1097055e2SEdward Tomasz Napierala /*- 2097055e2SEdward Tomasz Napierala * Copyright (c) 2010 The FreeBSD Foundation 3097055e2SEdward Tomasz Napierala * All rights reserved. 4097055e2SEdward Tomasz Napierala * 5097055e2SEdward Tomasz Napierala * This software was developed by Edward Tomasz Napierala under sponsorship 6097055e2SEdward Tomasz Napierala * from the FreeBSD Foundation. 7097055e2SEdward Tomasz Napierala * 8097055e2SEdward Tomasz Napierala * Redistribution and use in source and binary forms, with or without 9097055e2SEdward Tomasz Napierala * modification, are permitted provided that the following conditions 10097055e2SEdward Tomasz Napierala * are met: 11097055e2SEdward Tomasz Napierala * 1. Redistributions of source code must retain the above copyright 12097055e2SEdward Tomasz Napierala * notice, this list of conditions and the following disclaimer. 13097055e2SEdward Tomasz Napierala * 2. Redistributions in binary form must reproduce the above copyright 14097055e2SEdward Tomasz Napierala * notice, this list of conditions and the following disclaimer in the 15097055e2SEdward Tomasz Napierala * documentation and/or other materials provided with the distribution. 16097055e2SEdward Tomasz Napierala * 17097055e2SEdward Tomasz Napierala * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18097055e2SEdward Tomasz Napierala * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19097055e2SEdward Tomasz Napierala * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20097055e2SEdward Tomasz Napierala * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21097055e2SEdward Tomasz Napierala * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22097055e2SEdward Tomasz Napierala * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23097055e2SEdward Tomasz Napierala * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24097055e2SEdward Tomasz Napierala * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25097055e2SEdward Tomasz Napierala * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26097055e2SEdward Tomasz Napierala * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27097055e2SEdward Tomasz Napierala * SUCH DAMAGE. 28097055e2SEdward Tomasz Napierala * 29097055e2SEdward Tomasz Napierala * $FreeBSD$ 30097055e2SEdward Tomasz Napierala */ 31097055e2SEdward Tomasz Napierala 32097055e2SEdward Tomasz Napierala #include <sys/cdefs.h> 33097055e2SEdward Tomasz Napierala __FBSDID("$FreeBSD$"); 34097055e2SEdward Tomasz Napierala 35097055e2SEdward Tomasz Napierala #include "opt_kdtrace.h" 3636af9869SEdward Tomasz Napierala #include "opt_sched.h" 37097055e2SEdward Tomasz Napierala 38097055e2SEdward Tomasz Napierala #include <sys/param.h> 390e225211SAndriy Gapon #include <sys/systm.h> 40097055e2SEdward Tomasz Napierala #include <sys/eventhandler.h> 41097055e2SEdward Tomasz Napierala #include <sys/jail.h> 42097055e2SEdward Tomasz Napierala #include <sys/kernel.h> 43097055e2SEdward Tomasz Napierala #include <sys/kthread.h> 44097055e2SEdward Tomasz Napierala #include <sys/lock.h> 45097055e2SEdward Tomasz Napierala #include <sys/loginclass.h> 46097055e2SEdward Tomasz Napierala #include <sys/malloc.h> 47097055e2SEdward Tomasz Napierala #include <sys/mutex.h> 48097055e2SEdward Tomasz Napierala #include <sys/proc.h> 49097055e2SEdward Tomasz Napierala #include <sys/racct.h> 50097055e2SEdward Tomasz Napierala #include <sys/resourcevar.h> 51097055e2SEdward Tomasz Napierala #include <sys/sbuf.h> 52097055e2SEdward Tomasz Napierala #include <sys/sched.h> 53097055e2SEdward Tomasz Napierala #include <sys/sdt.h> 5436af9869SEdward Tomasz Napierala #include <sys/smp.h> 55097055e2SEdward Tomasz Napierala #include <sys/sx.h> 5636af9869SEdward Tomasz Napierala #include <sys/sysctl.h> 57097055e2SEdward Tomasz Napierala #include <sys/sysent.h> 58097055e2SEdward Tomasz Napierala #include <sys/sysproto.h> 59097055e2SEdward Tomasz Napierala #include <sys/umtx.h> 6036af9869SEdward Tomasz Napierala #include <machine/smp.h> 61097055e2SEdward Tomasz Napierala 62097055e2SEdward Tomasz Napierala #ifdef RCTL 63097055e2SEdward Tomasz Napierala #include <sys/rctl.h> 64097055e2SEdward Tomasz Napierala #endif 65097055e2SEdward Tomasz Napierala 66097055e2SEdward Tomasz Napierala #ifdef RACCT 67097055e2SEdward Tomasz Napierala 68097055e2SEdward Tomasz Napierala FEATURE(racct, "Resource Accounting"); 69097055e2SEdward Tomasz Napierala 7036af9869SEdward Tomasz Napierala /* 7136af9869SEdward Tomasz Napierala * Do not block processes that have their %cpu usage <= pcpu_threshold. 7236af9869SEdward Tomasz Napierala */ 7336af9869SEdward Tomasz Napierala static int pcpu_threshold = 1; 7436af9869SEdward Tomasz Napierala 7536af9869SEdward Tomasz Napierala SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW, 0, "Resource Accounting"); 7636af9869SEdward Tomasz Napierala SYSCTL_UINT(_kern_racct, OID_AUTO, pcpu_threshold, CTLFLAG_RW, &pcpu_threshold, 7736af9869SEdward Tomasz Napierala 0, "Processes with higher %cpu usage than this value can be throttled."); 7836af9869SEdward Tomasz Napierala 7936af9869SEdward Tomasz Napierala /* 8036af9869SEdward Tomasz Napierala * How many seconds it takes to use the scheduler %cpu calculations. When a 8136af9869SEdward Tomasz Napierala * process starts, we compute its %cpu usage by dividing its runtime by the 8236af9869SEdward Tomasz Napierala * process wall clock time. After RACCT_PCPU_SECS pass, we use the value 8336af9869SEdward Tomasz Napierala * provided by the scheduler. 8436af9869SEdward Tomasz Napierala */ 8536af9869SEdward Tomasz Napierala #define RACCT_PCPU_SECS 3 8636af9869SEdward Tomasz Napierala 87097055e2SEdward Tomasz Napierala static struct mtx racct_lock; 88097055e2SEdward Tomasz Napierala MTX_SYSINIT(racct_lock, &racct_lock, "racct lock", MTX_DEF); 89097055e2SEdward Tomasz Napierala 90097055e2SEdward Tomasz Napierala static uma_zone_t racct_zone; 91097055e2SEdward Tomasz Napierala 92097055e2SEdward Tomasz Napierala static void racct_sub_racct(struct racct *dest, const struct racct *src); 93097055e2SEdward Tomasz Napierala static void racct_sub_cred_locked(struct ucred *cred, int resource, 94097055e2SEdward Tomasz Napierala uint64_t amount); 95097055e2SEdward Tomasz Napierala static void racct_add_cred_locked(struct ucred *cred, int resource, 96097055e2SEdward Tomasz Napierala uint64_t amount); 97097055e2SEdward Tomasz Napierala 98097055e2SEdward Tomasz Napierala SDT_PROVIDER_DEFINE(racct); 99097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, add, add, "struct proc *", "int", 100097055e2SEdward Tomasz Napierala "uint64_t"); 101097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, add_failure, add-failure, 102097055e2SEdward Tomasz Napierala "struct proc *", "int", "uint64_t"); 103097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, add_cred, add-cred, "struct ucred *", 104097055e2SEdward Tomasz Napierala "int", "uint64_t"); 105097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, add_force, add-force, "struct proc *", 106097055e2SEdward Tomasz Napierala "int", "uint64_t"); 107097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, set, set, "struct proc *", "int", 108097055e2SEdward Tomasz Napierala "uint64_t"); 109097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, set_failure, set-failure, 110097055e2SEdward Tomasz Napierala "struct proc *", "int", "uint64_t"); 111097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, sub, sub, "struct proc *", "int", 112097055e2SEdward Tomasz Napierala "uint64_t"); 113097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, kernel, rusage, sub_cred, sub-cred, "struct ucred *", 114097055e2SEdward Tomasz Napierala "int", "uint64_t"); 115097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE1(racct, kernel, racct, create, create, "struct racct *"); 116097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE1(racct, kernel, racct, destroy, destroy, "struct racct *"); 117097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE2(racct, kernel, racct, join, join, "struct racct *", 118097055e2SEdward Tomasz Napierala "struct racct *"); 119097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE2(racct, kernel, racct, join_failure, join-failure, 120097055e2SEdward Tomasz Napierala "struct racct *", "struct racct *"); 121097055e2SEdward Tomasz Napierala SDT_PROBE_DEFINE2(racct, kernel, racct, leave, leave, "struct racct *", 122097055e2SEdward Tomasz Napierala "struct racct *"); 123097055e2SEdward Tomasz Napierala 124097055e2SEdward Tomasz Napierala int racct_types[] = { 125097055e2SEdward Tomasz Napierala [RACCT_CPU] = 12685a2f1b4SEdward Tomasz Napierala RACCT_IN_MILLIONS, 127097055e2SEdward Tomasz Napierala [RACCT_DATA] = 128097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 129097055e2SEdward Tomasz Napierala [RACCT_STACK] = 130097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 131097055e2SEdward Tomasz Napierala [RACCT_CORE] = 132097055e2SEdward Tomasz Napierala RACCT_DENIABLE, 133097055e2SEdward Tomasz Napierala [RACCT_RSS] = 134097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE, 135097055e2SEdward Tomasz Napierala [RACCT_MEMLOCK] = 136097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 137097055e2SEdward Tomasz Napierala [RACCT_NPROC] = 138097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 139097055e2SEdward Tomasz Napierala [RACCT_NOFILE] = 140097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 141097055e2SEdward Tomasz Napierala [RACCT_VMEM] = 142097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 143097055e2SEdward Tomasz Napierala [RACCT_NPTS] = 144097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 145097055e2SEdward Tomasz Napierala [RACCT_SWAP] = 146097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 147097055e2SEdward Tomasz Napierala [RACCT_NTHR] = 148097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 149097055e2SEdward Tomasz Napierala [RACCT_MSGQQUEUED] = 150097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 151097055e2SEdward Tomasz Napierala [RACCT_MSGQSIZE] = 152097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 153097055e2SEdward Tomasz Napierala [RACCT_NMSGQ] = 154097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 155097055e2SEdward Tomasz Napierala [RACCT_NSEM] = 156097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 157097055e2SEdward Tomasz Napierala [RACCT_NSEMOP] = 158097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 159097055e2SEdward Tomasz Napierala [RACCT_NSHM] = 160097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 161097055e2SEdward Tomasz Napierala [RACCT_SHMSIZE] = 162097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 163097055e2SEdward Tomasz Napierala [RACCT_WALLCLOCK] = 16436af9869SEdward Tomasz Napierala RACCT_IN_MILLIONS, 16536af9869SEdward Tomasz Napierala [RACCT_PCTCPU] = 16636af9869SEdward Tomasz Napierala RACCT_DECAYING | RACCT_DENIABLE | RACCT_IN_MILLIONS }; 16736af9869SEdward Tomasz Napierala 16836af9869SEdward Tomasz Napierala static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE; 16936af9869SEdward Tomasz Napierala 17036af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 17136af9869SEdward Tomasz Napierala /* 17236af9869SEdward Tomasz Napierala * Contains intermediate values for %cpu calculations to avoid using floating 17336af9869SEdward Tomasz Napierala * point in the kernel. 17436af9869SEdward Tomasz Napierala * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20) 17536af9869SEdward Tomasz Napierala * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to 17636af9869SEdward Tomasz Napierala * zero so the calculations are more straightforward. 17736af9869SEdward Tomasz Napierala */ 17836af9869SEdward Tomasz Napierala fixpt_t ccpu_exp[] = { 17936af9869SEdward Tomasz Napierala [0] = FSCALE * 1, 18036af9869SEdward Tomasz Napierala [1] = FSCALE * 0.95122942450071400909, 18136af9869SEdward Tomasz Napierala [2] = FSCALE * 0.90483741803595957316, 18236af9869SEdward Tomasz Napierala [3] = FSCALE * 0.86070797642505780722, 18336af9869SEdward Tomasz Napierala [4] = FSCALE * 0.81873075307798185866, 18436af9869SEdward Tomasz Napierala [5] = FSCALE * 0.77880078307140486824, 18536af9869SEdward Tomasz Napierala [6] = FSCALE * 0.74081822068171786606, 18636af9869SEdward Tomasz Napierala [7] = FSCALE * 0.70468808971871343435, 18736af9869SEdward Tomasz Napierala [8] = FSCALE * 0.67032004603563930074, 18836af9869SEdward Tomasz Napierala [9] = FSCALE * 0.63762815162177329314, 18936af9869SEdward Tomasz Napierala [10] = FSCALE * 0.60653065971263342360, 19036af9869SEdward Tomasz Napierala [11] = FSCALE * 0.57694981038048669531, 19136af9869SEdward Tomasz Napierala [12] = FSCALE * 0.54881163609402643262, 19236af9869SEdward Tomasz Napierala [13] = FSCALE * 0.52204577676101604789, 19336af9869SEdward Tomasz Napierala [14] = FSCALE * 0.49658530379140951470, 19436af9869SEdward Tomasz Napierala [15] = FSCALE * 0.47236655274101470713, 19536af9869SEdward Tomasz Napierala [16] = FSCALE * 0.44932896411722159143, 19636af9869SEdward Tomasz Napierala [17] = FSCALE * 0.42741493194872666992, 19736af9869SEdward Tomasz Napierala [18] = FSCALE * 0.40656965974059911188, 19836af9869SEdward Tomasz Napierala [19] = FSCALE * 0.38674102345450120691, 19936af9869SEdward Tomasz Napierala [20] = FSCALE * 0.36787944117144232159, 20036af9869SEdward Tomasz Napierala [21] = FSCALE * 0.34993774911115535467, 20136af9869SEdward Tomasz Napierala [22] = FSCALE * 0.33287108369807955328, 20236af9869SEdward Tomasz Napierala [23] = FSCALE * 0.31663676937905321821, 20336af9869SEdward Tomasz Napierala [24] = FSCALE * 0.30119421191220209664, 20436af9869SEdward Tomasz Napierala [25] = FSCALE * 0.28650479686019010032, 20536af9869SEdward Tomasz Napierala [26] = FSCALE * 0.27253179303401260312, 20636af9869SEdward Tomasz Napierala [27] = FSCALE * 0.25924026064589150757, 20736af9869SEdward Tomasz Napierala [28] = FSCALE * 0.24659696394160647693, 20836af9869SEdward Tomasz Napierala [29] = FSCALE * 0.23457028809379765313, 20936af9869SEdward Tomasz Napierala [30] = FSCALE * 0.22313016014842982893, 21036af9869SEdward Tomasz Napierala [31] = FSCALE * 0.21224797382674305771, 21136af9869SEdward Tomasz Napierala [32] = FSCALE * 0.20189651799465540848, 21236af9869SEdward Tomasz Napierala [33] = FSCALE * 0.19204990862075411423, 21336af9869SEdward Tomasz Napierala [34] = FSCALE * 0.18268352405273465022, 21436af9869SEdward Tomasz Napierala [35] = FSCALE * 0.17377394345044512668, 21536af9869SEdward Tomasz Napierala [36] = FSCALE * 0.16529888822158653829, 21636af9869SEdward Tomasz Napierala [37] = FSCALE * 0.15723716631362761621, 21736af9869SEdward Tomasz Napierala [38] = FSCALE * 0.14956861922263505264, 21836af9869SEdward Tomasz Napierala [39] = FSCALE * 0.14227407158651357185, 21936af9869SEdward Tomasz Napierala [40] = FSCALE * 0.13533528323661269189, 22036af9869SEdward Tomasz Napierala [41] = FSCALE * 0.12873490358780421886, 22136af9869SEdward Tomasz Napierala [42] = FSCALE * 0.12245642825298191021, 22236af9869SEdward Tomasz Napierala [43] = FSCALE * 0.11648415777349695786, 22336af9869SEdward Tomasz Napierala [44] = FSCALE * 0.11080315836233388333, 22436af9869SEdward Tomasz Napierala [45] = FSCALE * 0.10539922456186433678, 22536af9869SEdward Tomasz Napierala [46] = FSCALE * 0.10025884372280373372, 22636af9869SEdward Tomasz Napierala [47] = FSCALE * 0.09536916221554961888, 22736af9869SEdward Tomasz Napierala [48] = FSCALE * 0.09071795328941250337, 22836af9869SEdward Tomasz Napierala [49] = FSCALE * 0.08629358649937051097, 22936af9869SEdward Tomasz Napierala [50] = FSCALE * 0.08208499862389879516, 23036af9869SEdward Tomasz Napierala [51] = FSCALE * 0.07808166600115315231, 23136af9869SEdward Tomasz Napierala [52] = FSCALE * 0.07427357821433388042, 23236af9869SEdward Tomasz Napierala [53] = FSCALE * 0.07065121306042958674, 23336af9869SEdward Tomasz Napierala [54] = FSCALE * 0.06720551273974976512, 23436af9869SEdward Tomasz Napierala [55] = FSCALE * 0.06392786120670757270, 23536af9869SEdward Tomasz Napierala [56] = FSCALE * 0.06081006262521796499, 23636af9869SEdward Tomasz Napierala [57] = FSCALE * 0.05784432087483846296, 23736af9869SEdward Tomasz Napierala [58] = FSCALE * 0.05502322005640722902, 23836af9869SEdward Tomasz Napierala [59] = FSCALE * 0.05233970594843239308, 23936af9869SEdward Tomasz Napierala [60] = FSCALE * 0.04978706836786394297, 24036af9869SEdward Tomasz Napierala [61] = FSCALE * 0.04735892439114092119, 24136af9869SEdward Tomasz Napierala [62] = FSCALE * 0.04504920239355780606, 24236af9869SEdward Tomasz Napierala [63] = FSCALE * 0.04285212686704017991, 24336af9869SEdward Tomasz Napierala [64] = FSCALE * 0.04076220397836621516, 24436af9869SEdward Tomasz Napierala [65] = FSCALE * 0.03877420783172200988, 24536af9869SEdward Tomasz Napierala [66] = FSCALE * 0.03688316740124000544, 24636af9869SEdward Tomasz Napierala [67] = FSCALE * 0.03508435410084502588, 24736af9869SEdward Tomasz Napierala [68] = FSCALE * 0.03337326996032607948, 24836af9869SEdward Tomasz Napierala [69] = FSCALE * 0.03174563637806794323, 24936af9869SEdward Tomasz Napierala [70] = FSCALE * 0.03019738342231850073, 25036af9869SEdward Tomasz Napierala [71] = FSCALE * 0.02872463965423942912, 25136af9869SEdward Tomasz Napierala [72] = FSCALE * 0.02732372244729256080, 25236af9869SEdward Tomasz Napierala [73] = FSCALE * 0.02599112877875534358, 25336af9869SEdward Tomasz Napierala [74] = FSCALE * 0.02472352647033939120, 25436af9869SEdward Tomasz Napierala [75] = FSCALE * 0.02351774585600910823, 25536af9869SEdward Tomasz Napierala [76] = FSCALE * 0.02237077185616559577, 25636af9869SEdward Tomasz Napierala [77] = FSCALE * 0.02127973643837716938, 25736af9869SEdward Tomasz Napierala [78] = FSCALE * 0.02024191144580438847, 25836af9869SEdward Tomasz Napierala [79] = FSCALE * 0.01925470177538692429, 25936af9869SEdward Tomasz Napierala [80] = FSCALE * 0.01831563888873418029, 26036af9869SEdward Tomasz Napierala [81] = FSCALE * 0.01742237463949351138, 26136af9869SEdward Tomasz Napierala [82] = FSCALE * 0.01657267540176124754, 26236af9869SEdward Tomasz Napierala [83] = FSCALE * 0.01576441648485449082, 26336af9869SEdward Tomasz Napierala [84] = FSCALE * 0.01499557682047770621, 26436af9869SEdward Tomasz Napierala [85] = FSCALE * 0.01426423390899925527, 26536af9869SEdward Tomasz Napierala [86] = FSCALE * 0.01356855901220093175, 26636af9869SEdward Tomasz Napierala [87] = FSCALE * 0.01290681258047986886, 26736af9869SEdward Tomasz Napierala [88] = FSCALE * 0.01227733990306844117, 26836af9869SEdward Tomasz Napierala [89] = FSCALE * 0.01167856697039544521, 26936af9869SEdward Tomasz Napierala [90] = FSCALE * 0.01110899653824230649, 27036af9869SEdward Tomasz Napierala [91] = FSCALE * 0.01056720438385265337, 27136af9869SEdward Tomasz Napierala [92] = FSCALE * 0.01005183574463358164, 27236af9869SEdward Tomasz Napierala [93] = FSCALE * 0.00956160193054350793, 27336af9869SEdward Tomasz Napierala [94] = FSCALE * 0.00909527710169581709, 27436af9869SEdward Tomasz Napierala [95] = FSCALE * 0.00865169520312063417, 27536af9869SEdward Tomasz Napierala [96] = FSCALE * 0.00822974704902002884, 27636af9869SEdward Tomasz Napierala [97] = FSCALE * 0.00782837754922577143, 27736af9869SEdward Tomasz Napierala [98] = FSCALE * 0.00744658307092434051, 27836af9869SEdward Tomasz Napierala [99] = FSCALE * 0.00708340892905212004, 27936af9869SEdward Tomasz Napierala [100] = FSCALE * 0.00673794699908546709, 28036af9869SEdward Tomasz Napierala [101] = FSCALE * 0.00640933344625638184, 28136af9869SEdward Tomasz Napierala [102] = FSCALE * 0.00609674656551563610, 28236af9869SEdward Tomasz Napierala [103] = FSCALE * 0.00579940472684214321, 28336af9869SEdward Tomasz Napierala [104] = FSCALE * 0.00551656442076077241, 28436af9869SEdward Tomasz Napierala [105] = FSCALE * 0.00524751839918138427, 28536af9869SEdward Tomasz Napierala [106] = FSCALE * 0.00499159390691021621, 28636af9869SEdward Tomasz Napierala [107] = FSCALE * 0.00474815099941147558, 28736af9869SEdward Tomasz Napierala [108] = FSCALE * 0.00451658094261266798, 28836af9869SEdward Tomasz Napierala [109] = FSCALE * 0.00429630469075234057, 28936af9869SEdward Tomasz Napierala [110] = FSCALE * 0.00408677143846406699, 29036af9869SEdward Tomasz Napierala }; 29136af9869SEdward Tomasz Napierala #endif 29236af9869SEdward Tomasz Napierala 29336af9869SEdward Tomasz Napierala #define CCPU_EXP_MAX 110 29436af9869SEdward Tomasz Napierala 29536af9869SEdward Tomasz Napierala /* 29636af9869SEdward Tomasz Napierala * This function is analogical to the getpcpu() function in the ps(1) command. 29736af9869SEdward Tomasz Napierala * They should both calculate in the same way so that the racct %cpu 29836af9869SEdward Tomasz Napierala * calculations are consistent with the values showed by the ps(1) tool. 29936af9869SEdward Tomasz Napierala * The calculations are more complex in the 4BSD scheduler because of the value 30036af9869SEdward Tomasz Napierala * of the ccpu variable. In ULE it is defined to be zero which saves us some 30136af9869SEdward Tomasz Napierala * work. 30236af9869SEdward Tomasz Napierala */ 30336af9869SEdward Tomasz Napierala static uint64_t 30436af9869SEdward Tomasz Napierala racct_getpcpu(struct proc *p, u_int pcpu) 30536af9869SEdward Tomasz Napierala { 30636af9869SEdward Tomasz Napierala u_int swtime; 30736af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 30836af9869SEdward Tomasz Napierala fixpt_t pctcpu, pctcpu_next; 30936af9869SEdward Tomasz Napierala #endif 31036af9869SEdward Tomasz Napierala #ifdef SMP 31136af9869SEdward Tomasz Napierala struct pcpu *pc; 31236af9869SEdward Tomasz Napierala int found; 31336af9869SEdward Tomasz Napierala #endif 31436af9869SEdward Tomasz Napierala fixpt_t p_pctcpu; 31536af9869SEdward Tomasz Napierala struct thread *td; 31636af9869SEdward Tomasz Napierala 31736af9869SEdward Tomasz Napierala /* 31836af9869SEdward Tomasz Napierala * If the process is swapped out, we count its %cpu usage as zero. 31936af9869SEdward Tomasz Napierala * This behaviour is consistent with the userland ps(1) tool. 32036af9869SEdward Tomasz Napierala */ 32136af9869SEdward Tomasz Napierala if ((p->p_flag & P_INMEM) == 0) 32236af9869SEdward Tomasz Napierala return (0); 32336af9869SEdward Tomasz Napierala swtime = (ticks - p->p_swtick) / hz; 32436af9869SEdward Tomasz Napierala 32536af9869SEdward Tomasz Napierala /* 32636af9869SEdward Tomasz Napierala * For short-lived processes, the sched_pctcpu() returns small 32736af9869SEdward Tomasz Napierala * values even for cpu intensive processes. Therefore we use 32836af9869SEdward Tomasz Napierala * our own estimate in this case. 32936af9869SEdward Tomasz Napierala */ 33036af9869SEdward Tomasz Napierala if (swtime < RACCT_PCPU_SECS) 33136af9869SEdward Tomasz Napierala return (pcpu); 33236af9869SEdward Tomasz Napierala 33336af9869SEdward Tomasz Napierala p_pctcpu = 0; 33436af9869SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) { 33536af9869SEdward Tomasz Napierala if (td == PCPU_GET(idlethread)) 33636af9869SEdward Tomasz Napierala continue; 33736af9869SEdward Tomasz Napierala #ifdef SMP 33836af9869SEdward Tomasz Napierala found = 0; 33936af9869SEdward Tomasz Napierala STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 34036af9869SEdward Tomasz Napierala if (td == pc->pc_idlethread) { 34136af9869SEdward Tomasz Napierala found = 1; 34236af9869SEdward Tomasz Napierala break; 34336af9869SEdward Tomasz Napierala } 34436af9869SEdward Tomasz Napierala } 34536af9869SEdward Tomasz Napierala if (found) 34636af9869SEdward Tomasz Napierala continue; 34736af9869SEdward Tomasz Napierala #endif 34836af9869SEdward Tomasz Napierala thread_lock(td); 34936af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 35036af9869SEdward Tomasz Napierala pctcpu = sched_pctcpu(td); 35136af9869SEdward Tomasz Napierala /* Count also the yet unfinished second. */ 35236af9869SEdward Tomasz Napierala pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT; 35336af9869SEdward Tomasz Napierala pctcpu_next += sched_pctcpu_delta(td); 35436af9869SEdward Tomasz Napierala p_pctcpu += max(pctcpu, pctcpu_next); 35536af9869SEdward Tomasz Napierala #else 35636af9869SEdward Tomasz Napierala /* 35736af9869SEdward Tomasz Napierala * In ULE the %cpu statistics are updated on every 35836af9869SEdward Tomasz Napierala * sched_pctcpu() call. So special calculations to 35936af9869SEdward Tomasz Napierala * account for the latest (unfinished) second are 36036af9869SEdward Tomasz Napierala * not needed. 36136af9869SEdward Tomasz Napierala */ 36236af9869SEdward Tomasz Napierala p_pctcpu += sched_pctcpu(td); 36336af9869SEdward Tomasz Napierala #endif 36436af9869SEdward Tomasz Napierala thread_unlock(td); 36536af9869SEdward Tomasz Napierala } 36636af9869SEdward Tomasz Napierala 36736af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 36836af9869SEdward Tomasz Napierala if (swtime <= CCPU_EXP_MAX) 36936af9869SEdward Tomasz Napierala return ((100 * (uint64_t)p_pctcpu * 1000000) / 37036af9869SEdward Tomasz Napierala (FSCALE - ccpu_exp[swtime])); 37136af9869SEdward Tomasz Napierala #endif 37236af9869SEdward Tomasz Napierala 37336af9869SEdward Tomasz Napierala return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE); 37436af9869SEdward Tomasz Napierala } 375097055e2SEdward Tomasz Napierala 376097055e2SEdward Tomasz Napierala static void 377097055e2SEdward Tomasz Napierala racct_add_racct(struct racct *dest, const struct racct *src) 378097055e2SEdward Tomasz Napierala { 379097055e2SEdward Tomasz Napierala int i; 380097055e2SEdward Tomasz Napierala 381097055e2SEdward Tomasz Napierala mtx_assert(&racct_lock, MA_OWNED); 382097055e2SEdward Tomasz Napierala 383097055e2SEdward Tomasz Napierala /* 384097055e2SEdward Tomasz Napierala * Update resource usage in dest. 385097055e2SEdward Tomasz Napierala */ 386097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 387097055e2SEdward Tomasz Napierala KASSERT(dest->r_resources[i] >= 0, 388097055e2SEdward Tomasz Napierala ("racct propagation meltdown: dest < 0")); 389097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] >= 0, 390097055e2SEdward Tomasz Napierala ("racct propagation meltdown: src < 0")); 391097055e2SEdward Tomasz Napierala dest->r_resources[i] += src->r_resources[i]; 392097055e2SEdward Tomasz Napierala } 393097055e2SEdward Tomasz Napierala } 394097055e2SEdward Tomasz Napierala 395097055e2SEdward Tomasz Napierala static void 396097055e2SEdward Tomasz Napierala racct_sub_racct(struct racct *dest, const struct racct *src) 397097055e2SEdward Tomasz Napierala { 398097055e2SEdward Tomasz Napierala int i; 399097055e2SEdward Tomasz Napierala 400097055e2SEdward Tomasz Napierala mtx_assert(&racct_lock, MA_OWNED); 401097055e2SEdward Tomasz Napierala 402097055e2SEdward Tomasz Napierala /* 403097055e2SEdward Tomasz Napierala * Update resource usage in dest. 404097055e2SEdward Tomasz Napierala */ 405097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 406*84c9193bSEdward Tomasz Napierala if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) { 407097055e2SEdward Tomasz Napierala KASSERT(dest->r_resources[i] >= 0, 408097055e2SEdward Tomasz Napierala ("racct propagation meltdown: dest < 0")); 409097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] >= 0, 410097055e2SEdward Tomasz Napierala ("racct propagation meltdown: src < 0")); 411097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] <= dest->r_resources[i], 412097055e2SEdward Tomasz Napierala ("racct propagation meltdown: src > dest")); 413097055e2SEdward Tomasz Napierala } 41436af9869SEdward Tomasz Napierala if (RACCT_CAN_DROP(i)) { 415097055e2SEdward Tomasz Napierala dest->r_resources[i] -= src->r_resources[i]; 416097055e2SEdward Tomasz Napierala if (dest->r_resources[i] < 0) { 417*84c9193bSEdward Tomasz Napierala KASSERT(RACCT_IS_SLOPPY(i) || 418*84c9193bSEdward Tomasz Napierala RACCT_IS_DECAYING(i), 419097055e2SEdward Tomasz Napierala ("racct_sub_racct: usage < 0")); 420097055e2SEdward Tomasz Napierala dest->r_resources[i] = 0; 421097055e2SEdward Tomasz Napierala } 422097055e2SEdward Tomasz Napierala } 423097055e2SEdward Tomasz Napierala } 424097055e2SEdward Tomasz Napierala } 425097055e2SEdward Tomasz Napierala 426097055e2SEdward Tomasz Napierala void 427097055e2SEdward Tomasz Napierala racct_create(struct racct **racctp) 428097055e2SEdward Tomasz Napierala { 429097055e2SEdward Tomasz Napierala 430097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, racct, create, racctp, 0, 0, 0, 0); 431097055e2SEdward Tomasz Napierala 432097055e2SEdward Tomasz Napierala KASSERT(*racctp == NULL, ("racct already allocated")); 433097055e2SEdward Tomasz Napierala 434097055e2SEdward Tomasz Napierala *racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO); 435097055e2SEdward Tomasz Napierala } 436097055e2SEdward Tomasz Napierala 437097055e2SEdward Tomasz Napierala static void 438097055e2SEdward Tomasz Napierala racct_destroy_locked(struct racct **racctp) 439097055e2SEdward Tomasz Napierala { 440097055e2SEdward Tomasz Napierala int i; 441097055e2SEdward Tomasz Napierala struct racct *racct; 442097055e2SEdward Tomasz Napierala 443097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, racct, destroy, racctp, 0, 0, 0, 0); 444097055e2SEdward Tomasz Napierala 445097055e2SEdward Tomasz Napierala mtx_assert(&racct_lock, MA_OWNED); 446097055e2SEdward Tomasz Napierala KASSERT(racctp != NULL, ("NULL racctp")); 447097055e2SEdward Tomasz Napierala KASSERT(*racctp != NULL, ("NULL racct")); 448097055e2SEdward Tomasz Napierala 449097055e2SEdward Tomasz Napierala racct = *racctp; 450097055e2SEdward Tomasz Napierala 451097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 4524fe84775SEdward Tomasz Napierala if (RACCT_IS_SLOPPY(i)) 453097055e2SEdward Tomasz Napierala continue; 4544fe84775SEdward Tomasz Napierala if (!RACCT_IS_RECLAIMABLE(i)) 455097055e2SEdward Tomasz Napierala continue; 456097055e2SEdward Tomasz Napierala KASSERT(racct->r_resources[i] == 0, 457097055e2SEdward Tomasz Napierala ("destroying non-empty racct: " 458097055e2SEdward Tomasz Napierala "%ju allocated for resource %d\n", 459097055e2SEdward Tomasz Napierala racct->r_resources[i], i)); 460097055e2SEdward Tomasz Napierala } 461097055e2SEdward Tomasz Napierala uma_zfree(racct_zone, racct); 462097055e2SEdward Tomasz Napierala *racctp = NULL; 463097055e2SEdward Tomasz Napierala } 464097055e2SEdward Tomasz Napierala 465097055e2SEdward Tomasz Napierala void 466097055e2SEdward Tomasz Napierala racct_destroy(struct racct **racct) 467097055e2SEdward Tomasz Napierala { 468097055e2SEdward Tomasz Napierala 469097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 470097055e2SEdward Tomasz Napierala racct_destroy_locked(racct); 471097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 472097055e2SEdward Tomasz Napierala } 473097055e2SEdward Tomasz Napierala 474097055e2SEdward Tomasz Napierala /* 475097055e2SEdward Tomasz Napierala * Increase consumption of 'resource' by 'amount' for 'racct' 476097055e2SEdward Tomasz Napierala * and all its parents. Differently from other cases, 'amount' here 477097055e2SEdward Tomasz Napierala * may be less than zero. 478097055e2SEdward Tomasz Napierala */ 479097055e2SEdward Tomasz Napierala static void 480097055e2SEdward Tomasz Napierala racct_alloc_resource(struct racct *racct, int resource, 481097055e2SEdward Tomasz Napierala uint64_t amount) 482097055e2SEdward Tomasz Napierala { 483097055e2SEdward Tomasz Napierala 484097055e2SEdward Tomasz Napierala mtx_assert(&racct_lock, MA_OWNED); 485097055e2SEdward Tomasz Napierala KASSERT(racct != NULL, ("NULL racct")); 486097055e2SEdward Tomasz Napierala 487097055e2SEdward Tomasz Napierala racct->r_resources[resource] += amount; 488097055e2SEdward Tomasz Napierala if (racct->r_resources[resource] < 0) { 48936af9869SEdward Tomasz Napierala KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource), 490097055e2SEdward Tomasz Napierala ("racct_alloc_resource: usage < 0")); 491097055e2SEdward Tomasz Napierala racct->r_resources[resource] = 0; 492097055e2SEdward Tomasz Napierala } 49336af9869SEdward Tomasz Napierala 49436af9869SEdward Tomasz Napierala /* 49536af9869SEdward Tomasz Napierala * There are some cases where the racct %cpu resource would grow 49636af9869SEdward Tomasz Napierala * beyond 100%. 49736af9869SEdward Tomasz Napierala * For example in racct_proc_exit() we add the process %cpu usage 49836af9869SEdward Tomasz Napierala * to the ucred racct containers. If too many processes terminated 49936af9869SEdward Tomasz Napierala * in a short time span, the ucred %cpu resource could grow too much. 50036af9869SEdward Tomasz Napierala * Also, the 4BSD scheduler sometimes returns for a thread more than 50136af9869SEdward Tomasz Napierala * 100% cpu usage. So we set a boundary here to 100%. 50236af9869SEdward Tomasz Napierala */ 50336af9869SEdward Tomasz Napierala if ((resource == RACCT_PCTCPU) && 50436af9869SEdward Tomasz Napierala (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000)) 50536af9869SEdward Tomasz Napierala racct->r_resources[RACCT_PCTCPU] = 100 * 1000000; 506097055e2SEdward Tomasz Napierala } 507097055e2SEdward Tomasz Napierala 5082d8696d1SEdward Tomasz Napierala static int 5092d8696d1SEdward Tomasz Napierala racct_add_locked(struct proc *p, int resource, uint64_t amount) 510097055e2SEdward Tomasz Napierala { 511097055e2SEdward Tomasz Napierala #ifdef RCTL 512097055e2SEdward Tomasz Napierala int error; 513097055e2SEdward Tomasz Napierala #endif 514097055e2SEdward Tomasz Napierala 515097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, add, p, resource, amount, 0, 0); 516097055e2SEdward Tomasz Napierala 517097055e2SEdward Tomasz Napierala /* 518097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 519097055e2SEdward Tomasz Napierala */ 520097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 521097055e2SEdward Tomasz Napierala 522097055e2SEdward Tomasz Napierala #ifdef RCTL 523097055e2SEdward Tomasz Napierala error = rctl_enforce(p, resource, amount); 5244fe84775SEdward Tomasz Napierala if (error && RACCT_IS_DENIABLE(resource)) { 525097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, add_failure, p, resource, 526097055e2SEdward Tomasz Napierala amount, 0, 0); 527097055e2SEdward Tomasz Napierala return (error); 528097055e2SEdward Tomasz Napierala } 529097055e2SEdward Tomasz Napierala #endif 530097055e2SEdward Tomasz Napierala racct_alloc_resource(p->p_racct, resource, amount); 531097055e2SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, resource, amount); 532097055e2SEdward Tomasz Napierala 533097055e2SEdward Tomasz Napierala return (0); 534097055e2SEdward Tomasz Napierala } 535097055e2SEdward Tomasz Napierala 5362d8696d1SEdward Tomasz Napierala /* 5372d8696d1SEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for process 'p'. 5382d8696d1SEdward Tomasz Napierala * Return 0 if it's below limits, or errno, if it's not. 5392d8696d1SEdward Tomasz Napierala */ 5402d8696d1SEdward Tomasz Napierala int 5412d8696d1SEdward Tomasz Napierala racct_add(struct proc *p, int resource, uint64_t amount) 5422d8696d1SEdward Tomasz Napierala { 5432d8696d1SEdward Tomasz Napierala int error; 5442d8696d1SEdward Tomasz Napierala 5452d8696d1SEdward Tomasz Napierala mtx_lock(&racct_lock); 5462d8696d1SEdward Tomasz Napierala error = racct_add_locked(p, resource, amount); 5472d8696d1SEdward Tomasz Napierala mtx_unlock(&racct_lock); 5482d8696d1SEdward Tomasz Napierala return (error); 5492d8696d1SEdward Tomasz Napierala } 5502d8696d1SEdward Tomasz Napierala 551097055e2SEdward Tomasz Napierala static void 552097055e2SEdward Tomasz Napierala racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount) 553097055e2SEdward Tomasz Napierala { 554097055e2SEdward Tomasz Napierala struct prison *pr; 555097055e2SEdward Tomasz Napierala 556097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, add_cred, cred, resource, amount, 557097055e2SEdward Tomasz Napierala 0, 0); 558097055e2SEdward Tomasz Napierala 559097055e2SEdward Tomasz Napierala racct_alloc_resource(cred->cr_ruidinfo->ui_racct, resource, amount); 560097055e2SEdward Tomasz Napierala for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) 561a7ad07bfSEdward Tomasz Napierala racct_alloc_resource(pr->pr_prison_racct->prr_racct, resource, 562a7ad07bfSEdward Tomasz Napierala amount); 563097055e2SEdward Tomasz Napierala racct_alloc_resource(cred->cr_loginclass->lc_racct, resource, amount); 564097055e2SEdward Tomasz Napierala } 565097055e2SEdward Tomasz Napierala 566097055e2SEdward Tomasz Napierala /* 567097055e2SEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for credential 'cred'. 568097055e2SEdward Tomasz Napierala * Doesn't check for limits and never fails. 569097055e2SEdward Tomasz Napierala * 570097055e2SEdward Tomasz Napierala * XXX: Shouldn't this ever return an error? 571097055e2SEdward Tomasz Napierala */ 572097055e2SEdward Tomasz Napierala void 573097055e2SEdward Tomasz Napierala racct_add_cred(struct ucred *cred, int resource, uint64_t amount) 574097055e2SEdward Tomasz Napierala { 575097055e2SEdward Tomasz Napierala 576097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 577097055e2SEdward Tomasz Napierala racct_add_cred_locked(cred, resource, amount); 578097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 579097055e2SEdward Tomasz Napierala } 580097055e2SEdward Tomasz Napierala 581097055e2SEdward Tomasz Napierala /* 582097055e2SEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for process 'p'. 583097055e2SEdward Tomasz Napierala * Doesn't check for limits and never fails. 584097055e2SEdward Tomasz Napierala */ 585097055e2SEdward Tomasz Napierala void 586097055e2SEdward Tomasz Napierala racct_add_force(struct proc *p, int resource, uint64_t amount) 587097055e2SEdward Tomasz Napierala { 588097055e2SEdward Tomasz Napierala 589097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, add_force, p, resource, amount, 0, 0); 590097055e2SEdward Tomasz Napierala 591097055e2SEdward Tomasz Napierala /* 592097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 593097055e2SEdward Tomasz Napierala */ 594097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 595097055e2SEdward Tomasz Napierala 596097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 597097055e2SEdward Tomasz Napierala racct_alloc_resource(p->p_racct, resource, amount); 598097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 599097055e2SEdward Tomasz Napierala racct_add_cred(p->p_ucred, resource, amount); 600097055e2SEdward Tomasz Napierala } 601097055e2SEdward Tomasz Napierala 602097055e2SEdward Tomasz Napierala static int 603097055e2SEdward Tomasz Napierala racct_set_locked(struct proc *p, int resource, uint64_t amount) 604097055e2SEdward Tomasz Napierala { 60536af9869SEdward Tomasz Napierala int64_t old_amount, decayed_amount; 60636af9869SEdward Tomasz Napierala int64_t diff_proc, diff_cred; 607097055e2SEdward Tomasz Napierala #ifdef RCTL 608097055e2SEdward Tomasz Napierala int error; 609097055e2SEdward Tomasz Napierala #endif 610097055e2SEdward Tomasz Napierala 611097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, set, p, resource, amount, 0, 0); 612097055e2SEdward Tomasz Napierala 613097055e2SEdward Tomasz Napierala /* 614097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 615097055e2SEdward Tomasz Napierala */ 616097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 617097055e2SEdward Tomasz Napierala 61836af9869SEdward Tomasz Napierala old_amount = p->p_racct->r_resources[resource]; 61936af9869SEdward Tomasz Napierala /* 62036af9869SEdward Tomasz Napierala * The diffs may be negative. 62136af9869SEdward Tomasz Napierala */ 62236af9869SEdward Tomasz Napierala diff_proc = amount - old_amount; 62336af9869SEdward Tomasz Napierala if (RACCT_IS_DECAYING(resource)) { 62436af9869SEdward Tomasz Napierala /* 62536af9869SEdward Tomasz Napierala * Resources in per-credential racct containers may decay. 62636af9869SEdward Tomasz Napierala * If this is the case, we need to calculate the difference 62736af9869SEdward Tomasz Napierala * between the new amount and the proportional value of the 62836af9869SEdward Tomasz Napierala * old amount that has decayed in the ucred racct containers. 62936af9869SEdward Tomasz Napierala */ 63036af9869SEdward Tomasz Napierala decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE; 63136af9869SEdward Tomasz Napierala diff_cred = amount - decayed_amount; 63236af9869SEdward Tomasz Napierala } else 63336af9869SEdward Tomasz Napierala diff_cred = diff_proc; 634097055e2SEdward Tomasz Napierala #ifdef notyet 63536af9869SEdward Tomasz Napierala KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource), 63636af9869SEdward Tomasz Napierala ("racct_set: usage of non-droppable resource %d dropping", 637097055e2SEdward Tomasz Napierala resource)); 638097055e2SEdward Tomasz Napierala #endif 639097055e2SEdward Tomasz Napierala #ifdef RCTL 64036af9869SEdward Tomasz Napierala if (diff_proc > 0) { 64136af9869SEdward Tomasz Napierala error = rctl_enforce(p, resource, diff_proc); 6424fe84775SEdward Tomasz Napierala if (error && RACCT_IS_DENIABLE(resource)) { 643097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, set_failure, p, 644097055e2SEdward Tomasz Napierala resource, amount, 0, 0); 645097055e2SEdward Tomasz Napierala return (error); 646097055e2SEdward Tomasz Napierala } 647097055e2SEdward Tomasz Napierala } 648097055e2SEdward Tomasz Napierala #endif 64936af9869SEdward Tomasz Napierala racct_alloc_resource(p->p_racct, resource, diff_proc); 65036af9869SEdward Tomasz Napierala if (diff_cred > 0) 65136af9869SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, resource, diff_cred); 65236af9869SEdward Tomasz Napierala else if (diff_cred < 0) 65336af9869SEdward Tomasz Napierala racct_sub_cred_locked(p->p_ucred, resource, -diff_cred); 654097055e2SEdward Tomasz Napierala 655097055e2SEdward Tomasz Napierala return (0); 656097055e2SEdward Tomasz Napierala } 657097055e2SEdward Tomasz Napierala 658097055e2SEdward Tomasz Napierala /* 659097055e2SEdward Tomasz Napierala * Set allocation of 'resource' to 'amount' for process 'p'. 660097055e2SEdward Tomasz Napierala * Return 0 if it's below limits, or errno, if it's not. 661097055e2SEdward Tomasz Napierala * 662097055e2SEdward Tomasz Napierala * Note that decreasing the allocation always returns 0, 663097055e2SEdward Tomasz Napierala * even if it's above the limit. 664097055e2SEdward Tomasz Napierala */ 665097055e2SEdward Tomasz Napierala int 666097055e2SEdward Tomasz Napierala racct_set(struct proc *p, int resource, uint64_t amount) 667097055e2SEdward Tomasz Napierala { 668097055e2SEdward Tomasz Napierala int error; 669097055e2SEdward Tomasz Napierala 670097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 671097055e2SEdward Tomasz Napierala error = racct_set_locked(p, resource, amount); 672097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 673097055e2SEdward Tomasz Napierala return (error); 674097055e2SEdward Tomasz Napierala } 675097055e2SEdward Tomasz Napierala 67636af9869SEdward Tomasz Napierala static void 67736af9869SEdward Tomasz Napierala racct_set_force_locked(struct proc *p, int resource, uint64_t amount) 678097055e2SEdward Tomasz Napierala { 67936af9869SEdward Tomasz Napierala int64_t old_amount, decayed_amount; 68036af9869SEdward Tomasz Napierala int64_t diff_proc, diff_cred; 681097055e2SEdward Tomasz Napierala 682097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, set, p, resource, amount, 0, 0); 683097055e2SEdward Tomasz Napierala 684097055e2SEdward Tomasz Napierala /* 685097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 686097055e2SEdward Tomasz Napierala */ 687097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 688097055e2SEdward Tomasz Napierala 68936af9869SEdward Tomasz Napierala old_amount = p->p_racct->r_resources[resource]; 69036af9869SEdward Tomasz Napierala /* 69136af9869SEdward Tomasz Napierala * The diffs may be negative. 69236af9869SEdward Tomasz Napierala */ 69336af9869SEdward Tomasz Napierala diff_proc = amount - old_amount; 69436af9869SEdward Tomasz Napierala if (RACCT_IS_DECAYING(resource)) { 69536af9869SEdward Tomasz Napierala /* 69636af9869SEdward Tomasz Napierala * Resources in per-credential racct containers may decay. 69736af9869SEdward Tomasz Napierala * If this is the case, we need to calculate the difference 69836af9869SEdward Tomasz Napierala * between the new amount and the proportional value of the 69936af9869SEdward Tomasz Napierala * old amount that has decayed in the ucred racct containers. 70036af9869SEdward Tomasz Napierala */ 70136af9869SEdward Tomasz Napierala decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE; 70236af9869SEdward Tomasz Napierala diff_cred = amount - decayed_amount; 70336af9869SEdward Tomasz Napierala } else 70436af9869SEdward Tomasz Napierala diff_cred = diff_proc; 70536af9869SEdward Tomasz Napierala 70636af9869SEdward Tomasz Napierala racct_alloc_resource(p->p_racct, resource, diff_proc); 70736af9869SEdward Tomasz Napierala if (diff_cred > 0) 70836af9869SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, resource, diff_cred); 70936af9869SEdward Tomasz Napierala else if (diff_cred < 0) 71036af9869SEdward Tomasz Napierala racct_sub_cred_locked(p->p_ucred, resource, -diff_cred); 71136af9869SEdward Tomasz Napierala } 71236af9869SEdward Tomasz Napierala 71336af9869SEdward Tomasz Napierala void 71436af9869SEdward Tomasz Napierala racct_set_force(struct proc *p, int resource, uint64_t amount) 71536af9869SEdward Tomasz Napierala { 716097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 71736af9869SEdward Tomasz Napierala racct_set_force_locked(p, resource, amount); 718097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 719097055e2SEdward Tomasz Napierala } 720097055e2SEdward Tomasz Napierala 721097055e2SEdward Tomasz Napierala /* 722097055e2SEdward Tomasz Napierala * Returns amount of 'resource' the process 'p' can keep allocated. 723097055e2SEdward Tomasz Napierala * Allocating more than that would be denied, unless the resource 724097055e2SEdward Tomasz Napierala * is marked undeniable. Amount of already allocated resource does 725097055e2SEdward Tomasz Napierala * not matter. 726097055e2SEdward Tomasz Napierala */ 727097055e2SEdward Tomasz Napierala uint64_t 728097055e2SEdward Tomasz Napierala racct_get_limit(struct proc *p, int resource) 729097055e2SEdward Tomasz Napierala { 730097055e2SEdward Tomasz Napierala 731097055e2SEdward Tomasz Napierala #ifdef RCTL 732097055e2SEdward Tomasz Napierala return (rctl_get_limit(p, resource)); 733097055e2SEdward Tomasz Napierala #else 734097055e2SEdward Tomasz Napierala return (UINT64_MAX); 735097055e2SEdward Tomasz Napierala #endif 736097055e2SEdward Tomasz Napierala } 737097055e2SEdward Tomasz Napierala 738097055e2SEdward Tomasz Napierala /* 739097055e2SEdward Tomasz Napierala * Returns amount of 'resource' the process 'p' can keep allocated. 740097055e2SEdward Tomasz Napierala * Allocating more than that would be denied, unless the resource 741097055e2SEdward Tomasz Napierala * is marked undeniable. Amount of already allocated resource does 742097055e2SEdward Tomasz Napierala * matter. 743097055e2SEdward Tomasz Napierala */ 744097055e2SEdward Tomasz Napierala uint64_t 745097055e2SEdward Tomasz Napierala racct_get_available(struct proc *p, int resource) 746097055e2SEdward Tomasz Napierala { 747097055e2SEdward Tomasz Napierala 748097055e2SEdward Tomasz Napierala #ifdef RCTL 749097055e2SEdward Tomasz Napierala return (rctl_get_available(p, resource)); 750097055e2SEdward Tomasz Napierala #else 751097055e2SEdward Tomasz Napierala return (UINT64_MAX); 752097055e2SEdward Tomasz Napierala #endif 753097055e2SEdward Tomasz Napierala } 754097055e2SEdward Tomasz Napierala 755097055e2SEdward Tomasz Napierala /* 75636af9869SEdward Tomasz Napierala * Returns amount of the %cpu resource that process 'p' can add to its %cpu 75736af9869SEdward Tomasz Napierala * utilization. Adding more than that would lead to the process being 75836af9869SEdward Tomasz Napierala * throttled. 75936af9869SEdward Tomasz Napierala */ 76036af9869SEdward Tomasz Napierala static int64_t 76136af9869SEdward Tomasz Napierala racct_pcpu_available(struct proc *p) 76236af9869SEdward Tomasz Napierala { 76336af9869SEdward Tomasz Napierala 76436af9869SEdward Tomasz Napierala #ifdef RCTL 76536af9869SEdward Tomasz Napierala return (rctl_pcpu_available(p)); 76636af9869SEdward Tomasz Napierala #else 76736af9869SEdward Tomasz Napierala return (INT64_MAX); 76836af9869SEdward Tomasz Napierala #endif 76936af9869SEdward Tomasz Napierala } 77036af9869SEdward Tomasz Napierala 77136af9869SEdward Tomasz Napierala /* 772097055e2SEdward Tomasz Napierala * Decrease allocation of 'resource' by 'amount' for process 'p'. 773097055e2SEdward Tomasz Napierala */ 774097055e2SEdward Tomasz Napierala void 775097055e2SEdward Tomasz Napierala racct_sub(struct proc *p, int resource, uint64_t amount) 776097055e2SEdward Tomasz Napierala { 777097055e2SEdward Tomasz Napierala 778097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, sub, p, resource, amount, 0, 0); 779097055e2SEdward Tomasz Napierala 780097055e2SEdward Tomasz Napierala /* 781097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 782097055e2SEdward Tomasz Napierala */ 783097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 78436af9869SEdward Tomasz Napierala KASSERT(RACCT_CAN_DROP(resource), 78536af9869SEdward Tomasz Napierala ("racct_sub: called for non-droppable resource %d", resource)); 786097055e2SEdward Tomasz Napierala 787097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 788097055e2SEdward Tomasz Napierala KASSERT(amount <= p->p_racct->r_resources[resource], 789097055e2SEdward Tomasz Napierala ("racct_sub: freeing %ju of resource %d, which is more " 790097055e2SEdward Tomasz Napierala "than allocated %jd for %s (pid %d)", amount, resource, 791097055e2SEdward Tomasz Napierala (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid)); 792097055e2SEdward Tomasz Napierala 793097055e2SEdward Tomasz Napierala racct_alloc_resource(p->p_racct, resource, -amount); 794097055e2SEdward Tomasz Napierala racct_sub_cred_locked(p->p_ucred, resource, amount); 795097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 796097055e2SEdward Tomasz Napierala } 797097055e2SEdward Tomasz Napierala 798097055e2SEdward Tomasz Napierala static void 799097055e2SEdward Tomasz Napierala racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount) 800097055e2SEdward Tomasz Napierala { 801097055e2SEdward Tomasz Napierala struct prison *pr; 802097055e2SEdward Tomasz Napierala 803097055e2SEdward Tomasz Napierala SDT_PROBE(racct, kernel, rusage, sub_cred, cred, resource, amount, 804097055e2SEdward Tomasz Napierala 0, 0); 805097055e2SEdward Tomasz Napierala 806097055e2SEdward Tomasz Napierala #ifdef notyet 80736af9869SEdward Tomasz Napierala KASSERT(RACCT_CAN_DROP(resource), 80836af9869SEdward Tomasz Napierala ("racct_sub_cred: called for resource %d which can not drop", 809097055e2SEdward Tomasz Napierala resource)); 810097055e2SEdward Tomasz Napierala #endif 811097055e2SEdward Tomasz Napierala 812097055e2SEdward Tomasz Napierala racct_alloc_resource(cred->cr_ruidinfo->ui_racct, resource, -amount); 813097055e2SEdward Tomasz Napierala for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) 814a7ad07bfSEdward Tomasz Napierala racct_alloc_resource(pr->pr_prison_racct->prr_racct, resource, 815a7ad07bfSEdward Tomasz Napierala -amount); 816097055e2SEdward Tomasz Napierala racct_alloc_resource(cred->cr_loginclass->lc_racct, resource, -amount); 817097055e2SEdward Tomasz Napierala } 818097055e2SEdward Tomasz Napierala 819097055e2SEdward Tomasz Napierala /* 820097055e2SEdward Tomasz Napierala * Decrease allocation of 'resource' by 'amount' for credential 'cred'. 821097055e2SEdward Tomasz Napierala */ 822097055e2SEdward Tomasz Napierala void 823097055e2SEdward Tomasz Napierala racct_sub_cred(struct ucred *cred, int resource, uint64_t amount) 824097055e2SEdward Tomasz Napierala { 825097055e2SEdward Tomasz Napierala 826097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 827097055e2SEdward Tomasz Napierala racct_sub_cred_locked(cred, resource, amount); 828097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 829097055e2SEdward Tomasz Napierala } 830097055e2SEdward Tomasz Napierala 831097055e2SEdward Tomasz Napierala /* 832097055e2SEdward Tomasz Napierala * Inherit resource usage information from the parent process. 833097055e2SEdward Tomasz Napierala */ 834097055e2SEdward Tomasz Napierala int 835097055e2SEdward Tomasz Napierala racct_proc_fork(struct proc *parent, struct proc *child) 836097055e2SEdward Tomasz Napierala { 837097055e2SEdward Tomasz Napierala int i, error = 0; 838097055e2SEdward Tomasz Napierala 839097055e2SEdward Tomasz Napierala /* 840097055e2SEdward Tomasz Napierala * Create racct for the child process. 841097055e2SEdward Tomasz Napierala */ 842097055e2SEdward Tomasz Napierala racct_create(&child->p_racct); 843097055e2SEdward Tomasz Napierala 844097055e2SEdward Tomasz Napierala PROC_LOCK(parent); 845097055e2SEdward Tomasz Napierala PROC_LOCK(child); 846097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 847097055e2SEdward Tomasz Napierala 848c0c09362SEdward Tomasz Napierala #ifdef RCTL 849c0c09362SEdward Tomasz Napierala error = rctl_proc_fork(parent, child); 850c0c09362SEdward Tomasz Napierala if (error != 0) 851c0c09362SEdward Tomasz Napierala goto out; 852c0c09362SEdward Tomasz Napierala #endif 853c0c09362SEdward Tomasz Napierala 85436af9869SEdward Tomasz Napierala /* Init process cpu time. */ 85536af9869SEdward Tomasz Napierala child->p_prev_runtime = 0; 85636af9869SEdward Tomasz Napierala child->p_throttled = 0; 85736af9869SEdward Tomasz Napierala 858097055e2SEdward Tomasz Napierala /* 859097055e2SEdward Tomasz Napierala * Inherit resource usage. 860097055e2SEdward Tomasz Napierala */ 861097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 862097055e2SEdward Tomasz Napierala if (parent->p_racct->r_resources[i] == 0 || 8634fe84775SEdward Tomasz Napierala !RACCT_IS_INHERITABLE(i)) 864097055e2SEdward Tomasz Napierala continue; 865097055e2SEdward Tomasz Napierala 866097055e2SEdward Tomasz Napierala error = racct_set_locked(child, i, 867097055e2SEdward Tomasz Napierala parent->p_racct->r_resources[i]); 868ac6fafe6SEdward Tomasz Napierala if (error != 0) 869097055e2SEdward Tomasz Napierala goto out; 870097055e2SEdward Tomasz Napierala } 871097055e2SEdward Tomasz Napierala 8722d8696d1SEdward Tomasz Napierala error = racct_add_locked(child, RACCT_NPROC, 1); 8732d8696d1SEdward Tomasz Napierala error += racct_add_locked(child, RACCT_NTHR, 1); 8742d8696d1SEdward Tomasz Napierala 875097055e2SEdward Tomasz Napierala out: 876097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 877097055e2SEdward Tomasz Napierala PROC_UNLOCK(child); 878097055e2SEdward Tomasz Napierala PROC_UNLOCK(parent); 879097055e2SEdward Tomasz Napierala 880ab27d5d8SEdward Tomasz Napierala if (error != 0) 881ab27d5d8SEdward Tomasz Napierala racct_proc_exit(child); 882ab27d5d8SEdward Tomasz Napierala 883097055e2SEdward Tomasz Napierala return (error); 884097055e2SEdward Tomasz Napierala } 885097055e2SEdward Tomasz Napierala 88672a401d9SEdward Tomasz Napierala /* 88772a401d9SEdward Tomasz Napierala * Called at the end of fork1(), to handle rules that require the process 88872a401d9SEdward Tomasz Napierala * to be fully initialized. 88972a401d9SEdward Tomasz Napierala */ 89072a401d9SEdward Tomasz Napierala void 89172a401d9SEdward Tomasz Napierala racct_proc_fork_done(struct proc *child) 89272a401d9SEdward Tomasz Napierala { 89372a401d9SEdward Tomasz Napierala 89472a401d9SEdward Tomasz Napierala #ifdef RCTL 89572a401d9SEdward Tomasz Napierala PROC_LOCK(child); 89672a401d9SEdward Tomasz Napierala mtx_lock(&racct_lock); 89772a401d9SEdward Tomasz Napierala rctl_enforce(child, RACCT_NPROC, 0); 89872a401d9SEdward Tomasz Napierala rctl_enforce(child, RACCT_NTHR, 0); 89972a401d9SEdward Tomasz Napierala mtx_unlock(&racct_lock); 90072a401d9SEdward Tomasz Napierala PROC_UNLOCK(child); 90172a401d9SEdward Tomasz Napierala #endif 90272a401d9SEdward Tomasz Napierala } 90372a401d9SEdward Tomasz Napierala 904097055e2SEdward Tomasz Napierala void 905097055e2SEdward Tomasz Napierala racct_proc_exit(struct proc *p) 906097055e2SEdward Tomasz Napierala { 9072419d7f9SEdward Tomasz Napierala int i; 908097055e2SEdward Tomasz Napierala uint64_t runtime; 90936af9869SEdward Tomasz Napierala struct timeval wallclock; 91036af9869SEdward Tomasz Napierala uint64_t pct_estimate, pct; 911097055e2SEdward Tomasz Napierala 912097055e2SEdward Tomasz Napierala PROC_LOCK(p); 913097055e2SEdward Tomasz Napierala /* 914097055e2SEdward Tomasz Napierala * We don't need to calculate rux, proc_reap() has already done this. 915097055e2SEdward Tomasz Napierala */ 916097055e2SEdward Tomasz Napierala runtime = cputick2usec(p->p_rux.rux_runtime); 917097055e2SEdward Tomasz Napierala #ifdef notyet 918097055e2SEdward Tomasz Napierala KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime")); 919097055e2SEdward Tomasz Napierala #else 920097055e2SEdward Tomasz Napierala if (runtime < p->p_prev_runtime) 921097055e2SEdward Tomasz Napierala runtime = p->p_prev_runtime; 922097055e2SEdward Tomasz Napierala #endif 92336af9869SEdward Tomasz Napierala microuptime(&wallclock); 92436af9869SEdward Tomasz Napierala timevalsub(&wallclock, &p->p_stats->p_start); 92584590fd8SEdward Tomasz Napierala if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) { 92636af9869SEdward Tomasz Napierala pct_estimate = (1000000 * runtime * 100) / 92736af9869SEdward Tomasz Napierala ((uint64_t)wallclock.tv_sec * 1000000 + 92836af9869SEdward Tomasz Napierala wallclock.tv_usec); 92984590fd8SEdward Tomasz Napierala } else 93084590fd8SEdward Tomasz Napierala pct_estimate = 0; 93136af9869SEdward Tomasz Napierala pct = racct_getpcpu(p, pct_estimate); 93236af9869SEdward Tomasz Napierala 9332419d7f9SEdward Tomasz Napierala mtx_lock(&racct_lock); 9342419d7f9SEdward Tomasz Napierala racct_set_locked(p, RACCT_CPU, runtime); 93536af9869SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct); 936097055e2SEdward Tomasz Napierala 9372419d7f9SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 9382419d7f9SEdward Tomasz Napierala if (p->p_racct->r_resources[i] == 0) 9392419d7f9SEdward Tomasz Napierala continue; 9402419d7f9SEdward Tomasz Napierala if (!RACCT_IS_RECLAIMABLE(i)) 9412419d7f9SEdward Tomasz Napierala continue; 9422419d7f9SEdward Tomasz Napierala racct_set_locked(p, i, 0); 9432419d7f9SEdward Tomasz Napierala } 9442419d7f9SEdward Tomasz Napierala 9452419d7f9SEdward Tomasz Napierala mtx_unlock(&racct_lock); 946097055e2SEdward Tomasz Napierala PROC_UNLOCK(p); 947097055e2SEdward Tomasz Napierala 948097055e2SEdward Tomasz Napierala #ifdef RCTL 949097055e2SEdward Tomasz Napierala rctl_racct_release(p->p_racct); 950097055e2SEdward Tomasz Napierala #endif 951097055e2SEdward Tomasz Napierala racct_destroy(&p->p_racct); 952097055e2SEdward Tomasz Napierala } 953097055e2SEdward Tomasz Napierala 954097055e2SEdward Tomasz Napierala /* 955097055e2SEdward Tomasz Napierala * Called after credentials change, to move resource utilisation 956097055e2SEdward Tomasz Napierala * between raccts. 957097055e2SEdward Tomasz Napierala */ 958097055e2SEdward Tomasz Napierala void 959097055e2SEdward Tomasz Napierala racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred, 960097055e2SEdward Tomasz Napierala struct ucred *newcred) 961097055e2SEdward Tomasz Napierala { 962097055e2SEdward Tomasz Napierala struct uidinfo *olduip, *newuip; 963097055e2SEdward Tomasz Napierala struct loginclass *oldlc, *newlc; 964097055e2SEdward Tomasz Napierala struct prison *oldpr, *newpr, *pr; 965097055e2SEdward Tomasz Napierala 966097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_NOTOWNED); 967097055e2SEdward Tomasz Napierala 968097055e2SEdward Tomasz Napierala newuip = newcred->cr_ruidinfo; 969097055e2SEdward Tomasz Napierala olduip = oldcred->cr_ruidinfo; 970097055e2SEdward Tomasz Napierala newlc = newcred->cr_loginclass; 971097055e2SEdward Tomasz Napierala oldlc = oldcred->cr_loginclass; 972097055e2SEdward Tomasz Napierala newpr = newcred->cr_prison; 973097055e2SEdward Tomasz Napierala oldpr = oldcred->cr_prison; 974097055e2SEdward Tomasz Napierala 975097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 976097055e2SEdward Tomasz Napierala if (newuip != olduip) { 977097055e2SEdward Tomasz Napierala racct_sub_racct(olduip->ui_racct, p->p_racct); 978097055e2SEdward Tomasz Napierala racct_add_racct(newuip->ui_racct, p->p_racct); 979097055e2SEdward Tomasz Napierala } 980097055e2SEdward Tomasz Napierala if (newlc != oldlc) { 981097055e2SEdward Tomasz Napierala racct_sub_racct(oldlc->lc_racct, p->p_racct); 982097055e2SEdward Tomasz Napierala racct_add_racct(newlc->lc_racct, p->p_racct); 983097055e2SEdward Tomasz Napierala } 984097055e2SEdward Tomasz Napierala if (newpr != oldpr) { 985097055e2SEdward Tomasz Napierala for (pr = oldpr; pr != NULL; pr = pr->pr_parent) 986a7ad07bfSEdward Tomasz Napierala racct_sub_racct(pr->pr_prison_racct->prr_racct, 987a7ad07bfSEdward Tomasz Napierala p->p_racct); 988097055e2SEdward Tomasz Napierala for (pr = newpr; pr != NULL; pr = pr->pr_parent) 989a7ad07bfSEdward Tomasz Napierala racct_add_racct(pr->pr_prison_racct->prr_racct, 990a7ad07bfSEdward Tomasz Napierala p->p_racct); 991097055e2SEdward Tomasz Napierala } 992097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 993097055e2SEdward Tomasz Napierala 994097055e2SEdward Tomasz Napierala #ifdef RCTL 995097055e2SEdward Tomasz Napierala rctl_proc_ucred_changed(p, newcred); 996097055e2SEdward Tomasz Napierala #endif 997097055e2SEdward Tomasz Napierala } 998097055e2SEdward Tomasz Napierala 999c34bbd2aSEdward Tomasz Napierala void 1000c34bbd2aSEdward Tomasz Napierala racct_move(struct racct *dest, struct racct *src) 1001c34bbd2aSEdward Tomasz Napierala { 1002c34bbd2aSEdward Tomasz Napierala 1003c34bbd2aSEdward Tomasz Napierala mtx_lock(&racct_lock); 1004c34bbd2aSEdward Tomasz Napierala 1005c34bbd2aSEdward Tomasz Napierala racct_add_racct(dest, src); 1006c34bbd2aSEdward Tomasz Napierala racct_sub_racct(src, src); 1007c34bbd2aSEdward Tomasz Napierala 1008c34bbd2aSEdward Tomasz Napierala mtx_unlock(&racct_lock); 1009c34bbd2aSEdward Tomasz Napierala } 1010c34bbd2aSEdward Tomasz Napierala 1011097055e2SEdward Tomasz Napierala static void 101236af9869SEdward Tomasz Napierala racct_proc_throttle(struct proc *p) 101336af9869SEdward Tomasz Napierala { 101436af9869SEdward Tomasz Napierala struct thread *td; 101536af9869SEdward Tomasz Napierala #ifdef SMP 101636af9869SEdward Tomasz Napierala int cpuid; 101736af9869SEdward Tomasz Napierala #endif 101836af9869SEdward Tomasz Napierala 101936af9869SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 102036af9869SEdward Tomasz Napierala 102136af9869SEdward Tomasz Napierala /* 102236af9869SEdward Tomasz Napierala * Do not block kernel processes. Also do not block processes with 102336af9869SEdward Tomasz Napierala * low %cpu utilization to improve interactivity. 102436af9869SEdward Tomasz Napierala */ 102536af9869SEdward Tomasz Napierala if (((p->p_flag & (P_SYSTEM | P_KTHREAD)) != 0) || 102636af9869SEdward Tomasz Napierala (p->p_racct->r_resources[RACCT_PCTCPU] <= pcpu_threshold)) 102736af9869SEdward Tomasz Napierala return; 102836af9869SEdward Tomasz Napierala p->p_throttled = 1; 102936af9869SEdward Tomasz Napierala 103036af9869SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) { 103136af9869SEdward Tomasz Napierala switch (td->td_state) { 103236af9869SEdward Tomasz Napierala case TDS_RUNQ: 103336af9869SEdward Tomasz Napierala /* 103436af9869SEdward Tomasz Napierala * If the thread is on the scheduler run-queue, we can 103536af9869SEdward Tomasz Napierala * not just remove it from there. So we set the flag 103636af9869SEdward Tomasz Napierala * TDF_NEEDRESCHED for the thread, so that once it is 103736af9869SEdward Tomasz Napierala * running, it is taken off the cpu as soon as possible. 103836af9869SEdward Tomasz Napierala */ 103936af9869SEdward Tomasz Napierala thread_lock(td); 104036af9869SEdward Tomasz Napierala td->td_flags |= TDF_NEEDRESCHED; 104136af9869SEdward Tomasz Napierala thread_unlock(td); 104236af9869SEdward Tomasz Napierala break; 104336af9869SEdward Tomasz Napierala case TDS_RUNNING: 104436af9869SEdward Tomasz Napierala /* 104536af9869SEdward Tomasz Napierala * If the thread is running, we request a context 104636af9869SEdward Tomasz Napierala * switch for it by setting the TDF_NEEDRESCHED flag. 104736af9869SEdward Tomasz Napierala */ 104836af9869SEdward Tomasz Napierala thread_lock(td); 104936af9869SEdward Tomasz Napierala td->td_flags |= TDF_NEEDRESCHED; 105036af9869SEdward Tomasz Napierala #ifdef SMP 105136af9869SEdward Tomasz Napierala cpuid = td->td_oncpu; 105236af9869SEdward Tomasz Napierala if ((cpuid != NOCPU) && (td != curthread)) 105336af9869SEdward Tomasz Napierala ipi_cpu(cpuid, IPI_AST); 105436af9869SEdward Tomasz Napierala #endif 105536af9869SEdward Tomasz Napierala thread_unlock(td); 105636af9869SEdward Tomasz Napierala break; 105736af9869SEdward Tomasz Napierala default: 105836af9869SEdward Tomasz Napierala break; 105936af9869SEdward Tomasz Napierala } 106036af9869SEdward Tomasz Napierala } 106136af9869SEdward Tomasz Napierala } 106236af9869SEdward Tomasz Napierala 106336af9869SEdward Tomasz Napierala static void 106436af9869SEdward Tomasz Napierala racct_proc_wakeup(struct proc *p) 106536af9869SEdward Tomasz Napierala { 106636af9869SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 106736af9869SEdward Tomasz Napierala 106836af9869SEdward Tomasz Napierala if (p->p_throttled) { 106936af9869SEdward Tomasz Napierala p->p_throttled = 0; 107036af9869SEdward Tomasz Napierala wakeup(p->p_racct); 107136af9869SEdward Tomasz Napierala } 107236af9869SEdward Tomasz Napierala } 107336af9869SEdward Tomasz Napierala 107436af9869SEdward Tomasz Napierala static void 107536af9869SEdward Tomasz Napierala racct_decay_resource(struct racct *racct, void * res, void* dummy) 107636af9869SEdward Tomasz Napierala { 107736af9869SEdward Tomasz Napierala int resource; 107836af9869SEdward Tomasz Napierala int64_t r_old, r_new; 107936af9869SEdward Tomasz Napierala 108036af9869SEdward Tomasz Napierala resource = *(int *)res; 108136af9869SEdward Tomasz Napierala r_old = racct->r_resources[resource]; 108236af9869SEdward Tomasz Napierala 108336af9869SEdward Tomasz Napierala /* If there is nothing to decay, just exit. */ 108436af9869SEdward Tomasz Napierala if (r_old <= 0) 108536af9869SEdward Tomasz Napierala return; 108636af9869SEdward Tomasz Napierala 108736af9869SEdward Tomasz Napierala mtx_lock(&racct_lock); 108836af9869SEdward Tomasz Napierala r_new = r_old * RACCT_DECAY_FACTOR / FSCALE; 108936af9869SEdward Tomasz Napierala racct->r_resources[resource] = r_new; 109036af9869SEdward Tomasz Napierala mtx_unlock(&racct_lock); 109136af9869SEdward Tomasz Napierala } 109236af9869SEdward Tomasz Napierala 109336af9869SEdward Tomasz Napierala static void 109436af9869SEdward Tomasz Napierala racct_decay(int resource) 109536af9869SEdward Tomasz Napierala { 109636af9869SEdward Tomasz Napierala ui_racct_foreach(racct_decay_resource, &resource, NULL); 109736af9869SEdward Tomasz Napierala loginclass_racct_foreach(racct_decay_resource, &resource, NULL); 109836af9869SEdward Tomasz Napierala prison_racct_foreach(racct_decay_resource, &resource, NULL); 109936af9869SEdward Tomasz Napierala } 110036af9869SEdward Tomasz Napierala 110136af9869SEdward Tomasz Napierala static void 1102097055e2SEdward Tomasz Napierala racctd(void) 1103097055e2SEdward Tomasz Napierala { 1104097055e2SEdward Tomasz Napierala struct thread *td; 1105097055e2SEdward Tomasz Napierala struct proc *p; 1106097055e2SEdward Tomasz Napierala struct timeval wallclock; 1107097055e2SEdward Tomasz Napierala uint64_t runtime; 110836af9869SEdward Tomasz Napierala uint64_t pct, pct_estimate; 1109097055e2SEdward Tomasz Napierala 1110097055e2SEdward Tomasz Napierala for (;;) { 111136af9869SEdward Tomasz Napierala racct_decay(RACCT_PCTCPU); 111236af9869SEdward Tomasz Napierala 1113097055e2SEdward Tomasz Napierala sx_slock(&allproc_lock); 1114097055e2SEdward Tomasz Napierala 111536af9869SEdward Tomasz Napierala LIST_FOREACH(p, &zombproc, p_list) { 111636af9869SEdward Tomasz Napierala PROC_LOCK(p); 111736af9869SEdward Tomasz Napierala racct_set(p, RACCT_PCTCPU, 0); 111836af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 111936af9869SEdward Tomasz Napierala } 112036af9869SEdward Tomasz Napierala 1121097055e2SEdward Tomasz Napierala FOREACH_PROC_IN_SYSTEM(p) { 112236af9869SEdward Tomasz Napierala PROC_LOCK(p); 112336af9869SEdward Tomasz Napierala if (p->p_state != PRS_NORMAL) { 112436af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 1125097055e2SEdward Tomasz Napierala continue; 112636af9869SEdward Tomasz Napierala } 1127097055e2SEdward Tomasz Napierala 1128097055e2SEdward Tomasz Napierala microuptime(&wallclock); 1129097055e2SEdward Tomasz Napierala timevalsub(&wallclock, &p->p_stats->p_start); 1130097055e2SEdward Tomasz Napierala PROC_SLOCK(p); 11310a53cd57SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) 1132097055e2SEdward Tomasz Napierala ruxagg(p, td); 1133097055e2SEdward Tomasz Napierala runtime = cputick2usec(p->p_rux.rux_runtime); 1134097055e2SEdward Tomasz Napierala PROC_SUNLOCK(p); 1135097055e2SEdward Tomasz Napierala #ifdef notyet 1136097055e2SEdward Tomasz Napierala KASSERT(runtime >= p->p_prev_runtime, 1137097055e2SEdward Tomasz Napierala ("runtime < p_prev_runtime")); 1138097055e2SEdward Tomasz Napierala #else 1139097055e2SEdward Tomasz Napierala if (runtime < p->p_prev_runtime) 1140097055e2SEdward Tomasz Napierala runtime = p->p_prev_runtime; 1141097055e2SEdward Tomasz Napierala #endif 1142097055e2SEdward Tomasz Napierala p->p_prev_runtime = runtime; 114384590fd8SEdward Tomasz Napierala if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) { 114436af9869SEdward Tomasz Napierala pct_estimate = (1000000 * runtime * 100) / 114536af9869SEdward Tomasz Napierala ((uint64_t)wallclock.tv_sec * 1000000 + 114636af9869SEdward Tomasz Napierala wallclock.tv_usec); 114784590fd8SEdward Tomasz Napierala } else 114884590fd8SEdward Tomasz Napierala pct_estimate = 0; 114936af9869SEdward Tomasz Napierala pct = racct_getpcpu(p, pct_estimate); 1150097055e2SEdward Tomasz Napierala mtx_lock(&racct_lock); 115136af9869SEdward Tomasz Napierala racct_set_force_locked(p, RACCT_PCTCPU, pct); 1152097055e2SEdward Tomasz Napierala racct_set_locked(p, RACCT_CPU, runtime); 1153097055e2SEdward Tomasz Napierala racct_set_locked(p, RACCT_WALLCLOCK, 115459f513cdSJaakko Heinonen (uint64_t)wallclock.tv_sec * 1000000 + 115559f513cdSJaakko Heinonen wallclock.tv_usec); 1156097055e2SEdward Tomasz Napierala mtx_unlock(&racct_lock); 1157097055e2SEdward Tomasz Napierala PROC_UNLOCK(p); 1158097055e2SEdward Tomasz Napierala } 115936af9869SEdward Tomasz Napierala 116036af9869SEdward Tomasz Napierala /* 116136af9869SEdward Tomasz Napierala * To ensure that processes are throttled in a fair way, we need 116236af9869SEdward Tomasz Napierala * to iterate over all processes again and check the limits 116336af9869SEdward Tomasz Napierala * for %cpu resource only after ucred racct containers have been 116436af9869SEdward Tomasz Napierala * properly filled. 116536af9869SEdward Tomasz Napierala */ 116636af9869SEdward Tomasz Napierala FOREACH_PROC_IN_SYSTEM(p) { 116736af9869SEdward Tomasz Napierala PROC_LOCK(p); 116836af9869SEdward Tomasz Napierala if (p->p_state != PRS_NORMAL) { 116936af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 117036af9869SEdward Tomasz Napierala continue; 117136af9869SEdward Tomasz Napierala } 117236af9869SEdward Tomasz Napierala 117336af9869SEdward Tomasz Napierala if (racct_pcpu_available(p) <= 0) 117436af9869SEdward Tomasz Napierala racct_proc_throttle(p); 117536af9869SEdward Tomasz Napierala else if (p->p_throttled) 117636af9869SEdward Tomasz Napierala racct_proc_wakeup(p); 117736af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 117836af9869SEdward Tomasz Napierala } 1179097055e2SEdward Tomasz Napierala sx_sunlock(&allproc_lock); 1180097055e2SEdward Tomasz Napierala pause("-", hz); 1181097055e2SEdward Tomasz Napierala } 1182097055e2SEdward Tomasz Napierala } 1183097055e2SEdward Tomasz Napierala 1184097055e2SEdward Tomasz Napierala static struct kproc_desc racctd_kp = { 1185097055e2SEdward Tomasz Napierala "racctd", 1186097055e2SEdward Tomasz Napierala racctd, 1187097055e2SEdward Tomasz Napierala NULL 1188097055e2SEdward Tomasz Napierala }; 1189097055e2SEdward Tomasz Napierala SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, kproc_start, &racctd_kp); 1190097055e2SEdward Tomasz Napierala 1191097055e2SEdward Tomasz Napierala static void 1192097055e2SEdward Tomasz Napierala racct_init(void) 1193097055e2SEdward Tomasz Napierala { 1194097055e2SEdward Tomasz Napierala 1195097055e2SEdward Tomasz Napierala racct_zone = uma_zcreate("racct", sizeof(struct racct), 1196097055e2SEdward Tomasz Napierala NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1197097055e2SEdward Tomasz Napierala /* 1198097055e2SEdward Tomasz Napierala * XXX: Move this somewhere. 1199097055e2SEdward Tomasz Napierala */ 1200a7ad07bfSEdward Tomasz Napierala prison0.pr_prison_racct = prison_racct_find("0"); 1201097055e2SEdward Tomasz Napierala } 1202097055e2SEdward Tomasz Napierala SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL); 1203097055e2SEdward Tomasz Napierala 1204097055e2SEdward Tomasz Napierala #else /* !RACCT */ 1205097055e2SEdward Tomasz Napierala 1206097055e2SEdward Tomasz Napierala int 1207097055e2SEdward Tomasz Napierala racct_add(struct proc *p, int resource, uint64_t amount) 1208097055e2SEdward Tomasz Napierala { 1209097055e2SEdward Tomasz Napierala 1210097055e2SEdward Tomasz Napierala return (0); 1211097055e2SEdward Tomasz Napierala } 1212097055e2SEdward Tomasz Napierala 1213097055e2SEdward Tomasz Napierala void 1214097055e2SEdward Tomasz Napierala racct_add_cred(struct ucred *cred, int resource, uint64_t amount) 1215097055e2SEdward Tomasz Napierala { 1216097055e2SEdward Tomasz Napierala } 1217097055e2SEdward Tomasz Napierala 1218097055e2SEdward Tomasz Napierala void 1219097055e2SEdward Tomasz Napierala racct_add_force(struct proc *p, int resource, uint64_t amount) 1220097055e2SEdward Tomasz Napierala { 1221097055e2SEdward Tomasz Napierala 1222097055e2SEdward Tomasz Napierala return; 1223097055e2SEdward Tomasz Napierala } 1224097055e2SEdward Tomasz Napierala 1225097055e2SEdward Tomasz Napierala int 1226097055e2SEdward Tomasz Napierala racct_set(struct proc *p, int resource, uint64_t amount) 1227097055e2SEdward Tomasz Napierala { 1228097055e2SEdward Tomasz Napierala 1229097055e2SEdward Tomasz Napierala return (0); 1230097055e2SEdward Tomasz Napierala } 1231097055e2SEdward Tomasz Napierala 1232097055e2SEdward Tomasz Napierala void 1233c98fe0a5SEdward Tomasz Napierala racct_set_force(struct proc *p, int resource, uint64_t amount) 1234c98fe0a5SEdward Tomasz Napierala { 1235c98fe0a5SEdward Tomasz Napierala } 1236c98fe0a5SEdward Tomasz Napierala 1237c98fe0a5SEdward Tomasz Napierala void 1238097055e2SEdward Tomasz Napierala racct_sub(struct proc *p, int resource, uint64_t amount) 1239097055e2SEdward Tomasz Napierala { 1240097055e2SEdward Tomasz Napierala } 1241097055e2SEdward Tomasz Napierala 1242097055e2SEdward Tomasz Napierala void 1243097055e2SEdward Tomasz Napierala racct_sub_cred(struct ucred *cred, int resource, uint64_t amount) 1244097055e2SEdward Tomasz Napierala { 1245097055e2SEdward Tomasz Napierala } 1246097055e2SEdward Tomasz Napierala 1247097055e2SEdward Tomasz Napierala uint64_t 1248097055e2SEdward Tomasz Napierala racct_get_limit(struct proc *p, int resource) 1249097055e2SEdward Tomasz Napierala { 1250097055e2SEdward Tomasz Napierala 1251097055e2SEdward Tomasz Napierala return (UINT64_MAX); 1252097055e2SEdward Tomasz Napierala } 1253097055e2SEdward Tomasz Napierala 1254c98fe0a5SEdward Tomasz Napierala uint64_t 1255c98fe0a5SEdward Tomasz Napierala racct_get_available(struct proc *p, int resource) 1256c98fe0a5SEdward Tomasz Napierala { 1257c98fe0a5SEdward Tomasz Napierala 1258c98fe0a5SEdward Tomasz Napierala return (UINT64_MAX); 1259c98fe0a5SEdward Tomasz Napierala } 1260c98fe0a5SEdward Tomasz Napierala 1261097055e2SEdward Tomasz Napierala void 1262097055e2SEdward Tomasz Napierala racct_create(struct racct **racctp) 1263097055e2SEdward Tomasz Napierala { 1264097055e2SEdward Tomasz Napierala } 1265097055e2SEdward Tomasz Napierala 1266097055e2SEdward Tomasz Napierala void 1267097055e2SEdward Tomasz Napierala racct_destroy(struct racct **racctp) 1268097055e2SEdward Tomasz Napierala { 1269097055e2SEdward Tomasz Napierala } 1270097055e2SEdward Tomasz Napierala 1271097055e2SEdward Tomasz Napierala int 1272097055e2SEdward Tomasz Napierala racct_proc_fork(struct proc *parent, struct proc *child) 1273097055e2SEdward Tomasz Napierala { 1274097055e2SEdward Tomasz Napierala 1275097055e2SEdward Tomasz Napierala return (0); 1276097055e2SEdward Tomasz Napierala } 1277097055e2SEdward Tomasz Napierala 1278097055e2SEdward Tomasz Napierala void 127972a401d9SEdward Tomasz Napierala racct_proc_fork_done(struct proc *child) 128072a401d9SEdward Tomasz Napierala { 128172a401d9SEdward Tomasz Napierala } 128272a401d9SEdward Tomasz Napierala 128372a401d9SEdward Tomasz Napierala void 1284097055e2SEdward Tomasz Napierala racct_proc_exit(struct proc *p) 1285097055e2SEdward Tomasz Napierala { 1286097055e2SEdward Tomasz Napierala } 1287097055e2SEdward Tomasz Napierala 1288097055e2SEdward Tomasz Napierala #endif /* !RACCT */ 1289