1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2010 The FreeBSD Foundation
5 *
6 * This software was developed by Edward Tomasz Napierala under sponsorship
7 * from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 #include "opt_sched.h"
33
34 #include <sys/param.h>
35 #include <sys/buf.h>
36 #include <sys/systm.h>
37 #include <sys/eventhandler.h>
38 #include <sys/jail.h>
39 #include <sys/kernel.h>
40 #include <sys/kthread.h>
41 #include <sys/lock.h>
42 #include <sys/loginclass.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/racct.h>
47 #include <sys/resourcevar.h>
48 #include <sys/sbuf.h>
49 #include <sys/sched.h>
50 #include <sys/sdt.h>
51 #include <sys/smp.h>
52 #include <sys/sx.h>
53 #include <sys/sysctl.h>
54 #include <sys/sysproto.h>
55 #include <sys/umtxvar.h>
56 #include <machine/smp.h>
57
58 #ifdef RCTL
59 #include <sys/rctl.h>
60 #endif
61
62 FEATURE(racct, "Resource Accounting");
63
64 /*
65 * Do not block processes that have their %cpu usage <= pcpu_threshold.
66 */
67 static int pcpu_threshold = 1;
68 #ifdef RACCT_DEFAULT_TO_DISABLED
69 bool __read_frequently racct_enable = false;
70 #else
71 bool __read_frequently racct_enable = true;
72 #endif
73
74 SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
75 "Resource Accounting");
76 SYSCTL_BOOL(_kern_racct, OID_AUTO, enable, CTLFLAG_RDTUN, &racct_enable,
77 0, "Enable RACCT/RCTL");
78 SYSCTL_UINT(_kern_racct, OID_AUTO, pcpu_threshold, CTLFLAG_RW, &pcpu_threshold,
79 0, "Processes with higher %cpu usage than this value can be throttled.");
80
81 /*
82 * How many seconds it takes to use the scheduler %cpu calculations. When a
83 * process starts, we compute its %cpu usage by dividing its runtime by the
84 * process wall clock time. After RACCT_PCPU_SECS pass, we use the value
85 * provided by the scheduler.
86 */
87 #define RACCT_PCPU_SECS 3
88
89 struct mtx racct_lock;
90 MTX_SYSINIT(racct_lock, &racct_lock, "racct lock", MTX_DEF);
91
92 static uma_zone_t racct_zone;
93
94 static void racct_sub_racct(struct racct *dest, const struct racct *src);
95 static void racct_sub_cred_locked(struct ucred *cred, int resource,
96 uint64_t amount);
97 static void racct_add_cred_locked(struct ucred *cred, int resource,
98 uint64_t amount);
99
100 SDT_PROVIDER_DEFINE(racct);
101 SDT_PROBE_DEFINE3(racct, , rusage, add,
102 "struct proc *", "int", "uint64_t");
103 SDT_PROBE_DEFINE3(racct, , rusage, add__failure,
104 "struct proc *", "int", "uint64_t");
105 SDT_PROBE_DEFINE3(racct, , rusage, add__buf,
106 "struct proc *", "const struct buf *", "int");
107 SDT_PROBE_DEFINE3(racct, , rusage, add__cred,
108 "struct ucred *", "int", "uint64_t");
109 SDT_PROBE_DEFINE3(racct, , rusage, add__force,
110 "struct proc *", "int", "uint64_t");
111 SDT_PROBE_DEFINE3(racct, , rusage, set,
112 "struct proc *", "int", "uint64_t");
113 SDT_PROBE_DEFINE3(racct, , rusage, set__failure,
114 "struct proc *", "int", "uint64_t");
115 SDT_PROBE_DEFINE3(racct, , rusage, set__force,
116 "struct proc *", "int", "uint64_t");
117 SDT_PROBE_DEFINE3(racct, , rusage, sub,
118 "struct proc *", "int", "uint64_t");
119 SDT_PROBE_DEFINE3(racct, , rusage, sub__cred,
120 "struct ucred *", "int", "uint64_t");
121 SDT_PROBE_DEFINE1(racct, , racct, create,
122 "struct racct *");
123 SDT_PROBE_DEFINE1(racct, , racct, destroy,
124 "struct racct *");
125 SDT_PROBE_DEFINE2(racct, , racct, join,
126 "struct racct *", "struct racct *");
127 SDT_PROBE_DEFINE2(racct, , racct, join__failure,
128 "struct racct *", "struct racct *");
129 SDT_PROBE_DEFINE2(racct, , racct, leave,
130 "struct racct *", "struct racct *");
131
132 int racct_types[] = {
133 [RACCT_CPU] =
134 RACCT_IN_MILLIONS,
135 [RACCT_DATA] =
136 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
137 [RACCT_STACK] =
138 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
139 [RACCT_CORE] =
140 RACCT_DENIABLE,
141 [RACCT_RSS] =
142 RACCT_RECLAIMABLE,
143 [RACCT_MEMLOCK] =
144 RACCT_RECLAIMABLE | RACCT_DENIABLE,
145 [RACCT_NPROC] =
146 RACCT_RECLAIMABLE | RACCT_DENIABLE,
147 [RACCT_NOFILE] =
148 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
149 [RACCT_VMEM] =
150 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
151 [RACCT_NPTS] =
152 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
153 [RACCT_SWAP] =
154 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
155 [RACCT_NTHR] =
156 RACCT_RECLAIMABLE | RACCT_DENIABLE,
157 [RACCT_MSGQQUEUED] =
158 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
159 [RACCT_MSGQSIZE] =
160 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
161 [RACCT_NMSGQ] =
162 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
163 [RACCT_NSEM] =
164 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
165 [RACCT_NSEMOP] =
166 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
167 [RACCT_NSHM] =
168 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
169 [RACCT_SHMSIZE] =
170 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
171 [RACCT_WALLCLOCK] =
172 RACCT_IN_MILLIONS,
173 [RACCT_PCTCPU] =
174 RACCT_DECAYING | RACCT_DENIABLE | RACCT_IN_MILLIONS,
175 [RACCT_READBPS] =
176 RACCT_DECAYING,
177 [RACCT_WRITEBPS] =
178 RACCT_DECAYING,
179 [RACCT_READIOPS] =
180 RACCT_DECAYING,
181 [RACCT_WRITEIOPS] =
182 RACCT_DECAYING };
183
184 static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE;
185
186 #ifdef SCHED_4BSD
187 /*
188 * Contains intermediate values for %cpu calculations to avoid using floating
189 * point in the kernel.
190 * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20)
191 * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to
192 * zero so the calculations are more straightforward.
193 */
194 fixpt_t ccpu_exp[] = {
195 [0] = FSCALE * 1,
196 [1] = FSCALE * 0.95122942450071400909,
197 [2] = FSCALE * 0.90483741803595957316,
198 [3] = FSCALE * 0.86070797642505780722,
199 [4] = FSCALE * 0.81873075307798185866,
200 [5] = FSCALE * 0.77880078307140486824,
201 [6] = FSCALE * 0.74081822068171786606,
202 [7] = FSCALE * 0.70468808971871343435,
203 [8] = FSCALE * 0.67032004603563930074,
204 [9] = FSCALE * 0.63762815162177329314,
205 [10] = FSCALE * 0.60653065971263342360,
206 [11] = FSCALE * 0.57694981038048669531,
207 [12] = FSCALE * 0.54881163609402643262,
208 [13] = FSCALE * 0.52204577676101604789,
209 [14] = FSCALE * 0.49658530379140951470,
210 [15] = FSCALE * 0.47236655274101470713,
211 [16] = FSCALE * 0.44932896411722159143,
212 [17] = FSCALE * 0.42741493194872666992,
213 [18] = FSCALE * 0.40656965974059911188,
214 [19] = FSCALE * 0.38674102345450120691,
215 [20] = FSCALE * 0.36787944117144232159,
216 [21] = FSCALE * 0.34993774911115535467,
217 [22] = FSCALE * 0.33287108369807955328,
218 [23] = FSCALE * 0.31663676937905321821,
219 [24] = FSCALE * 0.30119421191220209664,
220 [25] = FSCALE * 0.28650479686019010032,
221 [26] = FSCALE * 0.27253179303401260312,
222 [27] = FSCALE * 0.25924026064589150757,
223 [28] = FSCALE * 0.24659696394160647693,
224 [29] = FSCALE * 0.23457028809379765313,
225 [30] = FSCALE * 0.22313016014842982893,
226 [31] = FSCALE * 0.21224797382674305771,
227 [32] = FSCALE * 0.20189651799465540848,
228 [33] = FSCALE * 0.19204990862075411423,
229 [34] = FSCALE * 0.18268352405273465022,
230 [35] = FSCALE * 0.17377394345044512668,
231 [36] = FSCALE * 0.16529888822158653829,
232 [37] = FSCALE * 0.15723716631362761621,
233 [38] = FSCALE * 0.14956861922263505264,
234 [39] = FSCALE * 0.14227407158651357185,
235 [40] = FSCALE * 0.13533528323661269189,
236 [41] = FSCALE * 0.12873490358780421886,
237 [42] = FSCALE * 0.12245642825298191021,
238 [43] = FSCALE * 0.11648415777349695786,
239 [44] = FSCALE * 0.11080315836233388333,
240 [45] = FSCALE * 0.10539922456186433678,
241 [46] = FSCALE * 0.10025884372280373372,
242 [47] = FSCALE * 0.09536916221554961888,
243 [48] = FSCALE * 0.09071795328941250337,
244 [49] = FSCALE * 0.08629358649937051097,
245 [50] = FSCALE * 0.08208499862389879516,
246 [51] = FSCALE * 0.07808166600115315231,
247 [52] = FSCALE * 0.07427357821433388042,
248 [53] = FSCALE * 0.07065121306042958674,
249 [54] = FSCALE * 0.06720551273974976512,
250 [55] = FSCALE * 0.06392786120670757270,
251 [56] = FSCALE * 0.06081006262521796499,
252 [57] = FSCALE * 0.05784432087483846296,
253 [58] = FSCALE * 0.05502322005640722902,
254 [59] = FSCALE * 0.05233970594843239308,
255 [60] = FSCALE * 0.04978706836786394297,
256 [61] = FSCALE * 0.04735892439114092119,
257 [62] = FSCALE * 0.04504920239355780606,
258 [63] = FSCALE * 0.04285212686704017991,
259 [64] = FSCALE * 0.04076220397836621516,
260 [65] = FSCALE * 0.03877420783172200988,
261 [66] = FSCALE * 0.03688316740124000544,
262 [67] = FSCALE * 0.03508435410084502588,
263 [68] = FSCALE * 0.03337326996032607948,
264 [69] = FSCALE * 0.03174563637806794323,
265 [70] = FSCALE * 0.03019738342231850073,
266 [71] = FSCALE * 0.02872463965423942912,
267 [72] = FSCALE * 0.02732372244729256080,
268 [73] = FSCALE * 0.02599112877875534358,
269 [74] = FSCALE * 0.02472352647033939120,
270 [75] = FSCALE * 0.02351774585600910823,
271 [76] = FSCALE * 0.02237077185616559577,
272 [77] = FSCALE * 0.02127973643837716938,
273 [78] = FSCALE * 0.02024191144580438847,
274 [79] = FSCALE * 0.01925470177538692429,
275 [80] = FSCALE * 0.01831563888873418029,
276 [81] = FSCALE * 0.01742237463949351138,
277 [82] = FSCALE * 0.01657267540176124754,
278 [83] = FSCALE * 0.01576441648485449082,
279 [84] = FSCALE * 0.01499557682047770621,
280 [85] = FSCALE * 0.01426423390899925527,
281 [86] = FSCALE * 0.01356855901220093175,
282 [87] = FSCALE * 0.01290681258047986886,
283 [88] = FSCALE * 0.01227733990306844117,
284 [89] = FSCALE * 0.01167856697039544521,
285 [90] = FSCALE * 0.01110899653824230649,
286 [91] = FSCALE * 0.01056720438385265337,
287 [92] = FSCALE * 0.01005183574463358164,
288 [93] = FSCALE * 0.00956160193054350793,
289 [94] = FSCALE * 0.00909527710169581709,
290 [95] = FSCALE * 0.00865169520312063417,
291 [96] = FSCALE * 0.00822974704902002884,
292 [97] = FSCALE * 0.00782837754922577143,
293 [98] = FSCALE * 0.00744658307092434051,
294 [99] = FSCALE * 0.00708340892905212004,
295 [100] = FSCALE * 0.00673794699908546709,
296 [101] = FSCALE * 0.00640933344625638184,
297 [102] = FSCALE * 0.00609674656551563610,
298 [103] = FSCALE * 0.00579940472684214321,
299 [104] = FSCALE * 0.00551656442076077241,
300 [105] = FSCALE * 0.00524751839918138427,
301 [106] = FSCALE * 0.00499159390691021621,
302 [107] = FSCALE * 0.00474815099941147558,
303 [108] = FSCALE * 0.00451658094261266798,
304 [109] = FSCALE * 0.00429630469075234057,
305 [110] = FSCALE * 0.00408677143846406699,
306 };
307 #endif
308
309 #define CCPU_EXP_MAX 110
310
311 /*
312 * This function is analogical to the getpcpu() function in the ps(1) command.
313 * They should both calculate in the same way so that the racct %cpu
314 * calculations are consistent with the values showed by the ps(1) tool.
315 * The calculations are more complex in the 4BSD scheduler because of the value
316 * of the ccpu variable. In ULE it is defined to be zero which saves us some
317 * work.
318 */
319 static uint64_t
racct_getpcpu(struct proc * p,u_int pcpu)320 racct_getpcpu(struct proc *p, u_int pcpu)
321 {
322 u_int swtime;
323 #ifdef SCHED_4BSD
324 fixpt_t pctcpu, pctcpu_next;
325 #endif
326 #ifdef SMP
327 struct pcpu *pc;
328 int found;
329 #endif
330 fixpt_t p_pctcpu;
331 struct thread *td;
332
333 ASSERT_RACCT_ENABLED();
334
335 swtime = (ticks - p->p_swtick) / hz;
336
337 /*
338 * For short-lived processes, the sched_pctcpu() returns small
339 * values even for cpu intensive processes. Therefore we use
340 * our own estimate in this case.
341 */
342 if (swtime < RACCT_PCPU_SECS)
343 return (pcpu);
344
345 p_pctcpu = 0;
346 FOREACH_THREAD_IN_PROC(p, td) {
347 if (td == PCPU_GET(idlethread))
348 continue;
349 #ifdef SMP
350 found = 0;
351 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
352 if (td == pc->pc_idlethread) {
353 found = 1;
354 break;
355 }
356 }
357 if (found)
358 continue;
359 #endif
360 thread_lock(td);
361 #ifdef SCHED_4BSD
362 pctcpu = sched_pctcpu(td);
363 /* Count also the yet unfinished second. */
364 pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT;
365 pctcpu_next += sched_pctcpu_delta(td);
366 p_pctcpu += max(pctcpu, pctcpu_next);
367 #else
368 /*
369 * In ULE the %cpu statistics are updated on every
370 * sched_pctcpu() call. So special calculations to
371 * account for the latest (unfinished) second are
372 * not needed.
373 */
374 p_pctcpu += sched_pctcpu(td);
375 #endif
376 thread_unlock(td);
377 }
378
379 #ifdef SCHED_4BSD
380 if (swtime <= CCPU_EXP_MAX)
381 return ((100 * (uint64_t)p_pctcpu * 1000000) /
382 (FSCALE - ccpu_exp[swtime]));
383 #endif
384
385 return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE);
386 }
387
388 static void
racct_add_racct(struct racct * dest,const struct racct * src)389 racct_add_racct(struct racct *dest, const struct racct *src)
390 {
391 int i;
392
393 ASSERT_RACCT_ENABLED();
394 RACCT_LOCK_ASSERT();
395
396 /*
397 * Update resource usage in dest.
398 */
399 for (i = 0; i <= RACCT_MAX; i++) {
400 KASSERT(dest->r_resources[i] >= 0,
401 ("%s: resource %d propagation meltdown: dest < 0",
402 __func__, i));
403 KASSERT(src->r_resources[i] >= 0,
404 ("%s: resource %d propagation meltdown: src < 0",
405 __func__, i));
406 dest->r_resources[i] += src->r_resources[i];
407 }
408 }
409
410 static void
racct_sub_racct(struct racct * dest,const struct racct * src)411 racct_sub_racct(struct racct *dest, const struct racct *src)
412 {
413 int i;
414
415 ASSERT_RACCT_ENABLED();
416 RACCT_LOCK_ASSERT();
417
418 /*
419 * Update resource usage in dest.
420 */
421 for (i = 0; i <= RACCT_MAX; i++) {
422 if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) {
423 KASSERT(dest->r_resources[i] >= 0,
424 ("%s: resource %d propagation meltdown: dest < 0",
425 __func__, i));
426 KASSERT(src->r_resources[i] >= 0,
427 ("%s: resource %d propagation meltdown: src < 0",
428 __func__, i));
429 KASSERT(src->r_resources[i] <= dest->r_resources[i],
430 ("%s: resource %d propagation meltdown: src > dest",
431 __func__, i));
432 }
433 if (RACCT_CAN_DROP(i)) {
434 dest->r_resources[i] -= src->r_resources[i];
435 if (dest->r_resources[i] < 0)
436 dest->r_resources[i] = 0;
437 }
438 }
439 }
440
441 void
racct_create(struct racct ** racctp)442 racct_create(struct racct **racctp)
443 {
444
445 if (!racct_enable)
446 return;
447
448 SDT_PROBE1(racct, , racct, create, racctp);
449
450 KASSERT(*racctp == NULL, ("racct already allocated"));
451
452 *racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO);
453 }
454
455 static void
racct_destroy_locked(struct racct ** racctp)456 racct_destroy_locked(struct racct **racctp)
457 {
458 struct racct *racct;
459 int i;
460
461 ASSERT_RACCT_ENABLED();
462
463 SDT_PROBE1(racct, , racct, destroy, racctp);
464
465 RACCT_LOCK_ASSERT();
466 KASSERT(racctp != NULL, ("NULL racctp"));
467 KASSERT(*racctp != NULL, ("NULL racct"));
468
469 racct = *racctp;
470
471 for (i = 0; i <= RACCT_MAX; i++) {
472 if (RACCT_IS_SLOPPY(i))
473 continue;
474 if (!RACCT_IS_RECLAIMABLE(i))
475 continue;
476 KASSERT(racct->r_resources[i] == 0,
477 ("destroying non-empty racct: "
478 "%ju allocated for resource %d\n",
479 racct->r_resources[i], i));
480 }
481 uma_zfree(racct_zone, racct);
482 *racctp = NULL;
483 }
484
485 void
racct_destroy(struct racct ** racct)486 racct_destroy(struct racct **racct)
487 {
488
489 if (!racct_enable)
490 return;
491
492 RACCT_LOCK();
493 racct_destroy_locked(racct);
494 RACCT_UNLOCK();
495 }
496
497 /*
498 * Increase consumption of 'resource' by 'amount' for 'racct',
499 * but not its parents. Differently from other cases, 'amount' here
500 * may be less than zero.
501 */
502 static void
racct_adjust_resource(struct racct * racct,int resource,int64_t amount)503 racct_adjust_resource(struct racct *racct, int resource,
504 int64_t amount)
505 {
506
507 ASSERT_RACCT_ENABLED();
508 RACCT_LOCK_ASSERT();
509 KASSERT(racct != NULL, ("NULL racct"));
510
511 racct->r_resources[resource] += amount;
512 if (racct->r_resources[resource] < 0) {
513 KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource),
514 ("%s: resource %d usage < 0", __func__, resource));
515 racct->r_resources[resource] = 0;
516 }
517
518 /*
519 * There are some cases where the racct %cpu resource would grow
520 * beyond 100% per core. For example in racct_proc_exit() we add
521 * the process %cpu usage to the ucred racct containers. If too
522 * many processes terminated in a short time span, the ucred %cpu
523 * resource could grow too much. Also, the 4BSD scheduler sometimes
524 * returns for a thread more than 100% cpu usage. So we set a sane
525 * boundary here to 100% * the maximum number of CPUs.
526 */
527 if ((resource == RACCT_PCTCPU) &&
528 (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000 * (int64_t)MAXCPU))
529 racct->r_resources[RACCT_PCTCPU] = 100 * 1000000 * (int64_t)MAXCPU;
530 }
531
532 static int
racct_add_locked(struct proc * p,int resource,uint64_t amount,int force)533 racct_add_locked(struct proc *p, int resource, uint64_t amount, int force)
534 {
535 #ifdef RCTL
536 int error;
537 #endif
538
539 ASSERT_RACCT_ENABLED();
540
541 /*
542 * We need proc lock to dereference p->p_ucred.
543 */
544 PROC_LOCK_ASSERT(p, MA_OWNED);
545
546 #ifdef RCTL
547 error = rctl_enforce(p, resource, amount);
548 if (error && !force && RACCT_IS_DENIABLE(resource)) {
549 SDT_PROBE3(racct, , rusage, add__failure, p, resource, amount);
550 return (error);
551 }
552 #endif
553 racct_adjust_resource(p->p_racct, resource, amount);
554 racct_add_cred_locked(p->p_ucred, resource, amount);
555
556 return (0);
557 }
558
559 /*
560 * Increase allocation of 'resource' by 'amount' for process 'p'.
561 * Return 0 if it's below limits, or errno, if it's not.
562 */
563 int
racct_add(struct proc * p,int resource,uint64_t amount)564 racct_add(struct proc *p, int resource, uint64_t amount)
565 {
566 int error;
567
568 if (!racct_enable)
569 return (0);
570
571 SDT_PROBE3(racct, , rusage, add, p, resource, amount);
572
573 RACCT_LOCK();
574 error = racct_add_locked(p, resource, amount, 0);
575 RACCT_UNLOCK();
576 return (error);
577 }
578
579 /*
580 * Increase allocation of 'resource' by 'amount' for process 'p'.
581 * Doesn't check for limits and never fails.
582 */
583 void
racct_add_force(struct proc * p,int resource,uint64_t amount)584 racct_add_force(struct proc *p, int resource, uint64_t amount)
585 {
586
587 if (!racct_enable)
588 return;
589
590 SDT_PROBE3(racct, , rusage, add__force, p, resource, amount);
591
592 RACCT_LOCK();
593 racct_add_locked(p, resource, amount, 1);
594 RACCT_UNLOCK();
595 }
596
597 static void
racct_add_cred_locked(struct ucred * cred,int resource,uint64_t amount)598 racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount)
599 {
600 struct prison *pr;
601
602 ASSERT_RACCT_ENABLED();
603
604 racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, amount);
605 for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
606 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
607 amount);
608 racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, amount);
609 }
610
611 /*
612 * Increase allocation of 'resource' by 'amount' for credential 'cred'.
613 * Doesn't check for limits and never fails.
614 */
615 void
racct_add_cred(struct ucred * cred,int resource,uint64_t amount)616 racct_add_cred(struct ucred *cred, int resource, uint64_t amount)
617 {
618
619 if (!racct_enable)
620 return;
621
622 SDT_PROBE3(racct, , rusage, add__cred, cred, resource, amount);
623
624 RACCT_LOCK();
625 racct_add_cred_locked(cred, resource, amount);
626 RACCT_UNLOCK();
627 }
628
629 /*
630 * Account for disk IO resource consumption. Checks for limits,
631 * but never fails, due to disk limits being undeniable.
632 */
633 void
racct_add_buf(struct proc * p,const struct buf * bp,int is_write)634 racct_add_buf(struct proc *p, const struct buf *bp, int is_write)
635 {
636
637 ASSERT_RACCT_ENABLED();
638 PROC_LOCK_ASSERT(p, MA_OWNED);
639
640 SDT_PROBE3(racct, , rusage, add__buf, p, bp, is_write);
641
642 RACCT_LOCK();
643 if (is_write) {
644 racct_add_locked(curproc, RACCT_WRITEBPS, bp->b_bcount, 1);
645 racct_add_locked(curproc, RACCT_WRITEIOPS, 1, 1);
646 } else {
647 racct_add_locked(curproc, RACCT_READBPS, bp->b_bcount, 1);
648 racct_add_locked(curproc, RACCT_READIOPS, 1, 1);
649 }
650 RACCT_UNLOCK();
651 }
652
653 static int
racct_set_locked(struct proc * p,int resource,uint64_t amount,int force)654 racct_set_locked(struct proc *p, int resource, uint64_t amount, int force)
655 {
656 int64_t old_amount, decayed_amount, diff_proc, diff_cred;
657 #ifdef RCTL
658 int error;
659 #endif
660
661 ASSERT_RACCT_ENABLED();
662
663 /*
664 * We need proc lock to dereference p->p_ucred.
665 */
666 PROC_LOCK_ASSERT(p, MA_OWNED);
667
668 old_amount = p->p_racct->r_resources[resource];
669 /*
670 * The diffs may be negative.
671 */
672 diff_proc = amount - old_amount;
673 if (resource == RACCT_PCTCPU) {
674 /*
675 * Resources in per-credential racct containers may decay.
676 * If this is the case, we need to calculate the difference
677 * between the new amount and the proportional value of the
678 * old amount that has decayed in the ucred racct containers.
679 */
680 decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE;
681 diff_cred = amount - decayed_amount;
682 } else
683 diff_cred = diff_proc;
684 #ifdef notyet
685 KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource),
686 ("%s: usage of non-droppable resource %d dropping", __func__,
687 resource));
688 #endif
689 #ifdef RCTL
690 if (diff_proc > 0) {
691 error = rctl_enforce(p, resource, diff_proc);
692 if (error && !force && RACCT_IS_DENIABLE(resource)) {
693 SDT_PROBE3(racct, , rusage, set__failure, p, resource,
694 amount);
695 return (error);
696 }
697 }
698 #endif
699 racct_adjust_resource(p->p_racct, resource, diff_proc);
700 if (diff_cred > 0)
701 racct_add_cred_locked(p->p_ucred, resource, diff_cred);
702 else if (diff_cred < 0)
703 racct_sub_cred_locked(p->p_ucred, resource, -diff_cred);
704
705 return (0);
706 }
707
708 /*
709 * Set allocation of 'resource' to 'amount' for process 'p'.
710 * Return 0 if it's below limits, or errno, if it's not.
711 *
712 * Note that decreasing the allocation always returns 0,
713 * even if it's above the limit.
714 */
715 int
racct_set_unlocked(struct proc * p,int resource,uint64_t amount)716 racct_set_unlocked(struct proc *p, int resource, uint64_t amount)
717 {
718 int error;
719
720 ASSERT_RACCT_ENABLED();
721 PROC_LOCK(p);
722 error = racct_set(p, resource, amount);
723 PROC_UNLOCK(p);
724 return (error);
725 }
726
727 int
racct_set(struct proc * p,int resource,uint64_t amount)728 racct_set(struct proc *p, int resource, uint64_t amount)
729 {
730 int error;
731
732 if (!racct_enable)
733 return (0);
734
735 SDT_PROBE3(racct, , rusage, set__force, p, resource, amount);
736
737 RACCT_LOCK();
738 error = racct_set_locked(p, resource, amount, 0);
739 RACCT_UNLOCK();
740 return (error);
741 }
742
743 void
racct_set_force(struct proc * p,int resource,uint64_t amount)744 racct_set_force(struct proc *p, int resource, uint64_t amount)
745 {
746
747 if (!racct_enable)
748 return;
749
750 SDT_PROBE3(racct, , rusage, set, p, resource, amount);
751
752 RACCT_LOCK();
753 racct_set_locked(p, resource, amount, 1);
754 RACCT_UNLOCK();
755 }
756
757 /*
758 * Returns amount of 'resource' the process 'p' can keep allocated.
759 * Allocating more than that would be denied, unless the resource
760 * is marked undeniable. Amount of already allocated resource does
761 * not matter.
762 */
763 uint64_t
racct_get_limit(struct proc * p,int resource)764 racct_get_limit(struct proc *p, int resource)
765 {
766 #ifdef RCTL
767 uint64_t available;
768
769 if (!racct_enable)
770 return (UINT64_MAX);
771
772 RACCT_LOCK();
773 available = rctl_get_limit(p, resource);
774 RACCT_UNLOCK();
775
776 return (available);
777 #else
778
779 return (UINT64_MAX);
780 #endif
781 }
782
783 /*
784 * Returns amount of 'resource' the process 'p' can keep allocated.
785 * Allocating more than that would be denied, unless the resource
786 * is marked undeniable. Amount of already allocated resource does
787 * matter.
788 */
789 uint64_t
racct_get_available(struct proc * p,int resource)790 racct_get_available(struct proc *p, int resource)
791 {
792 #ifdef RCTL
793 uint64_t available;
794
795 if (!racct_enable)
796 return (UINT64_MAX);
797
798 RACCT_LOCK();
799 available = rctl_get_available(p, resource);
800 RACCT_UNLOCK();
801
802 return (available);
803 #else
804
805 return (UINT64_MAX);
806 #endif
807 }
808
809 /*
810 * Returns amount of the %cpu resource that process 'p' can add to its %cpu
811 * utilization. Adding more than that would lead to the process being
812 * throttled.
813 */
814 static int64_t
racct_pcpu_available(struct proc * p)815 racct_pcpu_available(struct proc *p)
816 {
817 #ifdef RCTL
818 uint64_t available;
819
820 ASSERT_RACCT_ENABLED();
821
822 RACCT_LOCK();
823 available = rctl_pcpu_available(p);
824 RACCT_UNLOCK();
825
826 return (available);
827 #else
828
829 return (INT64_MAX);
830 #endif
831 }
832
833 /*
834 * Decrease allocation of 'resource' by 'amount' for process 'p'.
835 */
836 void
racct_sub(struct proc * p,int resource,uint64_t amount)837 racct_sub(struct proc *p, int resource, uint64_t amount)
838 {
839
840 if (!racct_enable)
841 return;
842
843 SDT_PROBE3(racct, , rusage, sub, p, resource, amount);
844
845 /*
846 * We need proc lock to dereference p->p_ucred.
847 */
848 PROC_LOCK_ASSERT(p, MA_OWNED);
849 KASSERT(RACCT_CAN_DROP(resource),
850 ("%s: called for non-droppable resource %d", __func__, resource));
851
852 RACCT_LOCK();
853 KASSERT(amount <= p->p_racct->r_resources[resource],
854 ("%s: freeing %ju of resource %d, which is more "
855 "than allocated %jd for %s (pid %d)", __func__, amount, resource,
856 (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid));
857
858 racct_adjust_resource(p->p_racct, resource, -amount);
859 racct_sub_cred_locked(p->p_ucred, resource, amount);
860 RACCT_UNLOCK();
861 }
862
863 static void
racct_sub_cred_locked(struct ucred * cred,int resource,uint64_t amount)864 racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount)
865 {
866 struct prison *pr;
867
868 ASSERT_RACCT_ENABLED();
869
870 racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, -amount);
871 for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
872 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
873 -amount);
874 racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, -amount);
875 }
876
877 /*
878 * Decrease allocation of 'resource' by 'amount' for credential 'cred'.
879 */
880 void
racct_sub_cred(struct ucred * cred,int resource,uint64_t amount)881 racct_sub_cred(struct ucred *cred, int resource, uint64_t amount)
882 {
883
884 if (!racct_enable)
885 return;
886
887 SDT_PROBE3(racct, , rusage, sub__cred, cred, resource, amount);
888
889 #ifdef notyet
890 KASSERT(RACCT_CAN_DROP(resource),
891 ("%s: called for resource %d which can not drop", __func__,
892 resource));
893 #endif
894
895 RACCT_LOCK();
896 racct_sub_cred_locked(cred, resource, amount);
897 RACCT_UNLOCK();
898 }
899
900 /*
901 * Inherit resource usage information from the parent process.
902 */
903 int
racct_proc_fork(struct proc * parent,struct proc * child)904 racct_proc_fork(struct proc *parent, struct proc *child)
905 {
906 int i, error = 0;
907
908 if (!racct_enable)
909 return (0);
910
911 /*
912 * Create racct for the child process.
913 */
914 racct_create(&child->p_racct);
915
916 PROC_LOCK(parent);
917 PROC_LOCK(child);
918 RACCT_LOCK();
919
920 #ifdef RCTL
921 error = rctl_proc_fork(parent, child);
922 if (error != 0)
923 goto out;
924 #endif
925
926 /* Init process cpu time. */
927 child->p_prev_runtime = 0;
928 child->p_throttled = 0;
929
930 /*
931 * Inherit resource usage.
932 */
933 for (i = 0; i <= RACCT_MAX; i++) {
934 if (parent->p_racct->r_resources[i] == 0 ||
935 !RACCT_IS_INHERITABLE(i))
936 continue;
937
938 error = racct_set_locked(child, i,
939 parent->p_racct->r_resources[i], 0);
940 if (error != 0)
941 goto out;
942 }
943
944 error = racct_add_locked(child, RACCT_NPROC, 1, 0);
945 error += racct_add_locked(child, RACCT_NTHR, 1, 0);
946
947 out:
948 RACCT_UNLOCK();
949 PROC_UNLOCK(child);
950 PROC_UNLOCK(parent);
951
952 if (error != 0)
953 racct_proc_exit(child);
954
955 return (error);
956 }
957
958 /*
959 * Called at the end of fork1(), to handle rules that require the process
960 * to be fully initialized.
961 */
962 void
racct_proc_fork_done(struct proc * child)963 racct_proc_fork_done(struct proc *child)
964 {
965
966 if (!racct_enable)
967 return;
968
969 #ifdef RCTL
970 PROC_LOCK(child);
971 RACCT_LOCK();
972 rctl_enforce(child, RACCT_NPROC, 0);
973 rctl_enforce(child, RACCT_NTHR, 0);
974 RACCT_UNLOCK();
975 PROC_UNLOCK(child);
976 #endif
977 }
978
979 void
racct_proc_exit(struct proc * p)980 racct_proc_exit(struct proc *p)
981 {
982 struct timeval wallclock;
983 uint64_t pct_estimate, pct, runtime;
984 int i;
985
986 if (!racct_enable)
987 return;
988
989 PROC_LOCK(p);
990 /*
991 * We don't need to calculate rux, proc_reap() has already done this.
992 */
993 runtime = cputick2usec(p->p_rux.rux_runtime);
994 #ifdef notyet
995 KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime"));
996 #else
997 if (runtime < p->p_prev_runtime)
998 runtime = p->p_prev_runtime;
999 #endif
1000 microuptime(&wallclock);
1001 timevalsub(&wallclock, &p->p_stats->p_start);
1002 if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
1003 pct_estimate = (1000000 * runtime * 100) /
1004 ((uint64_t)wallclock.tv_sec * 1000000 +
1005 wallclock.tv_usec);
1006 } else
1007 pct_estimate = 0;
1008 pct = racct_getpcpu(p, pct_estimate);
1009
1010 RACCT_LOCK();
1011 racct_set_locked(p, RACCT_CPU, runtime, 0);
1012 racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct);
1013
1014 KASSERT(p->p_racct->r_resources[RACCT_RSS] == 0,
1015 ("process reaped with %ju allocated for RSS\n",
1016 p->p_racct->r_resources[RACCT_RSS]));
1017 for (i = 0; i <= RACCT_MAX; i++) {
1018 if (p->p_racct->r_resources[i] == 0)
1019 continue;
1020 if (!RACCT_IS_RECLAIMABLE(i))
1021 continue;
1022 racct_set_locked(p, i, 0, 0);
1023 }
1024
1025 #ifdef RCTL
1026 rctl_racct_release(p->p_racct);
1027 #endif
1028 racct_destroy_locked(&p->p_racct);
1029 RACCT_UNLOCK();
1030 PROC_UNLOCK(p);
1031 }
1032
1033 /*
1034 * Called after credentials change, to move resource utilisation
1035 * between raccts.
1036 */
1037 void
racct_proc_ucred_changed(struct proc * p,struct ucred * oldcred,struct ucred * newcred)1038 racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred,
1039 struct ucred *newcred)
1040 {
1041 struct uidinfo *olduip, *newuip;
1042 struct loginclass *oldlc, *newlc;
1043 struct prison *oldpr, *newpr, *pr;
1044
1045 if (!racct_enable)
1046 return;
1047
1048 PROC_LOCK_ASSERT(p, MA_OWNED);
1049
1050 newuip = newcred->cr_ruidinfo;
1051 olduip = oldcred->cr_ruidinfo;
1052 newlc = newcred->cr_loginclass;
1053 oldlc = oldcred->cr_loginclass;
1054 newpr = newcred->cr_prison;
1055 oldpr = oldcred->cr_prison;
1056
1057 RACCT_LOCK();
1058 if (newuip != olduip) {
1059 racct_sub_racct(olduip->ui_racct, p->p_racct);
1060 racct_add_racct(newuip->ui_racct, p->p_racct);
1061 }
1062 if (newlc != oldlc) {
1063 racct_sub_racct(oldlc->lc_racct, p->p_racct);
1064 racct_add_racct(newlc->lc_racct, p->p_racct);
1065 }
1066 if (newpr != oldpr) {
1067 for (pr = oldpr; pr != NULL; pr = pr->pr_parent)
1068 racct_sub_racct(pr->pr_prison_racct->prr_racct,
1069 p->p_racct);
1070 for (pr = newpr; pr != NULL; pr = pr->pr_parent)
1071 racct_add_racct(pr->pr_prison_racct->prr_racct,
1072 p->p_racct);
1073 }
1074 RACCT_UNLOCK();
1075 }
1076
1077 void
racct_move(struct racct * dest,struct racct * src)1078 racct_move(struct racct *dest, struct racct *src)
1079 {
1080
1081 ASSERT_RACCT_ENABLED();
1082
1083 RACCT_LOCK();
1084 racct_add_racct(dest, src);
1085 racct_sub_racct(src, src);
1086 RACCT_UNLOCK();
1087 }
1088
1089 static void
ast_racct(struct thread * td,int tda __unused)1090 ast_racct(struct thread *td, int tda __unused)
1091 {
1092 struct proc *p;
1093
1094 ASSERT_RACCT_ENABLED();
1095
1096 p = td->td_proc;
1097 if (p->p_throttled == 0)
1098 return;
1099
1100 PROC_LOCK(p);
1101 while (p->p_throttled != 0) {
1102 msleep(p->p_racct, &p->p_mtx, 0, "racct",
1103 p->p_throttled < 0 ? 0 : p->p_throttled);
1104 if (p->p_throttled > 0)
1105 p->p_throttled = 0;
1106 }
1107 PROC_UNLOCK(p);
1108 }
1109
1110 /*
1111 * Make the process sleep in userret() for 'timeout' ticks. Setting
1112 * timeout to -1 makes it sleep until woken up by racct_proc_wakeup().
1113 */
1114 void
racct_proc_throttle(struct proc * p,int timeout)1115 racct_proc_throttle(struct proc *p, int timeout)
1116 {
1117 struct thread *td;
1118 #ifdef SMP
1119 int cpuid;
1120 #endif
1121
1122 KASSERT(timeout != 0, ("timeout %d", timeout));
1123 ASSERT_RACCT_ENABLED();
1124 PROC_LOCK_ASSERT(p, MA_OWNED);
1125
1126 /*
1127 * Do not block kernel processes. Also do not block processes with
1128 * low %cpu utilization to improve interactivity.
1129 */
1130 if ((p->p_flag & (P_SYSTEM | P_KPROC)) != 0)
1131 return;
1132
1133 if (p->p_throttled < 0 || (timeout > 0 && p->p_throttled > timeout))
1134 return;
1135
1136 p->p_throttled = timeout;
1137
1138 FOREACH_THREAD_IN_PROC(p, td) {
1139 thread_lock(td);
1140 ast_sched_locked(td, TDA_RACCT);
1141
1142 switch (TD_GET_STATE(td)) {
1143 case TDS_RUNQ:
1144 /*
1145 * If the thread is on the scheduler run-queue, we can
1146 * not just remove it from there. So we set the flag
1147 * TDA_SCHED for the thread, so that once it is
1148 * running, it is taken off the cpu as soon as possible.
1149 */
1150 ast_sched_locked(td, TDA_SCHED);
1151 break;
1152 case TDS_RUNNING:
1153 /*
1154 * If the thread is running, we request a context
1155 * switch for it by setting the TDA_SCHED flag.
1156 */
1157 ast_sched_locked(td, TDA_SCHED);
1158 #ifdef SMP
1159 cpuid = td->td_oncpu;
1160 if ((cpuid != NOCPU) && (td != curthread))
1161 ipi_cpu(cpuid, IPI_AST);
1162 #endif
1163 break;
1164 default:
1165 break;
1166 }
1167 thread_unlock(td);
1168 }
1169 }
1170
1171 static void
racct_proc_wakeup(struct proc * p)1172 racct_proc_wakeup(struct proc *p)
1173 {
1174
1175 ASSERT_RACCT_ENABLED();
1176
1177 PROC_LOCK_ASSERT(p, MA_OWNED);
1178
1179 if (p->p_throttled != 0) {
1180 p->p_throttled = 0;
1181 wakeup(p->p_racct);
1182 }
1183 }
1184
1185 static void
racct_decay_callback(struct racct * racct,void * dummy1,void * dummy2)1186 racct_decay_callback(struct racct *racct, void *dummy1, void *dummy2)
1187 {
1188 int64_t r_old, r_new;
1189
1190 ASSERT_RACCT_ENABLED();
1191 RACCT_LOCK_ASSERT();
1192
1193 #ifdef RCTL
1194 rctl_throttle_decay(racct, RACCT_READBPS);
1195 rctl_throttle_decay(racct, RACCT_WRITEBPS);
1196 rctl_throttle_decay(racct, RACCT_READIOPS);
1197 rctl_throttle_decay(racct, RACCT_WRITEIOPS);
1198 #endif
1199
1200 r_old = racct->r_resources[RACCT_PCTCPU];
1201
1202 /* If there is nothing to decay, just exit. */
1203 if (r_old <= 0)
1204 return;
1205
1206 r_new = r_old * RACCT_DECAY_FACTOR / FSCALE;
1207 racct->r_resources[RACCT_PCTCPU] = r_new;
1208 }
1209
1210 static void
racct_decay_pre(void)1211 racct_decay_pre(void)
1212 {
1213
1214 RACCT_LOCK();
1215 }
1216
1217 static void
racct_decay_post(void)1218 racct_decay_post(void)
1219 {
1220
1221 RACCT_UNLOCK();
1222 }
1223
1224 static void
racct_decay(void)1225 racct_decay(void)
1226 {
1227
1228 ASSERT_RACCT_ENABLED();
1229
1230 ui_racct_foreach(racct_decay_callback, racct_decay_pre,
1231 racct_decay_post, NULL, NULL);
1232 loginclass_racct_foreach(racct_decay_callback, racct_decay_pre,
1233 racct_decay_post, NULL, NULL);
1234 prison_racct_foreach(racct_decay_callback, racct_decay_pre,
1235 racct_decay_post, NULL, NULL);
1236 }
1237
1238 static void
racctd(void)1239 racctd(void)
1240 {
1241 struct thread *td;
1242 struct proc *p;
1243 struct timeval wallclock;
1244 uint64_t pct, pct_estimate, runtime;
1245
1246 ASSERT_RACCT_ENABLED();
1247
1248 for (;;) {
1249 racct_decay();
1250
1251 sx_slock(&allproc_lock);
1252
1253 FOREACH_PROC_IN_SYSTEM(p) {
1254 PROC_LOCK(p);
1255 if (p->p_state != PRS_NORMAL) {
1256 if (p->p_state == PRS_ZOMBIE)
1257 racct_set(p, RACCT_PCTCPU, 0);
1258 PROC_UNLOCK(p);
1259 continue;
1260 }
1261
1262 microuptime(&wallclock);
1263 timevalsub(&wallclock, &p->p_stats->p_start);
1264 PROC_STATLOCK(p);
1265 FOREACH_THREAD_IN_PROC(p, td)
1266 ruxagg(p, td);
1267 runtime = cputick2usec(p->p_rux.rux_runtime);
1268 PROC_STATUNLOCK(p);
1269 #ifdef notyet
1270 KASSERT(runtime >= p->p_prev_runtime,
1271 ("runtime < p_prev_runtime"));
1272 #else
1273 if (runtime < p->p_prev_runtime)
1274 runtime = p->p_prev_runtime;
1275 #endif
1276 p->p_prev_runtime = runtime;
1277 if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
1278 pct_estimate = (1000000 * runtime * 100) /
1279 ((uint64_t)wallclock.tv_sec * 1000000 +
1280 wallclock.tv_usec);
1281 } else
1282 pct_estimate = 0;
1283 pct = racct_getpcpu(p, pct_estimate);
1284 RACCT_LOCK();
1285 #ifdef RCTL
1286 rctl_throttle_decay(p->p_racct, RACCT_READBPS);
1287 rctl_throttle_decay(p->p_racct, RACCT_WRITEBPS);
1288 rctl_throttle_decay(p->p_racct, RACCT_READIOPS);
1289 rctl_throttle_decay(p->p_racct, RACCT_WRITEIOPS);
1290 #endif
1291 racct_set_locked(p, RACCT_PCTCPU, pct, 1);
1292 racct_set_locked(p, RACCT_CPU, runtime, 0);
1293 racct_set_locked(p, RACCT_WALLCLOCK,
1294 (uint64_t)wallclock.tv_sec * 1000000 +
1295 wallclock.tv_usec, 0);
1296 RACCT_UNLOCK();
1297 PROC_UNLOCK(p);
1298 }
1299
1300 /*
1301 * To ensure that processes are throttled in a fair way, we need
1302 * to iterate over all processes again and check the limits
1303 * for %cpu resource only after ucred racct containers have been
1304 * properly filled.
1305 */
1306 FOREACH_PROC_IN_SYSTEM(p) {
1307 PROC_LOCK(p);
1308 if (p->p_state != PRS_NORMAL) {
1309 PROC_UNLOCK(p);
1310 continue;
1311 }
1312
1313 if (racct_pcpu_available(p) <= 0) {
1314 if (p->p_racct->r_resources[RACCT_PCTCPU] >
1315 pcpu_threshold)
1316 racct_proc_throttle(p, -1);
1317 } else if (p->p_throttled == -1) {
1318 racct_proc_wakeup(p);
1319 }
1320 PROC_UNLOCK(p);
1321 }
1322 sx_sunlock(&allproc_lock);
1323 pause("-", hz);
1324 }
1325 }
1326
1327 static struct kproc_desc racctd_kp = {
1328 "racctd",
1329 racctd,
1330 NULL
1331 };
1332
1333 static void
racctd_init(void)1334 racctd_init(void)
1335 {
1336 if (!racct_enable)
1337 return;
1338
1339 kproc_start(&racctd_kp);
1340 }
1341 SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, racctd_init, NULL);
1342
1343 static void
racct_init(void)1344 racct_init(void)
1345 {
1346 if (!racct_enable)
1347 return;
1348
1349 racct_zone = uma_zcreate("racct", sizeof(struct racct),
1350 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1351 ast_register(TDA_RACCT, ASTR_ASTF_REQUIRED, 0, ast_racct);
1352
1353 /*
1354 * XXX: Move this somewhere.
1355 */
1356 prison0.pr_prison_racct = prison_racct_find("0");
1357 }
1358 SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL);
1359