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/jiffies.h>
  14 #include <linux/kernel.h>
  15 #include <linux/llc.h>
  16 #include <linux/module.h>
  17 #include <linux/in.h>
  18 #include <linux/rcupdate.h>
  19
  20 #include "openflow/openflow.h"
  21 #include "openflow/nicira-ext.h"
  22 #include "compat.h"
  23
  24 struct kmem_cache *flow_cache;
  25
  26 /* Internal function used to compare fields in flow. */
  27 static inline
  28 int flow_fields_match(const struct sw_flow_key *a, const struct sw_flow_key *b,
  29                       uint32_t w, uint32_t src_mask, uint32_t dst_mask)
  30 {
  31         return ((w & OFPFW_IN_PORT || a->in_port == b->in_port)
  32                 && (w & OFPFW_DL_VLAN || a->dl_vlan == b->dl_vlan)
  33                 && (w & OFPFW_DL_SRC || !memcmp(a->dl_src, b->dl_src, ETH_ALEN))
  34                 && (w & OFPFW_DL_DST || !memcmp(a->dl_dst, b->dl_dst, ETH_ALEN))
  35                 && (w & OFPFW_DL_TYPE || a->dl_type == b->dl_type)
  36                 && !((a->nw_src ^ b->nw_src) & src_mask)
  37                 && !((a->nw_dst ^ b->nw_dst) & dst_mask)
  38                 && (w & OFPFW_NW_PROTO || a->nw_proto == b->nw_proto)
  39                 && (w & OFPFW_TP_SRC || a->tp_src == b->tp_src)
  40                 && (w & OFPFW_TP_DST || a->tp_dst == b->tp_dst));
  41 }
  42
  43 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  44  * modulo wildcards in 'b', zero otherwise. */
  45 int flow_matches_1wild(const struct sw_flow_key *a,
  46                        const struct sw_flow_key *b)
  47 {
  48         return flow_fields_match(a, b, b->wildcards,
  49                                  b->nw_src_mask, b->nw_dst_mask);
  50 }
  51 EXPORT_SYMBOL(flow_matches_1wild);
  52
  53 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
  54  * modulo wildcards in 'a' or 'b', zero otherwise. */
  55 int flow_matches_2wild(const struct sw_flow_key *a,
  56                        const struct sw_flow_key *b)
  57 {
  58         return flow_fields_match(a, b,
  59                                  a->wildcards | b->wildcards,
  60                                  a->nw_src_mask & b->nw_src_mask,
  61                                  a->nw_dst_mask & b->nw_dst_mask);
  62 }
  63 EXPORT_SYMBOL(flow_matches_2wild);
  64
  65 /* Returns nonzero if 't' (the table entry's key) and 'd' (the key
  66  * describing the match) match, that is, if their fields are
  67  * equal modulo wildcards, zero otherwise.  If 'strict' is nonzero, the
  68  * wildcards must match in both 't_key' and 'd_key'.  Note that the
  69  * table's wildcards are ignored unless 'strict' is set. */
  70 int flow_matches_desc(const struct sw_flow_key *t, const struct sw_flow_key *d,
  71                 int strict)
  72 {
  73         if (strict && d->wildcards != t->wildcards)
  74                 return 0;
  75         return flow_matches_1wild(t, d);
  76 }
  77 EXPORT_SYMBOL(flow_matches_desc);
  78
  79 static uint32_t make_nw_mask(int n_wild_bits)
  80 {
  81         n_wild_bits &= (1u << OFPFW_NW_SRC_BITS) - 1;
  82         return n_wild_bits < 32 ? htonl(~((1u << n_wild_bits) - 1)) : 0;
  83 }
  84
  85 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
  86 {
  87         to->wildcards = ntohl(from->wildcards) & OFPFW_ALL;
  88         to->pad = 0;
  89         to->in_port = from->in_port;
  90         to->dl_vlan = from->dl_vlan;
  91         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
  92         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
  93         to->dl_type = from->dl_type;
  94
  95         to->nw_src = to->nw_dst = to->nw_proto = 0;
  96         to->tp_src = to->tp_dst = 0;
  97
  98 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
  99 #define OFPFW_NW (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO)
 100         if (to->wildcards & OFPFW_DL_TYPE) {
 101                 /* Can't sensibly match on network or transport headers if the
 102                  * data link type is unknown. */
 103                 to->wildcards |= OFPFW_NW | OFPFW_TP;
 104         } else if (from->dl_type == htons(ETH_P_IP)) {
 105                 to->nw_src   = from->nw_src;
 106                 to->nw_dst   = from->nw_dst;
 107                 to->nw_proto = from->nw_proto;
 108
 109                 if (to->wildcards & OFPFW_NW_PROTO) {
 110                         /* Can't sensibly match on transport headers if the
 111                          * network protocol is unknown. */
 112                         to->wildcards |= OFPFW_TP;
 113                 } else if (from->nw_proto == IPPROTO_TCP
 114                                 || from->nw_proto == IPPROTO_UDP
 115                                 || from->nw_proto == IPPROTO_ICMP) {
 116                         to->tp_src = from->tp_src;
 117                         to->tp_dst = from->tp_dst;
 118                 } else {
 119                         /* Transport layer fields are undefined.  Mark them as
 120                          * exact-match to allow such flows to reside in
 121                          * table-hash, instead of falling into table-linear. */
 122                         to->wildcards &= ~OFPFW_TP;
 123                 }
 124         } else {
 125                 /* Network and transport layer fields are undefined.  Mark them
 126                  * as exact-match to allow such flows to reside in table-hash,
 127                  * instead of falling into table-linear. */
 128                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
 129         }
 130
 131         /* We set these late because code above adjusts to->wildcards. */
 132         to->nw_src_mask = make_nw_mask(to->wildcards >> OFPFW_NW_SRC_SHIFT);
 133         to->nw_dst_mask = make_nw_mask(to->wildcards >> OFPFW_NW_DST_SHIFT);
 134 }
 135
 136 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
 137 {
 138         to->wildcards = htonl(from->wildcards);
 139         to->in_port   = from->in_port;
 140         to->dl_vlan   = from->dl_vlan;
 141         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
 142         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
 143         to->dl_type   = from->dl_type;
 144         to->nw_src    = from->nw_src;
 145         to->nw_dst    = from->nw_dst;
 146         to->nw_proto  = from->nw_proto;
 147         to->tp_src    = from->tp_src;
 148         to->tp_dst    = from->tp_dst;
 149         to->pad       = 0;
 150 }
 151
 152 int flow_timeout(struct sw_flow *flow)
 153 {
 154         if (flow->idle_timeout != OFP_FLOW_PERMANENT
 155             && time_after64(get_jiffies_64(), flow->used + flow->idle_timeout * HZ))
 156                 return NXFER_IDLE_TIMEOUT;
 157         else if (flow->hard_timeout != OFP_FLOW_PERMANENT
 158                  && time_after64(get_jiffies_64(),
 159                                flow->created + flow->hard_timeout * HZ))
 160                 return NXFER_HARD_TIMEOUT;
 161         else
 162                 return -1;
 163 }
 164 EXPORT_SYMBOL(flow_timeout);
 165
 166 /* Returns nonzero if 'flow' contains an output action to 'out_port' or
 167  * has the value OFPP_NONE. 'out_port' is in network-byte order. */
 168 int flow_has_out_port(struct sw_flow *flow, uint16_t out_port)
 169 {
 170         struct sw_flow_actions *sf_acts;
 171         size_t actions_len;
 172         uint8_t *p;
 173
 174         if (out_port == htons(OFPP_NONE))
 175                 return 1;
 176
 177         sf_acts = rcu_dereference(flow->sf_acts);
 178
 179         actions_len = sf_acts->actions_len;
 180         p = (uint8_t *)sf_acts->actions;
 181
 182         while (actions_len > 0) {
 183                 struct ofp_action_header *ah = (struct ofp_action_header *)p;
 184                 size_t len = ntohs(ah->len);
 185
 186                 if (ah->type == htons(OFPAT_OUTPUT)) {
 187                         struct ofp_action_output *oa = (struct ofp_action_output *)p;
 188                         if (oa->port == out_port)
 189                                 return 1;
 190                 }
 191
 192                 p += len;
 193                 actions_len -= len;
 194         }
 195
 196         return 0;
 197 }
 198 EXPORT_SYMBOL(flow_has_out_port);
 199
 200 /* Allocates and returns a new flow with room for 'actions_len' actions,
 201  * using allocation flags 'flags'.  Returns the new flow or a null pointer
 202  * on failure. */
 203 struct sw_flow *flow_alloc(size_t actions_len, gfp_t flags)
 204 {
 205         struct sw_flow_actions *sfa;
 206         size_t size = sizeof *sfa + actions_len;
 207         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 208         if (unlikely(!flow))
 209                 return NULL;
 210
 211         sfa = kmalloc(size, flags);
 212         if (unlikely(!sfa)) {
 213                 kmem_cache_free(flow_cache, flow);
 214                 return NULL;
 215         }
 216         sfa->actions_len = actions_len;
 217         flow->sf_acts = sfa;
 218
 219         return flow;
 220 }
 221
 222 /* Frees 'flow' immediately. */
 223 void flow_free(struct sw_flow *flow)
 224 {
 225         if (unlikely(!flow))
 226                 return;
 227         kfree(flow->sf_acts);
 228         kmem_cache_free(flow_cache, flow);
 229 }
 230 EXPORT_SYMBOL(flow_free);
 231
 232 /* RCU callback used by flow_deferred_free. */
 233 static void rcu_free_flow_callback(struct rcu_head *rcu)
 234 {
 235         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 236         flow_free(flow);
 237 }
 238
 239 /* Schedules 'flow' to be freed after the next RCU grace period.
 240  * The caller must hold rcu_read_lock for this to be sensible. */
 241 void flow_deferred_free(struct sw_flow *flow)
 242 {
 243         call_rcu(&flow->rcu, rcu_free_flow_callback);
 244 }
 245 EXPORT_SYMBOL(flow_deferred_free);
 246
 247 /* RCU callback used by flow_deferred_free_acts. */
 248 static void rcu_free_acts_callback(struct rcu_head *rcu)
 249 {
 250         struct sw_flow_actions *sf_acts = container_of(rcu,
 251                         struct sw_flow_actions, rcu);
 252         kfree(sf_acts);
 253 }
 254
 255 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 256  * The caller must hold rcu_read_lock for this to be sensible. */
 257 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 258 {
 259         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 260 }
 261 EXPORT_SYMBOL(flow_deferred_free_acts);
 262
 263 /* Copies 'actions' into a newly allocated structure for use by 'flow'
 264  * and safely frees the structure that defined the previous actions. */
 265 void flow_replace_acts(struct sw_flow *flow,
 266                 const struct ofp_action_header *actions, size_t actions_len)
 267 {
 268         struct sw_flow_actions *sfa;
 269         struct sw_flow_actions *orig_sfa = flow->sf_acts;
 270         size_t size = sizeof *sfa + actions_len;
 271
 272         sfa = kmalloc(size, GFP_ATOMIC);
 273         if (unlikely(!sfa))
 274                 return;
 275
 276         sfa->actions_len = actions_len;
 277         memcpy(sfa->actions, actions, actions_len);
 278
 279         rcu_assign_pointer(flow->sf_acts, sfa);
 280         flow_deferred_free_acts(orig_sfa);
 281
 282         return;
 283 }
 284 EXPORT_SYMBOL(flow_replace_acts);
 285
 286 /* Prints a representation of 'key' to the kernel log. */
 287 void print_flow(const struct sw_flow_key *key)
 288 {
 289         printk("wild%08x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 290                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 291                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 292                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 293                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 294                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 295                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 296                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 297                         ntohs(key->dl_type),
 298                         ((unsigned char *)&key->nw_src)[0],
 299                         ((unsigned char *)&key->nw_src)[1],
 300                         ((unsigned char *)&key->nw_src)[2],
 301                         ((unsigned char *)&key->nw_src)[3],
 302                         ((unsigned char *)&key->nw_dst)[0],
 303                         ((unsigned char *)&key->nw_dst)[1],
 304                         ((unsigned char *)&key->nw_dst)[2],
 305                         ((unsigned char *)&key->nw_dst)[3],
 306                         ntohs(key->tp_src), ntohs(key->tp_dst));
 307 }
 308 EXPORT_SYMBOL(print_flow);
 309
 310 #define SNAP_OUI_LEN 3
 311
 312 struct eth_snap_hdr
 313 {
 314         struct ethhdr eth;
 315         uint8_t  dsap;  /* Always 0xAA */
 316         uint8_t  ssap;  /* Always 0xAA */
 317         uint8_t  ctrl;
 318         uint8_t  oui[SNAP_OUI_LEN];
 319         uint16_t ethertype;
 320 } __attribute__ ((packed));
 321
 322 static int is_snap(const struct eth_snap_hdr *esh)
 323 {
 324         return (esh->dsap == LLC_SAP_SNAP
 325                 && esh->ssap == LLC_SAP_SNAP
 326                 && !memcmp(esh->oui, "\0\0\0", 3));
 327 }
 328
 329 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 330  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 331  * fragment, 0 otherwise. */
 332 int flow_extract(struct sk_buff *skb, uint16_t in_port,
 333                  struct sw_flow_key *key)
 334 {
 335         struct ethhdr *eth;
 336         struct eth_snap_hdr *esh;
 337         int retval = 0;
 338         int nh_ofs;
 339
 340         memset(key, 0, sizeof *key);
 341         key->dl_vlan = htons(OFP_VLAN_NONE);
 342         key->in_port = htons(in_port);
 343
 344         if (skb->len < sizeof *eth)
 345                 return 0;
 346         if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
 347                 return 0;
 348         }
 349
 350         skb_reset_mac_header(skb);
 351         eth = eth_hdr(skb);
 352         esh = (struct eth_snap_hdr *) eth;
 353         nh_ofs = sizeof *eth;
 354         if (likely(ntohs(eth->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF))
 355                 key->dl_type = eth->h_proto;
 356         else if (skb->len >= sizeof *esh && is_snap(esh)) {
 357                 key->dl_type = esh->ethertype;
 358                 nh_ofs = sizeof *esh;
 359         } else {
 360                 key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
 361                 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
 362                         nh_ofs += sizeof(struct llc_pdu_un);
 363                 }
 364         }
 365
 366         /* Check for a VLAN tag */
 367         if (key->dl_type == htons(ETH_P_8021Q) &&
 368             skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
 369                 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
 370                 key->dl_type = vh->h_vlan_encapsulated_proto;
 371                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 372                 nh_ofs += sizeof(struct vlan_hdr);
 373         }
 374         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 375         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 376         skb_set_network_header(skb, nh_ofs);
 377
 378         /* Network layer. */
 379         if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
 380                 struct iphdr *nh = ip_hdr(skb);
 381                 int th_ofs = nh_ofs + nh->ihl * 4;
 382                 key->nw_src = nh->saddr;
 383                 key->nw_dst = nh->daddr;
 384                 key->nw_proto = nh->protocol;
 385                 skb_set_transport_header(skb, th_ofs);
 386
 387                 /* Transport layer. */
 388                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 389                         if (key->nw_proto == IPPROTO_TCP) {
 390                                 if (tcphdr_ok(skb)) {
 391                                         struct tcphdr *tcp = tcp_hdr(skb);
 392                                         key->tp_src = tcp->source;
 393                                         key->tp_dst = tcp->dest;
 394                                 } else {
 395                                         /* Avoid tricking other code into
 396                                          * thinking that this packet has an L4
 397                                          * header. */
 398                                         key->nw_proto = 0;
 399                                 }
 400                         } else if (key->nw_proto == IPPROTO_UDP) {
 401                                 if (udphdr_ok(skb)) {
 402                                         struct udphdr *udp = udp_hdr(skb);
 403                                         key->tp_src = udp->source;
 404                                         key->tp_dst = udp->dest;
 405                                 } else {
 406                                         /* Avoid tricking other code into
 407                                          * thinking that this packet has an L4
 408                                          * header. */
 409                                         key->nw_proto = 0;
 410                                 }
 411                         } else if (key->nw_proto == IPPROTO_ICMP) {
 412                                 if (icmphdr_ok(skb)) {
 413                                         struct icmphdr *icmp = icmp_hdr(skb);
 414                                         /* The ICMP type and code fields use the 16-bit
 415                                          * transport port fields, so we need to store them
 416                                          * in 16-bit network byte order. */
 417                                         key->icmp_type = htons(icmp->type);
 418                                         key->icmp_code = htons(icmp->code);
 419                                 } else {
 420                                         /* Avoid tricking other code into
 421                                          * thinking that this packet has an L4
 422                                          * header. */
 423                                         key->nw_proto = 0;
 424                                 }
 425                         }
 426                 } else {
 427                         retval = 1;
 428                 }
 429         } else {
 430                 skb_reset_transport_header(skb);
 431         }
 432         return retval;
 433 }
 434
 435 /* Initializes the flow module.
 436  * Returns zero if successful or a negative error code. */
 437 int flow_init(void)
 438 {
 439         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 440                                         0, NULL);
 441         if (flow_cache == NULL)
 442                 return -ENOMEM;
 443
 444         return 0;
 445 }
 446
 447 /* Uninitializes the flow module. */
 448 void flow_exit(void)
 449 {
 450         kmem_cache_destroy(flow_cache);
 451 }
 452