pintos-os.org Git - openvswitch/blob - datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
   4  * Stanford Junior University
   5  */
   6
   7 #include "flow.h"
   8 #include <linux/netdevice.h>
   9 #include <linux/etherdevice.h>
  10 #include <linux/if_ether.h>
  11 #include <linux/if_vlan.h>
  12 #include <net/llc_pdu.h>
  13 #include <linux/ip.h>
  14 #include <linux/kernel.h>
  15 #include <linux/module.h>
  16 #include <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <linux/in.h>
  19 #include <linux/rcupdate.h>
  20
  21 #include "openflow.h"
  22 #include "compat.h"
  23 #include "snap.h"
  24
  25 struct kmem_cache *flow_cache;
  26
  27 /* Internal function used to compare fields in flow. */
  28 static inline
  29 int flow_fields_match(const struct sw_flow_key *a, const struct sw_flow_key *b,
  30                 uint16_t w)
  31 {
  32         return ((w & OFPFW_IN_PORT || a->in_port == b->in_port)
  33                 && (w & OFPFW_DL_VLAN || a->dl_vlan == b->dl_vlan)
  34                 && (w & OFPFW_DL_SRC || !memcmp(a->dl_src, b->dl_src, ETH_ALEN))
  35                 && (w & OFPFW_DL_DST || !memcmp(a->dl_dst, b->dl_dst, ETH_ALEN))
  36                 && (w & OFPFW_DL_TYPE || a->dl_type == b->dl_type)
  37                 && (w & OFPFW_NW_SRC || a->nw_src == b->nw_src)
  38                 && (w & OFPFW_NW_DST || a->nw_dst == b->nw_dst)
  39                 && (w & OFPFW_NW_PROTO || a->nw_proto == b->nw_proto)
  40                 && (w & OFPFW_TP_SRC || a->tp_src == b->tp_src)
  41                 && (w & OFPFW_TP_DST || a->tp_dst == b->tp_dst));
  42 }
  43
  44 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  45  * modulo wildcards, zero otherwise. */
  46 int flow_matches(const struct sw_flow_key *a, const struct sw_flow_key *b)
  47 {
  48         return flow_fields_match(a, b, (a->wildcards | b->wildcards));
  49 }
  50 EXPORT_SYMBOL(flow_matches);
  51
  52 /* Returns nonzero if 't' (the table entry's key) and 'd' (the key
  53  * describing the deletion) match, that is, if their fields are
  54  * equal modulo wildcards, zero otherwise.  If 'strict' is nonzero, the
  55  * wildcards must match in both 't_key' and 'd_key'.  Note that the
  56  * table's wildcards are ignored unless 'strict' is set. */
  57 int flow_del_matches(const struct sw_flow_key *t, const struct sw_flow_key *d, int strict)
  58 {
  59         if (strict && (t->wildcards != d->wildcards))
  60                 return 0;
  61
  62         return flow_fields_match(t, d, d->wildcards);
  63 }
  64 EXPORT_SYMBOL(flow_del_matches);
  65
  66 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
  67 {
  68         to->wildcards = ntohs(from->wildcards) & OFPFW_ALL;
  69         memset(to->pad, '\0', sizeof(to->pad));
  70         to->in_port = from->in_port;
  71         to->dl_vlan = from->dl_vlan;
  72         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
  73         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
  74         to->dl_type = from->dl_type;
  75
  76         to->nw_src = to->nw_dst = to->nw_proto = 0;
  77         to->tp_src = to->tp_dst = 0;
  78
  79 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
  80 #define OFPFW_NW (OFPFW_NW_SRC | OFPFW_NW_DST | OFPFW_NW_PROTO)
  81         if (to->wildcards & OFPFW_DL_TYPE) {
  82                 /* Can't sensibly match on network or transport headers if the
  83                  * data link type is unknown. */
  84                 to->wildcards |= OFPFW_NW | OFPFW_TP;
  85         } else if (from->dl_type == htons(ETH_P_IP)) {
  86                 to->nw_src   = from->nw_src;
  87                 to->nw_dst   = from->nw_dst;
  88                 to->nw_proto = from->nw_proto;
  89
  90                 if (to->wildcards & OFPFW_NW_PROTO) {
  91                         /* Can't sensibly match on transport headers if the
  92                          * network protocol is unknown. */
  93                         to->wildcards |= OFPFW_TP;
  94                 } else if (from->nw_proto == IPPROTO_TCP
  95                            || from->nw_proto == IPPROTO_UDP) {
  96                         to->tp_src = from->tp_src;
  97                         to->tp_dst = from->tp_dst;
  98                 } else {
  99                         /* Transport layer fields are undefined.  Mark them as
 100                          * exact-match to allow such flows to reside in
 101                          * table-hash, instead of falling into table-linear. */
 102                         to->wildcards &= ~OFPFW_TP;
 103                 }
 104         } else {
 105                 /* Network and transport layer fields are undefined.  Mark them
 106                  * as exact-match to allow such flows to reside in table-hash,
 107                  * instead of falling into table-linear. */
 108                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
 109         }
 110 }
 111
 112 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
 113 {
 114         to->wildcards = htons(from->wildcards);
 115         to->in_port   = from->in_port;
 116         to->dl_vlan   = from->dl_vlan;
 117         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
 118         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
 119         to->dl_type   = from->dl_type;
 120         to->nw_src        = from->nw_src;
 121         to->nw_dst        = from->nw_dst;
 122         to->nw_proto  = from->nw_proto;
 123         to->tp_src        = from->tp_src;
 124         to->tp_dst        = from->tp_dst;
 125         memset(to->pad, '\0', sizeof(to->pad));
 126 }
 127
 128 /* Returns true if 'flow' can be deleted and set up for a deferred free, false
 129  * if deletion has already been scheduled (by another thread).
 130  *
 131  * Caller must hold rcu_read_lock. */
 132 int flow_del(struct sw_flow *flow)
 133 {
 134         return !atomic_cmpxchg(&flow->deleted, 0, 1);
 135 }
 136 EXPORT_SYMBOL(flow_del);
 137
 138 /* Allocates and returns a new flow with 'n_actions' action, using allocation
 139  * flags 'flags'.  Returns the new flow or a null pointer on failure. */
 140 struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
 141 {
 142         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 143         if (unlikely(!flow))
 144                 return NULL;
 145
 146         flow->n_actions = n_actions;
 147         flow->actions = kmalloc(n_actions * sizeof *flow->actions,
 148                                 flags);
 149         if (unlikely(!flow->actions) && n_actions > 0) {
 150                 kmem_cache_free(flow_cache, flow);
 151                 return NULL;
 152         }
 153         return flow;
 154 }
 155
 156 /* Frees 'flow' immediately. */
 157 void flow_free(struct sw_flow *flow)
 158 {
 159         if (unlikely(!flow))
 160                 return;
 161         kfree(flow->actions);
 162         kmem_cache_free(flow_cache, flow);
 163 }
 164 EXPORT_SYMBOL(flow_free);
 165
 166 /* RCU callback used by flow_deferred_free. */
 167 static void rcu_callback(struct rcu_head *rcu)
 168 {
 169         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 170         flow_free(flow);
 171 }
 172
 173 /* Schedules 'flow' to be freed after the next RCU grace period.
 174  * The caller must hold rcu_read_lock for this to be sensible. */
 175 void flow_deferred_free(struct sw_flow *flow)
 176 {
 177         call_rcu(&flow->rcu, rcu_callback);
 178 }
 179 EXPORT_SYMBOL(flow_deferred_free);
 180
 181 /* Prints a representation of 'key' to the kernel log. */
 182 void print_flow(const struct sw_flow_key *key)
 183 {
 184         printk("wild%04x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 185                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 186                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 187                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 188                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 189                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 190                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 191                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 192                         ntohs(key->dl_type),
 193                         ((unsigned char *)&key->nw_src)[0],
 194                         ((unsigned char *)&key->nw_src)[1],
 195                         ((unsigned char *)&key->nw_src)[2],
 196                         ((unsigned char *)&key->nw_src)[3],
 197                         ((unsigned char *)&key->nw_dst)[0],
 198                         ((unsigned char *)&key->nw_dst)[1],
 199                         ((unsigned char *)&key->nw_dst)[2],
 200                         ((unsigned char *)&key->nw_dst)[3],
 201                         ntohs(key->tp_src), ntohs(key->tp_dst));
 202 }
 203 EXPORT_SYMBOL(print_flow);
 204
 205 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 206  * and initializes 'key' to match. */
 207 void flow_extract(struct sk_buff *skb, uint16_t in_port,
 208                   struct sw_flow_key *key)
 209 {
 210         struct ethhdr *mac;
 211         struct udphdr *th;
 212         int nh_ofs, th_ofs;
 213
 214         key->in_port = htons(in_port);
 215         key->wildcards = 0;
 216         memset(key->pad, '\0', sizeof(key->pad));
 217
 218         /* This code doesn't check that skb->len is long enough to contain the
 219          * MAC or network header.  With a 46-byte minimum length frame this
 220          * assumption is always correct. */
 221
 222         /* Doesn't verify checksums.  Should it? */
 223
 224         /* Data link layer.  We only support Ethernet. */
 225         mac = eth_hdr(skb);
 226         nh_ofs = sizeof(struct ethhdr);
 227         if (likely(ntohs(mac->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF)) {
 228                 /* This is an Ethernet II frame */
 229                 key->dl_type = mac->h_proto;
 230         } else {
 231                 /* This is an 802.2 frame */
 232                 if (snap_get_ethertype(skb, &key->dl_type) != -EINVAL) {
 233                         nh_ofs += sizeof(struct snap_hdr);
 234                 } else {
 235                         key->dl_type = OFP_DL_TYPE_NOT_ETH_TYPE;
 236                         nh_ofs += sizeof(struct llc_pdu_un);
 237                 }
 238         }
 239
 240         /* Check for a VLAN tag */
 241         if (likely(key->dl_type != htons(ETH_P_8021Q))) {
 242                 key->dl_vlan = htons(OFP_VLAN_NONE);
 243         } else {
 244                 struct vlan_hdr *vh = (struct vlan_hdr *)(skb_mac_header(skb) + nh_ofs);
 245                 key->dl_type = vh->h_vlan_encapsulated_proto;
 246                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 247                 nh_ofs += sizeof(*vh);
 248         }
 249         memcpy(key->dl_src, mac->h_source, ETH_ALEN);
 250         memcpy(key->dl_dst, mac->h_dest, ETH_ALEN);
 251         skb_set_network_header(skb, nh_ofs);
 252
 253         /* Network layer. */
 254         if (likely(key->dl_type == htons(ETH_P_IP))) {
 255                 struct iphdr *nh = ip_hdr(skb);
 256                 key->nw_src = nh->saddr;
 257                 key->nw_dst = nh->daddr;
 258                 key->nw_proto = nh->protocol;
 259                 th_ofs = nh_ofs + nh->ihl * 4;
 260                 skb_set_transport_header(skb, th_ofs);
 261
 262                 /* Transport layer. */
 263                 if ((key->nw_proto != IPPROTO_TCP && key->nw_proto != IPPROTO_UDP)
 264                                 || skb->len < th_ofs + sizeof(struct udphdr)) {
 265                         goto no_th;
 266                 }
 267                 th = udp_hdr(skb);
 268                 key->tp_src = th->source;
 269                 key->tp_dst = th->dest;
 270
 271                 return;
 272         }
 273
 274         key->nw_src = 0;
 275         key->nw_dst = 0;
 276         key->nw_proto = 0;
 277
 278 no_th:
 279         key->tp_src = 0;
 280         key->tp_dst = 0;
 281 }
 282
 283 /* Initializes the flow module.
 284  * Returns zero if successful or a negative error code. */
 285 int flow_init(void)
 286 {
 287         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 288                                         0, NULL);
 289         if (flow_cache == NULL)
 290                 return -ENOMEM;
 291
 292         return 0;
 293 }
 294
 295 /* Uninitializes the flow module. */
 296 void flow_exit(void)
 297 {
 298         kmem_cache_destroy(flow_cache);
 299 }
 300