The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/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  * Copyright (c) 2010 The FreeBSD Foundation
    3  * All rights reserved.
    4  *
    5  * This software was developed by Edward Tomasz Napierala under sponsorship
    6  * from the FreeBSD Foundation.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  *
   29  * $FreeBSD: releng/10.3/sys/kern/kern_racct.c 293473 2016-01-09 14:08:10Z dchagin $
   30  */
   31 
   32 #include <sys/cdefs.h>
   33 __FBSDID("$FreeBSD: releng/10.3/sys/kern/kern_racct.c 293473 2016-01-09 14:08:10Z dchagin $");
   34 
   35 #include "opt_kdtrace.h"
   36 #include "opt_sched.h"
   37 
   38 #include <sys/param.h>
   39 #include <sys/systm.h>
   40 #include <sys/eventhandler.h>
   41 #include <sys/jail.h>
   42 #include <sys/kernel.h>
   43 #include <sys/kthread.h>
   44 #include <sys/lock.h>
   45 #include <sys/loginclass.h>
   46 #include <sys/malloc.h>
   47 #include <sys/mutex.h>
   48 #include <sys/proc.h>
   49 #include <sys/racct.h>
   50 #include <sys/resourcevar.h>
   51 #include <sys/sbuf.h>
   52 #include <sys/sched.h>
   53 #include <sys/sdt.h>
   54 #include <sys/smp.h>
   55 #include <sys/sx.h>
   56 #include <sys/sysctl.h>
   57 #include <sys/sysent.h>
   58 #include <sys/sysproto.h>
   59 #include <sys/umtx.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 int racct_enable = 0;
   76 #else
   77 int racct_enable = 1;
   78 #endif
   79 
   80 SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW, 0, "Resource Accounting");
   81 TUNABLE_INT("kern.racct.enable", &racct_enable);
   82 SYSCTL_UINT(_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 static 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, kernel, rusage, add, "struct proc *", "int",
  108     "uint64_t");
  109 SDT_PROBE_DEFINE3(racct, kernel, rusage, add__failure,
  110     "struct proc *", "int", "uint64_t");
  111 SDT_PROBE_DEFINE3(racct, kernel, rusage, add__cred, "struct ucred *",
  112     "int", "uint64_t");
  113 SDT_PROBE_DEFINE3(racct, kernel, rusage, add__force, "struct proc *",
  114     "int", "uint64_t");
  115 SDT_PROBE_DEFINE3(racct, kernel, rusage, set, "struct proc *", "int",
  116     "uint64_t");
  117 SDT_PROBE_DEFINE3(racct, kernel, rusage, set__failure,
  118     "struct proc *", "int", "uint64_t");
  119 SDT_PROBE_DEFINE3(racct, kernel, rusage, sub, "struct proc *", "int",
  120     "uint64_t");
  121 SDT_PROBE_DEFINE3(racct, kernel, rusage, sub__cred, "struct ucred *",
  122     "int", "uint64_t");
  123 SDT_PROBE_DEFINE1(racct, kernel, racct, create, "struct racct *");
  124 SDT_PROBE_DEFINE1(racct, kernel, racct, destroy, "struct racct *");
  125 SDT_PROBE_DEFINE2(racct, kernel, racct, join, "struct racct *",
  126     "struct racct *");
  127 SDT_PROBE_DEFINE2(racct, kernel, racct, join__failure,
  128     "struct racct *", "struct racct *");
  129 SDT_PROBE_DEFINE2(racct, kernel, racct, leave, "struct racct *",
  130     "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 
  176 static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE;
  177 
  178 #ifdef SCHED_4BSD
  179 /*
  180  * Contains intermediate values for %cpu calculations to avoid using floating
  181  * point in the kernel.
  182  * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20)
  183  * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to
  184  * zero so the calculations are more straightforward.
  185  */
  186 fixpt_t ccpu_exp[] = {
  187         [0] = FSCALE * 1,
  188         [1] = FSCALE * 0.95122942450071400909,
  189         [2] = FSCALE * 0.90483741803595957316,
  190         [3] = FSCALE * 0.86070797642505780722,
  191         [4] = FSCALE * 0.81873075307798185866,
  192         [5] = FSCALE * 0.77880078307140486824,
  193         [6] = FSCALE * 0.74081822068171786606,
  194         [7] = FSCALE * 0.70468808971871343435,
  195         [8] = FSCALE * 0.67032004603563930074,
  196         [9] = FSCALE * 0.63762815162177329314,
  197         [10] = FSCALE * 0.60653065971263342360,
  198         [11] = FSCALE * 0.57694981038048669531,
  199         [12] = FSCALE * 0.54881163609402643262,
  200         [13] = FSCALE * 0.52204577676101604789,
  201         [14] = FSCALE * 0.49658530379140951470,
  202         [15] = FSCALE * 0.47236655274101470713,
  203         [16] = FSCALE * 0.44932896411722159143,
  204         [17] = FSCALE * 0.42741493194872666992,
  205         [18] = FSCALE * 0.40656965974059911188,
  206         [19] = FSCALE * 0.38674102345450120691,
  207         [20] = FSCALE * 0.36787944117144232159,
  208         [21] = FSCALE * 0.34993774911115535467,
  209         [22] = FSCALE * 0.33287108369807955328,
  210         [23] = FSCALE * 0.31663676937905321821,
  211         [24] = FSCALE * 0.30119421191220209664,
  212         [25] = FSCALE * 0.28650479686019010032,
  213         [26] = FSCALE * 0.27253179303401260312,
  214         [27] = FSCALE * 0.25924026064589150757,
  215         [28] = FSCALE * 0.24659696394160647693,
  216         [29] = FSCALE * 0.23457028809379765313,
  217         [30] = FSCALE * 0.22313016014842982893,
  218         [31] = FSCALE * 0.21224797382674305771,
  219         [32] = FSCALE * 0.20189651799465540848,
  220         [33] = FSCALE * 0.19204990862075411423,
  221         [34] = FSCALE * 0.18268352405273465022,
  222         [35] = FSCALE * 0.17377394345044512668,
  223         [36] = FSCALE * 0.16529888822158653829,
  224         [37] = FSCALE * 0.15723716631362761621,
  225         [38] = FSCALE * 0.14956861922263505264,
  226         [39] = FSCALE * 0.14227407158651357185,
  227         [40] = FSCALE * 0.13533528323661269189,
  228         [41] = FSCALE * 0.12873490358780421886,
  229         [42] = FSCALE * 0.12245642825298191021,
  230         [43] = FSCALE * 0.11648415777349695786,
  231         [44] = FSCALE * 0.11080315836233388333,
  232         [45] = FSCALE * 0.10539922456186433678,
  233         [46] = FSCALE * 0.10025884372280373372,
  234         [47] = FSCALE * 0.09536916221554961888,
  235         [48] = FSCALE * 0.09071795328941250337,
  236         [49] = FSCALE * 0.08629358649937051097,
  237         [50] = FSCALE * 0.08208499862389879516,
  238         [51] = FSCALE * 0.07808166600115315231,
  239         [52] = FSCALE * 0.07427357821433388042,
  240         [53] = FSCALE * 0.07065121306042958674,
  241         [54] = FSCALE * 0.06720551273974976512,
  242         [55] = FSCALE * 0.06392786120670757270,
  243         [56] = FSCALE * 0.06081006262521796499,
  244         [57] = FSCALE * 0.05784432087483846296,
  245         [58] = FSCALE * 0.05502322005640722902,
  246         [59] = FSCALE * 0.05233970594843239308,
  247         [60] = FSCALE * 0.04978706836786394297,
  248         [61] = FSCALE * 0.04735892439114092119,
  249         [62] = FSCALE * 0.04504920239355780606,
  250         [63] = FSCALE * 0.04285212686704017991,
  251         [64] = FSCALE * 0.04076220397836621516,
  252         [65] = FSCALE * 0.03877420783172200988,
  253         [66] = FSCALE * 0.03688316740124000544,
  254         [67] = FSCALE * 0.03508435410084502588,
  255         [68] = FSCALE * 0.03337326996032607948,
  256         [69] = FSCALE * 0.03174563637806794323,
  257         [70] = FSCALE * 0.03019738342231850073,
  258         [71] = FSCALE * 0.02872463965423942912,
  259         [72] = FSCALE * 0.02732372244729256080,
  260         [73] = FSCALE * 0.02599112877875534358,
  261         [74] = FSCALE * 0.02472352647033939120,
  262         [75] = FSCALE * 0.02351774585600910823,
  263         [76] = FSCALE * 0.02237077185616559577,
  264         [77] = FSCALE * 0.02127973643837716938,
  265         [78] = FSCALE * 0.02024191144580438847,
  266         [79] = FSCALE * 0.01925470177538692429,
  267         [80] = FSCALE * 0.01831563888873418029,
  268         [81] = FSCALE * 0.01742237463949351138,
  269         [82] = FSCALE * 0.01657267540176124754,
  270         [83] = FSCALE * 0.01576441648485449082,
  271         [84] = FSCALE * 0.01499557682047770621,
  272         [85] = FSCALE * 0.01426423390899925527,
  273         [86] = FSCALE * 0.01356855901220093175,
  274         [87] = FSCALE * 0.01290681258047986886,
  275         [88] = FSCALE * 0.01227733990306844117,
  276         [89] = FSCALE * 0.01167856697039544521,
  277         [90] = FSCALE * 0.01110899653824230649,
  278         [91] = FSCALE * 0.01056720438385265337,
  279         [92] = FSCALE * 0.01005183574463358164,
  280         [93] = FSCALE * 0.00956160193054350793,
  281         [94] = FSCALE * 0.00909527710169581709,
  282         [95] = FSCALE * 0.00865169520312063417,
  283         [96] = FSCALE * 0.00822974704902002884,
  284         [97] = FSCALE * 0.00782837754922577143,
  285         [98] = FSCALE * 0.00744658307092434051,
  286         [99] = FSCALE * 0.00708340892905212004,
  287         [100] = FSCALE * 0.00673794699908546709,
  288         [101] = FSCALE * 0.00640933344625638184,
  289         [102] = FSCALE * 0.00609674656551563610,
  290         [103] = FSCALE * 0.00579940472684214321,
  291         [104] = FSCALE * 0.00551656442076077241,
  292         [105] = FSCALE * 0.00524751839918138427,
  293         [106] = FSCALE * 0.00499159390691021621,
  294         [107] = FSCALE * 0.00474815099941147558,
  295         [108] = FSCALE * 0.00451658094261266798,
  296         [109] = FSCALE * 0.00429630469075234057,
  297         [110] = FSCALE * 0.00408677143846406699,
  298 };
  299 #endif
  300 
  301 #define CCPU_EXP_MAX    110
  302 
  303 /*
  304  * This function is analogical to the getpcpu() function in the ps(1) command.
  305  * They should both calculate in the same way so that the racct %cpu
  306  * calculations are consistent with the values showed by the ps(1) tool.
  307  * The calculations are more complex in the 4BSD scheduler because of the value
  308  * of the ccpu variable.  In ULE it is defined to be zero which saves us some
  309  * work.
  310  */
  311 static uint64_t
  312 racct_getpcpu(struct proc *p, u_int pcpu)
  313 {
  314         u_int swtime;
  315 #ifdef SCHED_4BSD
  316         fixpt_t pctcpu, pctcpu_next;
  317 #endif
  318 #ifdef SMP
  319         struct pcpu *pc;
  320         int found;
  321 #endif
  322         fixpt_t p_pctcpu;
  323         struct thread *td;
  324 
  325         ASSERT_RACCT_ENABLED();
  326 
  327         /*
  328          * If the process is swapped out, we count its %cpu usage as zero.
  329          * This behaviour is consistent with the userland ps(1) tool.
  330          */
  331         if ((p->p_flag & P_INMEM) == 0)
  332                 return (0);
  333         swtime = (ticks - p->p_swtick) / hz;
  334 
  335         /*
  336          * For short-lived processes, the sched_pctcpu() returns small
  337          * values even for cpu intensive processes.  Therefore we use
  338          * our own estimate in this case.
  339          */
  340         if (swtime < RACCT_PCPU_SECS)
  341                 return (pcpu);
  342 
  343         p_pctcpu = 0;
  344         FOREACH_THREAD_IN_PROC(p, td) {
  345                 if (td == PCPU_GET(idlethread))
  346                         continue;
  347 #ifdef SMP
  348                 found = 0;
  349                 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
  350                         if (td == pc->pc_idlethread) {
  351                                 found = 1;
  352                                 break;
  353                         }
  354                 }
  355                 if (found)
  356                         continue;
  357 #endif
  358                 thread_lock(td);
  359 #ifdef SCHED_4BSD
  360                 pctcpu = sched_pctcpu(td);
  361                 /* Count also the yet unfinished second. */
  362                 pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT;
  363                 pctcpu_next += sched_pctcpu_delta(td);
  364                 p_pctcpu += max(pctcpu, pctcpu_next);
  365 #else
  366                 /*
  367                  * In ULE the %cpu statistics are updated on every
  368                  * sched_pctcpu() call.  So special calculations to
  369                  * account for the latest (unfinished) second are
  370                  * not needed.
  371                  */
  372                 p_pctcpu += sched_pctcpu(td);
  373 #endif
  374                 thread_unlock(td);
  375         }
  376 
  377 #ifdef SCHED_4BSD
  378         if (swtime <= CCPU_EXP_MAX)
  379                 return ((100 * (uint64_t)p_pctcpu * 1000000) /
  380                     (FSCALE - ccpu_exp[swtime]));
  381 #endif
  382 
  383         return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE);
  384 }
  385 
  386 static void
  387 racct_add_racct(struct racct *dest, const struct racct *src)
  388 {
  389         int i;
  390 
  391         ASSERT_RACCT_ENABLED();
  392         mtx_assert(&racct_lock, MA_OWNED);
  393 
  394         /*
  395          * Update resource usage in dest.
  396          */
  397         for (i = 0; i <= RACCT_MAX; i++) {
  398                 KASSERT(dest->r_resources[i] >= 0,
  399                     ("%s: resource %d propagation meltdown: dest < 0",
  400                     __func__, i));
  401                 KASSERT(src->r_resources[i] >= 0,
  402                     ("%s: resource %d propagation meltdown: src < 0",
  403                     __func__, i));
  404                 dest->r_resources[i] += src->r_resources[i];
  405         }
  406 }
  407 
  408 static void
  409 racct_sub_racct(struct racct *dest, const struct racct *src)
  410 {
  411         int i;
  412 
  413         ASSERT_RACCT_ENABLED();
  414         mtx_assert(&racct_lock, MA_OWNED);
  415 
  416         /*
  417          * Update resource usage in dest.
  418          */
  419         for (i = 0; i <= RACCT_MAX; i++) {
  420                 if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) {
  421                         KASSERT(dest->r_resources[i] >= 0,
  422                             ("%s: resource %d propagation meltdown: dest < 0",
  423                             __func__, i));
  424                         KASSERT(src->r_resources[i] >= 0,
  425                             ("%s: resource %d propagation meltdown: src < 0",
  426                             __func__, i));
  427                         KASSERT(src->r_resources[i] <= dest->r_resources[i],
  428                             ("%s: resource %d propagation meltdown: src > dest",
  429                             __func__, i));
  430                 }
  431                 if (RACCT_CAN_DROP(i)) {
  432                         dest->r_resources[i] -= src->r_resources[i];
  433                         if (dest->r_resources[i] < 0) {
  434                                 KASSERT(RACCT_IS_SLOPPY(i) ||
  435                                     RACCT_IS_DECAYING(i),
  436                                     ("%s: resource %d usage < 0", __func__, i));
  437                                 dest->r_resources[i] = 0;
  438                         }
  439                 }
  440         }
  441 }
  442 
  443 void
  444 racct_create(struct racct **racctp)
  445 {
  446 
  447         if (!racct_enable)
  448                 return;
  449 
  450         SDT_PROBE1(racct, kernel, racct, create, racctp);
  451 
  452         KASSERT(*racctp == NULL, ("racct already allocated"));
  453 
  454         *racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO);
  455 }
  456 
  457 static void
  458 racct_destroy_locked(struct racct **racctp)
  459 {
  460         int i;
  461         struct racct *racct;
  462 
  463         ASSERT_RACCT_ENABLED();
  464 
  465         SDT_PROBE1(racct, kernel, racct, destroy, racctp);
  466 
  467         mtx_assert(&racct_lock, MA_OWNED);
  468         KASSERT(racctp != NULL, ("NULL racctp"));
  469         KASSERT(*racctp != NULL, ("NULL racct"));
  470 
  471         racct = *racctp;
  472 
  473         for (i = 0; i <= RACCT_MAX; i++) {
  474                 if (RACCT_IS_SLOPPY(i))
  475                         continue;
  476                 if (!RACCT_IS_RECLAIMABLE(i))
  477                         continue;
  478                 KASSERT(racct->r_resources[i] == 0,
  479                     ("destroying non-empty racct: "
  480                     "%ju allocated for resource %d\n",
  481                     racct->r_resources[i], i));
  482         }
  483         uma_zfree(racct_zone, racct);
  484         *racctp = NULL;
  485 }
  486 
  487 void
  488 racct_destroy(struct racct **racct)
  489 {
  490 
  491         if (!racct_enable)
  492                 return;
  493 
  494         mtx_lock(&racct_lock);
  495         racct_destroy_locked(racct);
  496         mtx_unlock(&racct_lock);
  497 }
  498 
  499 /*
  500  * Increase consumption of 'resource' by 'amount' for 'racct'
  501  * and all its parents.  Differently from other cases, 'amount' here
  502  * may be less than zero.
  503  */
  504 static void
  505 racct_adjust_resource(struct racct *racct, int resource,
  506     uint64_t amount)
  507 {
  508 
  509         ASSERT_RACCT_ENABLED();
  510         mtx_assert(&racct_lock, MA_OWNED);
  511         KASSERT(racct != NULL, ("NULL racct"));
  512 
  513         racct->r_resources[resource] += amount;
  514         if (racct->r_resources[resource] < 0) {
  515                 KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource),
  516                     ("%s: resource %d usage < 0", __func__, resource));
  517                 racct->r_resources[resource] = 0;
  518         }
  519         
  520         /*
  521          * There are some cases where the racct %cpu resource would grow
  522          * beyond 100% per core.  For example in racct_proc_exit() we add
  523          * the process %cpu usage to the ucred racct containers.  If too
  524          * many processes terminated in a short time span, the ucred %cpu
  525          * resource could grow too much.  Also, the 4BSD scheduler sometimes
  526          * returns for a thread more than 100% cpu usage. So we set a sane
  527          * boundary here to 100% * the maxumum number of CPUs.
  528          */
  529         if ((resource == RACCT_PCTCPU) &&
  530             (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000 * (int64_t)MAXCPU))
  531                 racct->r_resources[RACCT_PCTCPU] = 100 * 1000000 * (int64_t)MAXCPU;
  532 }
  533 
  534 static int
  535 racct_add_locked(struct proc *p, int resource, uint64_t amount)
  536 {
  537 #ifdef RCTL
  538         int error;
  539 #endif
  540 
  541         ASSERT_RACCT_ENABLED();
  542 
  543         SDT_PROBE3(racct, kernel, rusage, add, p, resource, amount);
  544 
  545         /*
  546          * We need proc lock to dereference p->p_ucred.
  547          */
  548         PROC_LOCK_ASSERT(p, MA_OWNED);
  549 
  550 #ifdef RCTL
  551         error = rctl_enforce(p, resource, amount);
  552         if (error && RACCT_IS_DENIABLE(resource)) {
  553                 SDT_PROBE3(racct, kernel, rusage, add__failure, p, resource,
  554                     amount);
  555                 return (error);
  556         }
  557 #endif
  558         racct_adjust_resource(p->p_racct, resource, amount);
  559         racct_add_cred_locked(p->p_ucred, resource, amount);
  560 
  561         return (0);
  562 }
  563 
  564 /*
  565  * Increase allocation of 'resource' by 'amount' for process 'p'.
  566  * Return 0 if it's below limits, or errno, if it's not.
  567  */
  568 int
  569 racct_add(struct proc *p, int resource, uint64_t amount)
  570 {
  571         int error;
  572 
  573         if (!racct_enable)
  574                 return (0);
  575 
  576         mtx_lock(&racct_lock);
  577         error = racct_add_locked(p, resource, amount);
  578         mtx_unlock(&racct_lock);
  579         return (error);
  580 }
  581 
  582 static void
  583 racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount)
  584 {
  585         struct prison *pr;
  586 
  587         ASSERT_RACCT_ENABLED();
  588 
  589         SDT_PROBE3(racct, kernel, rusage, add__cred, cred, resource, amount);
  590 
  591         racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, amount);
  592         for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
  593                 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
  594                     amount);
  595         racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, amount);
  596 }
  597 
  598 /*
  599  * Increase allocation of 'resource' by 'amount' for credential 'cred'.
  600  * Doesn't check for limits and never fails.
  601  *
  602  * XXX: Shouldn't this ever return an error?
  603  */
  604 void
  605 racct_add_cred(struct ucred *cred, int resource, uint64_t amount)
  606 {
  607 
  608         if (!racct_enable)
  609                 return;
  610 
  611         mtx_lock(&racct_lock);
  612         racct_add_cred_locked(cred, resource, amount);
  613         mtx_unlock(&racct_lock);
  614 }
  615 
  616 /*
  617  * Increase allocation of 'resource' by 'amount' for process 'p'.
  618  * Doesn't check for limits and never fails.
  619  */
  620 void
  621 racct_add_force(struct proc *p, int resource, uint64_t amount)
  622 {
  623 
  624         if (!racct_enable)
  625                 return;
  626 
  627         SDT_PROBE3(racct, kernel, rusage, add__force, p, resource, amount);
  628 
  629         /*
  630          * We need proc lock to dereference p->p_ucred.
  631          */
  632         PROC_LOCK_ASSERT(p, MA_OWNED);
  633 
  634         mtx_lock(&racct_lock);
  635         racct_adjust_resource(p->p_racct, resource, amount);
  636         mtx_unlock(&racct_lock);
  637         racct_add_cred(p->p_ucred, resource, amount);
  638 }
  639 
  640 static int
  641 racct_set_locked(struct proc *p, int resource, uint64_t amount)
  642 {
  643         int64_t old_amount, decayed_amount;
  644         int64_t diff_proc, diff_cred;
  645 #ifdef RCTL
  646         int error;
  647 #endif
  648 
  649         ASSERT_RACCT_ENABLED();
  650 
  651         SDT_PROBE3(racct, kernel, rusage, set, p, resource, amount);
  652 
  653         /*
  654          * We need proc lock to dereference p->p_ucred.
  655          */
  656         PROC_LOCK_ASSERT(p, MA_OWNED);
  657 
  658         old_amount = p->p_racct->r_resources[resource];
  659         /*
  660          * The diffs may be negative.
  661          */
  662         diff_proc = amount - old_amount;
  663         if (RACCT_IS_DECAYING(resource)) {
  664                 /*
  665                  * Resources in per-credential racct containers may decay.
  666                  * If this is the case, we need to calculate the difference
  667                  * between the new amount and the proportional value of the
  668                  * old amount that has decayed in the ucred racct containers.
  669                  */
  670                 decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE;
  671                 diff_cred = amount - decayed_amount;
  672         } else
  673                 diff_cred = diff_proc;
  674 #ifdef notyet
  675         KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource),
  676             ("%s: usage of non-droppable resource %d dropping", __func__,
  677              resource));
  678 #endif
  679 #ifdef RCTL
  680         if (diff_proc > 0) {
  681                 error = rctl_enforce(p, resource, diff_proc);
  682                 if (error && RACCT_IS_DENIABLE(resource)) {
  683                         SDT_PROBE3(racct, kernel, rusage, set__failure, p,
  684                             resource, amount);
  685                         return (error);
  686                 }
  687         }
  688 #endif
  689         racct_adjust_resource(p->p_racct, resource, diff_proc);
  690         if (diff_cred > 0)
  691                 racct_add_cred_locked(p->p_ucred, resource, diff_cred);
  692         else if (diff_cred < 0)
  693                 racct_sub_cred_locked(p->p_ucred, resource, -diff_cred);
  694 
  695         return (0);
  696 }
  697 
  698 /*
  699  * Set allocation of 'resource' to 'amount' for process 'p'.
  700  * Return 0 if it's below limits, or errno, if it's not.
  701  *
  702  * Note that decreasing the allocation always returns 0,
  703  * even if it's above the limit.
  704  */
  705 int
  706 racct_set(struct proc *p, int resource, uint64_t amount)
  707 {
  708         int error;
  709 
  710         if (!racct_enable)
  711                 return (0);
  712 
  713         mtx_lock(&racct_lock);
  714         error = racct_set_locked(p, resource, amount);
  715         mtx_unlock(&racct_lock);
  716         return (error);
  717 }
  718 
  719 static void
  720 racct_set_force_locked(struct proc *p, int resource, uint64_t amount)
  721 {
  722         int64_t old_amount, decayed_amount;
  723         int64_t diff_proc, diff_cred;
  724 
  725         ASSERT_RACCT_ENABLED();
  726 
  727         SDT_PROBE3(racct, kernel, rusage, set, p, resource, amount);
  728 
  729         /*
  730          * We need proc lock to dereference p->p_ucred.
  731          */
  732         PROC_LOCK_ASSERT(p, MA_OWNED);
  733 
  734         old_amount = p->p_racct->r_resources[resource];
  735         /*
  736          * The diffs may be negative.
  737          */
  738         diff_proc = amount - old_amount;
  739         if (RACCT_IS_DECAYING(resource)) {
  740                 /*
  741                  * Resources in per-credential racct containers may decay.
  742                  * If this is the case, we need to calculate the difference
  743                  * between the new amount and the proportional value of the
  744                  * old amount that has decayed in the ucred racct containers.
  745                  */
  746                 decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE;
  747                 diff_cred = amount - decayed_amount;
  748         } else
  749                 diff_cred = diff_proc;
  750 
  751         racct_adjust_resource(p->p_racct, resource, diff_proc);
  752         if (diff_cred > 0)
  753                 racct_add_cred_locked(p->p_ucred, resource, diff_cred);
  754         else if (diff_cred < 0)
  755                 racct_sub_cred_locked(p->p_ucred, resource, -diff_cred);
  756 }
  757 
  758 void
  759 racct_set_force(struct proc *p, int resource, uint64_t amount)
  760 {
  761 
  762         if (!racct_enable)
  763                 return;
  764 
  765         mtx_lock(&racct_lock);
  766         racct_set_force_locked(p, resource, amount);
  767         mtx_unlock(&racct_lock);
  768 }
  769 
  770 /*
  771  * Returns amount of 'resource' the process 'p' can keep allocated.
  772  * Allocating more than that would be denied, unless the resource
  773  * is marked undeniable.  Amount of already allocated resource does
  774  * not matter.
  775  */
  776 uint64_t
  777 racct_get_limit(struct proc *p, int resource)
  778 {
  779 
  780         if (!racct_enable)
  781                 return (UINT64_MAX);
  782 
  783 #ifdef RCTL
  784         return (rctl_get_limit(p, resource));
  785 #else
  786         return (UINT64_MAX);
  787 #endif
  788 }
  789 
  790 /*
  791  * Returns amount of 'resource' the process 'p' can keep allocated.
  792  * Allocating more than that would be denied, unless the resource
  793  * is marked undeniable.  Amount of already allocated resource does
  794  * matter.
  795  */
  796 uint64_t
  797 racct_get_available(struct proc *p, int resource)
  798 {
  799 
  800         if (!racct_enable)
  801                 return (UINT64_MAX);
  802 
  803 #ifdef RCTL
  804         return (rctl_get_available(p, resource));
  805 #else
  806         return (UINT64_MAX);
  807 #endif
  808 }
  809 
  810 /*
  811  * Returns amount of the %cpu resource that process 'p' can add to its %cpu
  812  * utilization.  Adding more than that would lead to the process being
  813  * throttled.
  814  */
  815 static int64_t
  816 racct_pcpu_available(struct proc *p)
  817 {
  818 
  819         ASSERT_RACCT_ENABLED();
  820 
  821 #ifdef RCTL
  822         return (rctl_pcpu_available(p));
  823 #else
  824         return (INT64_MAX);
  825 #endif
  826 }
  827 
  828 /*
  829  * Decrease allocation of 'resource' by 'amount' for process 'p'.
  830  */
  831 void
  832 racct_sub(struct proc *p, int resource, uint64_t amount)
  833 {
  834 
  835         if (!racct_enable)
  836                 return;
  837 
  838         SDT_PROBE3(racct, kernel, rusage, sub, p, resource, amount);
  839 
  840         /*
  841          * We need proc lock to dereference p->p_ucred.
  842          */
  843         PROC_LOCK_ASSERT(p, MA_OWNED);
  844         KASSERT(RACCT_CAN_DROP(resource),
  845             ("%s: called for non-droppable resource %d", __func__, resource));
  846 
  847         mtx_lock(&racct_lock);
  848         KASSERT(amount <= p->p_racct->r_resources[resource],
  849             ("%s: freeing %ju of resource %d, which is more "
  850              "than allocated %jd for %s (pid %d)", __func__, amount, resource,
  851             (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid));
  852 
  853         racct_adjust_resource(p->p_racct, resource, -amount);
  854         racct_sub_cred_locked(p->p_ucred, resource, amount);
  855         mtx_unlock(&racct_lock);
  856 }
  857 
  858 static void
  859 racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount)
  860 {
  861         struct prison *pr;
  862 
  863         ASSERT_RACCT_ENABLED();
  864 
  865         SDT_PROBE3(racct, kernel, rusage, sub__cred, cred, resource, amount);
  866 
  867 #ifdef notyet
  868         KASSERT(RACCT_CAN_DROP(resource),
  869             ("%s: called for resource %d which can not drop", __func__,
  870              resource));
  871 #endif
  872 
  873         racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, -amount);
  874         for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent)
  875                 racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource,
  876                     -amount);
  877         racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, -amount);
  878 }
  879 
  880 /*
  881  * Decrease allocation of 'resource' by 'amount' for credential 'cred'.
  882  */
  883 void
  884 racct_sub_cred(struct ucred *cred, int resource, uint64_t amount)
  885 {
  886 
  887         if (!racct_enable)
  888                 return;
  889 
  890         mtx_lock(&racct_lock);
  891         racct_sub_cred_locked(cred, resource, amount);
  892         mtx_unlock(&racct_lock);
  893 }
  894 
  895 /*
  896  * Inherit resource usage information from the parent process.
  897  */
  898 int
  899 racct_proc_fork(struct proc *parent, struct proc *child)
  900 {
  901         int i, error = 0;
  902 
  903         if (!racct_enable)
  904                 return (0);
  905 
  906         /*
  907          * Create racct for the child process.
  908          */
  909         racct_create(&child->p_racct);
  910 
  911         PROC_LOCK(parent);
  912         PROC_LOCK(child);
  913         mtx_lock(&racct_lock);
  914 
  915 #ifdef RCTL
  916         error = rctl_proc_fork(parent, child);
  917         if (error != 0)
  918                 goto out;
  919 #endif
  920 
  921         /* Init process cpu time. */
  922         child->p_prev_runtime = 0;
  923         child->p_throttled = 0;
  924 
  925         /*
  926          * Inherit resource usage.
  927          */
  928         for (i = 0; i <= RACCT_MAX; i++) {
  929                 if (parent->p_racct->r_resources[i] == 0 ||
  930                     !RACCT_IS_INHERITABLE(i))
  931                         continue;
  932 
  933                 error = racct_set_locked(child, i,
  934                     parent->p_racct->r_resources[i]);
  935                 if (error != 0)
  936                         goto out;
  937         }
  938 
  939         error = racct_add_locked(child, RACCT_NPROC, 1);
  940         error += racct_add_locked(child, RACCT_NTHR, 1);
  941 
  942 out:
  943         mtx_unlock(&racct_lock);
  944         PROC_UNLOCK(child);
  945         PROC_UNLOCK(parent);
  946 
  947         if (error != 0)
  948                 racct_proc_exit(child);
  949 
  950         return (error);
  951 }
  952 
  953 /*
  954  * Called at the end of fork1(), to handle rules that require the process
  955  * to be fully initialized.
  956  */
  957 void
  958 racct_proc_fork_done(struct proc *child)
  959 {
  960 
  961 #ifdef RCTL
  962         if (!racct_enable)
  963                 return;
  964 
  965         PROC_LOCK(child);
  966         mtx_lock(&racct_lock);
  967         rctl_enforce(child, RACCT_NPROC, 0);
  968         rctl_enforce(child, RACCT_NTHR, 0);
  969         mtx_unlock(&racct_lock);
  970         PROC_UNLOCK(child);
  971 #endif
  972 }
  973 
  974 void
  975 racct_proc_exit(struct proc *p)
  976 {
  977         int i;
  978         uint64_t runtime;
  979         struct timeval wallclock;
  980         uint64_t pct_estimate, pct;
  981 
  982         if (!racct_enable)
  983                 return;
  984 
  985         PROC_LOCK(p);
  986         /*
  987          * We don't need to calculate rux, proc_reap() has already done this.
  988          */
  989         runtime = cputick2usec(p->p_rux.rux_runtime);
  990 #ifdef notyet
  991         KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime"));
  992 #else
  993         if (runtime < p->p_prev_runtime)
  994                 runtime = p->p_prev_runtime;
  995 #endif
  996         microuptime(&wallclock);
  997         timevalsub(&wallclock, &p->p_stats->p_start);
  998         if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
  999                 pct_estimate = (1000000 * runtime * 100) /
 1000                     ((uint64_t)wallclock.tv_sec * 1000000 +
 1001                     wallclock.tv_usec);
 1002         } else
 1003                 pct_estimate = 0;
 1004         pct = racct_getpcpu(p, pct_estimate);
 1005 
 1006         mtx_lock(&racct_lock);
 1007         racct_set_locked(p, RACCT_CPU, runtime);
 1008         racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct);
 1009 
 1010         for (i = 0; i <= RACCT_MAX; i++) {
 1011                 if (p->p_racct->r_resources[i] == 0)
 1012                         continue;
 1013                 if (!RACCT_IS_RECLAIMABLE(i))
 1014                         continue;
 1015                 racct_set_locked(p, i, 0);
 1016         }
 1017 
 1018         mtx_unlock(&racct_lock);
 1019         PROC_UNLOCK(p);
 1020 
 1021 #ifdef RCTL
 1022         rctl_racct_release(p->p_racct);
 1023 #endif
 1024         racct_destroy(&p->p_racct);
 1025 }
 1026 
 1027 /*
 1028  * Called after credentials change, to move resource utilisation
 1029  * between raccts.
 1030  */
 1031 void
 1032 racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred,
 1033     struct ucred *newcred)
 1034 {
 1035         struct uidinfo *olduip, *newuip;
 1036         struct loginclass *oldlc, *newlc;
 1037         struct prison *oldpr, *newpr, *pr;
 1038 
 1039         if (!racct_enable)
 1040                 return;
 1041 
 1042         PROC_LOCK_ASSERT(p, MA_NOTOWNED);
 1043 
 1044         newuip = newcred->cr_ruidinfo;
 1045         olduip = oldcred->cr_ruidinfo;
 1046         newlc = newcred->cr_loginclass;
 1047         oldlc = oldcred->cr_loginclass;
 1048         newpr = newcred->cr_prison;
 1049         oldpr = oldcred->cr_prison;
 1050 
 1051         mtx_lock(&racct_lock);
 1052         if (newuip != olduip) {
 1053                 racct_sub_racct(olduip->ui_racct, p->p_racct);
 1054                 racct_add_racct(newuip->ui_racct, p->p_racct);
 1055         }
 1056         if (newlc != oldlc) {
 1057                 racct_sub_racct(oldlc->lc_racct, p->p_racct);
 1058                 racct_add_racct(newlc->lc_racct, p->p_racct);
 1059         }
 1060         if (newpr != oldpr) {
 1061                 for (pr = oldpr; pr != NULL; pr = pr->pr_parent)
 1062                         racct_sub_racct(pr->pr_prison_racct->prr_racct,
 1063                             p->p_racct);
 1064                 for (pr = newpr; pr != NULL; pr = pr->pr_parent)
 1065                         racct_add_racct(pr->pr_prison_racct->prr_racct,
 1066                             p->p_racct);
 1067         }
 1068         mtx_unlock(&racct_lock);
 1069 
 1070 #ifdef RCTL
 1071         rctl_proc_ucred_changed(p, newcred);
 1072 #endif
 1073 }
 1074 
 1075 void
 1076 racct_move(struct racct *dest, struct racct *src)
 1077 {
 1078 
 1079         ASSERT_RACCT_ENABLED();
 1080 
 1081         mtx_lock(&racct_lock);
 1082 
 1083         racct_add_racct(dest, src);
 1084         racct_sub_racct(src, src);
 1085 
 1086         mtx_unlock(&racct_lock);
 1087 }
 1088 
 1089 static void
 1090 racct_proc_throttle(struct proc *p)
 1091 {
 1092         struct thread *td;
 1093 #ifdef SMP
 1094         int cpuid;
 1095 #endif
 1096 
 1097         ASSERT_RACCT_ENABLED();
 1098         PROC_LOCK_ASSERT(p, MA_OWNED);
 1099 
 1100         /*
 1101          * Do not block kernel processes.  Also do not block processes with
 1102          * low %cpu utilization to improve interactivity.
 1103          */
 1104         if (((p->p_flag & (P_SYSTEM | P_KTHREAD)) != 0) ||
 1105             (p->p_racct->r_resources[RACCT_PCTCPU] <= pcpu_threshold))
 1106                 return;
 1107         p->p_throttled = 1;
 1108 
 1109         FOREACH_THREAD_IN_PROC(p, td) {
 1110                 thread_lock(td);
 1111                 switch (td->td_state) {
 1112                 case TDS_RUNQ:
 1113                         /*
 1114                          * If the thread is on the scheduler run-queue, we can
 1115                          * not just remove it from there.  So we set the flag
 1116                          * TDF_NEEDRESCHED for the thread, so that once it is
 1117                          * running, it is taken off the cpu as soon as possible.
 1118                          */
 1119                         td->td_flags |= TDF_NEEDRESCHED;
 1120                         break;
 1121                 case TDS_RUNNING:
 1122                         /*
 1123                          * If the thread is running, we request a context
 1124                          * switch for it by setting the TDF_NEEDRESCHED flag.
 1125                          */
 1126                         td->td_flags |= TDF_NEEDRESCHED;
 1127 #ifdef SMP
 1128                         cpuid = td->td_oncpu;
 1129                         if ((cpuid != NOCPU) && (td != curthread))
 1130                                 ipi_cpu(cpuid, IPI_AST);
 1131 #endif
 1132                         break;
 1133                 default:
 1134                         break;
 1135                 }
 1136                 thread_unlock(td);
 1137         }
 1138 }
 1139 
 1140 static void
 1141 racct_proc_wakeup(struct proc *p)
 1142 {
 1143 
 1144         ASSERT_RACCT_ENABLED();
 1145 
 1146         PROC_LOCK_ASSERT(p, MA_OWNED);
 1147 
 1148         if (p->p_throttled) {
 1149                 p->p_throttled = 0;
 1150                 wakeup(p->p_racct);
 1151         }
 1152 }
 1153 
 1154 static void
 1155 racct_decay_resource(struct racct *racct, void * res, void* dummy)
 1156 {
 1157         int resource;
 1158         int64_t r_old, r_new;
 1159 
 1160         ASSERT_RACCT_ENABLED();
 1161 
 1162         resource = *(int *)res;
 1163         r_old = racct->r_resources[resource];
 1164 
 1165         /* If there is nothing to decay, just exit. */
 1166         if (r_old <= 0)
 1167                 return;
 1168 
 1169         mtx_lock(&racct_lock);
 1170         r_new = r_old * RACCT_DECAY_FACTOR / FSCALE;
 1171         racct->r_resources[resource] = r_new;
 1172         mtx_unlock(&racct_lock);
 1173 }
 1174 
 1175 static void
 1176 racct_decay(int resource)
 1177 {
 1178 
 1179         ASSERT_RACCT_ENABLED();
 1180 
 1181         ui_racct_foreach(racct_decay_resource, &resource, NULL);
 1182         loginclass_racct_foreach(racct_decay_resource, &resource, NULL);
 1183         prison_racct_foreach(racct_decay_resource, &resource, NULL);
 1184 }
 1185 
 1186 static void
 1187 racctd(void)
 1188 {
 1189         struct thread *td;
 1190         struct proc *p;
 1191         struct timeval wallclock;
 1192         uint64_t runtime;
 1193         uint64_t pct, pct_estimate;
 1194 
 1195         ASSERT_RACCT_ENABLED();
 1196 
 1197         for (;;) {
 1198                 racct_decay(RACCT_PCTCPU);
 1199 
 1200                 sx_slock(&allproc_lock);
 1201 
 1202                 LIST_FOREACH(p, &zombproc, p_list) {
 1203                         PROC_LOCK(p);
 1204                         racct_set(p, RACCT_PCTCPU, 0);
 1205                         PROC_UNLOCK(p);
 1206                 }
 1207 
 1208                 FOREACH_PROC_IN_SYSTEM(p) {
 1209                         PROC_LOCK(p);
 1210                         if (p->p_state != PRS_NORMAL) {
 1211                                 PROC_UNLOCK(p);
 1212                                 continue;
 1213                         }
 1214 
 1215                         microuptime(&wallclock);
 1216                         timevalsub(&wallclock, &p->p_stats->p_start);
 1217                         PROC_STATLOCK(p);
 1218                         FOREACH_THREAD_IN_PROC(p, td)
 1219                                 ruxagg(p, td);
 1220                         runtime = cputick2usec(p->p_rux.rux_runtime);
 1221                         PROC_STATUNLOCK(p);
 1222 #ifdef notyet
 1223                         KASSERT(runtime >= p->p_prev_runtime,
 1224                             ("runtime < p_prev_runtime"));
 1225 #else
 1226                         if (runtime < p->p_prev_runtime)
 1227                                 runtime = p->p_prev_runtime;
 1228 #endif
 1229                         p->p_prev_runtime = runtime;
 1230                         if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) {
 1231                                 pct_estimate = (1000000 * runtime * 100) /
 1232                                     ((uint64_t)wallclock.tv_sec * 1000000 +
 1233                                     wallclock.tv_usec);
 1234                         } else
 1235                                 pct_estimate = 0;
 1236                         pct = racct_getpcpu(p, pct_estimate);
 1237                         mtx_lock(&racct_lock);
 1238                         racct_set_force_locked(p, RACCT_PCTCPU, pct);
 1239                         racct_set_locked(p, RACCT_CPU, runtime);
 1240                         racct_set_locked(p, RACCT_WALLCLOCK,
 1241                             (uint64_t)wallclock.tv_sec * 1000000 +
 1242                             wallclock.tv_usec);
 1243                         mtx_unlock(&racct_lock);
 1244                         PROC_UNLOCK(p);
 1245                 }
 1246 
 1247                 /*
 1248                  * To ensure that processes are throttled in a fair way, we need
 1249                  * to iterate over all processes again and check the limits
 1250                  * for %cpu resource only after ucred racct containers have been
 1251                  * properly filled.
 1252                  */
 1253                 FOREACH_PROC_IN_SYSTEM(p) {
 1254                         PROC_LOCK(p);
 1255                         if (p->p_state != PRS_NORMAL) {
 1256                                 PROC_UNLOCK(p);
 1257                                 continue;
 1258                         }
 1259 
 1260                         if (racct_pcpu_available(p) <= 0)
 1261                                 racct_proc_throttle(p);
 1262                         else if (p->p_throttled)
 1263                                 racct_proc_wakeup(p);
 1264                         PROC_UNLOCK(p);
 1265                 }
 1266                 sx_sunlock(&allproc_lock);
 1267                 pause("-", hz);
 1268         }
 1269 }
 1270 
 1271 static struct kproc_desc racctd_kp = {
 1272         "racctd",
 1273         racctd,
 1274         NULL
 1275 };
 1276 
 1277 static void
 1278 racctd_init(void)
 1279 {
 1280         if (!racct_enable)
 1281                 return;
 1282 
 1283         kproc_start(&racctd_kp);
 1284 }
 1285 SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, racctd_init, NULL);
 1286 
 1287 static void
 1288 racct_init(void)
 1289 {
 1290         if (!racct_enable)
 1291                 return;
 1292 
 1293         racct_zone = uma_zcreate("racct", sizeof(struct racct),
 1294             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 1295         /*
 1296          * XXX: Move this somewhere.
 1297          */
 1298         prison0.pr_prison_racct = prison_racct_find("");
 1299 }
 1300 SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL);
 1301 
 1302 #else /* !RACCT */
 1303 
 1304 int
 1305 racct_add(struct proc *p, int resource, uint64_t amount)
 1306 {
 1307 
 1308         return (0);
 1309 }
 1310 
 1311 void
 1312 racct_add_cred(struct ucred *cred, int resource, uint64_t amount)
 1313 {
 1314 }
 1315 
 1316 void
 1317 racct_add_force(struct proc *p, int resource, uint64_t amount)
 1318 {
 1319 
 1320         return;
 1321 }
 1322 
 1323 int
 1324 racct_set(struct proc *p, int resource, uint64_t amount)
 1325 {
 1326 
 1327         return (0);
 1328 }
 1329 
 1330 void
 1331 racct_set_force(struct proc *p, int resource, uint64_t amount)
 1332 {
 1333 }
 1334 
 1335 void
 1336 racct_sub(struct proc *p, int resource, uint64_t amount)
 1337 {
 1338 }
 1339 
 1340 void
 1341 racct_sub_cred(struct ucred *cred, int resource, uint64_t amount)
 1342 {
 1343 }
 1344 
 1345 uint64_t
 1346 racct_get_limit(struct proc *p, int resource)
 1347 {
 1348 
 1349         return (UINT64_MAX);
 1350 }
 1351 
 1352 uint64_t
 1353 racct_get_available(struct proc *p, int resource)
 1354 {
 1355 
 1356         return (UINT64_MAX);
 1357 }
 1358 
 1359 void
 1360 racct_create(struct racct **racctp)
 1361 {
 1362 }
 1363 
 1364 void
 1365 racct_destroy(struct racct **racctp)
 1366 {
 1367 }
 1368 
 1369 int
 1370 racct_proc_fork(struct proc *parent, struct proc *child)
 1371 {
 1372 
 1373         return (0);
 1374 }
 1375 
 1376 void
 1377 racct_proc_fork_done(struct proc *child)
 1378 {
 1379 }
 1380 
 1381 void
 1382 racct_proc_exit(struct proc *p)
 1383 {
 1384 }
 1385 
 1386 #endif /* !RACCT */

Cache object: 551fe4dfbe6f475da7a450247ba6fc7e


[ 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.