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 3536af9869SEdward Tomasz Napierala #include "opt_sched.h" 36097055e2SEdward Tomasz Napierala 37097055e2SEdward Tomasz Napierala #include <sys/param.h> 380e225211SAndriy Gapon #include <sys/systm.h> 39097055e2SEdward Tomasz Napierala #include <sys/eventhandler.h> 40097055e2SEdward Tomasz Napierala #include <sys/jail.h> 41097055e2SEdward Tomasz Napierala #include <sys/kernel.h> 42097055e2SEdward Tomasz Napierala #include <sys/kthread.h> 43097055e2SEdward Tomasz Napierala #include <sys/lock.h> 44097055e2SEdward Tomasz Napierala #include <sys/loginclass.h> 45097055e2SEdward Tomasz Napierala #include <sys/malloc.h> 46097055e2SEdward Tomasz Napierala #include <sys/mutex.h> 47097055e2SEdward Tomasz Napierala #include <sys/proc.h> 48097055e2SEdward Tomasz Napierala #include <sys/racct.h> 49097055e2SEdward Tomasz Napierala #include <sys/resourcevar.h> 50097055e2SEdward Tomasz Napierala #include <sys/sbuf.h> 51097055e2SEdward Tomasz Napierala #include <sys/sched.h> 52097055e2SEdward Tomasz Napierala #include <sys/sdt.h> 5336af9869SEdward Tomasz Napierala #include <sys/smp.h> 54097055e2SEdward Tomasz Napierala #include <sys/sx.h> 5536af9869SEdward Tomasz Napierala #include <sys/sysctl.h> 56097055e2SEdward Tomasz Napierala #include <sys/sysent.h> 57097055e2SEdward Tomasz Napierala #include <sys/sysproto.h> 58097055e2SEdward Tomasz Napierala #include <sys/umtx.h> 5936af9869SEdward Tomasz Napierala #include <machine/smp.h> 60097055e2SEdward Tomasz Napierala 61097055e2SEdward Tomasz Napierala #ifdef RCTL 62097055e2SEdward Tomasz Napierala #include <sys/rctl.h> 63097055e2SEdward Tomasz Napierala #endif 64097055e2SEdward Tomasz Napierala 65097055e2SEdward Tomasz Napierala #ifdef RACCT 66097055e2SEdward Tomasz Napierala 67097055e2SEdward Tomasz Napierala FEATURE(racct, "Resource Accounting"); 68097055e2SEdward Tomasz Napierala 6936af9869SEdward Tomasz Napierala /* 7036af9869SEdward Tomasz Napierala * Do not block processes that have their %cpu usage <= pcpu_threshold. 7136af9869SEdward Tomasz Napierala */ 7236af9869SEdward Tomasz Napierala static int pcpu_threshold = 1; 73ba8f0eb8SEdward Tomasz Napierala #ifdef RACCT_DEFAULT_TO_DISABLED 744b5c9cf6SEdward Tomasz Napierala int racct_enable = 0; 754b5c9cf6SEdward Tomasz Napierala #else 764b5c9cf6SEdward Tomasz Napierala int racct_enable = 1; 774b5c9cf6SEdward Tomasz Napierala #endif 7836af9869SEdward Tomasz Napierala 7936af9869SEdward Tomasz Napierala SYSCTL_NODE(_kern, OID_AUTO, racct, CTLFLAG_RW, 0, "Resource Accounting"); 804b5c9cf6SEdward Tomasz Napierala SYSCTL_UINT(_kern_racct, OID_AUTO, enable, CTLFLAG_RDTUN, &racct_enable, 814b5c9cf6SEdward Tomasz Napierala 0, "Enable RACCT/RCTL"); 8236af9869SEdward Tomasz Napierala SYSCTL_UINT(_kern_racct, OID_AUTO, pcpu_threshold, CTLFLAG_RW, &pcpu_threshold, 8336af9869SEdward Tomasz Napierala 0, "Processes with higher %cpu usage than this value can be throttled."); 8436af9869SEdward Tomasz Napierala 8536af9869SEdward Tomasz Napierala /* 8636af9869SEdward Tomasz Napierala * How many seconds it takes to use the scheduler %cpu calculations. When a 8736af9869SEdward Tomasz Napierala * process starts, we compute its %cpu usage by dividing its runtime by the 8836af9869SEdward Tomasz Napierala * process wall clock time. After RACCT_PCPU_SECS pass, we use the value 8936af9869SEdward Tomasz Napierala * provided by the scheduler. 9036af9869SEdward Tomasz Napierala */ 9136af9869SEdward Tomasz Napierala #define RACCT_PCPU_SECS 3 9236af9869SEdward Tomasz Napierala 93097055e2SEdward Tomasz Napierala static struct mtx racct_lock; 94097055e2SEdward Tomasz Napierala MTX_SYSINIT(racct_lock, &racct_lock, "racct lock", MTX_DEF); 95097055e2SEdward Tomasz Napierala 96*4c230cdaSEdward Tomasz Napierala #define RACCT_LOCK() mtx_lock(&racct_lock) 97*4c230cdaSEdward Tomasz Napierala #define RACCT_UNLOCK() mtx_unlock(&racct_lock) 98*4c230cdaSEdward Tomasz Napierala #define RACCT_LOCK_ASSERT() mtx_assert(&racct_lock, MA_OWNED) 99*4c230cdaSEdward Tomasz Napierala 100097055e2SEdward Tomasz Napierala static uma_zone_t racct_zone; 101097055e2SEdward Tomasz Napierala 102097055e2SEdward Tomasz Napierala static void racct_sub_racct(struct racct *dest, const struct racct *src); 103097055e2SEdward Tomasz Napierala static void racct_sub_cred_locked(struct ucred *cred, int resource, 104097055e2SEdward Tomasz Napierala uint64_t amount); 105097055e2SEdward Tomasz Napierala static void racct_add_cred_locked(struct ucred *cred, int resource, 106097055e2SEdward Tomasz Napierala uint64_t amount); 107097055e2SEdward Tomasz Napierala 108097055e2SEdward Tomasz Napierala SDT_PROVIDER_DEFINE(racct); 10936160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, add, 110097055e2SEdward Tomasz Napierala "struct proc *", "int", "uint64_t"); 11136160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, add__failure, 112097055e2SEdward Tomasz Napierala "struct proc *", "int", "uint64_t"); 11336160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, add__cred, 11436160958SMark Johnston "struct ucred *", "int", "uint64_t"); 11536160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, add__force, 11636160958SMark Johnston "struct proc *", "int", "uint64_t"); 11736160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, set, 11836160958SMark Johnston "struct proc *", "int", "uint64_t"); 11936160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, set__failure, 12036160958SMark Johnston "struct proc *", "int", "uint64_t"); 12110287198SEdward Tomasz Napierala SDT_PROBE_DEFINE3(racct, , rusage, set__force, 12210287198SEdward Tomasz Napierala "struct proc *", "int", "uint64_t"); 12336160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, sub, 12436160958SMark Johnston "struct proc *", "int", "uint64_t"); 12536160958SMark Johnston SDT_PROBE_DEFINE3(racct, , rusage, sub__cred, 12636160958SMark Johnston "struct ucred *", "int", "uint64_t"); 12736160958SMark Johnston SDT_PROBE_DEFINE1(racct, , racct, create, 128097055e2SEdward Tomasz Napierala "struct racct *"); 12936160958SMark Johnston SDT_PROBE_DEFINE1(racct, , racct, destroy, 13036160958SMark Johnston "struct racct *"); 13136160958SMark Johnston SDT_PROBE_DEFINE2(racct, , racct, join, 132097055e2SEdward Tomasz Napierala "struct racct *", "struct racct *"); 13336160958SMark Johnston SDT_PROBE_DEFINE2(racct, , racct, join__failure, 13436160958SMark Johnston "struct racct *", "struct racct *"); 13536160958SMark Johnston SDT_PROBE_DEFINE2(racct, , racct, leave, 13636160958SMark Johnston "struct racct *", "struct racct *"); 137097055e2SEdward Tomasz Napierala 138097055e2SEdward Tomasz Napierala int racct_types[] = { 139097055e2SEdward Tomasz Napierala [RACCT_CPU] = 14085a2f1b4SEdward Tomasz Napierala RACCT_IN_MILLIONS, 141097055e2SEdward Tomasz Napierala [RACCT_DATA] = 142097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 143097055e2SEdward Tomasz Napierala [RACCT_STACK] = 144097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 145097055e2SEdward Tomasz Napierala [RACCT_CORE] = 146097055e2SEdward Tomasz Napierala RACCT_DENIABLE, 147097055e2SEdward Tomasz Napierala [RACCT_RSS] = 148097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE, 149097055e2SEdward Tomasz Napierala [RACCT_MEMLOCK] = 150097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 151097055e2SEdward Tomasz Napierala [RACCT_NPROC] = 152097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 153097055e2SEdward Tomasz Napierala [RACCT_NOFILE] = 154097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 155097055e2SEdward Tomasz Napierala [RACCT_VMEM] = 156097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 157097055e2SEdward Tomasz Napierala [RACCT_NPTS] = 158097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 159097055e2SEdward Tomasz Napierala [RACCT_SWAP] = 160097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 161097055e2SEdward Tomasz Napierala [RACCT_NTHR] = 162097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE, 163097055e2SEdward Tomasz Napierala [RACCT_MSGQQUEUED] = 164097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 165097055e2SEdward Tomasz Napierala [RACCT_MSGQSIZE] = 166097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 167097055e2SEdward Tomasz Napierala [RACCT_NMSGQ] = 168097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 169097055e2SEdward Tomasz Napierala [RACCT_NSEM] = 170097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 171097055e2SEdward Tomasz Napierala [RACCT_NSEMOP] = 172097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_INHERITABLE | RACCT_DENIABLE, 173097055e2SEdward Tomasz Napierala [RACCT_NSHM] = 174097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 175097055e2SEdward Tomasz Napierala [RACCT_SHMSIZE] = 176097055e2SEdward Tomasz Napierala RACCT_RECLAIMABLE | RACCT_DENIABLE | RACCT_SLOPPY, 177097055e2SEdward Tomasz Napierala [RACCT_WALLCLOCK] = 17836af9869SEdward Tomasz Napierala RACCT_IN_MILLIONS, 17936af9869SEdward Tomasz Napierala [RACCT_PCTCPU] = 18036af9869SEdward Tomasz Napierala RACCT_DECAYING | RACCT_DENIABLE | RACCT_IN_MILLIONS }; 18136af9869SEdward Tomasz Napierala 18236af9869SEdward Tomasz Napierala static const fixpt_t RACCT_DECAY_FACTOR = 0.3 * FSCALE; 18336af9869SEdward Tomasz Napierala 18436af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 18536af9869SEdward Tomasz Napierala /* 18636af9869SEdward Tomasz Napierala * Contains intermediate values for %cpu calculations to avoid using floating 18736af9869SEdward Tomasz Napierala * point in the kernel. 18836af9869SEdward Tomasz Napierala * ccpu_exp[k] = FSCALE * (ccpu/FSCALE)^k = FSCALE * exp(-k/20) 18936af9869SEdward Tomasz Napierala * It is needed only for the 4BSD scheduler, because in ULE, the ccpu equals to 19036af9869SEdward Tomasz Napierala * zero so the calculations are more straightforward. 19136af9869SEdward Tomasz Napierala */ 19236af9869SEdward Tomasz Napierala fixpt_t ccpu_exp[] = { 19336af9869SEdward Tomasz Napierala [0] = FSCALE * 1, 19436af9869SEdward Tomasz Napierala [1] = FSCALE * 0.95122942450071400909, 19536af9869SEdward Tomasz Napierala [2] = FSCALE * 0.90483741803595957316, 19636af9869SEdward Tomasz Napierala [3] = FSCALE * 0.86070797642505780722, 19736af9869SEdward Tomasz Napierala [4] = FSCALE * 0.81873075307798185866, 19836af9869SEdward Tomasz Napierala [5] = FSCALE * 0.77880078307140486824, 19936af9869SEdward Tomasz Napierala [6] = FSCALE * 0.74081822068171786606, 20036af9869SEdward Tomasz Napierala [7] = FSCALE * 0.70468808971871343435, 20136af9869SEdward Tomasz Napierala [8] = FSCALE * 0.67032004603563930074, 20236af9869SEdward Tomasz Napierala [9] = FSCALE * 0.63762815162177329314, 20336af9869SEdward Tomasz Napierala [10] = FSCALE * 0.60653065971263342360, 20436af9869SEdward Tomasz Napierala [11] = FSCALE * 0.57694981038048669531, 20536af9869SEdward Tomasz Napierala [12] = FSCALE * 0.54881163609402643262, 20636af9869SEdward Tomasz Napierala [13] = FSCALE * 0.52204577676101604789, 20736af9869SEdward Tomasz Napierala [14] = FSCALE * 0.49658530379140951470, 20836af9869SEdward Tomasz Napierala [15] = FSCALE * 0.47236655274101470713, 20936af9869SEdward Tomasz Napierala [16] = FSCALE * 0.44932896411722159143, 21036af9869SEdward Tomasz Napierala [17] = FSCALE * 0.42741493194872666992, 21136af9869SEdward Tomasz Napierala [18] = FSCALE * 0.40656965974059911188, 21236af9869SEdward Tomasz Napierala [19] = FSCALE * 0.38674102345450120691, 21336af9869SEdward Tomasz Napierala [20] = FSCALE * 0.36787944117144232159, 21436af9869SEdward Tomasz Napierala [21] = FSCALE * 0.34993774911115535467, 21536af9869SEdward Tomasz Napierala [22] = FSCALE * 0.33287108369807955328, 21636af9869SEdward Tomasz Napierala [23] = FSCALE * 0.31663676937905321821, 21736af9869SEdward Tomasz Napierala [24] = FSCALE * 0.30119421191220209664, 21836af9869SEdward Tomasz Napierala [25] = FSCALE * 0.28650479686019010032, 21936af9869SEdward Tomasz Napierala [26] = FSCALE * 0.27253179303401260312, 22036af9869SEdward Tomasz Napierala [27] = FSCALE * 0.25924026064589150757, 22136af9869SEdward Tomasz Napierala [28] = FSCALE * 0.24659696394160647693, 22236af9869SEdward Tomasz Napierala [29] = FSCALE * 0.23457028809379765313, 22336af9869SEdward Tomasz Napierala [30] = FSCALE * 0.22313016014842982893, 22436af9869SEdward Tomasz Napierala [31] = FSCALE * 0.21224797382674305771, 22536af9869SEdward Tomasz Napierala [32] = FSCALE * 0.20189651799465540848, 22636af9869SEdward Tomasz Napierala [33] = FSCALE * 0.19204990862075411423, 22736af9869SEdward Tomasz Napierala [34] = FSCALE * 0.18268352405273465022, 22836af9869SEdward Tomasz Napierala [35] = FSCALE * 0.17377394345044512668, 22936af9869SEdward Tomasz Napierala [36] = FSCALE * 0.16529888822158653829, 23036af9869SEdward Tomasz Napierala [37] = FSCALE * 0.15723716631362761621, 23136af9869SEdward Tomasz Napierala [38] = FSCALE * 0.14956861922263505264, 23236af9869SEdward Tomasz Napierala [39] = FSCALE * 0.14227407158651357185, 23336af9869SEdward Tomasz Napierala [40] = FSCALE * 0.13533528323661269189, 23436af9869SEdward Tomasz Napierala [41] = FSCALE * 0.12873490358780421886, 23536af9869SEdward Tomasz Napierala [42] = FSCALE * 0.12245642825298191021, 23636af9869SEdward Tomasz Napierala [43] = FSCALE * 0.11648415777349695786, 23736af9869SEdward Tomasz Napierala [44] = FSCALE * 0.11080315836233388333, 23836af9869SEdward Tomasz Napierala [45] = FSCALE * 0.10539922456186433678, 23936af9869SEdward Tomasz Napierala [46] = FSCALE * 0.10025884372280373372, 24036af9869SEdward Tomasz Napierala [47] = FSCALE * 0.09536916221554961888, 24136af9869SEdward Tomasz Napierala [48] = FSCALE * 0.09071795328941250337, 24236af9869SEdward Tomasz Napierala [49] = FSCALE * 0.08629358649937051097, 24336af9869SEdward Tomasz Napierala [50] = FSCALE * 0.08208499862389879516, 24436af9869SEdward Tomasz Napierala [51] = FSCALE * 0.07808166600115315231, 24536af9869SEdward Tomasz Napierala [52] = FSCALE * 0.07427357821433388042, 24636af9869SEdward Tomasz Napierala [53] = FSCALE * 0.07065121306042958674, 24736af9869SEdward Tomasz Napierala [54] = FSCALE * 0.06720551273974976512, 24836af9869SEdward Tomasz Napierala [55] = FSCALE * 0.06392786120670757270, 24936af9869SEdward Tomasz Napierala [56] = FSCALE * 0.06081006262521796499, 25036af9869SEdward Tomasz Napierala [57] = FSCALE * 0.05784432087483846296, 25136af9869SEdward Tomasz Napierala [58] = FSCALE * 0.05502322005640722902, 25236af9869SEdward Tomasz Napierala [59] = FSCALE * 0.05233970594843239308, 25336af9869SEdward Tomasz Napierala [60] = FSCALE * 0.04978706836786394297, 25436af9869SEdward Tomasz Napierala [61] = FSCALE * 0.04735892439114092119, 25536af9869SEdward Tomasz Napierala [62] = FSCALE * 0.04504920239355780606, 25636af9869SEdward Tomasz Napierala [63] = FSCALE * 0.04285212686704017991, 25736af9869SEdward Tomasz Napierala [64] = FSCALE * 0.04076220397836621516, 25836af9869SEdward Tomasz Napierala [65] = FSCALE * 0.03877420783172200988, 25936af9869SEdward Tomasz Napierala [66] = FSCALE * 0.03688316740124000544, 26036af9869SEdward Tomasz Napierala [67] = FSCALE * 0.03508435410084502588, 26136af9869SEdward Tomasz Napierala [68] = FSCALE * 0.03337326996032607948, 26236af9869SEdward Tomasz Napierala [69] = FSCALE * 0.03174563637806794323, 26336af9869SEdward Tomasz Napierala [70] = FSCALE * 0.03019738342231850073, 26436af9869SEdward Tomasz Napierala [71] = FSCALE * 0.02872463965423942912, 26536af9869SEdward Tomasz Napierala [72] = FSCALE * 0.02732372244729256080, 26636af9869SEdward Tomasz Napierala [73] = FSCALE * 0.02599112877875534358, 26736af9869SEdward Tomasz Napierala [74] = FSCALE * 0.02472352647033939120, 26836af9869SEdward Tomasz Napierala [75] = FSCALE * 0.02351774585600910823, 26936af9869SEdward Tomasz Napierala [76] = FSCALE * 0.02237077185616559577, 27036af9869SEdward Tomasz Napierala [77] = FSCALE * 0.02127973643837716938, 27136af9869SEdward Tomasz Napierala [78] = FSCALE * 0.02024191144580438847, 27236af9869SEdward Tomasz Napierala [79] = FSCALE * 0.01925470177538692429, 27336af9869SEdward Tomasz Napierala [80] = FSCALE * 0.01831563888873418029, 27436af9869SEdward Tomasz Napierala [81] = FSCALE * 0.01742237463949351138, 27536af9869SEdward Tomasz Napierala [82] = FSCALE * 0.01657267540176124754, 27636af9869SEdward Tomasz Napierala [83] = FSCALE * 0.01576441648485449082, 27736af9869SEdward Tomasz Napierala [84] = FSCALE * 0.01499557682047770621, 27836af9869SEdward Tomasz Napierala [85] = FSCALE * 0.01426423390899925527, 27936af9869SEdward Tomasz Napierala [86] = FSCALE * 0.01356855901220093175, 28036af9869SEdward Tomasz Napierala [87] = FSCALE * 0.01290681258047986886, 28136af9869SEdward Tomasz Napierala [88] = FSCALE * 0.01227733990306844117, 28236af9869SEdward Tomasz Napierala [89] = FSCALE * 0.01167856697039544521, 28336af9869SEdward Tomasz Napierala [90] = FSCALE * 0.01110899653824230649, 28436af9869SEdward Tomasz Napierala [91] = FSCALE * 0.01056720438385265337, 28536af9869SEdward Tomasz Napierala [92] = FSCALE * 0.01005183574463358164, 28636af9869SEdward Tomasz Napierala [93] = FSCALE * 0.00956160193054350793, 28736af9869SEdward Tomasz Napierala [94] = FSCALE * 0.00909527710169581709, 28836af9869SEdward Tomasz Napierala [95] = FSCALE * 0.00865169520312063417, 28936af9869SEdward Tomasz Napierala [96] = FSCALE * 0.00822974704902002884, 29036af9869SEdward Tomasz Napierala [97] = FSCALE * 0.00782837754922577143, 29136af9869SEdward Tomasz Napierala [98] = FSCALE * 0.00744658307092434051, 29236af9869SEdward Tomasz Napierala [99] = FSCALE * 0.00708340892905212004, 29336af9869SEdward Tomasz Napierala [100] = FSCALE * 0.00673794699908546709, 29436af9869SEdward Tomasz Napierala [101] = FSCALE * 0.00640933344625638184, 29536af9869SEdward Tomasz Napierala [102] = FSCALE * 0.00609674656551563610, 29636af9869SEdward Tomasz Napierala [103] = FSCALE * 0.00579940472684214321, 29736af9869SEdward Tomasz Napierala [104] = FSCALE * 0.00551656442076077241, 29836af9869SEdward Tomasz Napierala [105] = FSCALE * 0.00524751839918138427, 29936af9869SEdward Tomasz Napierala [106] = FSCALE * 0.00499159390691021621, 30036af9869SEdward Tomasz Napierala [107] = FSCALE * 0.00474815099941147558, 30136af9869SEdward Tomasz Napierala [108] = FSCALE * 0.00451658094261266798, 30236af9869SEdward Tomasz Napierala [109] = FSCALE * 0.00429630469075234057, 30336af9869SEdward Tomasz Napierala [110] = FSCALE * 0.00408677143846406699, 30436af9869SEdward Tomasz Napierala }; 30536af9869SEdward Tomasz Napierala #endif 30636af9869SEdward Tomasz Napierala 30736af9869SEdward Tomasz Napierala #define CCPU_EXP_MAX 110 30836af9869SEdward Tomasz Napierala 30936af9869SEdward Tomasz Napierala /* 31036af9869SEdward Tomasz Napierala * This function is analogical to the getpcpu() function in the ps(1) command. 31136af9869SEdward Tomasz Napierala * They should both calculate in the same way so that the racct %cpu 31236af9869SEdward Tomasz Napierala * calculations are consistent with the values showed by the ps(1) tool. 31336af9869SEdward Tomasz Napierala * The calculations are more complex in the 4BSD scheduler because of the value 31436af9869SEdward Tomasz Napierala * of the ccpu variable. In ULE it is defined to be zero which saves us some 31536af9869SEdward Tomasz Napierala * work. 31636af9869SEdward Tomasz Napierala */ 31736af9869SEdward Tomasz Napierala static uint64_t 31836af9869SEdward Tomasz Napierala racct_getpcpu(struct proc *p, u_int pcpu) 31936af9869SEdward Tomasz Napierala { 32036af9869SEdward Tomasz Napierala u_int swtime; 32136af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 32236af9869SEdward Tomasz Napierala fixpt_t pctcpu, pctcpu_next; 32336af9869SEdward Tomasz Napierala #endif 32436af9869SEdward Tomasz Napierala #ifdef SMP 32536af9869SEdward Tomasz Napierala struct pcpu *pc; 32636af9869SEdward Tomasz Napierala int found; 32736af9869SEdward Tomasz Napierala #endif 32836af9869SEdward Tomasz Napierala fixpt_t p_pctcpu; 32936af9869SEdward Tomasz Napierala struct thread *td; 33036af9869SEdward Tomasz Napierala 3314b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 3324b5c9cf6SEdward Tomasz Napierala 33336af9869SEdward Tomasz Napierala /* 33436af9869SEdward Tomasz Napierala * If the process is swapped out, we count its %cpu usage as zero. 33536af9869SEdward Tomasz Napierala * This behaviour is consistent with the userland ps(1) tool. 33636af9869SEdward Tomasz Napierala */ 33736af9869SEdward Tomasz Napierala if ((p->p_flag & P_INMEM) == 0) 33836af9869SEdward Tomasz Napierala return (0); 33936af9869SEdward Tomasz Napierala swtime = (ticks - p->p_swtick) / hz; 34036af9869SEdward Tomasz Napierala 34136af9869SEdward Tomasz Napierala /* 34236af9869SEdward Tomasz Napierala * For short-lived processes, the sched_pctcpu() returns small 34336af9869SEdward Tomasz Napierala * values even for cpu intensive processes. Therefore we use 34436af9869SEdward Tomasz Napierala * our own estimate in this case. 34536af9869SEdward Tomasz Napierala */ 34636af9869SEdward Tomasz Napierala if (swtime < RACCT_PCPU_SECS) 34736af9869SEdward Tomasz Napierala return (pcpu); 34836af9869SEdward Tomasz Napierala 34936af9869SEdward Tomasz Napierala p_pctcpu = 0; 35036af9869SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) { 35136af9869SEdward Tomasz Napierala if (td == PCPU_GET(idlethread)) 35236af9869SEdward Tomasz Napierala continue; 35336af9869SEdward Tomasz Napierala #ifdef SMP 35436af9869SEdward Tomasz Napierala found = 0; 35536af9869SEdward Tomasz Napierala STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 35636af9869SEdward Tomasz Napierala if (td == pc->pc_idlethread) { 35736af9869SEdward Tomasz Napierala found = 1; 35836af9869SEdward Tomasz Napierala break; 35936af9869SEdward Tomasz Napierala } 36036af9869SEdward Tomasz Napierala } 36136af9869SEdward Tomasz Napierala if (found) 36236af9869SEdward Tomasz Napierala continue; 36336af9869SEdward Tomasz Napierala #endif 36436af9869SEdward Tomasz Napierala thread_lock(td); 36536af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 36636af9869SEdward Tomasz Napierala pctcpu = sched_pctcpu(td); 36736af9869SEdward Tomasz Napierala /* Count also the yet unfinished second. */ 36836af9869SEdward Tomasz Napierala pctcpu_next = (pctcpu * ccpu_exp[1]) >> FSHIFT; 36936af9869SEdward Tomasz Napierala pctcpu_next += sched_pctcpu_delta(td); 37036af9869SEdward Tomasz Napierala p_pctcpu += max(pctcpu, pctcpu_next); 37136af9869SEdward Tomasz Napierala #else 37236af9869SEdward Tomasz Napierala /* 37336af9869SEdward Tomasz Napierala * In ULE the %cpu statistics are updated on every 37436af9869SEdward Tomasz Napierala * sched_pctcpu() call. So special calculations to 37536af9869SEdward Tomasz Napierala * account for the latest (unfinished) second are 37636af9869SEdward Tomasz Napierala * not needed. 37736af9869SEdward Tomasz Napierala */ 37836af9869SEdward Tomasz Napierala p_pctcpu += sched_pctcpu(td); 37936af9869SEdward Tomasz Napierala #endif 38036af9869SEdward Tomasz Napierala thread_unlock(td); 38136af9869SEdward Tomasz Napierala } 38236af9869SEdward Tomasz Napierala 38336af9869SEdward Tomasz Napierala #ifdef SCHED_4BSD 38436af9869SEdward Tomasz Napierala if (swtime <= CCPU_EXP_MAX) 38536af9869SEdward Tomasz Napierala return ((100 * (uint64_t)p_pctcpu * 1000000) / 38636af9869SEdward Tomasz Napierala (FSCALE - ccpu_exp[swtime])); 38736af9869SEdward Tomasz Napierala #endif 38836af9869SEdward Tomasz Napierala 38936af9869SEdward Tomasz Napierala return ((100 * (uint64_t)p_pctcpu * 1000000) / FSCALE); 39036af9869SEdward Tomasz Napierala } 391097055e2SEdward Tomasz Napierala 392097055e2SEdward Tomasz Napierala static void 393097055e2SEdward Tomasz Napierala racct_add_racct(struct racct *dest, const struct racct *src) 394097055e2SEdward Tomasz Napierala { 395097055e2SEdward Tomasz Napierala int i; 396097055e2SEdward Tomasz Napierala 3974b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 398*4c230cdaSEdward Tomasz Napierala RACCT_LOCK_ASSERT(); 399097055e2SEdward Tomasz Napierala 400097055e2SEdward Tomasz Napierala /* 401097055e2SEdward Tomasz Napierala * Update resource usage in dest. 402097055e2SEdward Tomasz Napierala */ 403097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 404097055e2SEdward Tomasz Napierala KASSERT(dest->r_resources[i] >= 0, 405baf85d0aSEdward Tomasz Napierala ("%s: resource %d propagation meltdown: dest < 0", 406baf85d0aSEdward Tomasz Napierala __func__, i)); 407097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] >= 0, 408baf85d0aSEdward Tomasz Napierala ("%s: resource %d propagation meltdown: src < 0", 409baf85d0aSEdward Tomasz Napierala __func__, i)); 410097055e2SEdward Tomasz Napierala dest->r_resources[i] += src->r_resources[i]; 411097055e2SEdward Tomasz Napierala } 412097055e2SEdward Tomasz Napierala } 413097055e2SEdward Tomasz Napierala 414097055e2SEdward Tomasz Napierala static void 415097055e2SEdward Tomasz Napierala racct_sub_racct(struct racct *dest, const struct racct *src) 416097055e2SEdward Tomasz Napierala { 417097055e2SEdward Tomasz Napierala int i; 418097055e2SEdward Tomasz Napierala 4194b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 420*4c230cdaSEdward Tomasz Napierala RACCT_LOCK_ASSERT(); 421097055e2SEdward Tomasz Napierala 422097055e2SEdward Tomasz Napierala /* 423097055e2SEdward Tomasz Napierala * Update resource usage in dest. 424097055e2SEdward Tomasz Napierala */ 425097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 42684c9193bSEdward Tomasz Napierala if (!RACCT_IS_SLOPPY(i) && !RACCT_IS_DECAYING(i)) { 427097055e2SEdward Tomasz Napierala KASSERT(dest->r_resources[i] >= 0, 428baf85d0aSEdward Tomasz Napierala ("%s: resource %d propagation meltdown: dest < 0", 429baf85d0aSEdward Tomasz Napierala __func__, i)); 430097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] >= 0, 431baf85d0aSEdward Tomasz Napierala ("%s: resource %d propagation meltdown: src < 0", 432baf85d0aSEdward Tomasz Napierala __func__, i)); 433097055e2SEdward Tomasz Napierala KASSERT(src->r_resources[i] <= dest->r_resources[i], 434baf85d0aSEdward Tomasz Napierala ("%s: resource %d propagation meltdown: src > dest", 435baf85d0aSEdward Tomasz Napierala __func__, i)); 436097055e2SEdward Tomasz Napierala } 43736af9869SEdward Tomasz Napierala if (RACCT_CAN_DROP(i)) { 438097055e2SEdward Tomasz Napierala dest->r_resources[i] -= src->r_resources[i]; 439097055e2SEdward Tomasz Napierala if (dest->r_resources[i] < 0) { 44084c9193bSEdward Tomasz Napierala KASSERT(RACCT_IS_SLOPPY(i) || 44184c9193bSEdward Tomasz Napierala RACCT_IS_DECAYING(i), 442baf85d0aSEdward Tomasz Napierala ("%s: resource %d usage < 0", __func__, i)); 443097055e2SEdward Tomasz Napierala dest->r_resources[i] = 0; 444097055e2SEdward Tomasz Napierala } 445097055e2SEdward Tomasz Napierala } 446097055e2SEdward Tomasz Napierala } 447097055e2SEdward Tomasz Napierala } 448097055e2SEdward Tomasz Napierala 449097055e2SEdward Tomasz Napierala void 450097055e2SEdward Tomasz Napierala racct_create(struct racct **racctp) 451097055e2SEdward Tomasz Napierala { 452097055e2SEdward Tomasz Napierala 4534b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 4544b5c9cf6SEdward Tomasz Napierala return; 4554b5c9cf6SEdward Tomasz Napierala 45636160958SMark Johnston SDT_PROBE1(racct, , racct, create, racctp); 457097055e2SEdward Tomasz Napierala 458097055e2SEdward Tomasz Napierala KASSERT(*racctp == NULL, ("racct already allocated")); 459097055e2SEdward Tomasz Napierala 460097055e2SEdward Tomasz Napierala *racctp = uma_zalloc(racct_zone, M_WAITOK | M_ZERO); 461097055e2SEdward Tomasz Napierala } 462097055e2SEdward Tomasz Napierala 463097055e2SEdward Tomasz Napierala static void 464097055e2SEdward Tomasz Napierala racct_destroy_locked(struct racct **racctp) 465097055e2SEdward Tomasz Napierala { 466097055e2SEdward Tomasz Napierala int i; 467097055e2SEdward Tomasz Napierala struct racct *racct; 468097055e2SEdward Tomasz Napierala 4694b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 4704b5c9cf6SEdward Tomasz Napierala 47136160958SMark Johnston SDT_PROBE1(racct, , racct, destroy, racctp); 472097055e2SEdward Tomasz Napierala 473*4c230cdaSEdward Tomasz Napierala RACCT_LOCK_ASSERT(); 474097055e2SEdward Tomasz Napierala KASSERT(racctp != NULL, ("NULL racctp")); 475097055e2SEdward Tomasz Napierala KASSERT(*racctp != NULL, ("NULL racct")); 476097055e2SEdward Tomasz Napierala 477097055e2SEdward Tomasz Napierala racct = *racctp; 478097055e2SEdward Tomasz Napierala 479097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 4804fe84775SEdward Tomasz Napierala if (RACCT_IS_SLOPPY(i)) 481097055e2SEdward Tomasz Napierala continue; 4824fe84775SEdward Tomasz Napierala if (!RACCT_IS_RECLAIMABLE(i)) 483097055e2SEdward Tomasz Napierala continue; 484097055e2SEdward Tomasz Napierala KASSERT(racct->r_resources[i] == 0, 485097055e2SEdward Tomasz Napierala ("destroying non-empty racct: " 486097055e2SEdward Tomasz Napierala "%ju allocated for resource %d\n", 487097055e2SEdward Tomasz Napierala racct->r_resources[i], i)); 488097055e2SEdward Tomasz Napierala } 489097055e2SEdward Tomasz Napierala uma_zfree(racct_zone, racct); 490097055e2SEdward Tomasz Napierala *racctp = NULL; 491097055e2SEdward Tomasz Napierala } 492097055e2SEdward Tomasz Napierala 493097055e2SEdward Tomasz Napierala void 494097055e2SEdward Tomasz Napierala racct_destroy(struct racct **racct) 495097055e2SEdward Tomasz Napierala { 496097055e2SEdward Tomasz Napierala 4974b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 4984b5c9cf6SEdward Tomasz Napierala return; 4994b5c9cf6SEdward Tomasz Napierala 500*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 501097055e2SEdward Tomasz Napierala racct_destroy_locked(racct); 502*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 503097055e2SEdward Tomasz Napierala } 504097055e2SEdward Tomasz Napierala 505097055e2SEdward Tomasz Napierala /* 506af1a7b25SEdward Tomasz Napierala * Increase consumption of 'resource' by 'amount' for 'racct', 507af1a7b25SEdward Tomasz Napierala * but not its parents. Differently from other cases, 'amount' here 508097055e2SEdward Tomasz Napierala * may be less than zero. 509097055e2SEdward Tomasz Napierala */ 510097055e2SEdward Tomasz Napierala static void 5113768a5dfSJeremie Le Hen racct_adjust_resource(struct racct *racct, int resource, 512a8e0740eSEdward Tomasz Napierala int64_t amount) 513097055e2SEdward Tomasz Napierala { 514097055e2SEdward Tomasz Napierala 5154b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 516*4c230cdaSEdward Tomasz Napierala RACCT_LOCK_ASSERT(); 517097055e2SEdward Tomasz Napierala KASSERT(racct != NULL, ("NULL racct")); 518097055e2SEdward Tomasz Napierala 519097055e2SEdward Tomasz Napierala racct->r_resources[resource] += amount; 520097055e2SEdward Tomasz Napierala if (racct->r_resources[resource] < 0) { 52136af9869SEdward Tomasz Napierala KASSERT(RACCT_IS_SLOPPY(resource) || RACCT_IS_DECAYING(resource), 522baf85d0aSEdward Tomasz Napierala ("%s: resource %d usage < 0", __func__, resource)); 523097055e2SEdward Tomasz Napierala racct->r_resources[resource] = 0; 524097055e2SEdward Tomasz Napierala } 52536af9869SEdward Tomasz Napierala 52636af9869SEdward Tomasz Napierala /* 52736af9869SEdward Tomasz Napierala * There are some cases where the racct %cpu resource would grow 52809766dd5SJosh Paetzel * beyond 100% per core. For example in racct_proc_exit() we add 52909766dd5SJosh Paetzel * the process %cpu usage to the ucred racct containers. If too 53009766dd5SJosh Paetzel * many processes terminated in a short time span, the ucred %cpu 53109766dd5SJosh Paetzel * resource could grow too much. Also, the 4BSD scheduler sometimes 53209766dd5SJosh Paetzel * returns for a thread more than 100% cpu usage. So we set a sane 53309766dd5SJosh Paetzel * boundary here to 100% * the maxumum number of CPUs. 53436af9869SEdward Tomasz Napierala */ 53536af9869SEdward Tomasz Napierala if ((resource == RACCT_PCTCPU) && 53609766dd5SJosh Paetzel (racct->r_resources[RACCT_PCTCPU] > 100 * 1000000 * (int64_t)MAXCPU)) 53709766dd5SJosh Paetzel racct->r_resources[RACCT_PCTCPU] = 100 * 1000000 * (int64_t)MAXCPU; 538097055e2SEdward Tomasz Napierala } 539097055e2SEdward Tomasz Napierala 5402d8696d1SEdward Tomasz Napierala static int 5417cea96f6SEdward Tomasz Napierala racct_add_locked(struct proc *p, int resource, uint64_t amount, int force) 542097055e2SEdward Tomasz Napierala { 543097055e2SEdward Tomasz Napierala #ifdef RCTL 544097055e2SEdward Tomasz Napierala int error; 545097055e2SEdward Tomasz Napierala #endif 546097055e2SEdward Tomasz Napierala 5474b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 5484b5c9cf6SEdward Tomasz Napierala 549097055e2SEdward Tomasz Napierala /* 550097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 551097055e2SEdward Tomasz Napierala */ 552097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 553097055e2SEdward Tomasz Napierala 554097055e2SEdward Tomasz Napierala #ifdef RCTL 555097055e2SEdward Tomasz Napierala error = rctl_enforce(p, resource, amount); 556659e7466SEdward Tomasz Napierala if (error && !force && RACCT_IS_DENIABLE(resource)) { 55736160958SMark Johnston SDT_PROBE3(racct, , rusage, add__failure, p, resource, amount); 558097055e2SEdward Tomasz Napierala return (error); 559097055e2SEdward Tomasz Napierala } 560097055e2SEdward Tomasz Napierala #endif 5613768a5dfSJeremie Le Hen racct_adjust_resource(p->p_racct, resource, amount); 562097055e2SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, resource, amount); 563097055e2SEdward Tomasz Napierala 564097055e2SEdward Tomasz Napierala return (0); 565097055e2SEdward Tomasz Napierala } 566097055e2SEdward Tomasz Napierala 5672d8696d1SEdward Tomasz Napierala /* 5682d8696d1SEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for process 'p'. 5692d8696d1SEdward Tomasz Napierala * Return 0 if it's below limits, or errno, if it's not. 5702d8696d1SEdward Tomasz Napierala */ 5712d8696d1SEdward Tomasz Napierala int 5722d8696d1SEdward Tomasz Napierala racct_add(struct proc *p, int resource, uint64_t amount) 5732d8696d1SEdward Tomasz Napierala { 5742d8696d1SEdward Tomasz Napierala int error; 5752d8696d1SEdward Tomasz Napierala 5764b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 5774b5c9cf6SEdward Tomasz Napierala return (0); 5784b5c9cf6SEdward Tomasz Napierala 5797cea96f6SEdward Tomasz Napierala SDT_PROBE3(racct, , rusage, add, p, resource, amount); 5807cea96f6SEdward Tomasz Napierala 581*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 5827cea96f6SEdward Tomasz Napierala error = racct_add_locked(p, resource, amount, 0); 583*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 5842d8696d1SEdward Tomasz Napierala return (error); 5852d8696d1SEdward Tomasz Napierala } 5862d8696d1SEdward Tomasz Napierala 5870b9f1ecbSEdward Tomasz Napierala /* 5880b9f1ecbSEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for process 'p'. 5890b9f1ecbSEdward Tomasz Napierala * Doesn't check for limits and never fails. 5900b9f1ecbSEdward Tomasz Napierala */ 5910b9f1ecbSEdward Tomasz Napierala void 5920b9f1ecbSEdward Tomasz Napierala racct_add_force(struct proc *p, int resource, uint64_t amount) 5930b9f1ecbSEdward Tomasz Napierala { 5940b9f1ecbSEdward Tomasz Napierala 5950b9f1ecbSEdward Tomasz Napierala if (!racct_enable) 5960b9f1ecbSEdward Tomasz Napierala return; 5970b9f1ecbSEdward Tomasz Napierala 5980b9f1ecbSEdward Tomasz Napierala SDT_PROBE3(racct, , rusage, add__force, p, resource, amount); 5990b9f1ecbSEdward Tomasz Napierala 600*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 6010b9f1ecbSEdward Tomasz Napierala racct_add_locked(p, resource, amount, 1); 602*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 6030b9f1ecbSEdward Tomasz Napierala } 6040b9f1ecbSEdward Tomasz Napierala 605097055e2SEdward Tomasz Napierala static void 606097055e2SEdward Tomasz Napierala racct_add_cred_locked(struct ucred *cred, int resource, uint64_t amount) 607097055e2SEdward Tomasz Napierala { 608097055e2SEdward Tomasz Napierala struct prison *pr; 609097055e2SEdward Tomasz Napierala 6104b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 6114b5c9cf6SEdward Tomasz Napierala 61236160958SMark Johnston SDT_PROBE3(racct, , rusage, add__cred, cred, resource, amount); 613097055e2SEdward Tomasz Napierala 6143768a5dfSJeremie Le Hen racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, amount); 615097055e2SEdward Tomasz Napierala for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) 6163768a5dfSJeremie Le Hen racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource, 617a7ad07bfSEdward Tomasz Napierala amount); 6183768a5dfSJeremie Le Hen racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, amount); 619097055e2SEdward Tomasz Napierala } 620097055e2SEdward Tomasz Napierala 621097055e2SEdward Tomasz Napierala /* 622097055e2SEdward Tomasz Napierala * Increase allocation of 'resource' by 'amount' for credential 'cred'. 623097055e2SEdward Tomasz Napierala * Doesn't check for limits and never fails. 624097055e2SEdward Tomasz Napierala */ 625097055e2SEdward Tomasz Napierala void 626097055e2SEdward Tomasz Napierala racct_add_cred(struct ucred *cred, int resource, uint64_t amount) 627097055e2SEdward Tomasz Napierala { 628097055e2SEdward Tomasz Napierala 6294b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 6304b5c9cf6SEdward Tomasz Napierala return; 6314b5c9cf6SEdward Tomasz Napierala 632*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 633097055e2SEdward Tomasz Napierala racct_add_cred_locked(cred, resource, amount); 634*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 635097055e2SEdward Tomasz Napierala } 636097055e2SEdward Tomasz Napierala 637097055e2SEdward Tomasz Napierala static int 63810287198SEdward Tomasz Napierala racct_set_locked(struct proc *p, int resource, uint64_t amount, int force) 639097055e2SEdward Tomasz Napierala { 64036af9869SEdward Tomasz Napierala int64_t old_amount, decayed_amount; 64136af9869SEdward Tomasz Napierala int64_t diff_proc, diff_cred; 642097055e2SEdward Tomasz Napierala #ifdef RCTL 643097055e2SEdward Tomasz Napierala int error; 644097055e2SEdward Tomasz Napierala #endif 645097055e2SEdward Tomasz Napierala 6464b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 6474b5c9cf6SEdward Tomasz Napierala 648097055e2SEdward Tomasz Napierala /* 649097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 650097055e2SEdward Tomasz Napierala */ 651097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 652097055e2SEdward Tomasz Napierala 65336af9869SEdward Tomasz Napierala old_amount = p->p_racct->r_resources[resource]; 65436af9869SEdward Tomasz Napierala /* 65536af9869SEdward Tomasz Napierala * The diffs may be negative. 65636af9869SEdward Tomasz Napierala */ 65736af9869SEdward Tomasz Napierala diff_proc = amount - old_amount; 65836af9869SEdward Tomasz Napierala if (RACCT_IS_DECAYING(resource)) { 65936af9869SEdward Tomasz Napierala /* 66036af9869SEdward Tomasz Napierala * Resources in per-credential racct containers may decay. 66136af9869SEdward Tomasz Napierala * If this is the case, we need to calculate the difference 66236af9869SEdward Tomasz Napierala * between the new amount and the proportional value of the 66336af9869SEdward Tomasz Napierala * old amount that has decayed in the ucred racct containers. 66436af9869SEdward Tomasz Napierala */ 66536af9869SEdward Tomasz Napierala decayed_amount = old_amount * RACCT_DECAY_FACTOR / FSCALE; 66636af9869SEdward Tomasz Napierala diff_cred = amount - decayed_amount; 66736af9869SEdward Tomasz Napierala } else 66836af9869SEdward Tomasz Napierala diff_cred = diff_proc; 669097055e2SEdward Tomasz Napierala #ifdef notyet 67036af9869SEdward Tomasz Napierala KASSERT(diff_proc >= 0 || RACCT_CAN_DROP(resource), 671baf85d0aSEdward Tomasz Napierala ("%s: usage of non-droppable resource %d dropping", __func__, 672097055e2SEdward Tomasz Napierala resource)); 673097055e2SEdward Tomasz Napierala #endif 674097055e2SEdward Tomasz Napierala #ifdef RCTL 675659e7466SEdward Tomasz Napierala if (diff_proc > 0) { 67636af9869SEdward Tomasz Napierala error = rctl_enforce(p, resource, diff_proc); 677659e7466SEdward Tomasz Napierala if (error && !force && RACCT_IS_DENIABLE(resource)) { 67836160958SMark Johnston SDT_PROBE3(racct, , rusage, set__failure, p, resource, 67936160958SMark Johnston amount); 680097055e2SEdward Tomasz Napierala return (error); 681097055e2SEdward Tomasz Napierala } 682097055e2SEdward Tomasz Napierala } 683097055e2SEdward Tomasz Napierala #endif 6843768a5dfSJeremie Le Hen racct_adjust_resource(p->p_racct, resource, diff_proc); 68536af9869SEdward Tomasz Napierala if (diff_cred > 0) 68636af9869SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, resource, diff_cred); 68736af9869SEdward Tomasz Napierala else if (diff_cred < 0) 68836af9869SEdward Tomasz Napierala racct_sub_cred_locked(p->p_ucred, resource, -diff_cred); 689097055e2SEdward Tomasz Napierala 690097055e2SEdward Tomasz Napierala return (0); 691097055e2SEdward Tomasz Napierala } 692097055e2SEdward Tomasz Napierala 693097055e2SEdward Tomasz Napierala /* 694097055e2SEdward Tomasz Napierala * Set allocation of 'resource' to 'amount' for process 'p'. 695097055e2SEdward Tomasz Napierala * Return 0 if it's below limits, or errno, if it's not. 696097055e2SEdward Tomasz Napierala * 697097055e2SEdward Tomasz Napierala * Note that decreasing the allocation always returns 0, 698097055e2SEdward Tomasz Napierala * even if it's above the limit. 699097055e2SEdward Tomasz Napierala */ 700097055e2SEdward Tomasz Napierala int 701097055e2SEdward Tomasz Napierala racct_set(struct proc *p, int resource, uint64_t amount) 702097055e2SEdward Tomasz Napierala { 703097055e2SEdward Tomasz Napierala int error; 704097055e2SEdward Tomasz Napierala 7054b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 7064b5c9cf6SEdward Tomasz Napierala return (0); 7074b5c9cf6SEdward Tomasz Napierala 70810287198SEdward Tomasz Napierala SDT_PROBE3(racct, , rusage, set__force, p, resource, amount); 70910287198SEdward Tomasz Napierala 710*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 71110287198SEdward Tomasz Napierala error = racct_set_locked(p, resource, amount, 0); 712*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 713097055e2SEdward Tomasz Napierala return (error); 714097055e2SEdward Tomasz Napierala } 715097055e2SEdward Tomasz Napierala 7160b9f1ecbSEdward Tomasz Napierala void 7170b9f1ecbSEdward Tomasz Napierala racct_set_force(struct proc *p, int resource, uint64_t amount) 7180b9f1ecbSEdward Tomasz Napierala { 7190b9f1ecbSEdward Tomasz Napierala 7200b9f1ecbSEdward Tomasz Napierala if (!racct_enable) 7210b9f1ecbSEdward Tomasz Napierala return; 7220b9f1ecbSEdward Tomasz Napierala 7230b9f1ecbSEdward Tomasz Napierala SDT_PROBE3(racct, , rusage, set, p, resource, amount); 7240b9f1ecbSEdward Tomasz Napierala 725*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 7260b9f1ecbSEdward Tomasz Napierala racct_set_locked(p, resource, amount, 1); 727*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 7280b9f1ecbSEdward Tomasz Napierala } 7290b9f1ecbSEdward Tomasz Napierala 730097055e2SEdward Tomasz Napierala /* 731097055e2SEdward Tomasz Napierala * Returns amount of 'resource' the process 'p' can keep allocated. 732097055e2SEdward Tomasz Napierala * Allocating more than that would be denied, unless the resource 733097055e2SEdward Tomasz Napierala * is marked undeniable. Amount of already allocated resource does 734097055e2SEdward Tomasz Napierala * not matter. 735097055e2SEdward Tomasz Napierala */ 736097055e2SEdward Tomasz Napierala uint64_t 737097055e2SEdward Tomasz Napierala racct_get_limit(struct proc *p, int resource) 738097055e2SEdward Tomasz Napierala { 739097055e2SEdward Tomasz Napierala 7404b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 7414b5c9cf6SEdward Tomasz Napierala return (UINT64_MAX); 7424b5c9cf6SEdward Tomasz Napierala 743097055e2SEdward Tomasz Napierala #ifdef RCTL 744097055e2SEdward Tomasz Napierala return (rctl_get_limit(p, resource)); 745097055e2SEdward Tomasz Napierala #else 746097055e2SEdward Tomasz Napierala return (UINT64_MAX); 747097055e2SEdward Tomasz Napierala #endif 748097055e2SEdward Tomasz Napierala } 749097055e2SEdward Tomasz Napierala 750097055e2SEdward Tomasz Napierala /* 751097055e2SEdward Tomasz Napierala * Returns amount of 'resource' the process 'p' can keep allocated. 752097055e2SEdward Tomasz Napierala * Allocating more than that would be denied, unless the resource 753097055e2SEdward Tomasz Napierala * is marked undeniable. Amount of already allocated resource does 754097055e2SEdward Tomasz Napierala * matter. 755097055e2SEdward Tomasz Napierala */ 756097055e2SEdward Tomasz Napierala uint64_t 757097055e2SEdward Tomasz Napierala racct_get_available(struct proc *p, int resource) 758097055e2SEdward Tomasz Napierala { 759097055e2SEdward Tomasz Napierala 7604b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 7614b5c9cf6SEdward Tomasz Napierala return (UINT64_MAX); 7624b5c9cf6SEdward Tomasz Napierala 763097055e2SEdward Tomasz Napierala #ifdef RCTL 764097055e2SEdward Tomasz Napierala return (rctl_get_available(p, resource)); 765097055e2SEdward Tomasz Napierala #else 766097055e2SEdward Tomasz Napierala return (UINT64_MAX); 767097055e2SEdward Tomasz Napierala #endif 768097055e2SEdward Tomasz Napierala } 769097055e2SEdward Tomasz Napierala 770097055e2SEdward Tomasz Napierala /* 77136af9869SEdward Tomasz Napierala * Returns amount of the %cpu resource that process 'p' can add to its %cpu 77236af9869SEdward Tomasz Napierala * utilization. Adding more than that would lead to the process being 77336af9869SEdward Tomasz Napierala * throttled. 77436af9869SEdward Tomasz Napierala */ 77536af9869SEdward Tomasz Napierala static int64_t 77636af9869SEdward Tomasz Napierala racct_pcpu_available(struct proc *p) 77736af9869SEdward Tomasz Napierala { 77836af9869SEdward Tomasz Napierala 7794b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 7804b5c9cf6SEdward Tomasz Napierala 78136af9869SEdward Tomasz Napierala #ifdef RCTL 78236af9869SEdward Tomasz Napierala return (rctl_pcpu_available(p)); 78336af9869SEdward Tomasz Napierala #else 78436af9869SEdward Tomasz Napierala return (INT64_MAX); 78536af9869SEdward Tomasz Napierala #endif 78636af9869SEdward Tomasz Napierala } 78736af9869SEdward Tomasz Napierala 78836af9869SEdward Tomasz Napierala /* 789097055e2SEdward Tomasz Napierala * Decrease allocation of 'resource' by 'amount' for process 'p'. 790097055e2SEdward Tomasz Napierala */ 791097055e2SEdward Tomasz Napierala void 792097055e2SEdward Tomasz Napierala racct_sub(struct proc *p, int resource, uint64_t amount) 793097055e2SEdward Tomasz Napierala { 794097055e2SEdward Tomasz Napierala 7954b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 7964b5c9cf6SEdward Tomasz Napierala return; 7974b5c9cf6SEdward Tomasz Napierala 79836160958SMark Johnston SDT_PROBE3(racct, , rusage, sub, p, resource, amount); 799097055e2SEdward Tomasz Napierala 800097055e2SEdward Tomasz Napierala /* 801097055e2SEdward Tomasz Napierala * We need proc lock to dereference p->p_ucred. 802097055e2SEdward Tomasz Napierala */ 803097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 80436af9869SEdward Tomasz Napierala KASSERT(RACCT_CAN_DROP(resource), 805baf85d0aSEdward Tomasz Napierala ("%s: called for non-droppable resource %d", __func__, resource)); 806097055e2SEdward Tomasz Napierala 807*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 808097055e2SEdward Tomasz Napierala KASSERT(amount <= p->p_racct->r_resources[resource], 809baf85d0aSEdward Tomasz Napierala ("%s: freeing %ju of resource %d, which is more " 810baf85d0aSEdward Tomasz Napierala "than allocated %jd for %s (pid %d)", __func__, amount, resource, 811097055e2SEdward Tomasz Napierala (intmax_t)p->p_racct->r_resources[resource], p->p_comm, p->p_pid)); 812097055e2SEdward Tomasz Napierala 8133768a5dfSJeremie Le Hen racct_adjust_resource(p->p_racct, resource, -amount); 814097055e2SEdward Tomasz Napierala racct_sub_cred_locked(p->p_ucred, resource, amount); 815*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 816097055e2SEdward Tomasz Napierala } 817097055e2SEdward Tomasz Napierala 818097055e2SEdward Tomasz Napierala static void 819097055e2SEdward Tomasz Napierala racct_sub_cred_locked(struct ucred *cred, int resource, uint64_t amount) 820097055e2SEdward Tomasz Napierala { 821097055e2SEdward Tomasz Napierala struct prison *pr; 822097055e2SEdward Tomasz Napierala 8234b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 8244b5c9cf6SEdward Tomasz Napierala 82536160958SMark Johnston SDT_PROBE3(racct, , rusage, sub__cred, cred, resource, amount); 826097055e2SEdward Tomasz Napierala 827097055e2SEdward Tomasz Napierala #ifdef notyet 82836af9869SEdward Tomasz Napierala KASSERT(RACCT_CAN_DROP(resource), 829baf85d0aSEdward Tomasz Napierala ("%s: called for resource %d which can not drop", __func__, 830097055e2SEdward Tomasz Napierala resource)); 831097055e2SEdward Tomasz Napierala #endif 832097055e2SEdward Tomasz Napierala 8333768a5dfSJeremie Le Hen racct_adjust_resource(cred->cr_ruidinfo->ui_racct, resource, -amount); 834097055e2SEdward Tomasz Napierala for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) 8353768a5dfSJeremie Le Hen racct_adjust_resource(pr->pr_prison_racct->prr_racct, resource, 836a7ad07bfSEdward Tomasz Napierala -amount); 8373768a5dfSJeremie Le Hen racct_adjust_resource(cred->cr_loginclass->lc_racct, resource, -amount); 838097055e2SEdward Tomasz Napierala } 839097055e2SEdward Tomasz Napierala 840097055e2SEdward Tomasz Napierala /* 841097055e2SEdward Tomasz Napierala * Decrease allocation of 'resource' by 'amount' for credential 'cred'. 842097055e2SEdward Tomasz Napierala */ 843097055e2SEdward Tomasz Napierala void 844097055e2SEdward Tomasz Napierala racct_sub_cred(struct ucred *cred, int resource, uint64_t amount) 845097055e2SEdward Tomasz Napierala { 846097055e2SEdward Tomasz Napierala 8474b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 8484b5c9cf6SEdward Tomasz Napierala return; 8494b5c9cf6SEdward Tomasz Napierala 850*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 851097055e2SEdward Tomasz Napierala racct_sub_cred_locked(cred, resource, amount); 852*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 853097055e2SEdward Tomasz Napierala } 854097055e2SEdward Tomasz Napierala 855097055e2SEdward Tomasz Napierala /* 856097055e2SEdward Tomasz Napierala * Inherit resource usage information from the parent process. 857097055e2SEdward Tomasz Napierala */ 858097055e2SEdward Tomasz Napierala int 859097055e2SEdward Tomasz Napierala racct_proc_fork(struct proc *parent, struct proc *child) 860097055e2SEdward Tomasz Napierala { 861097055e2SEdward Tomasz Napierala int i, error = 0; 862097055e2SEdward Tomasz Napierala 8634b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 8644b5c9cf6SEdward Tomasz Napierala return (0); 8654b5c9cf6SEdward Tomasz Napierala 866097055e2SEdward Tomasz Napierala /* 867097055e2SEdward Tomasz Napierala * Create racct for the child process. 868097055e2SEdward Tomasz Napierala */ 869097055e2SEdward Tomasz Napierala racct_create(&child->p_racct); 870097055e2SEdward Tomasz Napierala 871097055e2SEdward Tomasz Napierala PROC_LOCK(parent); 872097055e2SEdward Tomasz Napierala PROC_LOCK(child); 873*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 874097055e2SEdward Tomasz Napierala 875c0c09362SEdward Tomasz Napierala #ifdef RCTL 876c0c09362SEdward Tomasz Napierala error = rctl_proc_fork(parent, child); 877c0c09362SEdward Tomasz Napierala if (error != 0) 878c0c09362SEdward Tomasz Napierala goto out; 879c0c09362SEdward Tomasz Napierala #endif 880c0c09362SEdward Tomasz Napierala 88136af9869SEdward Tomasz Napierala /* Init process cpu time. */ 88236af9869SEdward Tomasz Napierala child->p_prev_runtime = 0; 88336af9869SEdward Tomasz Napierala child->p_throttled = 0; 88436af9869SEdward Tomasz Napierala 885097055e2SEdward Tomasz Napierala /* 886097055e2SEdward Tomasz Napierala * Inherit resource usage. 887097055e2SEdward Tomasz Napierala */ 888097055e2SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 889097055e2SEdward Tomasz Napierala if (parent->p_racct->r_resources[i] == 0 || 8904fe84775SEdward Tomasz Napierala !RACCT_IS_INHERITABLE(i)) 891097055e2SEdward Tomasz Napierala continue; 892097055e2SEdward Tomasz Napierala 893097055e2SEdward Tomasz Napierala error = racct_set_locked(child, i, 89410287198SEdward Tomasz Napierala parent->p_racct->r_resources[i], 0); 895ac6fafe6SEdward Tomasz Napierala if (error != 0) 896097055e2SEdward Tomasz Napierala goto out; 897097055e2SEdward Tomasz Napierala } 898097055e2SEdward Tomasz Napierala 8997cea96f6SEdward Tomasz Napierala error = racct_add_locked(child, RACCT_NPROC, 1, 0); 9007cea96f6SEdward Tomasz Napierala error += racct_add_locked(child, RACCT_NTHR, 1, 0); 9012d8696d1SEdward Tomasz Napierala 902097055e2SEdward Tomasz Napierala out: 903*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 904097055e2SEdward Tomasz Napierala PROC_UNLOCK(child); 905097055e2SEdward Tomasz Napierala PROC_UNLOCK(parent); 906097055e2SEdward Tomasz Napierala 907ab27d5d8SEdward Tomasz Napierala if (error != 0) 908ab27d5d8SEdward Tomasz Napierala racct_proc_exit(child); 909ab27d5d8SEdward Tomasz Napierala 910097055e2SEdward Tomasz Napierala return (error); 911097055e2SEdward Tomasz Napierala } 912097055e2SEdward Tomasz Napierala 91372a401d9SEdward Tomasz Napierala /* 91472a401d9SEdward Tomasz Napierala * Called at the end of fork1(), to handle rules that require the process 91572a401d9SEdward Tomasz Napierala * to be fully initialized. 91672a401d9SEdward Tomasz Napierala */ 91772a401d9SEdward Tomasz Napierala void 91872a401d9SEdward Tomasz Napierala racct_proc_fork_done(struct proc *child) 91972a401d9SEdward Tomasz Napierala { 92072a401d9SEdward Tomasz Napierala 921813361c1SMateusz Guzik PROC_LOCK_ASSERT(child, MA_OWNED); 92272a401d9SEdward Tomasz Napierala #ifdef RCTL 9234b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 9244b5c9cf6SEdward Tomasz Napierala return; 9254b5c9cf6SEdward Tomasz Napierala 926*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 92772a401d9SEdward Tomasz Napierala rctl_enforce(child, RACCT_NPROC, 0); 92872a401d9SEdward Tomasz Napierala rctl_enforce(child, RACCT_NTHR, 0); 929*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 93072a401d9SEdward Tomasz Napierala #endif 93172a401d9SEdward Tomasz Napierala } 93272a401d9SEdward Tomasz Napierala 933097055e2SEdward Tomasz Napierala void 934097055e2SEdward Tomasz Napierala racct_proc_exit(struct proc *p) 935097055e2SEdward Tomasz Napierala { 9362419d7f9SEdward Tomasz Napierala int i; 937097055e2SEdward Tomasz Napierala uint64_t runtime; 93836af9869SEdward Tomasz Napierala struct timeval wallclock; 93936af9869SEdward Tomasz Napierala uint64_t pct_estimate, pct; 940097055e2SEdward Tomasz Napierala 9414b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 9424b5c9cf6SEdward Tomasz Napierala return; 9434b5c9cf6SEdward Tomasz Napierala 944097055e2SEdward Tomasz Napierala PROC_LOCK(p); 945097055e2SEdward Tomasz Napierala /* 946097055e2SEdward Tomasz Napierala * We don't need to calculate rux, proc_reap() has already done this. 947097055e2SEdward Tomasz Napierala */ 948097055e2SEdward Tomasz Napierala runtime = cputick2usec(p->p_rux.rux_runtime); 949097055e2SEdward Tomasz Napierala #ifdef notyet 950097055e2SEdward Tomasz Napierala KASSERT(runtime >= p->p_prev_runtime, ("runtime < p_prev_runtime")); 951097055e2SEdward Tomasz Napierala #else 952097055e2SEdward Tomasz Napierala if (runtime < p->p_prev_runtime) 953097055e2SEdward Tomasz Napierala runtime = p->p_prev_runtime; 954097055e2SEdward Tomasz Napierala #endif 95536af9869SEdward Tomasz Napierala microuptime(&wallclock); 95636af9869SEdward Tomasz Napierala timevalsub(&wallclock, &p->p_stats->p_start); 95784590fd8SEdward Tomasz Napierala if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) { 95836af9869SEdward Tomasz Napierala pct_estimate = (1000000 * runtime * 100) / 95936af9869SEdward Tomasz Napierala ((uint64_t)wallclock.tv_sec * 1000000 + 96036af9869SEdward Tomasz Napierala wallclock.tv_usec); 96184590fd8SEdward Tomasz Napierala } else 96284590fd8SEdward Tomasz Napierala pct_estimate = 0; 96336af9869SEdward Tomasz Napierala pct = racct_getpcpu(p, pct_estimate); 96436af9869SEdward Tomasz Napierala 965*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 96610287198SEdward Tomasz Napierala racct_set_locked(p, RACCT_CPU, runtime, 0); 96736af9869SEdward Tomasz Napierala racct_add_cred_locked(p->p_ucred, RACCT_PCTCPU, pct); 968097055e2SEdward Tomasz Napierala 9692419d7f9SEdward Tomasz Napierala for (i = 0; i <= RACCT_MAX; i++) { 9702419d7f9SEdward Tomasz Napierala if (p->p_racct->r_resources[i] == 0) 9712419d7f9SEdward Tomasz Napierala continue; 9722419d7f9SEdward Tomasz Napierala if (!RACCT_IS_RECLAIMABLE(i)) 9732419d7f9SEdward Tomasz Napierala continue; 97410287198SEdward Tomasz Napierala racct_set_locked(p, i, 0, 0); 9752419d7f9SEdward Tomasz Napierala } 9762419d7f9SEdward Tomasz Napierala 977*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 978097055e2SEdward Tomasz Napierala PROC_UNLOCK(p); 979097055e2SEdward Tomasz Napierala 980097055e2SEdward Tomasz Napierala #ifdef RCTL 981097055e2SEdward Tomasz Napierala rctl_racct_release(p->p_racct); 982097055e2SEdward Tomasz Napierala #endif 983097055e2SEdward Tomasz Napierala racct_destroy(&p->p_racct); 984097055e2SEdward Tomasz Napierala } 985097055e2SEdward Tomasz Napierala 986097055e2SEdward Tomasz Napierala /* 987097055e2SEdward Tomasz Napierala * Called after credentials change, to move resource utilisation 988097055e2SEdward Tomasz Napierala * between raccts. 989097055e2SEdward Tomasz Napierala */ 990097055e2SEdward Tomasz Napierala void 991097055e2SEdward Tomasz Napierala racct_proc_ucred_changed(struct proc *p, struct ucred *oldcred, 992097055e2SEdward Tomasz Napierala struct ucred *newcred) 993097055e2SEdward Tomasz Napierala { 994097055e2SEdward Tomasz Napierala struct uidinfo *olduip, *newuip; 995097055e2SEdward Tomasz Napierala struct loginclass *oldlc, *newlc; 996097055e2SEdward Tomasz Napierala struct prison *oldpr, *newpr, *pr; 997097055e2SEdward Tomasz Napierala 9984b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 9994b5c9cf6SEdward Tomasz Napierala return; 10004b5c9cf6SEdward Tomasz Napierala 1001097055e2SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_NOTOWNED); 1002097055e2SEdward Tomasz Napierala 1003097055e2SEdward Tomasz Napierala newuip = newcred->cr_ruidinfo; 1004097055e2SEdward Tomasz Napierala olduip = oldcred->cr_ruidinfo; 1005097055e2SEdward Tomasz Napierala newlc = newcred->cr_loginclass; 1006097055e2SEdward Tomasz Napierala oldlc = oldcred->cr_loginclass; 1007097055e2SEdward Tomasz Napierala newpr = newcred->cr_prison; 1008097055e2SEdward Tomasz Napierala oldpr = oldcred->cr_prison; 1009097055e2SEdward Tomasz Napierala 1010*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 1011097055e2SEdward Tomasz Napierala if (newuip != olduip) { 1012097055e2SEdward Tomasz Napierala racct_sub_racct(olduip->ui_racct, p->p_racct); 1013097055e2SEdward Tomasz Napierala racct_add_racct(newuip->ui_racct, p->p_racct); 1014097055e2SEdward Tomasz Napierala } 1015097055e2SEdward Tomasz Napierala if (newlc != oldlc) { 1016097055e2SEdward Tomasz Napierala racct_sub_racct(oldlc->lc_racct, p->p_racct); 1017097055e2SEdward Tomasz Napierala racct_add_racct(newlc->lc_racct, p->p_racct); 1018097055e2SEdward Tomasz Napierala } 1019097055e2SEdward Tomasz Napierala if (newpr != oldpr) { 1020097055e2SEdward Tomasz Napierala for (pr = oldpr; pr != NULL; pr = pr->pr_parent) 1021a7ad07bfSEdward Tomasz Napierala racct_sub_racct(pr->pr_prison_racct->prr_racct, 1022a7ad07bfSEdward Tomasz Napierala p->p_racct); 1023097055e2SEdward Tomasz Napierala for (pr = newpr; pr != NULL; pr = pr->pr_parent) 1024a7ad07bfSEdward Tomasz Napierala racct_add_racct(pr->pr_prison_racct->prr_racct, 1025a7ad07bfSEdward Tomasz Napierala p->p_racct); 1026097055e2SEdward Tomasz Napierala } 1027*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 1028097055e2SEdward Tomasz Napierala 1029097055e2SEdward Tomasz Napierala #ifdef RCTL 1030097055e2SEdward Tomasz Napierala rctl_proc_ucred_changed(p, newcred); 1031097055e2SEdward Tomasz Napierala #endif 1032097055e2SEdward Tomasz Napierala } 1033097055e2SEdward Tomasz Napierala 1034c34bbd2aSEdward Tomasz Napierala void 1035c34bbd2aSEdward Tomasz Napierala racct_move(struct racct *dest, struct racct *src) 1036c34bbd2aSEdward Tomasz Napierala { 1037c34bbd2aSEdward Tomasz Napierala 10384b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 10394b5c9cf6SEdward Tomasz Napierala 1040*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 1041c34bbd2aSEdward Tomasz Napierala racct_add_racct(dest, src); 1042c34bbd2aSEdward Tomasz Napierala racct_sub_racct(src, src); 1043*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 1044c34bbd2aSEdward Tomasz Napierala } 1045c34bbd2aSEdward Tomasz Napierala 1046097055e2SEdward Tomasz Napierala static void 104736af9869SEdward Tomasz Napierala racct_proc_throttle(struct proc *p) 104836af9869SEdward Tomasz Napierala { 104936af9869SEdward Tomasz Napierala struct thread *td; 105036af9869SEdward Tomasz Napierala #ifdef SMP 105136af9869SEdward Tomasz Napierala int cpuid; 105236af9869SEdward Tomasz Napierala #endif 105336af9869SEdward Tomasz Napierala 10544b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 105536af9869SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 105636af9869SEdward Tomasz Napierala 105736af9869SEdward Tomasz Napierala /* 105836af9869SEdward Tomasz Napierala * Do not block kernel processes. Also do not block processes with 105936af9869SEdward Tomasz Napierala * low %cpu utilization to improve interactivity. 106036af9869SEdward Tomasz Napierala */ 1061db57c70aSKonstantin Belousov if (((p->p_flag & (P_SYSTEM | P_KPROC)) != 0) || 106236af9869SEdward Tomasz Napierala (p->p_racct->r_resources[RACCT_PCTCPU] <= pcpu_threshold)) 106336af9869SEdward Tomasz Napierala return; 106436af9869SEdward Tomasz Napierala p->p_throttled = 1; 106536af9869SEdward Tomasz Napierala 106636af9869SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) { 106716befafdSEdward Tomasz Napierala thread_lock(td); 106836af9869SEdward Tomasz Napierala switch (td->td_state) { 106936af9869SEdward Tomasz Napierala case TDS_RUNQ: 107036af9869SEdward Tomasz Napierala /* 107136af9869SEdward Tomasz Napierala * If the thread is on the scheduler run-queue, we can 107236af9869SEdward Tomasz Napierala * not just remove it from there. So we set the flag 107336af9869SEdward Tomasz Napierala * TDF_NEEDRESCHED for the thread, so that once it is 107436af9869SEdward Tomasz Napierala * running, it is taken off the cpu as soon as possible. 107536af9869SEdward Tomasz Napierala */ 107636af9869SEdward Tomasz Napierala td->td_flags |= TDF_NEEDRESCHED; 107736af9869SEdward Tomasz Napierala break; 107836af9869SEdward Tomasz Napierala case TDS_RUNNING: 107936af9869SEdward Tomasz Napierala /* 108036af9869SEdward Tomasz Napierala * If the thread is running, we request a context 108136af9869SEdward Tomasz Napierala * switch for it by setting the TDF_NEEDRESCHED flag. 108236af9869SEdward Tomasz Napierala */ 108336af9869SEdward Tomasz Napierala td->td_flags |= TDF_NEEDRESCHED; 108436af9869SEdward Tomasz Napierala #ifdef SMP 108536af9869SEdward Tomasz Napierala cpuid = td->td_oncpu; 108636af9869SEdward Tomasz Napierala if ((cpuid != NOCPU) && (td != curthread)) 108736af9869SEdward Tomasz Napierala ipi_cpu(cpuid, IPI_AST); 108836af9869SEdward Tomasz Napierala #endif 108936af9869SEdward Tomasz Napierala break; 109036af9869SEdward Tomasz Napierala default: 109136af9869SEdward Tomasz Napierala break; 109236af9869SEdward Tomasz Napierala } 109316befafdSEdward Tomasz Napierala thread_unlock(td); 109436af9869SEdward Tomasz Napierala } 109536af9869SEdward Tomasz Napierala } 109636af9869SEdward Tomasz Napierala 109736af9869SEdward Tomasz Napierala static void 109836af9869SEdward Tomasz Napierala racct_proc_wakeup(struct proc *p) 109936af9869SEdward Tomasz Napierala { 11004b5c9cf6SEdward Tomasz Napierala 11014b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 11024b5c9cf6SEdward Tomasz Napierala 110336af9869SEdward Tomasz Napierala PROC_LOCK_ASSERT(p, MA_OWNED); 110436af9869SEdward Tomasz Napierala 110536af9869SEdward Tomasz Napierala if (p->p_throttled) { 110636af9869SEdward Tomasz Napierala p->p_throttled = 0; 110736af9869SEdward Tomasz Napierala wakeup(p->p_racct); 110836af9869SEdward Tomasz Napierala } 110936af9869SEdward Tomasz Napierala } 111036af9869SEdward Tomasz Napierala 111136af9869SEdward Tomasz Napierala static void 1112862d03fbSEdward Tomasz Napierala racct_decay_callback(struct racct *racct, void *dummy1, void *dummy2) 111336af9869SEdward Tomasz Napierala { 111436af9869SEdward Tomasz Napierala int64_t r_old, r_new; 111536af9869SEdward Tomasz Napierala 11164b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 1117*4c230cdaSEdward Tomasz Napierala RACCT_LOCK_ASSERT(); 11184b5c9cf6SEdward Tomasz Napierala 1119862d03fbSEdward Tomasz Napierala r_old = racct->r_resources[RACCT_PCTCPU]; 112036af9869SEdward Tomasz Napierala 112136af9869SEdward Tomasz Napierala /* If there is nothing to decay, just exit. */ 112236af9869SEdward Tomasz Napierala if (r_old <= 0) 112336af9869SEdward Tomasz Napierala return; 112436af9869SEdward Tomasz Napierala 112536af9869SEdward Tomasz Napierala r_new = r_old * RACCT_DECAY_FACTOR / FSCALE; 1126862d03fbSEdward Tomasz Napierala racct->r_resources[RACCT_PCTCPU] = r_new; 112715db3c07SEdward Tomasz Napierala } 112815db3c07SEdward Tomasz Napierala 112915db3c07SEdward Tomasz Napierala static void 113015db3c07SEdward Tomasz Napierala racct_decay_pre(void) 113115db3c07SEdward Tomasz Napierala { 113215db3c07SEdward Tomasz Napierala 1133*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 113415db3c07SEdward Tomasz Napierala } 113515db3c07SEdward Tomasz Napierala 113615db3c07SEdward Tomasz Napierala static void 113715db3c07SEdward Tomasz Napierala racct_decay_post(void) 113815db3c07SEdward Tomasz Napierala { 113915db3c07SEdward Tomasz Napierala 1140*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 114136af9869SEdward Tomasz Napierala } 114236af9869SEdward Tomasz Napierala 114336af9869SEdward Tomasz Napierala static void 1144097e0da7SEdward Tomasz Napierala racct_decay(void) 114536af9869SEdward Tomasz Napierala { 11464b5c9cf6SEdward Tomasz Napierala 11474b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 11484b5c9cf6SEdward Tomasz Napierala 1149862d03fbSEdward Tomasz Napierala ui_racct_foreach(racct_decay_callback, racct_decay_pre, 1150862d03fbSEdward Tomasz Napierala racct_decay_post, NULL, NULL); 1151862d03fbSEdward Tomasz Napierala loginclass_racct_foreach(racct_decay_callback, racct_decay_pre, 1152862d03fbSEdward Tomasz Napierala racct_decay_post, NULL, NULL); 1153862d03fbSEdward Tomasz Napierala prison_racct_foreach(racct_decay_callback, racct_decay_pre, 1154862d03fbSEdward Tomasz Napierala racct_decay_post, NULL, NULL); 115536af9869SEdward Tomasz Napierala } 115636af9869SEdward Tomasz Napierala 115736af9869SEdward Tomasz Napierala static void 1158097055e2SEdward Tomasz Napierala racctd(void) 1159097055e2SEdward Tomasz Napierala { 1160097055e2SEdward Tomasz Napierala struct thread *td; 1161097055e2SEdward Tomasz Napierala struct proc *p; 1162097055e2SEdward Tomasz Napierala struct timeval wallclock; 1163097055e2SEdward Tomasz Napierala uint64_t runtime; 116436af9869SEdward Tomasz Napierala uint64_t pct, pct_estimate; 1165097055e2SEdward Tomasz Napierala 11664b5c9cf6SEdward Tomasz Napierala ASSERT_RACCT_ENABLED(); 11674b5c9cf6SEdward Tomasz Napierala 1168097055e2SEdward Tomasz Napierala for (;;) { 1169862d03fbSEdward Tomasz Napierala racct_decay(); 117036af9869SEdward Tomasz Napierala 1171097055e2SEdward Tomasz Napierala sx_slock(&allproc_lock); 1172097055e2SEdward Tomasz Napierala 117336af9869SEdward Tomasz Napierala LIST_FOREACH(p, &zombproc, p_list) { 117436af9869SEdward Tomasz Napierala PROC_LOCK(p); 117536af9869SEdward Tomasz Napierala racct_set(p, RACCT_PCTCPU, 0); 117636af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 117736af9869SEdward Tomasz Napierala } 117836af9869SEdward Tomasz Napierala 1179097055e2SEdward Tomasz Napierala FOREACH_PROC_IN_SYSTEM(p) { 118036af9869SEdward Tomasz Napierala PROC_LOCK(p); 118136af9869SEdward Tomasz Napierala if (p->p_state != PRS_NORMAL) { 118236af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 1183097055e2SEdward Tomasz Napierala continue; 118436af9869SEdward Tomasz Napierala } 1185097055e2SEdward Tomasz Napierala 1186097055e2SEdward Tomasz Napierala microuptime(&wallclock); 1187097055e2SEdward Tomasz Napierala timevalsub(&wallclock, &p->p_stats->p_start); 11885c7bebf9SKonstantin Belousov PROC_STATLOCK(p); 11890a53cd57SEdward Tomasz Napierala FOREACH_THREAD_IN_PROC(p, td) 1190097055e2SEdward Tomasz Napierala ruxagg(p, td); 1191097055e2SEdward Tomasz Napierala runtime = cputick2usec(p->p_rux.rux_runtime); 11925c7bebf9SKonstantin Belousov PROC_STATUNLOCK(p); 1193097055e2SEdward Tomasz Napierala #ifdef notyet 1194097055e2SEdward Tomasz Napierala KASSERT(runtime >= p->p_prev_runtime, 1195097055e2SEdward Tomasz Napierala ("runtime < p_prev_runtime")); 1196097055e2SEdward Tomasz Napierala #else 1197097055e2SEdward Tomasz Napierala if (runtime < p->p_prev_runtime) 1198097055e2SEdward Tomasz Napierala runtime = p->p_prev_runtime; 1199097055e2SEdward Tomasz Napierala #endif 1200097055e2SEdward Tomasz Napierala p->p_prev_runtime = runtime; 120184590fd8SEdward Tomasz Napierala if (wallclock.tv_sec > 0 || wallclock.tv_usec > 0) { 120236af9869SEdward Tomasz Napierala pct_estimate = (1000000 * runtime * 100) / 120336af9869SEdward Tomasz Napierala ((uint64_t)wallclock.tv_sec * 1000000 + 120436af9869SEdward Tomasz Napierala wallclock.tv_usec); 120584590fd8SEdward Tomasz Napierala } else 120684590fd8SEdward Tomasz Napierala pct_estimate = 0; 120736af9869SEdward Tomasz Napierala pct = racct_getpcpu(p, pct_estimate); 1208*4c230cdaSEdward Tomasz Napierala RACCT_LOCK(); 120910287198SEdward Tomasz Napierala racct_set_locked(p, RACCT_PCTCPU, pct, 1); 121010287198SEdward Tomasz Napierala racct_set_locked(p, RACCT_CPU, runtime, 0); 1211097055e2SEdward Tomasz Napierala racct_set_locked(p, RACCT_WALLCLOCK, 121259f513cdSJaakko Heinonen (uint64_t)wallclock.tv_sec * 1000000 + 121310287198SEdward Tomasz Napierala wallclock.tv_usec, 0); 1214*4c230cdaSEdward Tomasz Napierala RACCT_UNLOCK(); 1215097055e2SEdward Tomasz Napierala PROC_UNLOCK(p); 1216097055e2SEdward Tomasz Napierala } 121736af9869SEdward Tomasz Napierala 121836af9869SEdward Tomasz Napierala /* 121936af9869SEdward Tomasz Napierala * To ensure that processes are throttled in a fair way, we need 122036af9869SEdward Tomasz Napierala * to iterate over all processes again and check the limits 122136af9869SEdward Tomasz Napierala * for %cpu resource only after ucred racct containers have been 122236af9869SEdward Tomasz Napierala * properly filled. 122336af9869SEdward Tomasz Napierala */ 122436af9869SEdward Tomasz Napierala FOREACH_PROC_IN_SYSTEM(p) { 122536af9869SEdward Tomasz Napierala PROC_LOCK(p); 122636af9869SEdward Tomasz Napierala if (p->p_state != PRS_NORMAL) { 122736af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 122836af9869SEdward Tomasz Napierala continue; 122936af9869SEdward Tomasz Napierala } 123036af9869SEdward Tomasz Napierala 123136af9869SEdward Tomasz Napierala if (racct_pcpu_available(p) <= 0) 123236af9869SEdward Tomasz Napierala racct_proc_throttle(p); 123336af9869SEdward Tomasz Napierala else if (p->p_throttled) 123436af9869SEdward Tomasz Napierala racct_proc_wakeup(p); 123536af9869SEdward Tomasz Napierala PROC_UNLOCK(p); 123636af9869SEdward Tomasz Napierala } 1237097055e2SEdward Tomasz Napierala sx_sunlock(&allproc_lock); 1238097055e2SEdward Tomasz Napierala pause("-", hz); 1239097055e2SEdward Tomasz Napierala } 1240097055e2SEdward Tomasz Napierala } 1241097055e2SEdward Tomasz Napierala 1242097055e2SEdward Tomasz Napierala static struct kproc_desc racctd_kp = { 1243097055e2SEdward Tomasz Napierala "racctd", 1244097055e2SEdward Tomasz Napierala racctd, 1245097055e2SEdward Tomasz Napierala NULL 1246097055e2SEdward Tomasz Napierala }; 12474b5c9cf6SEdward Tomasz Napierala 12484b5c9cf6SEdward Tomasz Napierala static void 12494b5c9cf6SEdward Tomasz Napierala racctd_init(void) 12504b5c9cf6SEdward Tomasz Napierala { 12514b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 12524b5c9cf6SEdward Tomasz Napierala return; 12534b5c9cf6SEdward Tomasz Napierala 12544b5c9cf6SEdward Tomasz Napierala kproc_start(&racctd_kp); 12554b5c9cf6SEdward Tomasz Napierala } 12564b5c9cf6SEdward Tomasz Napierala SYSINIT(racctd, SI_SUB_RACCTD, SI_ORDER_FIRST, racctd_init, NULL); 1257097055e2SEdward Tomasz Napierala 1258097055e2SEdward Tomasz Napierala static void 1259097055e2SEdward Tomasz Napierala racct_init(void) 1260097055e2SEdward Tomasz Napierala { 12614b5c9cf6SEdward Tomasz Napierala if (!racct_enable) 12624b5c9cf6SEdward Tomasz Napierala return; 1263097055e2SEdward Tomasz Napierala 1264097055e2SEdward Tomasz Napierala racct_zone = uma_zcreate("racct", sizeof(struct racct), 1265097055e2SEdward Tomasz Napierala NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1266097055e2SEdward Tomasz Napierala /* 1267097055e2SEdward Tomasz Napierala * XXX: Move this somewhere. 1268097055e2SEdward Tomasz Napierala */ 1269a7ad07bfSEdward Tomasz Napierala prison0.pr_prison_racct = prison_racct_find("0"); 1270097055e2SEdward Tomasz Napierala } 1271097055e2SEdward Tomasz Napierala SYSINIT(racct, SI_SUB_RACCT, SI_ORDER_FIRST, racct_init, NULL); 1272097055e2SEdward Tomasz Napierala 1273097055e2SEdward Tomasz Napierala #endif /* !RACCT */ 1274