCK();
756}
757
758/*
759 * Returns amount of 'resource' the process 'p' can keep allocated.
760 * Allocating more than that would be denied, unless the resource
761 * is marked undeniable.  Amount of already allocated resource does
762 * not matter.
763 */
764uint64_t
765racct_get_limit(struct proc *p, int resource)
766{
767#ifdef RCTL
768	uint64_t available;
769
770	if (!racct_enable)
771		return (UINT64_MAX);
772
773	RACCT_LOCK();
774	available = rctl_get_limit(p, resource);
775	RACCT_UNLOCK();
776
777	return (available);
778#else
779
780	return (UINT64_MAX);
781#endif
782}
783
784/*
785 * Returns amount of 'resource' the process 'p' can keep allocated.
786 * Allocating more than that would be denied, unless the resource
787 * is marked undeniable.  Amount of already allocated resource does
788 * matter.
789 */
790uint64_t
791racct_get_available(struct proc *p, int resource)
792{
793#ifdef RCTL
794	uint64_t available;
795
796	if (!racct_enable)
797		return (UINT64_MAX);
798
799	RACCT_LOCK();
800	available = rctl_get_available(p, resource);
801	RACCT_UNLOCK();
802
803	return (available);
804#else
805
806	return (UINT64_MAX);
807#endif
808}
809
810/*
811 * Returns amount of the %cpu resource that process 'p' can add to its %cpu
812 * utilization.  Adding more than that would lead to the process being
813 * throttled.
814 */
815static int64_t
816racct_pcpu_available(struct proc *p)
817{
818#ifdef RCTL
819	uint64_t available;
820
821	ASSERT_RACCT_ENABLED();
822
823	RACCT_LOCK();
824	available = rctl_pcpu_available(p);
825	RACCT_UNLOCK();
826
827	return (available);
828#else
829
830	return (INT64_MAX);
831#endif
832}
833
834/*
835 * Decrease allocation of 'resource' by 'amount' for process 'p'.
836 */
837void
838racct_sub(struct proc *p, int resource, uint64_t amount)
839{
840
841	if (!racct_enable)
842		return;
843
844	SDT_PROBE3(racct, , rusage, sub, p, resource, amount);
845
846	/*
847	 * We need proc lock to dereference p->p_ucred.
848	 */
849	PROC_LOCK_ASSERT(p, MA_OWNED);
850	KASSERT(RACCT_CAN_DROP(resource),
851	    ("%s: called for non-droppable resource %d", __func__, resource));
852
853	RACCT_LOCK();
854	KASSERT(amount <= p->p_racct->r_resources[resource],
855	    ("%s: freeing %ju of resource %d, which is more "
856	     "than allocated %jd for %s (pid %d)", __func__, amount, resource,
857	    (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid));
858
859	racct_adjust_resource(p->p_racct, resource, -amount);
860	racct_sub_cred_locked(p->p_ucred, resource, amount);
861	RACCT_UNLOCK();
862}
863
864static void
865racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount)
866{
867	struct prison *pr;
868
869	ASSERT_RACCT_ENABLED();
870
871	racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, -amount);
872	for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
873		racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
874		    -amount);
875	racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, -amount);
876}
877
878/*
879 * Decrease allocation of 'resource' by 'amount' for credential 'cred'.
880 */
881void
882racct_sub_cred(struct ucred *cred, int resource, uint64_t amount)
883{
884
885	if (!racct_enable)
886		return;
887
888	SDT_PROBE3(racct, , rusage, sub__cred, cred, resource, amount);
889
890#ifdef notyet
891	KASSERT(RACCT_CAN_DROP(resource),
892	    ("%s: called for resource %d which can not drop", __func__,
893	     resource));
894#endif
895
896	RACCT_LOCK();
897	racct_sub_cred_locked(cred, resource, amount);
898	RACCT_UNLOCK();
899}
900
901/*
902 * Inherit resource usage information from the parent process.
903 */
904int
905racct_proc_fork(struct proc *parent, struct proc *child)
906{
907	int i, error = 0;
908
909	if (!racct_enable)
910		return (0);
911
912	/*
913	 * Create racct for the child process.
914	 */
915	racct_create(&child->p_racct);
916
917	PROC_LOCK(parent);
918	PROC_LOCK(child);
919	RACCT_LOCK();
920
921#ifdef RCTL
922	error = rctl_proc_fork(parent, child);
923	if (error != 0)
924		goto out;
925#endif
926
927	/* Init process cpu time. */
928	child->p_prev_runtime = 0;
929	child->p_throttled = 0;
930
931	/*
932	 * Inherit resource usage.
933	 */
934	for (i = 0; i <= RACCT_MAX; i++) {
935		if (parent->p_racct->r_resources[i] == 0 ||
936		    !RACCT_IS_INHERITABLE(i))
937			continue;
938
939		error = racct_set_locked(child, i,
940		    parent->p_racct->r_resources[i], 0);
941		if (error != 0)
942			goto out;
943	}
944
945	error = racct_add_locked(child, RACCT_NPROC, 1, 0);
946	error += racct_add_locked(child, RACCT_NTHR, 1, 0);
947
948out:
949	RACCT_UNLOCK();
950	PROC_UNLOCK(child);
951	PROC_UNLOCK(parent);
952
953	if (error != 0)
954		racct_proc_exit(child);
955
956	return (error);
957}
958
959/*
960 * Called at the end of fork1(), to handle rules that require the process
961 * to be fully initialized.
962 */
963void
964racct_proc_fork_done(struct proc *child)
965{
966
967	if (!racct_enable)
968		return;
969
970	PROC_LOCK_ASSERT(child, MA_OWNED);
971
972#ifdef RCTL
973	RACCT_LOCK();
974	rctl_enforce(child, RACCT_NPROC, 0);
975	rctl_enforce(child, RACCT_NTHR, 0);
976	RACCT_UNLOCK();
977#endif
978}
979
980void
981racct_proc_exit(struct proc *p)
982{
983	struct timeval wallclock;
984	uint64_t pct_estimate, pct, runtime;
985	int i;
986
987	if (!racct_enable)
988		return;
989
990	PROC_LOCK(p);
991	/*
992	 * We don't need to calculate rux, proc_reap() has already done this.
993	 */
994	runtime = cputick2usec(p->p_rux.rux_runtime);
995#ifdef notyet
996	KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime"));
997#else
998	if (runtime < p->p_prev_runtime)
999		runtime = p->p_prev_runtime;
1000#endif
1001	microuptime(&wallclock);
1002	timevalsub(&wallclock, &p->p_stats->p_start);
1003	if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
1004		pct_estimate = (1000000 * runtime * 100) /
1005		    ((uint64_t)wallclock.tv_sec * 1000000 +
1006		    wallclock.tv_usec);
1007	} else
1008		pct_estimate = 0;
1009	pct = racct_getpcpu(p, pct_estimate);
1010
1011	RACCT_LOCK();
1012	racct_set_locked(p, RACCT_CPU, runtime, 0);
1013	racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct);
1014
1015	KASSERT(p->p_racct->r_resources[RACCT_RSS] == 0,
1016	    ("process reaped with %ju allocated for RSS\n",
1017	    p->p_racct->r_resources[RACCT_RSS]));
1018	for (i = 0; i <= RACCT_MAX; i++) {
1019		if (p->p_racct->r_resources[i] == 0)
1020			continue;
1021		if (!RACCT_IS_RECLAIMABLE(i))
1022			continue;
1023		racct_set_locked(p, i, 0, 0);
1024	}
1025
1026#ifdef RCTL
1027	rctl_racct_release(p->p_racct);
1028#endif
1029	racct_destroy_locked(&p->p_racct);
1030	RACCT_UNLOCK();
1031	PROC_UNLOCK(p);
1032}
1033
1034/*
1035 * Called after credentials change, to move resource utilisation
1036 * between raccts.
1037 */
1038void
1039racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred,
1040    struct ucred *newcred)
1041{
1042	struct uidinfo *olduip, *newuip;
1043	struct loginclass *oldlc, *newlc;
1044	struct prison *oldpr, *newpr, *pr;
1045
1046	if (!racct_enable)
1047		return;
1048
1049	PROC_LOCK_ASSERT(p, MA_OWNED);
1050
1051	newuip = newcred->cr_ruidinfo;
1052	olduip = oldcred->cr_ruidinfo;
1053	newlc = newcred->cr_loginclass;
1054	oldlc = oldcred->cr_loginclass;
1055	newpr = newcred->cr_prison;
1056	oldpr = oldcred->cr_prison;
1057
1058	RACCT_LOCK();
1059	if (newuip != olduip) {
1060		racct_sub_racct(olduip->ui_racct, p->p_racct);
1061		racct_add_racct(newuip->ui_racct, p->p_racct);
1062	}
1063	if (newlc != oldlc) {
1064		racct_sub_racct(oldlc->lc_racct, p->p_racct);
1065		racct_add_racct(newlc->lc_racct, p->p_racct);
1066	}
1067	if (newpr != oldpr) {
1068		for (pr = oldpr; pr != NULL; pr = pr->pr_parent)
1069			racct_sub_racct(pr->pr_prison_racct->prr_racct,
1070			    p->p_racct);
1071		for (pr = newpr; pr != NULL; pr = pr->pr_parent)
1072			racct_add_racct(pr->pr_prison_racct->prr_racct,
1073			    p->p_racct);
1074	}
1075	RACCT_UNLOCK();
1076}
1077
1078void
1079racct_move(struct racct *dest, struct racct *src)
1080{
1081
1082	ASSERT_RACCT_ENABLED();
1083
1084	RACCT_LOCK();
1085	racct_add_racct(dest, src);
1086	racct_sub_racct(src, src);
1087	RACCT_UNLOCK();
1088}
1089
1090/*
1091 * Make the process sleep in userret() for 'timeout' ticks.  Setting
1092 * timeout to -1 makes it sleep until woken up by racct_proc_wakeup().
1093 */
1094void
1095racct_proc_throttle(struct proc *p, int timeout)
1096{
1097	struct thread *td;
1098#ifdef SMP
1099	int cpuid;
1100#endif
1101
1102	KASSERT(timeout != 0, ("timeout %d", timeout));
1103	ASSERT_RACCT_ENABLED();
1104	PROC_LOCK_ASSERT(p, MA_OWNED);
1105
1106	/*
1107	 * Do not block kernel processes.  Also do not block processes with
1108	 * low %cpu utilization to improve interactivity.
1109	 */
1110	if ((p->p_flag & (P_SYSTEM | P_KPROC)) != 0)
1111		return;
1112
1113	if (p->p_throttled < 0 || (timeout > 0 && p->p_throttled > timeout))
1114		return;
1115
1116	p->p_throttled = timeout;
1117
1118	FOREACH_THREAD_IN_PROC(p, td) {
1119		thread_lock(td);
1120		switch (td->td_state) {
1121		case TDS_RUNQ:
1122			/*
1123			 * If the thread is on the scheduler run-queue, we can
1124			 * not just remove it from there.  So we set the flag
1125			 * TDF_NEEDRESCHED for the thread, so that once it is
1126			 * running, it is taken off the cpu as soon as possible.
1127			 */
1128			td->td_flags |= TDF_NEEDRESCHED;
1129			break;
1130		case TDS_RUNNING:
1131			/*
1132			 * If the thread is running, we request a context
1133			 * switch for it by setting the TDF_NEEDRESCHED flag.
1134			 */
1135			td->td_flags |= TDF_NEEDRESCHED;
1136#ifdef SMP
1137			cpuid = td->td_oncpu;
1138			if ((cpuid != NOCPU) && (td != curthread))
1139				ipi_cpu(cpuid, IPI_AST);
1140#endif
1141			break;
1142		default:
1143			break;
1144		}
1145		thread_unlock(td);
1146	}
1147}
1148
1149static void
1150racct_proc_wakeup(struct proc *p)
1151{
1152
1153	ASSERT_RACCT_ENABLED();
1154
1155	PROC_LOCK_ASSERT(p, MA_OWNED);
1156
1157	if (p->p_throttled != 0) {
1158		p->p_throttled = 0;
1159		wakeup(p->p_racct);
1160	}
1161}
1162
1163static void
1164racct_decay_callback(struct racct *racct, void *dummy1, void *dummy2)
1165{
1166	int64_t r_old, r_new;
1167
1168	ASSERT_RACCT_ENABLED();
1169	RACCT_LOCK_ASSERT();
1170
1171#ifdef RCTL
1172	rctl_throttle_decay(racct, RACCT_READBPS);
1173	rctl_throttle_decay(racct, RACCT_WRITEBPS);
1174	rctl_throttle_decay(racct, RACCT_READIOPS);
1175	rctl_throttle_decay(racct, RACCT_WRITEIOPS);
1176#endif
1177
1178	r_old = racct->r_resources[RACCT_PCTCPU];
1179
1180	/* If there is nothing to decay, just exit. */
1181	if (r_old <= 0)
1182		return;
1183
1184	r_new = r_old * RACCT_DECAY_FACTOR / FSCALE;
1185	racct->r_resources[RACCT_PCTCPU] = r_new;
1186}
1187
1188static void
1189racct_decay_pre(void)
1190{
1191
1192	RACCT_LOCK();
1193}
1194
1195static void
1196racct_decay_post(void)
1197{
1198
1199	RACCT_UNLOCK();
1200}
1201
1202static void
1203racct_decay(void)
1204{
1205
1206	ASSERT_RACCT_ENABLED();
1207
1208	ui_racct_foreach(racct_decay_callback, racct_decay_pre,
1209	    racct_decay_post, NULL, NULL);
1210	loginclass_racct_foreach(racct_decay_callback, racct_decay_pre,
1211	    racct_decay_post, NULL, NULL);
1212	prison_racct_foreach(racct_decay_callback, racct_decay_pre,
1213	    racct_decay_post, NULL, NULL);
1214}
1215
1216static void
1217racctd(void)
1218{
1219	struct thread *td;
1220	struct proc *p;
1221	struct timeval wallclock;
1222	uint64_t pct, pct_estimate, runtime;
1223
1224	ASSERT_RACCT_ENABLED();
1225
1226	for (;;) {
1227		racct_decay();
1228
1229		sx_slock(&allproc_lock);
1230
1231		LIST_FOREACH(p, &zombproc, p_list) {
1232			PROC_LOCK(p);
1233			racct_set(p, RACCT_PCTCPU, 0);
1234			PROC_UNLOCK(p);
1235		}
1236
1237		FOREACH_PROC_IN_SYSTEM(p) {
1238			PROC_LOCK(p);
1239			if (p->p_state != PRS_NORMAL) {
1240				PROC_UNLOCK(p);
1241				continue;
1242			}
1243
1244			microuptime(&wallclock);
1245			timevalsub(&wallclock, &p->p_stats->p_start);
1246			PROC_STATLOCK(p);
1247			FOREACH_THREAD_IN_PROC(p, td)
1248				ruxagg(p, td);
1249			runtime = cputick2usec(p->p_rux.rux_runtime);
1250			PROC_STATUNLOCK(p);
1251#ifdef notyet
1252			KASSERT(runtime >= p->p_prev_runtime,
1253			    ("runtime < p_prev_runtime"));
1254#else
1255			if (runtime < p->p_prev_runtime)
1256				runtime = p->p_prev_runtime;
1257#endif
1258			p->p_prev_runtime = runtime;
1259			if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
1260				pct_estimate = (1000000 * runtime * 100) /
1261				    ((uint64_t)wallclock.tv_sec * 1000000 +
1262				    wallclock.tv_usec);
1263			} else
1264				pct_estimate = 0;
1265			pct = racct_getpcpu(p, pct_estimate);
1266			RACCT_LOCK();
1267#ifdef RCTL
1268			rctl_throttle_decay(p->p_racct, RACCT_READBPS);
1269			rctl_throttle_decay(p->p_racct, RACCT_WRITEBPS);
1270			rctl_throttle_decay(p->p_racct, RACCT_READIOPS);
1271			rctl_throttle_decay(p->p_racct, RACCT_WRITEIOPS);
1272#endif
1273			racct_set_locked(p, RACCT_PCTCPU, pct, 1);
1274			racct_set_locked(p, RACCT_CPU, runtime, 0);
1275			racct_set_locked(p, RACCT_WALLCLOCK,
1276			    (uint64_t)wallclock.tv_sec * 1000000 +
1277			    wallclock.tv_usec, 0);
1278			RACCT_UNLOCK();
1279			PROC_UNLOCK(p);
1280		}
1281
1282		/*
1283		 * To ensure that processes are throttled in a fair way, we need
1284		 * to iterate over all processes again and check the limits
1285		 * for %cpu resource only after ucred racct containers have been
1286		 * properly filled.
1287		 */
1288		FOREACH_PROC_IN_SYSTEM(p) {
1289			PROC_LOCK(p);
1290			if (p->p_state != PRS_NORMAL) {
1291				PROC_UNLOCK(p);
1292				continue;
1293			}
1294
1295			if (racct_pcpu_available(p) <= 0) {
1296				if (p->p_racct->r_resources[RACCT_PCTCPU] >
1297				    pcpu_threshold)
1298					racct_proc_throttle(p, -1);
1299			} else if (p->p_throttled == -1) {
1300				racct_proc_wakeup(p);
1301			}
1302			PROC_UNLOCK(p);
1303		}
1304		sx_sunlock(&allproc_lock);
1305		pause("-", hz);
1306	}
1307}
1308
1309static struct kproc_desc racctd_kp = {
1310	"racctd",
1311	racctd,
1312	NULL
1313};
1314
1315static void
1316racctd_init(void)
1317{
1318	if (!racct_enable)
1319		return;
1320
1321	kproc_start(&racctd_kp);
1322}
1323SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, racctd_init, NULL);
1324
1325static void
1326racct_init(void)
1327{
1328	if (!racct_enable)
1329		return;
1330
1331	racct_zone = uma_zcreate("racct", sizeof(struct racct),
1332	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1333	/*
1334	 * XXX: Move this somewhere.
1335	 */
1336	prison0.pr_prison_racct = prison_racct_find("0");
1337}
1338SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL);
1339
1340#endif /* !RACCT */
1341