lib/multipath.c

   1 /*
   2  * Copyright (c) 2010, 2011 Nicira Networks.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "multipath.h"
  20 #include <arpa/inet.h>
  21 #include <inttypes.h>
  22 #include <sys/types.h>
  23 #include <netinet/in.h>
  24 #include "dynamic-string.h"
  25 #include "nx-match.h"
  26 #include "ofp-util.h"
  27 #include "openflow/nicira-ext.h"
  28 #include "packets.h"
  29 #include "vlog.h"
  30
  31 VLOG_DEFINE_THIS_MODULE(multipath);
  32
  33 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  34 \f
  35 /* multipath_check(). */
  36 int
  37 multipath_check(const struct nx_action_multipath *mp)
  38 {
  39     uint32_t dst = ntohl(mp->dst);
  40     int ofs = nxm_decode_ofs(mp->ofs_nbits);
  41     int n_bits = nxm_decode_n_bits(mp->ofs_nbits);
  42
  43     if (mp->fields != htons(NX_MP_FIELDS_ETH_SRC)
  44         && mp->fields != htons(NX_MP_FIELDS_SYMMETRIC_L4)) {
  45         VLOG_WARN_RL(&rl, "unsupported fields %"PRIu16, ntohs(mp->fields));
  46     } else if (mp->algorithm != htons(NX_MP_ALG_MODULO_N)
  47                && mp->algorithm != htons(NX_MP_ALG_HASH_THRESHOLD)
  48                && mp->algorithm != htons(NX_MP_ALG_HRW)
  49                && mp->algorithm != htons(NX_MP_ALG_ITER_HASH)) {
  50         VLOG_WARN_RL(&rl, "unsupported algorithm %"PRIu16,
  51                      ntohs(mp->algorithm));
  52     } else if (!NXM_IS_NX_REG(dst) || NXM_NX_REG_IDX(dst) >= FLOW_N_REGS) {
  53         VLOG_WARN_RL(&rl, "unsupported destination field %#"PRIx32, dst);
  54     } else if (ofs + n_bits > nxm_field_bits(dst)) {
  55         VLOG_WARN_RL(&rl, "destination overflows output field");
  56     } else if (n_bits < 16 && ntohs(mp->max_link) > (1u << n_bits)) {
  57         VLOG_WARN_RL(&rl, "max_link overflows output field");
  58     } else {
  59         return 0;
  60     }
  61
  62     return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_ARGUMENT);
  63 }
  64 \f
  65 /* multipath_execute(). */
  66
  67 static uint32_t multipath_hash(const struct flow *, enum nx_mp_fields,
  68                                uint16_t basis);
  69 static uint16_t multipath_algorithm(uint32_t hash, enum nx_mp_algorithm,
  70                                     unsigned int n_links, unsigned int arg);
  71
  72 void
  73 multipath_execute(const struct nx_action_multipath *mp, struct flow *flow)
  74 {
  75     /* Calculate value to store. */
  76     uint32_t hash = multipath_hash(flow, ntohs(mp->fields), ntohs(mp->basis));
  77     uint16_t link = multipath_algorithm(hash, ntohs(mp->algorithm),
  78                                         ntohs(mp->max_link) + 1,
  79                                         ntohl(mp->arg));
  80
  81     /* Store it. */
  82     uint32_t *reg = &flow->regs[NXM_NX_REG_IDX(ntohl(mp->dst))];
  83     int ofs = nxm_decode_ofs(mp->ofs_nbits);
  84     int n_bits = nxm_decode_n_bits(mp->ofs_nbits);
  85     uint32_t mask = n_bits == 32 ? UINT32_MAX : (UINT32_C(1) << n_bits) - 1;
  86     *reg = (*reg & ~(mask << ofs)) | (link << ofs);
  87 }
  88
  89 static uint32_t
  90 multipath_hash(const struct flow *flow, enum nx_mp_fields fields,
  91                uint16_t basis)
  92 {
  93     switch (fields) {
  94     case NX_MP_FIELDS_ETH_SRC:
  95         return hash_bytes(flow->dl_src, sizeof flow->dl_src, basis);
  96
  97     case NX_MP_FIELDS_SYMMETRIC_L4:
  98         return flow_hash_symmetric_l4(flow, basis);
  99     }
 100
 101     NOT_REACHED();
 102 }
 103
 104 static uint16_t
 105 algorithm_hrw(uint32_t hash, unsigned int n_links)
 106 {
 107     uint32_t best_weight;
 108     uint16_t best_link;
 109     unsigned int link;
 110
 111     best_link = 0;
 112     best_weight = hash_2words(hash, 0);
 113     for (link = 1; link < n_links; link++) {
 114         uint32_t weight = hash_2words(hash, link);
 115         if (weight > best_weight) {
 116             best_link = link;
 117             best_weight = weight;
 118         }
 119     }
 120     return best_link;
 121 }
 122
 123 /* Works for 'x' in the range [1,65536], which is all we need.  */
 124 static unsigned int
 125 round_up_pow2(unsigned int x)
 126 {
 127     x--;
 128     x |= x >> 1;
 129     x |= x >> 2;
 130     x |= x >> 4;
 131     x |= x >> 8;
 132     return x + 1;
 133 }
 134
 135 static uint16_t
 136 algorithm_iter_hash(uint32_t hash, unsigned int n_links, unsigned int modulo)
 137 {
 138     uint16_t link;
 139     int i;
 140
 141     if (modulo < n_links || modulo / 2 > n_links) {
 142         modulo = round_up_pow2(n_links);
 143     }
 144
 145     i = 0;
 146     do {
 147         link = hash_2words(hash, i++) % modulo;
 148     } while (link >= n_links);
 149
 150     return link;
 151 }
 152
 153 static uint16_t
 154 multipath_algorithm(uint32_t hash, enum nx_mp_algorithm algorithm,
 155                     unsigned int n_links, unsigned int arg)
 156 {
 157     switch (algorithm) {
 158     case NX_MP_ALG_MODULO_N:
 159         return hash % n_links;
 160
 161     case NX_MP_ALG_HASH_THRESHOLD:
 162         if (n_links == 1) {
 163             return 0;
 164         }
 165         return hash / (UINT32_MAX / n_links + 1);
 166
 167     case NX_MP_ALG_HRW:
 168         return (n_links <= 64
 169                 ? algorithm_hrw(hash, n_links)
 170                 : algorithm_iter_hash(hash, n_links, 0));
 171
 172     case NX_MP_ALG_ITER_HASH:
 173         return algorithm_iter_hash(hash, n_links, arg);
 174     }
 175
 176     NOT_REACHED();
 177 }
 178 \f
 179 /* multipath_parse(). */
 180
 181 void
 182 multipath_parse(struct nx_action_multipath *mp, const char *s_)
 183 {
 184     char *s = xstrdup(s_);
 185     char *save_ptr = NULL;
 186     char *fields, *basis, *algorithm, *n_links_str, *arg, *dst;
 187     uint32_t header;
 188     int ofs, n_bits;
 189     int n_links;
 190
 191     fields = strtok_r(s, ", ", &save_ptr);
 192     basis = strtok_r(NULL, ", ", &save_ptr);
 193     algorithm = strtok_r(NULL, ", ", &save_ptr);
 194     n_links_str = strtok_r(NULL, ", ", &save_ptr);
 195     arg = strtok_r(NULL, ", ", &save_ptr);
 196     dst = strtok_r(NULL, ", ", &save_ptr);
 197     if (!dst) {
 198         ovs_fatal(0, "%s: not enough arguments to multipath action", s_);
 199     }
 200
 201     memset(mp, 0, sizeof *mp);
 202     mp->type = htons(OFPAT_VENDOR);
 203     mp->len = htons(sizeof *mp);
 204     mp->vendor = htonl(NX_VENDOR_ID);
 205     mp->subtype = htons(NXAST_MULTIPATH);
 206     if (!strcasecmp(fields, "eth_src")) {
 207         mp->fields = htons(NX_MP_FIELDS_ETH_SRC);
 208     } else if (!strcasecmp(fields, "symmetric_l4")) {
 209         mp->fields = htons(NX_MP_FIELDS_SYMMETRIC_L4);
 210     } else {
 211         ovs_fatal(0, "%s: unknown fields `%s'", s_, fields);
 212     }
 213     mp->basis = htons(atoi(basis));
 214     if (!strcasecmp(algorithm, "modulo_n")) {
 215         mp->algorithm = htons(NX_MP_ALG_MODULO_N);
 216     } else if (!strcasecmp(algorithm, "hash_threshold")) {
 217         mp->algorithm = htons(NX_MP_ALG_HASH_THRESHOLD);
 218     } else if (!strcasecmp(algorithm, "hrw")) {
 219         mp->algorithm = htons(NX_MP_ALG_HRW);
 220     } else if (!strcasecmp(algorithm, "iter_hash")) {
 221         mp->algorithm = htons(NX_MP_ALG_ITER_HASH);
 222     } else {
 223         ovs_fatal(0, "%s: unknown algorithm `%s'", s_, algorithm);
 224     }
 225     n_links = atoi(n_links_str);
 226     if (n_links < 1 || n_links > 65536) {
 227         ovs_fatal(0, "%s: n_links %d is not in valid range 1 to 65536",
 228                   s_, n_links);
 229     }
 230     mp->max_link = htons(n_links - 1);
 231     mp->arg = htonl(atoi(arg));
 232
 233     nxm_parse_field_bits(dst, &header, &ofs, &n_bits);
 234     if (!NXM_IS_NX_REG(header) || NXM_NX_REG_IDX(header) >= FLOW_N_REGS) {
 235         ovs_fatal(0, "%s: destination field must be register", s_);
 236     }
 237     if (n_bits < 16 && n_links > (1u << n_bits)) {
 238         ovs_fatal(0, "%s: %d-bit destination field has %u possible values, "
 239                   "less than specified n_links %d",
 240                   s_, n_bits, 1u << n_bits, n_links);
 241     }
 242     mp->ofs_nbits = nxm_encode_ofs_nbits(ofs, n_bits);
 243     mp->dst = htonl(header);
 244
 245     free(s);
 246 }
 247
 248 void
 249 multipath_format(const struct nx_action_multipath *mp, struct ds *s)
 250 {
 251     const char *fields, *algorithm;
 252
 253     uint16_t mp_fields    = ntohs(mp->fields);
 254     uint16_t mp_algorithm = ntohs(mp->algorithm);
 255
 256     switch ((enum nx_mp_fields) mp_fields) {
 257     case NX_MP_FIELDS_ETH_SRC:
 258         fields = "eth_src";
 259         break;
 260     case NX_MP_FIELDS_SYMMETRIC_L4:
 261         fields = "symmetric_l4";
 262         break;
 263     default:
 264         fields = "<unknown>";
 265     }
 266
 267     switch ((enum nx_mp_algorithm) mp_algorithm) {
 268     case NX_MP_ALG_MODULO_N:
 269         algorithm = "modulo_n";
 270         break;
 271     case NX_MP_ALG_HASH_THRESHOLD:
 272         algorithm = "hash_threshold";
 273         break;
 274     case NX_MP_ALG_HRW:
 275         algorithm = "hrw";
 276         break;
 277     case NX_MP_ALG_ITER_HASH:
 278         algorithm = "iter_hash";
 279         break;
 280     default:
 281         algorithm = "<unknown>";
 282     }
 283
 284     ds_put_format(s, "multipath(%s,%"PRIu16",%s,%d,%"PRIu16",",
 285                   fields, ntohs(mp->basis), algorithm, ntohs(mp->max_link) + 1,
 286                   ntohl(mp->arg));
 287     nxm_format_field_bits(s, ntohl(mp->dst), nxm_decode_ofs(mp->ofs_nbits),
 288                           nxm_decode_n_bits(mp->ofs_nbits));
 289     ds_put_char(s, ')');
 290 }