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 /* Allocates and returns a new flow with room for 'actions_len' actions,
 171  * using allocation flags 'flags'.  Returns the new flow or a null pointer
 172  * on failure. */
 173 struct sw_flow *flow_alloc(size_t actions_len, gfp_t flags)
 174 {
 175         struct sw_flow_actions *sfa;
 176         size_t size = sizeof *sfa + actions_len;
 177         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 178         if (unlikely(!flow))
 179                 return NULL;
 180
 181         sfa = kmalloc(size, flags);
 182         if (unlikely(!sfa)) {
 183                 kmem_cache_free(flow_cache, flow);
 184                 return NULL;
 185         }
 186         sfa->actions_len = actions_len;
 187         flow->sf_acts = sfa;
 188
 189         return flow;
 190 }
 191
 192 /* Frees 'flow' immediately. */
 193 void flow_free(struct sw_flow *flow)
 194 {
 195         if (unlikely(!flow))
 196                 return;
 197         kfree(flow->sf_acts);
 198         kmem_cache_free(flow_cache, flow);
 199 }
 200 EXPORT_SYMBOL(flow_free);
 201
 202 /* RCU callback used by flow_deferred_free. */
 203 static void rcu_free_flow_callback(struct rcu_head *rcu)
 204 {
 205         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 206         flow_free(flow);
 207 }
 208
 209 /* Schedules 'flow' to be freed after the next RCU grace period.
 210  * The caller must hold rcu_read_lock for this to be sensible. */
 211 void flow_deferred_free(struct sw_flow *flow)
 212 {
 213         call_rcu(&flow->rcu, rcu_free_flow_callback);
 214 }
 215 EXPORT_SYMBOL(flow_deferred_free);
 216
 217 /* RCU callback used by flow_deferred_free_acts. */
 218 static void rcu_free_acts_callback(struct rcu_head *rcu)
 219 {
 220         struct sw_flow_actions *sf_acts = container_of(rcu,
 221                         struct sw_flow_actions, rcu);
 222         kfree(sf_acts);
 223 }
 224
 225 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 226  * The caller must hold rcu_read_lock for this to be sensible. */
 227 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 228 {
 229         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 230 }
 231 EXPORT_SYMBOL(flow_deferred_free_acts);
 232
 233 /* Copies 'actions' into a newly allocated structure for use by 'flow'
 234  * and safely frees the structure that defined the previous actions. */
 235 void flow_replace_acts(struct sw_flow *flow,
 236                 const struct ofp_action_header *actions, size_t actions_len)
 237 {
 238         struct sw_flow_actions *sfa;
 239         struct sw_flow_actions *orig_sfa = flow->sf_acts;
 240         size_t size = sizeof *sfa + actions_len;
 241
 242         sfa = kmalloc(size, GFP_ATOMIC);
 243         if (unlikely(!sfa))
 244                 return;
 245
 246         sfa->actions_len = actions_len;
 247         memcpy(sfa->actions, actions, actions_len);
 248
 249         rcu_assign_pointer(flow->sf_acts, sfa);
 250         flow_deferred_free_acts(orig_sfa);
 251
 252         return;
 253 }
 254 EXPORT_SYMBOL(flow_replace_acts);
 255
 256 /* Prints a representation of 'key' to the kernel log. */
 257 void print_flow(const struct sw_flow_key *key)
 258 {
 259         printk("wild%08x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 260                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 261                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 262                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 263                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 264                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 265                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 266                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 267                         ntohs(key->dl_type),
 268                         ((unsigned char *)&key->nw_src)[0],
 269                         ((unsigned char *)&key->nw_src)[1],
 270                         ((unsigned char *)&key->nw_src)[2],
 271                         ((unsigned char *)&key->nw_src)[3],
 272                         ((unsigned char *)&key->nw_dst)[0],
 273                         ((unsigned char *)&key->nw_dst)[1],
 274                         ((unsigned char *)&key->nw_dst)[2],
 275                         ((unsigned char *)&key->nw_dst)[3],
 276                         ntohs(key->tp_src), ntohs(key->tp_dst));
 277 }
 278 EXPORT_SYMBOL(print_flow);
 279
 280 #define SNAP_OUI_LEN 3
 281
 282 struct eth_snap_hdr
 283 {
 284         struct ethhdr eth;
 285         uint8_t  dsap;  /* Always 0xAA */
 286         uint8_t  ssap;  /* Always 0xAA */
 287         uint8_t  ctrl;
 288         uint8_t  oui[SNAP_OUI_LEN];
 289         uint16_t ethertype;
 290 } __attribute__ ((packed));
 291
 292 static int is_snap(const struct eth_snap_hdr *esh)
 293 {
 294         return (esh->dsap == LLC_SAP_SNAP
 295                 && esh->ssap == LLC_SAP_SNAP
 296                 && !memcmp(esh->oui, "\0\0\0", 3));
 297 }
 298
 299 static int iphdr_ok(struct sk_buff *skb)
 300 {
 301         int nh_ofs = skb_network_offset(skb);
 302         if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
 303                 int ip_len = ip_hdrlen(skb);
 304                 return (ip_len >= sizeof(struct iphdr)
 305                         && pskb_may_pull(skb, nh_ofs + ip_len));
 306         }
 307         return 0;
 308 }
 309
 310 static int tcphdr_ok(struct sk_buff *skb)
 311 {
 312         int th_ofs = skb_transport_offset(skb);
 313         if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
 314                 int tcp_len = tcp_hdrlen(skb);
 315                 return (tcp_len >= sizeof(struct tcphdr)
 316                         && skb->len >= th_ofs + tcp_len);
 317         }
 318         return 0;
 319 }
 320
 321 static int udphdr_ok(struct sk_buff *skb)
 322 {
 323         int th_ofs = skb_transport_offset(skb);
 324         return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
 325 }
 326
 327 static int icmphdr_ok(struct sk_buff *skb)
 328 {
 329         int th_ofs = skb_transport_offset(skb);
 330         return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
 331 }
 332
 333 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 334  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 335  * fragment, 0 otherwise. */
 336 int flow_extract(struct sk_buff *skb, uint16_t in_port,
 337                  struct sw_flow_key *key)
 338 {
 339         struct ethhdr *eth;
 340         struct eth_snap_hdr *esh;
 341         int retval = 0;
 342         int nh_ofs;
 343
 344         memset(key, 0, sizeof *key);
 345         key->dl_vlan = htons(OFP_VLAN_NONE);
 346         key->in_port = htons(in_port);
 347
 348         if (skb->len < sizeof *eth)
 349                 return 0;
 350         if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
 351                 return 0;
 352         }
 353
 354         skb_reset_mac_header(skb);
 355         eth = eth_hdr(skb);
 356         esh = (struct eth_snap_hdr *) eth;
 357         nh_ofs = sizeof *eth;
 358         if (likely(ntohs(eth->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF))
 359                 key->dl_type = eth->h_proto;
 360         else if (skb->len >= sizeof *esh && is_snap(esh)) {
 361                 key->dl_type = esh->ethertype;
 362                 nh_ofs = sizeof *esh;
 363         } else {
 364                 key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
 365                 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
 366                         nh_ofs += sizeof(struct llc_pdu_un);
 367                 }
 368         }
 369
 370         /* Check for a VLAN tag */
 371         if (key->dl_type == htons(ETH_P_8021Q) &&
 372             skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
 373                 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
 374                 key->dl_type = vh->h_vlan_encapsulated_proto;
 375                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 376                 nh_ofs += sizeof(struct vlan_hdr);
 377         }
 378         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 379         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 380         skb_set_network_header(skb, nh_ofs);
 381
 382         /* Network layer. */
 383         if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
 384                 struct iphdr *nh = ip_hdr(skb);
 385                 int th_ofs = nh_ofs + nh->ihl * 4;
 386                 key->nw_src = nh->saddr;
 387                 key->nw_dst = nh->daddr;
 388                 key->nw_proto = nh->protocol;
 389                 skb_set_transport_header(skb, th_ofs);
 390
 391                 /* Transport layer. */
 392                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 393                         if (key->nw_proto == IPPROTO_TCP) {
 394                                 if (tcphdr_ok(skb)) {
 395                                         struct tcphdr *tcp = tcp_hdr(skb);
 396                                         key->tp_src = tcp->source;
 397                                         key->tp_dst = tcp->dest;
 398                                 } else {
 399                                         /* Avoid tricking other code into
 400                                          * thinking that this packet has an L4
 401                                          * header. */
 402                                         key->nw_proto = 0;
 403                                 }
 404                         } else if (key->nw_proto == IPPROTO_UDP) {
 405                                 if (udphdr_ok(skb)) {
 406                                         struct udphdr *udp = udp_hdr(skb);
 407                                         key->tp_src = udp->source;
 408                                         key->tp_dst = udp->dest;
 409                                 } else {
 410                                         /* Avoid tricking other code into
 411                                          * thinking that this packet has an L4
 412                                          * header. */
 413                                         key->nw_proto = 0;
 414                                 }
 415                         } else if (key->nw_proto == IPPROTO_ICMP) {
 416                                 if (icmphdr_ok(skb)) {
 417                                         struct icmphdr *icmp = icmp_hdr(skb);
 418                                         /* The ICMP type and code fields use the 16-bit
 419                                          * transport port fields, so we need to store them
 420                                          * in 16-bit network byte order. */
 421                                         key->icmp_type = htons(icmp->type);
 422                                         key->icmp_code = htons(icmp->code);
 423                                 } else {
 424                                         /* Avoid tricking other code into
 425                                          * thinking that this packet has an L4
 426                                          * header. */
 427                                         key->nw_proto = 0;
 428                                 }
 429                         }
 430                 } else {
 431                         retval = 1;
 432                 }
 433         } else {
 434                 skb_reset_transport_header(skb);
 435         }
 436         return retval;
 437 }
 438
 439 /* Initializes the flow module.
 440  * Returns zero if successful or a negative error code. */
 441 int flow_init(void)
 442 {
 443         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 444                                         0, NULL);
 445         if (flow_cache == NULL)
 446                 return -ENOMEM;
 447
 448         return 0;
 449 }
 450
 451 /* Uninitializes the flow module. */
 452 void flow_exit(void)
 453 {
 454         kmem_cache_destroy(flow_cache);
 455 }
 456