tc/tc_core.c

   1 /*
   2  * tc_core.c            TC core library.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  */
  12
  13 #include <stdio.h>
  14 #include <stdlib.h>
  15 #include <stdint.h>
  16 #include <unistd.h>
  17 #include <fcntl.h>
  18 #include <math.h>
  19 #include <sys/socket.h>
  20 #include <netinet/in.h>
  21 #include <arpa/inet.h>
  22 #include <string.h>
  23
  24 #include "tc_core.h"
  25 #include <linux/atm.h>
  26
  27 static double tick_in_usec = 1;
  28 static double clock_factor = 1;
  29
  30 int tc_core_time2big(unsigned int time)
  31 {
  32         __u64 t = time;
  33
  34         t *= tick_in_usec;
  35         return (t >> 32) != 0;
  36 }
  37
  38
  39 unsigned int tc_core_time2tick(unsigned int time)
  40 {
  41         return time*tick_in_usec;
  42 }
  43
  44 unsigned int tc_core_tick2time(unsigned int tick)
  45 {
  46         return tick/tick_in_usec;
  47 }
  48
  49 unsigned int tc_core_time2ktime(unsigned int time)
  50 {
  51         return time * clock_factor;
  52 }
  53
  54 unsigned int tc_core_ktime2time(unsigned int ktime)
  55 {
  56         return ktime / clock_factor;
  57 }
  58
  59 unsigned int tc_calc_xmittime(__u64 rate, unsigned int size)
  60 {
  61         return tc_core_time2tick(TIME_UNITS_PER_SEC*((double)size/(double)rate));
  62 }
  63
  64 unsigned int tc_calc_xmitsize(__u64 rate, unsigned int ticks)
  65 {
  66         return ((double)rate*tc_core_tick2time(ticks))/TIME_UNITS_PER_SEC;
  67 }
  68
  69 /*
  70  * The align to ATM cells is used for determining the (ATM) SAR
  71  * alignment overhead at the ATM layer. (SAR = Segmentation And
  72  * Reassembly).  This is for example needed when scheduling packet on
  73  * an ADSL connection.  Note that the extra ATM-AAL overhead is _not_
  74  * included in this calculation. This overhead is added in the kernel
  75  * before doing the rate table lookup, as this gives better precision
  76  * (as the table will always be aligned for 48 bytes).
  77  *  --Hawk, d.7/11-2004. <hawk@diku.dk>
  78  */
  79 static unsigned int tc_align_to_atm(unsigned int size)
  80 {
  81         int linksize, cells;
  82
  83         cells = size / ATM_CELL_PAYLOAD;
  84         if ((size % ATM_CELL_PAYLOAD) > 0)
  85                 cells++;
  86
  87         linksize = cells * ATM_CELL_SIZE; /* Use full cell size to add ATM tax */
  88         return linksize;
  89 }
  90
  91 static unsigned int tc_adjust_size(unsigned int sz, unsigned int mpu, enum link_layer linklayer)
  92 {
  93         if (sz < mpu)
  94                 sz = mpu;
  95
  96         switch (linklayer) {
  97         case LINKLAYER_ATM:
  98                 return tc_align_to_atm(sz);
  99         case LINKLAYER_ETHERNET:
 100         default:
 101                 /* No size adjustments on Ethernet */
 102                 return sz;
 103         }
 104 }
 105
 106 /* Notice, the rate table calculated here, have gotten replaced in the
 107  * kernel and is no-longer used for lookups.
 108  *
 109  * This happened in kernel release v3.8 caused by kernel
 110  *  - commit 56b765b79 ("htb: improved accuracy at high rates").
 111  * This change unfortunately caused breakage of tc overhead and
 112  * linklayer parameters.
 113  *
 114  * Kernel overhead handling got fixed in kernel v3.10 by
 115  * - commit 01cb71d2d47 (net_sched: restore "overhead xxx" handling)
 116  *
 117  * Kernel linklayer handling got fixed in kernel v3.11 by
 118  * - commit 8a8e3d84b17 (net_sched: restore "linklayer atm" handling)
 119  */
 120
 121 /*
 122    rtab[pkt_len>>cell_log] = pkt_xmit_time
 123  */
 124
 125 int tc_calc_rtable(struct tc_ratespec *r, __u32 *rtab,
 126                    int cell_log, unsigned int mtu,
 127                    enum link_layer linklayer)
 128 {
 129         int i;
 130         unsigned int sz;
 131         unsigned int bps = r->rate;
 132         unsigned int mpu = r->mpu;
 133
 134         if (mtu == 0)
 135                 mtu = 2047;
 136
 137         if (cell_log < 0) {
 138                 cell_log = 0;
 139                 while ((mtu >> cell_log) > 255)
 140                         cell_log++;
 141         }
 142
 143         for (i = 0; i < 256; i++) {
 144                 sz = tc_adjust_size((i + 1) << cell_log, mpu, linklayer);
 145                 rtab[i] = tc_calc_xmittime(bps, sz);
 146         }
 147
 148         r->cell_align =  -1;
 149         r->cell_log = cell_log;
 150         r->linklayer = (linklayer & TC_LINKLAYER_MASK);
 151         return cell_log;
 152 }
 153
 154 /*
 155    stab[pkt_len>>cell_log] = pkt_xmit_size>>size_log
 156  */
 157
 158 int tc_calc_size_table(struct tc_sizespec *s, __u16 **stab)
 159 {
 160         int i;
 161         enum link_layer linklayer = s->linklayer;
 162         unsigned int sz;
 163
 164         if (linklayer <= LINKLAYER_ETHERNET && s->mpu == 0) {
 165                 /* don't need data table in this case (only overhead set) */
 166                 s->mtu = 0;
 167                 s->tsize = 0;
 168                 s->cell_log = 0;
 169                 s->cell_align = 0;
 170                 *stab = NULL;
 171                 return 0;
 172         }
 173
 174         if (s->mtu == 0)
 175                 s->mtu = 2047;
 176         if (s->tsize == 0)
 177                 s->tsize = 512;
 178
 179         s->cell_log = 0;
 180         while ((s->mtu >> s->cell_log) > s->tsize - 1)
 181                 s->cell_log++;
 182
 183         *stab = malloc(s->tsize * sizeof(__u16));
 184         if (!*stab)
 185                 return -1;
 186
 187 again:
 188         for (i = s->tsize - 1; i >= 0; i--) {
 189                 sz = tc_adjust_size((i + 1) << s->cell_log, s->mpu, linklayer);
 190                 if ((sz >> s->size_log) > UINT16_MAX) {
 191                         s->size_log++;
 192                         goto again;
 193                 }
 194                 (*stab)[i] = sz >> s->size_log;
 195         }
 196
 197         s->cell_align = -1; /* Due to the sz calc */
 198         return 0;
 199 }
 200
 201 int tc_core_init(void)
 202 {
 203         FILE *fp;
 204         __u32 clock_res;
 205         __u32 t2us;
 206         __u32 us2t;
 207
 208         fp = fopen("/proc/net/psched", "r");
 209         if (fp == NULL)
 210                 return -1;
 211
 212         if (fscanf(fp, "%08x%08x%08x", &t2us, &us2t, &clock_res) != 3) {
 213                 fclose(fp);
 214                 return -1;
 215         }
 216         fclose(fp);
 217
 218         /* compatibility hack: for old iproute binaries (ignoring
 219          * the kernel clock resolution) the kernel advertises a
 220          * tick multiplier of 1000 in case of nano-second resolution,
 221          * which really is 1. */
 222         if (clock_res == 1000000000)
 223                 t2us = us2t;
 224
 225         clock_factor  = (double)clock_res / TIME_UNITS_PER_SEC;
 226         tick_in_usec = (double)t2us / us2t * clock_factor;
 227         return 0;
 228 }