kern_racct.c

Version: - FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
    1 /*-
    2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
    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  * $FreeBSD$
   31  */
   32 
   33 #include <sys/cdefs.h>
   34 __FBSDID("$FreeBSD$");
   35 
   36 #include "opt_sched.h"
   37 
   38 #include <sys/param.h>
   39 #include <sys/buf.h>
   40 #include <sys/systm.h>
   41 #include <sys/eventhandler.h>
   42 #include <sys/jail.h>
   43 #include <sys/kernel.h>
   44 #include <sys/kthread.h>
   45 #include <sys/lock.h>
   46 #include <sys/loginclass.h>
   47 #include <sys/malloc.h>
   48 #include <sys/mutex.h>
   49 #include <sys/proc.h>
   50 #include <sys/racct.h>
   51 #include <sys/resourcevar.h>
   52 #include <sys/sbuf.h>
   53 #include <sys/sched.h>
   54 #include <sys/sdt.h>
   55 #include <sys/smp.h>
   56 #include <sys/sx.h>
   57 #include <sys/sysctl.h>
   58 #include <sys/sysproto.h>
   59 #include <sys/umtxvar.h>
   60 #include <machine/smp.h>
   61 
   62 #ifdef RCTL
   63 #include <sys/rctl.h>
   64 #endif
   65 
   66 #ifdef RACCT
   67 
   68 FEATURE(racct, "Resource Accounting");
   69 
   70 /*
   71  * Do not block processes that have their %cpu usage <= pcpu_threshold.
   72  */
   73 static int pcpu_threshold = 1;
   74 #ifdef RACCT_DEFAULT_TO_DISABLED
   75 bool __read_frequently racct_enable = false;
   76 #else
   77 bool __read_frequently racct_enable = true;
   78 #endif
   79 
   80 SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
   81     "Resource Accounting");
   82 SYSCTL_BOOL(_kern_racct, OID_AUTO, enable, CTLFLAG_RDTUN, &racct_enable,
   83     0, "Enable RACCT/RCTL");
   84 SYSCTL_UINT(_kern_racct, OID_AUTO, pcpu_threshold, CTLFLAG_RW, &pcpu_threshold,
   85     0, "Processes with higher %cpu usage than this value can be throttled.");
   86 
   87 /*
   88  * How many seconds it takes to use the scheduler %cpu calculations.  When a
   89  * process starts, we compute its %cpu usage by dividing its runtime by the
   90  * process wall clock time.  After RACCT_PCPU_SECS pass, we use the value
   91  * provided by the scheduler.
   92  */
   93 #define RACCT_PCPU_SECS         3
   94 
   95 struct mtx racct_lock;
   96 MTX_SYSINIT(racct_lock, &racct_lock, "racct lock", MTX_DEF);
   97 
   98 static uma_zone_t racct_zone;
   99 
  100 static void racct_sub_racct(struct racct *dest, const struct racct *src);
  101 static void racct_sub_cred_locked(struct ucred *cred, int resource,
  102                 uint64_t amount);
  103 static void racct_add_cred_locked(struct ucred *cred, int resource,
  104                 uint64_t amount);
  105 
  106 SDT_PROVIDER_DEFINE(racct);
  107 SDT_PROBE_DEFINE3(racct, , rusage, add,
  108     "struct proc *", "int", "uint64_t");
  109 SDT_PROBE_DEFINE3(racct, , rusage, add__failure,
  110     "struct proc *", "int", "uint64_t");
  111 SDT_PROBE_DEFINE3(racct, , rusage, add__buf,
  112     "struct proc *", "const struct buf *", "int");
  113 SDT_PROBE_DEFINE3(racct, , rusage, add__cred,
  114     "struct ucred *", "int", "uint64_t");
  115 SDT_PROBE_DEFINE3(racct, , rusage, add__force,
  116     "struct proc *", "int", "uint64_t");
  117 SDT_PROBE_DEFINE3(racct, , rusage, set,
  118     "struct proc *", "int", "uint64_t");
  119 SDT_PROBE_DEFINE3(racct, , rusage, set__failure,
  120     "struct proc *", "int", "uint64_t");
  121 SDT_PROBE_DEFINE3(racct, , rusage, set__force,
  122     "struct proc *", "int", "uint64_t");
  123 SDT_PROBE_DEFINE3(racct, , rusage, sub,
  124     "struct proc *", "int", "uint64_t");
  125 SDT_PROBE_DEFINE3(racct, , rusage, sub__cred,
  126     "struct ucred *", "int", "uint64_t");
  127 SDT_PROBE_DEFINE1(racct, , racct, create,
  128     "struct racct *");
  129 SDT_PROBE_DEFINE1(racct, , racct, destroy,
  130     "struct racct *");
  131 SDT_PROBE_DEFINE2(racct, , racct, join,
  132     "struct racct *", "struct racct *");
  133 SDT_PROBE_DEFINE2(racct, , racct, join__failure,
  134     "struct racct *", "struct racct *");
  135 SDT_PROBE_DEFINE2(racct, , racct, leave,
  136     "struct racct *", "struct racct *");
  137 
  138 int racct_types[] = {
  139         [RACCT_CPU] =
  140                 RACCT_IN_MILLIONS,
  141         [RACCT_DATA] =
  142                 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
  143         [RACCT_STACK] =
  144                 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
  145         [RACCT_CORE] =
  146                 RACCT_DENIABLE,
  147         [RACCT_RSS] =
  148                 RACCT_RECLAIMABLE,
  149         [RACCT_MEMLOCK] =
  150                 RACCT_RECLAIMABLE | RACCT_DENIABLE,
  151         [RACCT_NPROC] =
  152                 RACCT_RECLAIMABLE | RACCT_DENIABLE,
  153         [RACCT_NOFILE] =
  154                 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
  155         [RACCT_VMEM] =
  156                 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
  157         [RACCT_NPTS] =
  158                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  159         [RACCT_SWAP] =
  160                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  161         [RACCT_NTHR] =
  162                 RACCT_RECLAIMABLE | RACCT_DENIABLE,
  163         [RACCT_MSGQQUEUED] =
  164                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  165         [RACCT_MSGQSIZE] =
  166                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  167         [RACCT_NMSGQ] =
  168                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  169         [RACCT_NSEM] =
  170                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  171         [RACCT_NSEMOP] =
  172                 RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE,
  173         [RACCT_NSHM] =
  174                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  175         [RACCT_SHMSIZE] =
  176                 RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY,
  177         [RACCT_WALLCLOCK] =
  178                 RACCT_IN_MILLIONS,
  179         [RACCT_PCTCPU] =
  180                 RACCT_DECAYING | RACCT_DENIABLE | RACCT_IN_MILLIONS,
  181         [RACCT_READBPS] =
  182                 RACCT_DECAYING,
  183         [RACCT_WRITEBPS] =
  184                 RACCT_DECAYING,
  185         [RACCT_READIOPS] =
  186                 RACCT_DECAYING,
  187         [RACCT_WRITEIOPS] =
  188                 RACCT_DECAYING };
  189 
  190 static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE;
  191 
  192 #ifdef SCHED_4BSD
  193 /*
  194  * Contains intermediate values for %cpu calculations to avoid using floating
  195  * point in the kernel.
  196  * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20)
  197  * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to
  198  * zero so the calculations are more straightforward.
  199  */
  200 fixpt_t ccpu_exp[] = {
  201         [0] = FSCALE * 1,
  202         [1] = FSCALE * 0.95122942450071400909,
  203         [2] = FSCALE * 0.90483741803595957316,
  204         [3] = FSCALE * 0.86070797642505780722,
  205         [4] = FSCALE * 0.81873075307798185866,
  206         [5] = FSCALE * 0.77880078307140486824,
  207         [6] = FSCALE * 0.74081822068171786606,
  208         [7] = FSCALE * 0.70468808971871343435,
  209         [8] = FSCALE * 0.67032004603563930074,
  210         [9] = FSCALE * 0.63762815162177329314,
  211         [10] = FSCALE * 0.60653065971263342360,
  212         [11] = FSCALE * 0.57694981038048669531,
  213         [12] = FSCALE * 0.54881163609402643262,
  214         [13] = FSCALE * 0.52204577676101604789,
  215         [14] = FSCALE * 0.49658530379140951470,
  216         [15] = FSCALE * 0.47236655274101470713,
  217         [16] = FSCALE * 0.44932896411722159143,
  218         [17] = FSCALE * 0.42741493194872666992,
  219         [18] = FSCALE * 0.40656965974059911188,
  220         [19] = FSCALE * 0.38674102345450120691,
  221         [20] = FSCALE * 0.36787944117144232159,
  222         [21] = FSCALE * 0.34993774911115535467,
  223         [22] = FSCALE * 0.33287108369807955328,
  224         [23] = FSCALE * 0.31663676937905321821,
  225         [24] = FSCALE * 0.30119421191220209664,
  226         [25] = FSCALE * 0.28650479686019010032,
  227         [26] = FSCALE * 0.27253179303401260312,
  228         [27] = FSCALE * 0.25924026064589150757,
  229         [28] = FSCALE * 0.24659696394160647693,
  230         [29] = FSCALE * 0.23457028809379765313,
  231         [30] = FSCALE * 0.22313016014842982893,
  232         [31] = FSCALE * 0.21224797382674305771,
  233         [32] = FSCALE * 0.20189651799465540848,
  234         [33] = FSCALE * 0.19204990862075411423,
  235         [34] = FSCALE * 0.18268352405273465022,
  236         [35] = FSCALE * 0.17377394345044512668,
  237         [36] = FSCALE * 0.16529888822158653829,
  238         [37] = FSCALE * 0.15723716631362761621,
  239         [38] = FSCALE * 0.14956861922263505264,
  240         [39] = FSCALE * 0.14227407158651357185,
  241         [40] = FSCALE * 0.13533528323661269189,
  242         [41] = FSCALE * 0.12873490358780421886,
  243         [42] = FSCALE * 0.12245642825298191021,
  244         [43] = FSCALE * 0.11648415777349695786,
  245         [44] = FSCALE * 0.11080315836233388333,
  246         [45] = FSCALE * 0.10539922456186433678,
  247         [46] = FSCALE * 0.10025884372280373372,
  248         [47] = FSCALE * 0.09536916221554961888,
  249         [48] = FSCALE * 0.09071795328941250337,
  250         [49] = FSCALE * 0.08629358649937051097,
  251         [50] = FSCALE * 0.08208499862389879516,
  252         [51] = FSCALE * 0.07808166600115315231,
  253         [52] = FSCALE * 0.07427357821433388042,
  254         [53] = FSCALE * 0.07065121306042958674,
  255         [54] = FSCALE * 0.06720551273974976512,
  256         [55] = FSCALE * 0.06392786120670757270,
  257         [56] = FSCALE * 0.06081006262521796499,
  258         [57] = FSCALE * 0.05784432087483846296,
  259         [58] = FSCALE * 0.05502322005640722902,
  260         [59] = FSCALE * 0.05233970594843239308,
  261         [60] = FSCALE * 0.04978706836786394297,
  262         [61] = FSCALE * 0.04735892439114092119,
  263         [62] = FSCALE * 0.04504920239355780606,
  264         [63] = FSCALE * 0.04285212686704017991,
  265         [64] = FSCALE * 0.04076220397836621516,
  266         [65] = FSCALE * 0.03877420783172200988,
  267         [66] = FSCALE * 0.03688316740124000544,
  268         [67] = FSCALE * 0.03508435410084502588,
  269         [68] = FSCALE * 0.03337326996032607948,
  270         [69] = FSCALE * 0.03174563637806794323,
  271         [70] = FSCALE * 0.03019738342231850073,
  272         [71] = FSCALE * 0.02872463965423942912,
  273         [72] = FSCALE * 0.02732372244729256080,
  274         [73] = FSCALE * 0.02599112877875534358,
  275         [74] = FSCALE * 0.02472352647033939120,
  276         [75] = FSCALE * 0.02351774585600910823,
  277         [76] = FSCALE * 0.02237077185616559577,
  278         [77] = FSCALE * 0.02127973643837716938,
  279         [78] = FSCALE * 0.02024191144580438847,
  280         [79] = FSCALE * 0.01925470177538692429,
  281         [80] = FSCALE * 0.01831563888873418029,
  282         [81] = FSCALE * 0.01742237463949351138,
  283         [82] = FSCALE * 0.01657267540176124754,
  284         [83] = FSCALE * 0.01576441648485449082,
  285         [84] = FSCALE * 0.01499557682047770621,
  286         [85] = FSCALE * 0.01426423390899925527,
  287         [86] = FSCALE * 0.01356855901220093175,
  288         [87] = FSCALE * 0.01290681258047986886,
  289         [88] = FSCALE * 0.01227733990306844117,
  290         [89] = FSCALE * 0.01167856697039544521,
  291         [90] = FSCALE * 0.01110899653824230649,
  292         [91] = FSCALE * 0.01056720438385265337,
  293         [92] = FSCALE * 0.01005183574463358164,
  294         [93] = FSCALE * 0.00956160193054350793,
  295         [94] = FSCALE * 0.00909527710169581709,
  296         [95] = FSCALE * 0.00865169520312063417,
  297         [96] = FSCALE * 0.00822974704902002884,
  298         [97] = FSCALE * 0.00782837754922577143,
  299         [98] = FSCALE * 0.00744658307092434051,
  300         [99] = FSCALE * 0.00708340892905212004,
  301         [100] = FSCALE * 0.00673794699908546709,
  302         [101] = FSCALE * 0.00640933344625638184,
  303         [102] = FSCALE * 0.00609674656551563610,
  304         [103] = FSCALE * 0.00579940472684214321,
  305         [104] = FSCALE * 0.00551656442076077241,
  306         [105] = FSCALE * 0.00524751839918138427,
  307         [106] = FSCALE * 0.00499159390691021621,
  308         [107] = FSCALE * 0.00474815099941147558,
  309         [108] = FSCALE * 0.00451658094261266798,
  310         [109] = FSCALE * 0.00429630469075234057,
  311         [110] = FSCALE * 0.00408677143846406699,
  312 };
  313 #endif
  314 
  315 #define CCPU_EXP_MAX    110
  316 
  317 /*
  318  * This function is analogical to the getpcpu() function in the ps(1) command.
  319  * They should both calculate in the same way so that the racct %cpu
  320  * calculations are consistent with the values showed by the ps(1) tool.
  321  * The calculations are more complex in the 4BSD scheduler because of the value
  322  * of the ccpu variable.  In ULE it is defined to be zero which saves us some
  323  * work.
  324  */
  325 static uint64_t
  326 racct_getpcpu(struct proc *p, u_int pcpu)
  327 {
  328         u_int swtime;
  329 #ifdef SCHED_4BSD
  330         fixpt_t pctcpu, pctcpu_next;
  331 #endif
  332 #ifdef SMP
  333         struct pcpu *pc;
  334         int found;
  335 #endif
  336         fixpt_t p_pctcpu;
  337         struct thread *td;
  338 
  339         ASSERT_RACCT_ENABLED();
  340 
  341         /*
  342          * If the process is swapped out, we count its %cpu usage as zero.
  343          * This behaviour is consistent with the userland ps(1) tool.
  344          */
  345         if ((p->p_flag & P_INMEM) == 0)
  346                 return (0);
  347         swtime = (ticks - p->p_swtick) / hz;
  348 
  349         /*
  350          * For short-lived processes, the sched_pctcpu() returns small
  351          * values even for cpu intensive processes.  Therefore we use
  352          * our own estimate in this case.
  353          */
  354         if (swtime < RACCT_PCPU_SECS)
  355                 return (pcpu);
  356 
  357         p_pctcpu = 0;
  358         FOREACH_THREAD_IN_PROC(p, td) {
  359                 if (td == PCPU_GET(idlethread))
  360                         continue;
  361 #ifdef SMP
  362                 found = 0;
  363                 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
  364                         if (td == pc->pc_idlethread) {
  365                                 found = 1;
  366                                 break;
  367                         }
  368                 }
  369                 if (found)
  370                         continue;
  371 #endif
  372                 thread_lock(td);
  373 #ifdef SCHED_4BSD
  374                 pctcpu = sched_pctcpu(td);
  375                 /* Count also the yet unfinished second. */
  376                 pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT;
  377                 pctcpu_next += sched_pctcpu_delta(td);
  378                 p_pctcpu += max(pctcpu, pctcpu_next);
  379 #else
  380                 /*
  381                  * In ULE the %cpu statistics are updated on every
  382                  * sched_pctcpu() call.  So special calculations to
  383                  * account for the latest (unfinished) second are
  384                  * not needed.
  385                  */
  386                 p_pctcpu += sched_pctcpu(td);
  387 #endif
  388                 thread_unlock(td);
  389         }
  390 
  391 #ifdef SCHED_4BSD
  392         if (swtime <= CCPU_EXP_MAX)
  393                 return ((100 * (uint64_t)p_pctcpu * 1000000) /
  394                     (FSCALE - ccpu_exp[swtime]));
  395 #endif
  396 
  397         return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE);
  398 }
  399 
  400 static void
  401 racct_add_racct(struct racct *dest, const struct racct *src)
  402 {
  403         int i;
  404 
  405         ASSERT_RACCT_ENABLED();
  406         RACCT_LOCK_ASSERT();
  407 
  408         /*
  409          * Update resource usage in dest.
  410          */
  411         for (i = 0; i <= RACCT_MAX; i++) {
  412                 KASSERT(dest->r_resources[i] >= 0,
  413                     ("%s: resource %d propagation meltdown: dest < 0",
  414                     __func__, i));
  415                 KASSERT(src->r_resources[i] >= 0,
  416                     ("%s: resource %d propagation meltdown: src < 0",
  417                     __func__, i));
  418                 dest->r_resources[i] += src->r_resources[i];
  419         }
  420 }
  421 
  422 static void
  423 racct_sub_racct(struct racct *dest, const struct racct *src)
  424 {
  425         int i;
  426 
  427         ASSERT_RACCT_ENABLED();
  428         RACCT_LOCK_ASSERT();
  429 
  430         /*
  431          * Update resource usage in dest.
  432          */
  433         for (i = 0; i <= RACCT_MAX; i++) {
  434                 if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) {
  435                         KASSERT(dest->r_resources[i] >= 0,
  436                             ("%s: resource %d propagation meltdown: dest < 0",
  437                             __func__, i));
  438                         KASSERT(src->r_resources[i] >= 0,
  439                             ("%s: resource %d propagation meltdown: src < 0",
  440                             __func__, i));
  441                         KASSERT(src->r_resources[i] <= dest->r_resources[i],
  442                             ("%s: resource %d propagation meltdown: src > dest",
  443                             __func__, i));
  444                 }
  445                 if (RACCT_CAN_DROP(i)) {
  446                         dest->r_resources[i] -= src->r_resources[i];
  447                         if (dest->r_resources[i] < 0)
  448                                 dest->r_resources[i] = 0;
  449                 }
  450         }
  451 }
  452 
  453 void
  454 racct_create(struct racct **racctp)
  455 {
  456 
  457         if (!racct_enable)
  458                 return;
  459 
  460         SDT_PROBE1(racct, , racct, create, racctp);
  461 
  462         KASSERT(*racctp == NULL, ("racct already allocated"));
  463 
  464         *racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO);
  465 }
  466 
  467 static void
  468 racct_destroy_locked(struct racct **racctp)
  469 {
  470         struct racct *racct;
  471         int i;
  472 
  473         ASSERT_RACCT_ENABLED();
  474 
  475         SDT_PROBE1(racct, , racct, destroy, racctp);
  476 
  477         RACCT_LOCK_ASSERT();
  478         KASSERT(racctp != NULL, ("NULL racctp"));
  479         KASSERT(*racctp != NULL, ("NULL racct"));
  480 
  481         racct = *racctp;
  482 
  483         for (i = 0; i <= RACCT_MAX; i++) {
  484                 if (RACCT_IS_SLOPPY(i))
  485                         continue;
  486                 if (!RACCT_IS_RECLAIMABLE(i))
  487                         continue;
  488                 KASSERT(racct->r_resources[i] == 0,
  489                     ("destroying non-empty racct: "
  490                     "%ju allocated for resource %d\n",
  491                     racct->r_resources[i], i));
  492         }
  493         uma_zfree(racct_zone, racct);
  494         *racctp = NULL;
  495 }
  496 
  497 void
  498 racct_destroy(struct racct **racct)
  499 {
  500 
  501         if (!racct_enable)
  502                 return;
  503 
  504         RACCT_LOCK();
  505         racct_destroy_locked(racct);
  506         RACCT_UNLOCK();
  507 }
  508 
  509 /*
  510  * Increase consumption of 'resource' by 'amount' for 'racct',
  511  * but not its parents.  Differently from other cases, 'amount' here
  512  * may be less than zero.
  513  */
  514 static void
  515 racct_adjust_resource(struct racct *racct, int resource,
  516     int64_t amount)
  517 {
  518 
  519         ASSERT_RACCT_ENABLED();
  520         RACCT_LOCK_ASSERT();
  521         KASSERT(racct != NULL, ("NULL racct"));
  522 
  523         racct->r_resources[resource] += amount;
  524         if (racct->r_resources[resource] < 0) {
  525                 KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource),
  526                     ("%s: resource %d usage < 0", __func__, resource));
  527                 racct->r_resources[resource] = 0;
  528         }
  529 
  530         /*
  531          * There are some cases where the racct %cpu resource would grow
  532          * beyond 100% per core.  For example in racct_proc_exit() we add
  533          * the process %cpu usage to the ucred racct containers.  If too
  534          * many processes terminated in a short time span, the ucred %cpu
  535          * resource could grow too much.  Also, the 4BSD scheduler sometimes
  536          * returns for a thread more than 100% cpu usage. So we set a sane
  537          * boundary here to 100% * the maximum number of CPUs.
  538          */
  539         if ((resource == RACCT_PCTCPU) &&
  540             (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000 * (int64_t)MAXCPU))
  541                 racct->r_resources[RACCT_PCTCPU] = 100 * 1000000 * (int64_t)MAXCPU;
  542 }
  543 
  544 static int
  545 racct_add_locked(struct proc *p, int resource, uint64_t amount, int force)
  546 {
  547 #ifdef RCTL
  548         int error;
  549 #endif
  550 
  551         ASSERT_RACCT_ENABLED();
  552 
  553         /*
  554          * We need proc lock to dereference p->p_ucred.
  555          */
  556         PROC_LOCK_ASSERT(p, MA_OWNED);
  557 
  558 #ifdef RCTL
  559         error = rctl_enforce(p, resource, amount);
  560         if (error && !force && RACCT_IS_DENIABLE(resource)) {
  561                 SDT_PROBE3(racct, , rusage, add__failure, p, resource, amount);
  562                 return (error);
  563         }
  564 #endif
  565         racct_adjust_resource(p->p_racct, resource, amount);
  566         racct_add_cred_locked(p->p_ucred, resource, amount);
  567 
  568         return (0);
  569 }
  570 
  571 /*
  572  * Increase allocation of 'resource' by 'amount' for process 'p'.
  573  * Return 0 if it's below limits, or errno, if it's not.
  574  */
  575 int
  576 racct_add(struct proc *p, int resource, uint64_t amount)
  577 {
  578         int error;
  579 
  580         if (!racct_enable)
  581                 return (0);
  582 
  583         SDT_PROBE3(racct, , rusage, add, p, resource, amount);
  584 
  585         RACCT_LOCK();
  586         error = racct_add_locked(p, resource, amount, 0);
  587         RACCT_UNLOCK();
  588         return (error);
  589 }
  590 
  591 /*
  592  * Increase allocation of 'resource' by 'amount' for process 'p'.
  593  * Doesn't check for limits and never fails.
  594  */
  595 void
  596 racct_add_force(struct proc *p, int resource, uint64_t amount)
  597 {
  598 
  599         if (!racct_enable)
  600                 return;
  601 
  602         SDT_PROBE3(racct, , rusage, add__force, p, resource, amount);
  603 
  604         RACCT_LOCK();
  605         racct_add_locked(p, resource, amount, 1);
  606         RACCT_UNLOCK();
  607 }
  608 
  609 static void
  610 racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount)
  611 {
  612         struct prison *pr;
  613 
  614         ASSERT_RACCT_ENABLED();
  615 
  616         racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, amount);
  617         for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
  618                 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
  619                     amount);
  620         racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, amount);
  621 }
  622 
  623 /*
  624  * Increase allocation of 'resource' by 'amount' for credential 'cred'.
  625  * Doesn't check for limits and never fails.
  626  */
  627 void
  628 racct_add_cred(struct ucred *cred, int resource, uint64_t amount)
  629 {
  630 
  631         if (!racct_enable)
  632                 return;
  633 
  634         SDT_PROBE3(racct, , rusage, add__cred, cred, resource, amount);
  635 
  636         RACCT_LOCK();
  637         racct_add_cred_locked(cred, resource, amount);
  638         RACCT_UNLOCK();
  639 }
  640 
  641 /*
  642  * Account for disk IO resource consumption.  Checks for limits,
  643  * but never fails, due to disk limits being undeniable.
  644  */
  645 void
  646 racct_add_buf(struct proc *p, const struct buf *bp, int is_write)
  647 {
  648 
  649         ASSERT_RACCT_ENABLED();
  650         PROC_LOCK_ASSERT(p, MA_OWNED);
  651 
  652         SDT_PROBE3(racct, , rusage, add__buf, p, bp, is_write);
  653 
  654         RACCT_LOCK();
  655         if (is_write) {
  656                 racct_add_locked(curproc, RACCT_WRITEBPS, bp->b_bcount, 1);
  657                 racct_add_locked(curproc, RACCT_WRITEIOPS, 1, 1);
  658         } else {
  659                 racct_add_locked(curproc, RACCT_READBPS, bp->b_bcount, 1);
  660                 racct_add_locked(curproc, RACCT_READIOPS, 1, 1);
  661         }
  662         RACCT_UNLOCK();
  663 }
  664 
  665 static int
  666 racct_set_locked(struct proc *p, int resource, uint64_t amount, int force)
  667 {
  668         int64_t old_amount, decayed_amount, diff_proc, diff_cred;
  669 #ifdef RCTL
  670         int error;
  671 #endif
  672 
  673         ASSERT_RACCT_ENABLED();
  674 
  675         /*
  676          * We need proc lock to dereference p->p_ucred.
  677          */
  678         PROC_LOCK_ASSERT(p, MA_OWNED);
  679 
  680         old_amount = p->p_racct->r_resources[resource];
  681         /*
  682          * The diffs may be negative.
  683          */
  684         diff_proc = amount - old_amount;
  685         if (resource == RACCT_PCTCPU) {
  686                 /*
  687                  * Resources in per-credential racct containers may decay.
  688                  * If this is the case, we need to calculate the difference
  689                  * between the new amount and the proportional value of the
  690                  * old amount that has decayed in the ucred racct containers.
  691                  */
  692                 decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE;
  693                 diff_cred = amount - decayed_amount;
  694         } else
  695                 diff_cred = diff_proc;
  696 #ifdef notyet
  697         KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource),
  698             ("%s: usage of non-droppable resource %d dropping", __func__,
  699              resource));
  700 #endif
  701 #ifdef RCTL
  702         if (diff_proc > 0) {
  703                 error = rctl_enforce(p, resource, diff_proc);
  704                 if (error && !force && RACCT_IS_DENIABLE(resource)) {
  705                         SDT_PROBE3(racct, , rusage, set__failure, p, resource,
  706                             amount);
  707                         return (error);
  708                 }
  709         }
  710 #endif
  711         racct_adjust_resource(p->p_racct, resource, diff_proc);
  712         if (diff_cred > 0)
  713                 racct_add_cred_locked(p->p_ucred, resource, diff_cred);
  714         else if (diff_cred < 0)
  715                 racct_sub_cred_locked(p->p_ucred, resource, -diff_cred);
  716 
  717         return (0);
  718 }
  719 
  720 /*
  721  * Set allocation of 'resource' to 'amount' for process 'p'.
  722  * Return 0 if it's below limits, or errno, if it's not.
  723  *
  724  * Note that decreasing the allocation always returns 0,
  725  * even if it's above the limit.
  726  */
  727 int
  728 racct_set_unlocked(struct proc *p, int resource, uint64_t amount)
  729 {
  730         int error;
  731 
  732         ASSERT_RACCT_ENABLED();
  733         PROC_LOCK(p);
  734         error = racct_set(p, resource, amount);
  735         PROC_UNLOCK(p);
  736         return (error);
  737 }
  738 
  739 int
  740 racct_set(struct proc *p, int resource, uint64_t amount)
  741 {
  742         int error;
  743 
  744         if (!racct_enable)
  745                 return (0);
  746 
  747         SDT_PROBE3(racct, , rusage, set__force, p, resource, amount);
  748 
  749         RACCT_LOCK();
  750         error = racct_set_locked(p, resource, amount, 0);
  751         RACCT_UNLOCK();
  752         return (error);
  753 }
  754 
  755 void
  756 racct_set_force(struct proc *p, int resource, uint64_t amount)
  757 {
  758 
  759         if (!racct_enable)
  760                 return;
  761 
  762         SDT_PROBE3(racct, , rusage, set, p, resource, amount);
  763 
  764         RACCT_LOCK();
  765         racct_set_locked(p, resource, amount, 1);
  766         RACCT_UNLOCK();
  767 }
  768 
  769 /*
  770  * Returns amount of 'resource' the process 'p' can keep allocated.
  771  * Allocating more than that would be denied, unless the resource
  772  * is marked undeniable.  Amount of already allocated resource does
  773  * not matter.
  774  */
  775 uint64_t
  776 racct_get_limit(struct proc *p, int resource)
  777 {
  778 #ifdef RCTL
  779         uint64_t available;
  780 
  781         if (!racct_enable)
  782                 return (UINT64_MAX);
  783 
  784         RACCT_LOCK();
  785         available = rctl_get_limit(p, resource);
  786         RACCT_UNLOCK();
  787 
  788         return (available);
  789 #else
  790 
  791         return (UINT64_MAX);
  792 #endif
  793 }
  794 
  795 /*
  796  * Returns amount of 'resource' the process 'p' can keep allocated.
  797  * Allocating more than that would be denied, unless the resource
  798  * is marked undeniable.  Amount of already allocated resource does
  799  * matter.
  800  */
  801 uint64_t
  802 racct_get_available(struct proc *p, int resource)
  803 {
  804 #ifdef RCTL
  805         uint64_t available;
  806 
  807         if (!racct_enable)
  808                 return (UINT64_MAX);
  809 
  810         RACCT_LOCK();
  811         available = rctl_get_available(p, resource);
  812         RACCT_UNLOCK();
  813 
  814         return (available);
  815 #else
  816 
  817         return (UINT64_MAX);
  818 #endif
  819 }
  820 
  821 /*
  822  * Returns amount of the %cpu resource that process 'p' can add to its %cpu
  823  * utilization.  Adding more than that would lead to the process being
  824  * throttled.
  825  */
  826 static int64_t
  827 racct_pcpu_available(struct proc *p)
  828 {
  829 #ifdef RCTL
  830         uint64_t available;
  831 
  832         ASSERT_RACCT_ENABLED();
  833 
  834         RACCT_LOCK();
  835         available = rctl_pcpu_available(p);
  836         RACCT_UNLOCK();
  837 
  838         return (available);
  839 #else
  840 
  841         return (INT64_MAX);
  842 #endif
  843 }
  844 
  845 /*
  846  * Decrease allocation of 'resource' by 'amount' for process 'p'.
  847  */
  848 void
  849 racct_sub(struct proc *p, int resource, uint64_t amount)
  850 {
  851 
  852         if (!racct_enable)
  853                 return;
  854 
  855         SDT_PROBE3(racct, , rusage, sub, p, resource, amount);
  856 
  857         /*
  858          * We need proc lock to dereference p->p_ucred.
  859          */
  860         PROC_LOCK_ASSERT(p, MA_OWNED);
  861         KASSERT(RACCT_CAN_DROP(resource),
  862             ("%s: called for non-droppable resource %d", __func__, resource));
  863 
  864         RACCT_LOCK();
  865         KASSERT(amount <= p->p_racct->r_resources[resource],
  866             ("%s: freeing %ju of resource %d, which is more "
  867              "than allocated %jd for %s (pid %d)", __func__, amount, resource,
  868             (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid));
  869 
  870         racct_adjust_resource(p->p_racct, resource, -amount);
  871         racct_sub_cred_locked(p->p_ucred, resource, amount);
  872         RACCT_UNLOCK();
  873 }
  874 
  875 static void
  876 racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount)
  877 {
  878         struct prison *pr;
  879 
  880         ASSERT_RACCT_ENABLED();
  881 
  882         racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, -amount);
  883         for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
  884                 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
  885                     -amount);
  886         racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, -amount);
  887 }
  888 
  889 /*
  890  * Decrease allocation of 'resource' by 'amount' for credential 'cred'.
  891  */
  892 void
  893 racct_sub_cred(struct ucred *cred, int resource, uint64_t amount)
  894 {
  895 
  896         if (!racct_enable)
  897                 return;
  898 
  899         SDT_PROBE3(racct, , rusage, sub__cred, cred, resource, amount);
  900 
  901 #ifdef notyet
  902         KASSERT(RACCT_CAN_DROP(resource),
  903             ("%s: called for resource %d which can not drop", __func__,
  904              resource));
  905 #endif
  906 
  907         RACCT_LOCK();
  908         racct_sub_cred_locked(cred, resource, amount);
  909         RACCT_UNLOCK();
  910 }
  911 
  912 /*
  913  * Inherit resource usage information from the parent process.
  914  */
  915 int
  916 racct_proc_fork(struct proc *parent, struct proc *child)
  917 {
  918         int i, error = 0;
  919 
  920         if (!racct_enable)
  921                 return (0);
  922 
  923         /*
  924          * Create racct for the child process.
  925          */
  926         racct_create(&child->p_racct);
  927 
  928         PROC_LOCK(parent);
  929         PROC_LOCK(child);
  930         RACCT_LOCK();
  931 
  932 #ifdef RCTL
  933         error = rctl_proc_fork(parent, child);
  934         if (error != 0)
  935                 goto out;
  936 #endif
  937 
  938         /* Init process cpu time. */
  939         child->p_prev_runtime = 0;
  940         child->p_throttled = 0;
  941 
  942         /*
  943          * Inherit resource usage.
  944          */
  945         for (i = 0; i <= RACCT_MAX; i++) {
  946                 if (parent->p_racct->r_resources[i] == 0 ||
  947                     !RACCT_IS_INHERITABLE(i))
  948                         continue;
  949 
  950                 error = racct_set_locked(child, i,
  951                     parent->p_racct->r_resources[i], 0);
  952                 if (error != 0)
  953                         goto out;
  954         }
  955 
  956         error = racct_add_locked(child, RACCT_NPROC, 1, 0);
  957         error += racct_add_locked(child, RACCT_NTHR, 1, 0);
  958 
  959 out:
  960         RACCT_UNLOCK();
  961         PROC_UNLOCK(child);
  962         PROC_UNLOCK(parent);
  963 
  964         if (error != 0)
  965                 racct_proc_exit(child);
  966 
  967         return (error);
  968 }
  969 
  970 /*
  971  * Called at the end of fork1(), to handle rules that require the process
  972  * to be fully initialized.
  973  */
  974 void
  975 racct_proc_fork_done(struct proc *child)
  976 {
  977 
  978         if (!racct_enable)
  979                 return;
  980 
  981 #ifdef RCTL
  982         PROC_LOCK(child);
  983         RACCT_LOCK();
  984         rctl_enforce(child, RACCT_NPROC, 0);
  985         rctl_enforce(child, RACCT_NTHR, 0);
  986         RACCT_UNLOCK();
  987         PROC_UNLOCK(child);
  988 #endif
  989 }
  990 
  991 void
  992 racct_proc_exit(struct proc *p)
  993 {
  994         struct timeval wallclock;
  995         uint64_t pct_estimate, pct, runtime;
  996         int i;
  997 
  998         if (!racct_enable)
  999                 return;
 1000 
 1001         PROC_LOCK(p);
 1002         /*
 1003          * We don't need to calculate rux, proc_reap() has already done this.
 1004          */
 1005         runtime = cputick2usec(p->p_rux.rux_runtime);
 1006 #ifdef notyet
 1007         KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime"));
 1008 #else
 1009         if (runtime < p->p_prev_runtime)
 1010                 runtime = p->p_prev_runtime;
 1011 #endif
 1012         microuptime(&wallclock);
 1013         timevalsub(&wallclock, &p->p_stats->p_start);
 1014         if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
 1015                 pct_estimate = (1000000 * runtime * 100) /
 1016                     ((uint64_t)wallclock.tv_sec * 1000000 +
 1017                     wallclock.tv_usec);
 1018         } else
 1019                 pct_estimate = 0;
 1020         pct = racct_getpcpu(p, pct_estimate);
 1021 
 1022         RACCT_LOCK();
 1023         racct_set_locked(p, RACCT_CPU, runtime, 0);
 1024         racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct);
 1025 
 1026         KASSERT(p->p_racct->r_resources[RACCT_RSS] == 0,
 1027             ("process reaped with %ju allocated for RSS\n",
 1028             p->p_racct->r_resources[RACCT_RSS]));
 1029         for (i = 0; i <= RACCT_MAX; i++) {
 1030                 if (p->p_racct->r_resources[i] == 0)
 1031                         continue;
 1032                 if (!RACCT_IS_RECLAIMABLE(i))
 1033                         continue;
 1034                 racct_set_locked(p, i, 0, 0);
 1035         }
 1036 
 1037 #ifdef RCTL
 1038         rctl_racct_release(p->p_racct);
 1039 #endif
 1040         racct_destroy_locked(&p->p_racct);
 1041         RACCT_UNLOCK();
 1042         PROC_UNLOCK(p);
 1043 }
 1044 
 1045 /*
 1046  * Called after credentials change, to move resource utilisation
 1047  * between raccts.
 1048  */
 1049 void
 1050 racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred,
 1051     struct ucred *newcred)
 1052 {
 1053         struct uidinfo *olduip, *newuip;
 1054         struct loginclass *oldlc, *newlc;
 1055         struct prison *oldpr, *newpr, *pr;
 1056 
 1057         if (!racct_enable)
 1058                 return;
 1059 
 1060         PROC_LOCK_ASSERT(p, MA_OWNED);
 1061 
 1062         newuip = newcred->cr_ruidinfo;
 1063         olduip = oldcred->cr_ruidinfo;
 1064         newlc = newcred->cr_loginclass;
 1065         oldlc = oldcred->cr_loginclass;
 1066         newpr = newcred->cr_prison;
 1067         oldpr = oldcred->cr_prison;
 1068 
 1069         RACCT_LOCK();
 1070         if (newuip != olduip) {
 1071                 racct_sub_racct(olduip->ui_racct, p->p_racct);
 1072                 racct_add_racct(newuip->ui_racct, p->p_racct);
 1073         }
 1074         if (newlc != oldlc) {
 1075                 racct_sub_racct(oldlc->lc_racct, p->p_racct);
 1076                 racct_add_racct(newlc->lc_racct, p->p_racct);
 1077         }
 1078         if (newpr != oldpr) {
 1079                 for (pr = oldpr; pr != NULL; pr = pr->pr_parent)
 1080                         racct_sub_racct(pr->pr_prison_racct->prr_racct,
 1081                             p->p_racct);
 1082                 for (pr = newpr; pr != NULL; pr = pr->pr_parent)
 1083                         racct_add_racct(pr->pr_prison_racct->prr_racct,
 1084                             p->p_racct);
 1085         }
 1086         RACCT_UNLOCK();
 1087 }
 1088 
 1089 void
 1090 racct_move(struct racct *dest, struct racct *src)
 1091 {
 1092 
 1093         ASSERT_RACCT_ENABLED();
 1094 
 1095         RACCT_LOCK();
 1096         racct_add_racct(dest, src);
 1097         racct_sub_racct(src, src);
 1098         RACCT_UNLOCK();
 1099 }
 1100 
 1101 void
 1102 racct_proc_throttled(struct proc *p)
 1103 {
 1104 
 1105         ASSERT_RACCT_ENABLED();
 1106 
 1107         PROC_LOCK(p);
 1108         while (p->p_throttled != 0) {
 1109                 msleep(p->p_racct, &p->p_mtx, 0, "racct",
 1110                     p->p_throttled < 0 ? 0 : p->p_throttled);
 1111                 if (p->p_throttled > 0)
 1112                         p->p_throttled = 0;
 1113         }
 1114         PROC_UNLOCK(p);
 1115 }
 1116 
 1117 /*
 1118  * Make the process sleep in userret() for 'timeout' ticks.  Setting
 1119  * timeout to -1 makes it sleep until woken up by racct_proc_wakeup().
 1120  */
 1121 void
 1122 racct_proc_throttle(struct proc *p, int timeout)
 1123 {
 1124         struct thread *td;
 1125 #ifdef SMP
 1126         int cpuid;
 1127 #endif
 1128 
 1129         KASSERT(timeout != 0, ("timeout %d", timeout));
 1130         ASSERT_RACCT_ENABLED();
 1131         PROC_LOCK_ASSERT(p, MA_OWNED);
 1132 
 1133         /*
 1134          * Do not block kernel processes.  Also do not block processes with
 1135          * low %cpu utilization to improve interactivity.
 1136          */
 1137         if ((p->p_flag & (P_SYSTEM | P_KPROC)) != 0)
 1138                 return;
 1139 
 1140         if (p->p_throttled < 0 || (timeout > 0 && p->p_throttled > timeout))
 1141                 return;
 1142 
 1143         p->p_throttled = timeout;
 1144 
 1145         FOREACH_THREAD_IN_PROC(p, td) {
 1146                 thread_lock(td);
 1147                 td->td_flags |= TDF_ASTPENDING;
 1148 
 1149                 switch (td->td_state) {
 1150                 case TDS_RUNQ:
 1151                         /*
 1152                          * If the thread is on the scheduler run-queue, we can
 1153                          * not just remove it from there.  So we set the flag
 1154                          * TDF_NEEDRESCHED for the thread, so that once it is
 1155                          * running, it is taken off the cpu as soon as possible.
 1156                          */
 1157                         td->td_flags |= TDF_NEEDRESCHED;
 1158                         break;
 1159                 case TDS_RUNNING:
 1160                         /*
 1161                          * If the thread is running, we request a context
 1162                          * switch for it by setting the TDF_NEEDRESCHED flag.
 1163                          */
 1164                         td->td_flags |= TDF_NEEDRESCHED;
 1165 #ifdef SMP
 1166                         cpuid = td->td_oncpu;
 1167                         if ((cpuid != NOCPU) && (td != curthread))
 1168                                 ipi_cpu(cpuid, IPI_AST);
 1169 #endif
 1170                         break;
 1171                 default:
 1172                         break;
 1173                 }
 1174                 thread_unlock(td);
 1175         }
 1176 }
 1177 
 1178 static void
 1179 racct_proc_wakeup(struct proc *p)
 1180 {
 1181 
 1182         ASSERT_RACCT_ENABLED();
 1183 
 1184         PROC_LOCK_ASSERT(p, MA_OWNED);
 1185 
 1186         if (p->p_throttled != 0) {
 1187                 p->p_throttled = 0;
 1188                 wakeup(p->p_racct);
 1189         }
 1190 }
 1191 
 1192 static void
 1193 racct_decay_callback(struct racct *racct, void *dummy1, void *dummy2)
 1194 {
 1195         int64_t r_old, r_new;
 1196 
 1197         ASSERT_RACCT_ENABLED();
 1198         RACCT_LOCK_ASSERT();
 1199 
 1200 #ifdef RCTL
 1201         rctl_throttle_decay(racct, RACCT_READBPS);
 1202         rctl_throttle_decay(racct, RACCT_WRITEBPS);
 1203         rctl_throttle_decay(racct, RACCT_READIOPS);
 1204         rctl_throttle_decay(racct, RACCT_WRITEIOPS);
 1205 #endif
 1206 
 1207         r_old = racct->r_resources[RACCT_PCTCPU];
 1208 
 1209         /* If there is nothing to decay, just exit. */
 1210         if (r_old <= 0)
 1211                 return;
 1212 
 1213         r_new = r_old * RACCT_DECAY_FACTOR / FSCALE;
 1214         racct->r_resources[RACCT_PCTCPU] = r_new;
 1215 }
 1216 
 1217 static void
 1218 racct_decay_pre(void)
 1219 {
 1220 
 1221         RACCT_LOCK();
 1222 }
 1223 
 1224 static void
 1225 racct_decay_post(void)
 1226 {
 1227 
 1228         RACCT_UNLOCK();
 1229 }
 1230 
 1231 static void
 1232 racct_decay(void)
 1233 {
 1234 
 1235         ASSERT_RACCT_ENABLED();
 1236 
 1237         ui_racct_foreach(racct_decay_callback, racct_decay_pre,
 1238             racct_decay_post, NULL, NULL);
 1239         loginclass_racct_foreach(racct_decay_callback, racct_decay_pre,
 1240             racct_decay_post, NULL, NULL);
 1241         prison_racct_foreach(racct_decay_callback, racct_decay_pre,
 1242             racct_decay_post, NULL, NULL);
 1243 }
 1244 
 1245 static void
 1246 racctd(void)
 1247 {
 1248         struct thread *td;
 1249         struct proc *p;
 1250         struct timeval wallclock;
 1251         uint64_t pct, pct_estimate, runtime;
 1252 
 1253         ASSERT_RACCT_ENABLED();
 1254 
 1255         for (;;) {
 1256                 racct_decay();
 1257 
 1258                 sx_slock(&allproc_lock);
 1259 
 1260                 FOREACH_PROC_IN_SYSTEM(p) {
 1261                         PROC_LOCK(p);
 1262                         if (p->p_state != PRS_NORMAL) {
 1263                                 if (p->p_state == PRS_ZOMBIE)
 1264                                         racct_set(p, RACCT_PCTCPU, 0);
 1265                                 PROC_UNLOCK(p);
 1266                                 continue;
 1267                         }
 1268 
 1269                         microuptime(&wallclock);
 1270                         timevalsub(&wallclock, &p->p_stats->p_start);
 1271                         PROC_STATLOCK(p);
 1272                         FOREACH_THREAD_IN_PROC(p, td)
 1273                                 ruxagg(p, td);
 1274                         runtime = cputick2usec(p->p_rux.rux_runtime);
 1275                         PROC_STATUNLOCK(p);
 1276 #ifdef notyet
 1277                         KASSERT(runtime >= p->p_prev_runtime,
 1278                             ("runtime < p_prev_runtime"));
 1279 #else
 1280                         if (runtime < p->p_prev_runtime)
 1281                                 runtime = p->p_prev_runtime;
 1282 #endif
 1283                         p->p_prev_runtime = runtime;
 1284                         if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
 1285                                 pct_estimate = (1000000 * runtime * 100) /
 1286                                     ((uint64_t)wallclock.tv_sec * 1000000 +
 1287                                     wallclock.tv_usec);
 1288                         } else
 1289                                 pct_estimate = 0;
 1290                         pct = racct_getpcpu(p, pct_estimate);
 1291                         RACCT_LOCK();
 1292 #ifdef RCTL
 1293                         rctl_throttle_decay(p->p_racct, RACCT_READBPS);
 1294                         rctl_throttle_decay(p->p_racct, RACCT_WRITEBPS);
 1295                         rctl_throttle_decay(p->p_racct, RACCT_READIOPS);
 1296                         rctl_throttle_decay(p->p_racct, RACCT_WRITEIOPS);
 1297 #endif
 1298                         racct_set_locked(p, RACCT_PCTCPU, pct, 1);
 1299                         racct_set_locked(p, RACCT_CPU, runtime, 0);
 1300                         racct_set_locked(p, RACCT_WALLCLOCK,
 1301                             (uint64_t)wallclock.tv_sec * 1000000 +
 1302                             wallclock.tv_usec, 0);
 1303                         RACCT_UNLOCK();
 1304                         PROC_UNLOCK(p);
 1305                 }
 1306 
 1307                 /*
 1308                  * To ensure that processes are throttled in a fair way, we need
 1309                  * to iterate over all processes again and check the limits
 1310                  * for %cpu resource only after ucred racct containers have been
 1311                  * properly filled.
 1312                  */
 1313                 FOREACH_PROC_IN_SYSTEM(p) {
 1314                         PROC_LOCK(p);
 1315                         if (p->p_state != PRS_NORMAL) {
 1316                                 PROC_UNLOCK(p);
 1317                                 continue;
 1318                         }
 1319 
 1320                         if (racct_pcpu_available(p) <= 0) {
 1321                                 if (p->p_racct->r_resources[RACCT_PCTCPU] >
 1322                                     pcpu_threshold)
 1323                                         racct_proc_throttle(p, -1);
 1324                         } else if (p->p_throttled == -1) {
 1325                                 racct_proc_wakeup(p);
 1326                         }
 1327                         PROC_UNLOCK(p);
 1328                 }
 1329                 sx_sunlock(&allproc_lock);
 1330                 pause("-", hz);
 1331         }
 1332 }
 1333 
 1334 static struct kproc_desc racctd_kp = {
 1335         "racctd",
 1336         racctd,
 1337         NULL
 1338 };
 1339 
 1340 static void
 1341 racctd_init(void)
 1342 {
 1343         if (!racct_enable)
 1344                 return;
 1345 
 1346         kproc_start(&racctd_kp);
 1347 }
 1348 SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, racctd_init, NULL);
 1349 
 1350 static void
 1351 racct_init(void)
 1352 {
 1353         if (!racct_enable)
 1354                 return;
 1355 
 1356         racct_zone = uma_zcreate("racct", sizeof(struct racct),
 1357             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 1358         /*
 1359          * XXX: Move this somewhere.
 1360          */
 1361         prison0.pr_prison_racct = prison_racct_find("");
 1362 }
 1363 SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL);
 1364 
 1365 #endif /* !RACCT */
Cache object: 46dfab63c25f1ad162b3242eead19732
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.
FreeBSD/Linux Kernel Cross Reference sys/kern/kern_racct.c

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_racct.c
