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/jiffies.h>
  15 #include <linux/kernel.h>
  16 #include <linux/module.h>
  17 #include <linux/tcp.h>
  18 #include <linux/udp.h>
  19 #include <linux/in.h>
  20 #include <linux/rcupdate.h>
  21 #include <net/ip.h>
  22
  23 #include "openflow.h"
  24 #include "compat.h"
  25 #include "snap.h"
  26
  27 struct kmem_cache *flow_cache;
  28
  29 /* Internal function used to compare fields in flow. */
  30 static inline
  31 int flow_fields_match(const struct sw_flow_key *a, const struct sw_flow_key *b,
  32                       uint32_t w, uint32_t src_mask, uint32_t dst_mask)
  33 {
  34         return ((w & OFPFW_IN_PORT || a->in_port == b->in_port)
  35                 && (w & OFPFW_DL_VLAN || a->dl_vlan == b->dl_vlan)
  36                 && (w & OFPFW_DL_SRC || !memcmp(a->dl_src, b->dl_src, ETH_ALEN))
  37                 && (w & OFPFW_DL_DST || !memcmp(a->dl_dst, b->dl_dst, ETH_ALEN))
  38                 && (w & OFPFW_DL_TYPE || a->dl_type == b->dl_type)
  39                 && !((a->nw_src ^ b->nw_src) & src_mask)
  40                 && !((a->nw_dst ^ b->nw_dst) & dst_mask)
  41                 && (w & OFPFW_NW_PROTO || a->nw_proto == b->nw_proto)
  42                 && (w & OFPFW_TP_SRC || a->tp_src == b->tp_src)
  43                 && (w & OFPFW_TP_DST || a->tp_dst == b->tp_dst));
  44 }
  45
  46 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  47  * modulo wildcards in 'b', zero otherwise. */
  48 int flow_matches_1wild(const struct sw_flow_key *a,
  49                        const struct sw_flow_key *b)
  50 {
  51         return flow_fields_match(a, b, b->wildcards,
  52                                  b->nw_src_mask, b->nw_dst_mask);
  53 }
  54 EXPORT_SYMBOL(flow_matches_1wild);
  55
  56 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  57  * modulo wildcards in 'a' or 'b', zero otherwise. */
  58 int flow_matches_2wild(const struct sw_flow_key *a,
  59                        const struct sw_flow_key *b)
  60 {
  61         return flow_fields_match(a, b,
  62                                  a->wildcards | b->wildcards,
  63                                  a->nw_src_mask & b->nw_src_mask,
  64                                  a->nw_dst_mask & b->nw_dst_mask);
  65 }
  66 EXPORT_SYMBOL(flow_matches_2wild);
  67
  68 /* Returns nonzero if 't' (the table entry's key) and 'd' (the key
  69  * describing the deletion) match, that is, if their fields are
  70  * equal modulo wildcards, zero otherwise.  If 'strict' is nonzero, the
  71  * wildcards must match in both 't_key' and 'd_key'.  Note that the
  72  * table's wildcards are ignored unless 'strict' is set. */
  73 int flow_del_matches(const struct sw_flow_key *t, const struct sw_flow_key *d, int strict)
  74 {
  75         if (strict && d->wildcards != t->wildcards)
  76                 return 0;
  77         return flow_matches_1wild(t, d);
  78 }
  79 EXPORT_SYMBOL(flow_del_matches);
  80
  81 static uint32_t make_nw_mask(int n_wild_bits)
  82 {
  83         n_wild_bits &= (1u << OFPFW_NW_SRC_BITS) - 1;
  84         return n_wild_bits < 32 ? htonl(~((1u << n_wild_bits) - 1)) : 0;
  85 }
  86
  87 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
  88 {
  89         to->wildcards = ntohl(from->wildcards) & OFPFW_ALL;
  90         to->pad = 0;
  91         to->in_port = from->in_port;
  92         to->dl_vlan = from->dl_vlan;
  93         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
  94         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
  95         to->dl_type = from->dl_type;
  96
  97         to->nw_src = to->nw_dst = to->nw_proto = 0;
  98         to->tp_src = to->tp_dst = 0;
  99
 100 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
 101 #define OFPFW_NW (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO)
 102         if (to->wildcards & OFPFW_DL_TYPE) {
 103                 /* Can't sensibly match on network or transport headers if the
 104                  * data link type is unknown. */
 105                 to->wildcards |= OFPFW_NW | OFPFW_TP;
 106         } else if (from->dl_type == htons(ETH_P_IP)) {
 107                 to->nw_src   = from->nw_src;
 108                 to->nw_dst   = from->nw_dst;
 109                 to->nw_proto = from->nw_proto;
 110
 111                 if (to->wildcards & OFPFW_NW_PROTO) {
 112                         /* Can't sensibly match on transport headers if the
 113                          * network protocol is unknown. */
 114                         to->wildcards |= OFPFW_TP;
 115                 } else if (from->nw_proto == IPPROTO_TCP
 116                            || from->nw_proto == IPPROTO_UDP) {
 117                         to->tp_src = from->tp_src;
 118                         to->tp_dst = from->tp_dst;
 119                 } else {
 120                         /* Transport layer fields are undefined.  Mark them as
 121                          * exact-match to allow such flows to reside in
 122                          * table-hash, instead of falling into table-linear. */
 123                         to->wildcards &= ~OFPFW_TP;
 124                 }
 125         } else {
 126                 /* Network and transport layer fields are undefined.  Mark them
 127                  * as exact-match to allow such flows to reside in table-hash,
 128                  * instead of falling into table-linear. */
 129                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
 130         }
 131
 132         /* We set these late because code above adjusts to->wildcards. */
 133         to->nw_src_mask = make_nw_mask(to->wildcards >> OFPFW_NW_SRC_SHIFT);
 134         to->nw_dst_mask = make_nw_mask(to->wildcards >> OFPFW_NW_DST_SHIFT);
 135 }
 136
 137 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
 138 {
 139         to->wildcards = htonl(from->wildcards);
 140         to->in_port   = from->in_port;
 141         to->dl_vlan   = from->dl_vlan;
 142         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
 143         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
 144         to->dl_type   = from->dl_type;
 145         to->nw_src        = from->nw_src;
 146         to->nw_dst        = from->nw_dst;
 147         to->nw_proto  = from->nw_proto;
 148         to->tp_src        = from->tp_src;
 149         to->tp_dst        = from->tp_dst;
 150         to->pad           = 0;
 151 }
 152
 153 int flow_timeout(struct sw_flow *flow)
 154 {
 155         if (flow->idle_timeout != OFP_FLOW_PERMANENT
 156             && time_after(jiffies, flow->used + flow->idle_timeout * HZ))
 157                 return OFPER_IDLE_TIMEOUT;
 158         else if (flow->hard_timeout != OFP_FLOW_PERMANENT
 159                  && time_after(jiffies,
 160                                flow->init_time + flow->hard_timeout * HZ))
 161                 return OFPER_HARD_TIMEOUT;
 162         else
 163                 return -1;
 164 }
 165 EXPORT_SYMBOL(flow_timeout);
 166
 167 /* Allocates and returns a new flow with 'n_actions' action, using allocation
 168  * flags 'flags'.  Returns the new flow or a null pointer on failure. */
 169 struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
 170 {
 171         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 172         if (unlikely(!flow))
 173                 return NULL;
 174
 175         flow->n_actions = n_actions;
 176         flow->actions = kmalloc(n_actions * sizeof *flow->actions,
 177                                 flags);
 178         if (unlikely(!flow->actions) && n_actions > 0) {
 179                 kmem_cache_free(flow_cache, flow);
 180                 return NULL;
 181         }
 182         return flow;
 183 }
 184
 185 /* Frees 'flow' immediately. */
 186 void flow_free(struct sw_flow *flow)
 187 {
 188         if (unlikely(!flow))
 189                 return;
 190         kfree(flow->actions);
 191         kmem_cache_free(flow_cache, flow);
 192 }
 193 EXPORT_SYMBOL(flow_free);
 194
 195 /* RCU callback used by flow_deferred_free. */
 196 static void rcu_callback(struct rcu_head *rcu)
 197 {
 198         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 199         flow_free(flow);
 200 }
 201
 202 /* Schedules 'flow' to be freed after the next RCU grace period.
 203  * The caller must hold rcu_read_lock for this to be sensible. */
 204 void flow_deferred_free(struct sw_flow *flow)
 205 {
 206         call_rcu(&flow->rcu, rcu_callback);
 207 }
 208 EXPORT_SYMBOL(flow_deferred_free);
 209
 210 /* Prints a representation of 'key' to the kernel log. */
 211 void print_flow(const struct sw_flow_key *key)
 212 {
 213         printk("wild%08x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 214                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 215                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 216                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 217                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 218                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 219                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 220                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 221                         ntohs(key->dl_type),
 222                         ((unsigned char *)&key->nw_src)[0],
 223                         ((unsigned char *)&key->nw_src)[1],
 224                         ((unsigned char *)&key->nw_src)[2],
 225                         ((unsigned char *)&key->nw_src)[3],
 226                         ((unsigned char *)&key->nw_dst)[0],
 227                         ((unsigned char *)&key->nw_dst)[1],
 228                         ((unsigned char *)&key->nw_dst)[2],
 229                         ((unsigned char *)&key->nw_dst)[3],
 230                         ntohs(key->tp_src), ntohs(key->tp_dst));
 231 }
 232 EXPORT_SYMBOL(print_flow);
 233
 234 static int tcphdr_ok(struct sk_buff *skb)
 235 {
 236         int th_ofs = skb_transport_offset(skb);
 237         if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
 238                 int tcp_len = tcp_hdrlen(skb);
 239                 return (tcp_len >= sizeof(struct tcphdr)
 240                         && skb->len >= th_ofs + tcp_len);
 241         }
 242         return 0;
 243 }
 244
 245 static int udphdr_ok(struct sk_buff *skb)
 246 {
 247         int th_ofs = skb_transport_offset(skb);
 248         return skb->len >= th_ofs + sizeof(struct udphdr);
 249 }
 250
 251 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 252  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 253  * fragment, 0 otherwise. */
 254 int flow_extract(struct sk_buff *skb, uint16_t in_port,
 255                  struct sw_flow_key *key)
 256 {
 257         struct ethhdr *mac;
 258         int nh_ofs, th_ofs;
 259         int retval = 0;
 260
 261         key->in_port = htons(in_port);
 262         key->pad = 0;
 263         key->wildcards = 0;
 264         key->nw_src_mask = 0;
 265         key->nw_dst_mask = 0;
 266
 267         /* This code doesn't check that skb->len is long enough to contain the
 268          * MAC or network header.  With a 46-byte minimum length frame this
 269          * assumption is always correct. */
 270
 271         /* Doesn't verify checksums.  Should it? */
 272
 273         /* Data link layer.  We only support Ethernet. */
 274         mac = eth_hdr(skb);
 275         nh_ofs = sizeof(struct ethhdr);
 276         if (likely(ntohs(mac->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF)) {
 277                 /* This is an Ethernet II frame */
 278                 key->dl_type = mac->h_proto;
 279         } else {
 280                 /* This is an 802.2 frame */
 281                 if (snap_get_ethertype(skb, &key->dl_type) != -EINVAL) {
 282                         nh_ofs += sizeof(struct snap_hdr);
 283                 } else {
 284                         key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
 285                         nh_ofs += sizeof(struct llc_pdu_un);
 286                 }
 287         }
 288
 289         /* Check for a VLAN tag */
 290         if (likely(key->dl_type != htons(ETH_P_8021Q))) {
 291                 key->dl_vlan = htons(OFP_VLAN_NONE);
 292         } else {
 293                 struct vlan_hdr *vh = (struct vlan_hdr *)(skb_mac_header(skb) + nh_ofs);
 294                 key->dl_type = vh->h_vlan_encapsulated_proto;
 295                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 296                 nh_ofs += sizeof(*vh);
 297         }
 298         memcpy(key->dl_src, mac->h_source, ETH_ALEN);
 299         memcpy(key->dl_dst, mac->h_dest, ETH_ALEN);
 300         skb_set_network_header(skb, nh_ofs);
 301
 302         /* Network layer. */
 303         if (likely(key->dl_type == htons(ETH_P_IP))) {
 304                 struct iphdr *nh = ip_hdr(skb);
 305                 key->nw_src = nh->saddr;
 306                 key->nw_dst = nh->daddr;
 307                 key->nw_proto = nh->protocol;
 308                 th_ofs = nh_ofs + nh->ihl * 4;
 309                 skb_set_transport_header(skb, th_ofs);
 310
 311                 /* Transport layer. */
 312                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 313                         if (key->nw_proto == IPPROTO_TCP) {
 314                                 if (tcphdr_ok(skb)) {
 315                                         struct tcphdr *tcp = tcp_hdr(skb);
 316                                         key->tp_src = tcp->source;
 317                                         key->tp_dst = tcp->dest;
 318                                 } else {
 319                                         /* Avoid tricking other code into
 320                                          * thinking that this packet has an L4
 321                                          * header. */
 322                                         goto no_proto;
 323                                 }
 324                         } else if (key->nw_proto == IPPROTO_UDP) {
 325                                 if (udphdr_ok(skb)) {
 326                                         struct udphdr *udp = udp_hdr(skb);
 327                                         key->tp_src = udp->source;
 328                                         key->tp_dst = udp->dest;
 329                                 } else {
 330                                         /* Avoid tricking other code into
 331                                          * thinking that this packet has an L4
 332                                          * header. */
 333                                         goto no_proto;
 334                                 }
 335                         } else {
 336                                 goto no_th;
 337                         }
 338                 } else {
 339                         retval = 1;
 340                         goto no_th;
 341                 }
 342
 343                 return 0;
 344         }
 345
 346         key->nw_src = 0;
 347         key->nw_dst = 0;
 348
 349 no_proto:
 350         key->nw_proto = 0;
 351
 352 no_th:
 353         key->tp_src = 0;
 354         key->tp_dst = 0;
 355         return retval;
 356 }
 357
 358 /* Initializes the flow module.
 359  * Returns zero if successful or a negative error code. */
 360 int flow_init(void)
 361 {
 362         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 363                                         0, NULL);
 364         if (flow_cache == NULL)
 365                 return -ENOMEM;
 366
 367         return 0;
 368 }
 369
 370 /* Uninitializes the flow module. */
 371 void flow_exit(void)
 372 {
 373         kmem_cache_destroy(flow_cache);
 374 }
 375