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/llc.h>
  17 #include <linux/module.h>
  18 #include <linux/tcp.h>
  19 #include <linux/udp.h>
  20 #include <linux/icmp.h>
  21 #include <linux/in.h>
  22 #include <linux/rcupdate.h>
  23 #include <net/ip.h>
  24
  25 #include "openflow/openflow.h"
  26 #include "compat.h"
  27
  28 struct kmem_cache *flow_cache;
  29
  30 /* Internal function used to compare fields in flow. */
  31 static inline
  32 int flow_fields_match(const struct sw_flow_key *a, const struct sw_flow_key *b,
  33                       uint32_t w, uint32_t src_mask, uint32_t dst_mask)
  34 {
  35         return ((w & OFPFW_IN_PORT || a->in_port == b->in_port)
  36                 && (w & OFPFW_DL_VLAN || a->dl_vlan == b->dl_vlan)
  37                 && (w & OFPFW_DL_SRC || !memcmp(a->dl_src, b->dl_src, ETH_ALEN))
  38                 && (w & OFPFW_DL_DST || !memcmp(a->dl_dst, b->dl_dst, ETH_ALEN))
  39                 && (w & OFPFW_DL_TYPE || a->dl_type == b->dl_type)
  40                 && !((a->nw_src ^ b->nw_src) & src_mask)
  41                 && !((a->nw_dst ^ b->nw_dst) & dst_mask)
  42                 && (w & OFPFW_NW_PROTO || a->nw_proto == b->nw_proto)
  43                 && (w & OFPFW_TP_SRC || a->tp_src == b->tp_src)
  44                 && (w & OFPFW_TP_DST || a->tp_dst == b->tp_dst));
  45 }
  46
  47 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  48  * modulo wildcards in 'b', zero otherwise. */
  49 int flow_matches_1wild(const struct sw_flow_key *a,
  50                        const struct sw_flow_key *b)
  51 {
  52         return flow_fields_match(a, b, b->wildcards,
  53                                  b->nw_src_mask, b->nw_dst_mask);
  54 }
  55 EXPORT_SYMBOL(flow_matches_1wild);
  56
  57 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  58  * modulo wildcards in 'a' or 'b', zero otherwise. */
  59 int flow_matches_2wild(const struct sw_flow_key *a,
  60                        const struct sw_flow_key *b)
  61 {
  62         return flow_fields_match(a, b,
  63                                  a->wildcards | b->wildcards,
  64                                  a->nw_src_mask & b->nw_src_mask,
  65                                  a->nw_dst_mask & b->nw_dst_mask);
  66 }
  67 EXPORT_SYMBOL(flow_matches_2wild);
  68
  69 /* Returns nonzero if 't' (the table entry's key) and 'd' (the key
  70  * describing the match) match, that is, if their fields are
  71  * equal modulo wildcards, zero otherwise.  If 'strict' is nonzero, the
  72  * wildcards must match in both 't_key' and 'd_key'.  Note that the
  73  * table's wildcards are ignored unless 'strict' is set. */
  74 int flow_matches_desc(const struct sw_flow_key *t, const struct sw_flow_key *d,
  75                 int strict)
  76 {
  77         if (strict && d->wildcards != t->wildcards)
  78                 return 0;
  79         return flow_matches_1wild(t, d);
  80 }
  81 EXPORT_SYMBOL(flow_matches_desc);
  82
  83 static uint32_t make_nw_mask(int n_wild_bits)
  84 {
  85         n_wild_bits &= (1u << OFPFW_NW_SRC_BITS) - 1;
  86         return n_wild_bits < 32 ? htonl(~((1u << n_wild_bits) - 1)) : 0;
  87 }
  88
  89 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
  90 {
  91         to->wildcards = ntohl(from->wildcards) & OFPFW_ALL;
  92         to->pad = 0;
  93         to->in_port = from->in_port;
  94         to->dl_vlan = from->dl_vlan;
  95         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
  96         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
  97         to->dl_type = from->dl_type;
  98
  99         to->nw_src = to->nw_dst = to->nw_proto = 0;
 100         to->tp_src = to->tp_dst = 0;
 101
 102 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
 103 #define OFPFW_NW (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO)
 104         if (to->wildcards & OFPFW_DL_TYPE) {
 105                 /* Can't sensibly match on network or transport headers if the
 106                  * data link type is unknown. */
 107                 to->wildcards |= OFPFW_NW | OFPFW_TP;
 108         } else if (from->dl_type == htons(ETH_P_IP)) {
 109                 to->nw_src   = from->nw_src;
 110                 to->nw_dst   = from->nw_dst;
 111                 to->nw_proto = from->nw_proto;
 112
 113                 if (to->wildcards & OFPFW_NW_PROTO) {
 114                         /* Can't sensibly match on transport headers if the
 115                          * network protocol is unknown. */
 116                         to->wildcards |= OFPFW_TP;
 117                 } else if (from->nw_proto == IPPROTO_TCP
 118                                 || from->nw_proto == IPPROTO_UDP
 119                                 || from->nw_proto == IPPROTO_ICMP) {
 120                         to->tp_src = from->tp_src;
 121                         to->tp_dst = from->tp_dst;
 122                 } else {
 123                         /* Transport layer fields are undefined.  Mark them as
 124                          * exact-match to allow such flows to reside in
 125                          * table-hash, instead of falling into table-linear. */
 126                         to->wildcards &= ~OFPFW_TP;
 127                 }
 128         } else {
 129                 /* Network and transport layer fields are undefined.  Mark them
 130                  * as exact-match to allow such flows to reside in table-hash,
 131                  * instead of falling into table-linear. */
 132                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
 133         }
 134
 135         /* We set these late because code above adjusts to->wildcards. */
 136         to->nw_src_mask = make_nw_mask(to->wildcards >> OFPFW_NW_SRC_SHIFT);
 137         to->nw_dst_mask = make_nw_mask(to->wildcards >> OFPFW_NW_DST_SHIFT);
 138 }
 139
 140 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
 141 {
 142         to->wildcards = htonl(from->wildcards);
 143         to->in_port   = from->in_port;
 144         to->dl_vlan   = from->dl_vlan;
 145         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
 146         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
 147         to->dl_type   = from->dl_type;
 148         to->nw_src    = from->nw_src;
 149         to->nw_dst    = from->nw_dst;
 150         to->nw_proto  = from->nw_proto;
 151         to->tp_src    = from->tp_src;
 152         to->tp_dst    = from->tp_dst;
 153         to->pad       = 0;
 154 }
 155
 156 int flow_timeout(struct sw_flow *flow)
 157 {
 158         if (flow->idle_timeout != OFP_FLOW_PERMANENT
 159             && time_after(jiffies, flow->used + flow->idle_timeout * HZ))
 160                 return OFPER_IDLE_TIMEOUT;
 161         else if (flow->hard_timeout != OFP_FLOW_PERMANENT
 162                  && time_after(jiffies,
 163                                flow->init_time + flow->hard_timeout * HZ))
 164                 return OFPER_HARD_TIMEOUT;
 165         else
 166                 return -1;
 167 }
 168 EXPORT_SYMBOL(flow_timeout);
 169
 170 /* Returns nonzero if 'flow' contains an output action to 'out_port' or
 171  * has the value OFPP_NONE. 'out_port' is in network-byte order. */
 172 int flow_has_out_port(struct sw_flow *flow, uint16_t out_port)
 173 {
 174         struct sw_flow_actions *sf_acts;
 175         size_t actions_len;
 176         uint8_t *p;
 177
 178         if (out_port == htons(OFPP_NONE))
 179                 return 1;
 180
 181         sf_acts = rcu_dereference(flow->sf_acts);
 182
 183         actions_len = sf_acts->actions_len;
 184         p = (uint8_t *)sf_acts->actions;
 185
 186         while (actions_len > 0) {
 187                 struct ofp_action_header *ah = (struct ofp_action_header *)p;
 188                 size_t len = ntohs(ah->len);
 189
 190                 if (ah->type == htons(OFPAT_OUTPUT)) {
 191                         struct ofp_action_output *oa = (struct ofp_action_output *)p;
 192                         if (oa->port == out_port)
 193                                 return 1;
 194                 }
 195
 196                 p += len;
 197                 actions_len -= len;
 198         }
 199
 200         return 0;
 201 }
 202 EXPORT_SYMBOL(flow_has_out_port);
 203
 204 /* Allocates and returns a new flow with room for 'actions_len' actions,
 205  * using allocation flags 'flags'.  Returns the new flow or a null pointer
 206  * on failure. */
 207 struct sw_flow *flow_alloc(size_t actions_len, gfp_t flags)
 208 {
 209         struct sw_flow_actions *sfa;
 210         size_t size = sizeof *sfa + actions_len;
 211         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 212         if (unlikely(!flow))
 213                 return NULL;
 214
 215         sfa = kmalloc(size, flags);
 216         if (unlikely(!sfa)) {
 217                 kmem_cache_free(flow_cache, flow);
 218                 return NULL;
 219         }
 220         sfa->actions_len = actions_len;
 221         flow->sf_acts = sfa;
 222
 223         return flow;
 224 }
 225
 226 /* Frees 'flow' immediately. */
 227 void flow_free(struct sw_flow *flow)
 228 {
 229         if (unlikely(!flow))
 230                 return;
 231         kfree(flow->sf_acts);
 232         kmem_cache_free(flow_cache, flow);
 233 }
 234 EXPORT_SYMBOL(flow_free);
 235
 236 /* RCU callback used by flow_deferred_free. */
 237 static void rcu_free_flow_callback(struct rcu_head *rcu)
 238 {
 239         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 240         flow_free(flow);
 241 }
 242
 243 /* Schedules 'flow' to be freed after the next RCU grace period.
 244  * The caller must hold rcu_read_lock for this to be sensible. */
 245 void flow_deferred_free(struct sw_flow *flow)
 246 {
 247         call_rcu(&flow->rcu, rcu_free_flow_callback);
 248 }
 249 EXPORT_SYMBOL(flow_deferred_free);
 250
 251 /* RCU callback used by flow_deferred_free_acts. */
 252 static void rcu_free_acts_callback(struct rcu_head *rcu)
 253 {
 254         struct sw_flow_actions *sf_acts = container_of(rcu,
 255                         struct sw_flow_actions, rcu);
 256         kfree(sf_acts);
 257 }
 258
 259 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 260  * The caller must hold rcu_read_lock for this to be sensible. */
 261 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 262 {
 263         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 264 }
 265 EXPORT_SYMBOL(flow_deferred_free_acts);
 266
 267 /* Copies 'actions' into a newly allocated structure for use by 'flow'
 268  * and safely frees the structure that defined the previous actions. */
 269 void flow_replace_acts(struct sw_flow *flow,
 270                 const struct ofp_action_header *actions, size_t actions_len)
 271 {
 272         struct sw_flow_actions *sfa;
 273         struct sw_flow_actions *orig_sfa = flow->sf_acts;
 274         size_t size = sizeof *sfa + actions_len;
 275
 276         sfa = kmalloc(size, GFP_ATOMIC);
 277         if (unlikely(!sfa))
 278                 return;
 279
 280         sfa->actions_len = actions_len;
 281         memcpy(sfa->actions, actions, actions_len);
 282
 283         rcu_assign_pointer(flow->sf_acts, sfa);
 284         flow_deferred_free_acts(orig_sfa);
 285
 286         return;
 287 }
 288 EXPORT_SYMBOL(flow_replace_acts);
 289
 290 /* Prints a representation of 'key' to the kernel log. */
 291 void print_flow(const struct sw_flow_key *key)
 292 {
 293         printk("wild%08x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 294                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 295                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 296                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 297                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 298                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 299                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 300                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 301                         ntohs(key->dl_type),
 302                         ((unsigned char *)&key->nw_src)[0],
 303                         ((unsigned char *)&key->nw_src)[1],
 304                         ((unsigned char *)&key->nw_src)[2],
 305                         ((unsigned char *)&key->nw_src)[3],
 306                         ((unsigned char *)&key->nw_dst)[0],
 307                         ((unsigned char *)&key->nw_dst)[1],
 308                         ((unsigned char *)&key->nw_dst)[2],
 309                         ((unsigned char *)&key->nw_dst)[3],
 310                         ntohs(key->tp_src), ntohs(key->tp_dst));
 311 }
 312 EXPORT_SYMBOL(print_flow);
 313
 314 #define SNAP_OUI_LEN 3
 315
 316 struct eth_snap_hdr
 317 {
 318         struct ethhdr eth;
 319         uint8_t  dsap;  /* Always 0xAA */
 320         uint8_t  ssap;  /* Always 0xAA */
 321         uint8_t  ctrl;
 322         uint8_t  oui[SNAP_OUI_LEN];
 323         uint16_t ethertype;
 324 } __attribute__ ((packed));
 325
 326 static int is_snap(const struct eth_snap_hdr *esh)
 327 {
 328         return (esh->dsap == LLC_SAP_SNAP
 329                 && esh->ssap == LLC_SAP_SNAP
 330                 && !memcmp(esh->oui, "\0\0\0", 3));
 331 }
 332
 333 static int iphdr_ok(struct sk_buff *skb)
 334 {
 335         int nh_ofs = skb_network_offset(skb);
 336         if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
 337                 int ip_len = ip_hdrlen(skb);
 338                 return (ip_len >= sizeof(struct iphdr)
 339                         && pskb_may_pull(skb, nh_ofs + ip_len));
 340         }
 341         return 0;
 342 }
 343
 344 static int tcphdr_ok(struct sk_buff *skb)
 345 {
 346         int th_ofs = skb_transport_offset(skb);
 347         if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
 348                 int tcp_len = tcp_hdrlen(skb);
 349                 return (tcp_len >= sizeof(struct tcphdr)
 350                         && skb->len >= th_ofs + tcp_len);
 351         }
 352         return 0;
 353 }
 354
 355 static int udphdr_ok(struct sk_buff *skb)
 356 {
 357         int th_ofs = skb_transport_offset(skb);
 358         return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
 359 }
 360
 361 static int icmphdr_ok(struct sk_buff *skb)
 362 {
 363         int th_ofs = skb_transport_offset(skb);
 364         return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
 365 }
 366
 367 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 368  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 369  * fragment, 0 otherwise. */
 370 int flow_extract(struct sk_buff *skb, uint16_t in_port,
 371                  struct sw_flow_key *key)
 372 {
 373         struct ethhdr *eth;
 374         struct eth_snap_hdr *esh;
 375         int retval = 0;
 376         int nh_ofs;
 377
 378         memset(key, 0, sizeof *key);
 379         key->dl_vlan = htons(OFP_VLAN_NONE);
 380         key->in_port = htons(in_port);
 381
 382         if (skb->len < sizeof *eth)
 383                 return 0;
 384         if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
 385                 return 0;
 386         }
 387
 388         skb_reset_mac_header(skb);
 389         eth = eth_hdr(skb);
 390         esh = (struct eth_snap_hdr *) eth;
 391         nh_ofs = sizeof *eth;
 392         if (likely(ntohs(eth->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF))
 393                 key->dl_type = eth->h_proto;
 394         else if (skb->len >= sizeof *esh && is_snap(esh)) {
 395                 key->dl_type = esh->ethertype;
 396                 nh_ofs = sizeof *esh;
 397         } else {
 398                 key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
 399                 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
 400                         nh_ofs += sizeof(struct llc_pdu_un);
 401                 }
 402         }
 403
 404         /* Check for a VLAN tag */
 405         if (key->dl_type == htons(ETH_P_8021Q) &&
 406             skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
 407                 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
 408                 key->dl_type = vh->h_vlan_encapsulated_proto;
 409                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 410                 nh_ofs += sizeof(struct vlan_hdr);
 411         }
 412         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 413         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 414         skb_set_network_header(skb, nh_ofs);
 415
 416         /* Network layer. */
 417         if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
 418                 struct iphdr *nh = ip_hdr(skb);
 419                 int th_ofs = nh_ofs + nh->ihl * 4;
 420                 key->nw_src = nh->saddr;
 421                 key->nw_dst = nh->daddr;
 422                 key->nw_proto = nh->protocol;
 423                 skb_set_transport_header(skb, th_ofs);
 424
 425                 /* Transport layer. */
 426                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 427                         if (key->nw_proto == IPPROTO_TCP) {
 428                                 if (tcphdr_ok(skb)) {
 429                                         struct tcphdr *tcp = tcp_hdr(skb);
 430                                         key->tp_src = tcp->source;
 431                                         key->tp_dst = tcp->dest;
 432                                 } else {
 433                                         /* Avoid tricking other code into
 434                                          * thinking that this packet has an L4
 435                                          * header. */
 436                                         key->nw_proto = 0;
 437                                 }
 438                         } else if (key->nw_proto == IPPROTO_UDP) {
 439                                 if (udphdr_ok(skb)) {
 440                                         struct udphdr *udp = udp_hdr(skb);
 441                                         key->tp_src = udp->source;
 442                                         key->tp_dst = udp->dest;
 443                                 } else {
 444                                         /* Avoid tricking other code into
 445                                          * thinking that this packet has an L4
 446                                          * header. */
 447                                         key->nw_proto = 0;
 448                                 }
 449                         } else if (key->nw_proto == IPPROTO_ICMP) {
 450                                 if (icmphdr_ok(skb)) {
 451                                         struct icmphdr *icmp = icmp_hdr(skb);
 452                                         /* The ICMP type and code fields use the 16-bit
 453                                          * transport port fields, so we need to store them
 454                                          * in 16-bit network byte order. */
 455                                         key->icmp_type = htons(icmp->type);
 456                                         key->icmp_code = htons(icmp->code);
 457                                 } else {
 458                                         /* Avoid tricking other code into
 459                                          * thinking that this packet has an L4
 460                                          * header. */
 461                                         key->nw_proto = 0;
 462                                 }
 463                         }
 464                 } else {
 465                         retval = 1;
 466                 }
 467         } else {
 468                 skb_reset_transport_header(skb);
 469         }
 470         return retval;
 471 }
 472
 473 /* Initializes the flow module.
 474  * Returns zero if successful or a negative error code. */
 475 int flow_init(void)
 476 {
 477         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 478                                         0, NULL);
 479         if (flow_cache == NULL)
 480                 return -ENOMEM;
 481
 482         return 0;
 483 }
 484
 485 /* Uninitializes the flow module. */
 486 void flow_exit(void)
 487 {
 488         kmem_cache_destroy(flow_cache);
 489 }
 490