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 match) 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_matches_desc(const struct sw_flow_key *t, const struct sw_flow_key *d,
  74                 int strict)
  75 {
  76         if (strict && d->wildcards != t->wildcards)
  77                 return 0;
  78         return flow_matches_1wild(t, d);
  79 }
  80 EXPORT_SYMBOL(flow_matches_desc);
  81
  82 static uint32_t make_nw_mask(int n_wild_bits)
  83 {
  84         n_wild_bits &= (1u << OFPFW_NW_SRC_BITS) - 1;
  85         return n_wild_bits < 32 ? htonl(~((1u << n_wild_bits) - 1)) : 0;
  86 }
  87
  88 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
  89 {
  90         to->wildcards = ntohl(from->wildcards) & OFPFW_ALL;
  91         to->pad = 0;
  92         to->in_port = from->in_port;
  93         to->dl_vlan = from->dl_vlan;
  94         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
  95         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
  96         to->dl_type = from->dl_type;
  97
  98         to->nw_src = to->nw_dst = to->nw_proto = 0;
  99         to->tp_src = to->tp_dst = 0;
 100
 101 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
 102 #define OFPFW_NW (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO)
 103         if (to->wildcards & OFPFW_DL_TYPE) {
 104                 /* Can't sensibly match on network or transport headers if the
 105                  * data link type is unknown. */
 106                 to->wildcards |= OFPFW_NW | OFPFW_TP;
 107         } else if (from->dl_type == htons(ETH_P_IP)) {
 108                 to->nw_src   = from->nw_src;
 109                 to->nw_dst   = from->nw_dst;
 110                 to->nw_proto = from->nw_proto;
 111
 112                 if (to->wildcards & OFPFW_NW_PROTO) {
 113                         /* Can't sensibly match on transport headers if the
 114                          * network protocol is unknown. */
 115                         to->wildcards |= OFPFW_TP;
 116                 } else if (from->nw_proto == IPPROTO_TCP
 117                            || from->nw_proto == IPPROTO_UDP) {
 118                         to->tp_src = from->tp_src;
 119                         to->tp_dst = from->tp_dst;
 120                 } else {
 121                         /* Transport layer fields are undefined.  Mark them as
 122                          * exact-match to allow such flows to reside in
 123                          * table-hash, instead of falling into table-linear. */
 124                         to->wildcards &= ~OFPFW_TP;
 125                 }
 126         } else {
 127                 /* Network and transport layer fields are undefined.  Mark them
 128                  * as exact-match to allow such flows to reside in table-hash,
 129                  * instead of falling into table-linear. */
 130                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
 131         }
 132
 133         /* We set these late because code above adjusts to->wildcards. */
 134         to->nw_src_mask = make_nw_mask(to->wildcards >> OFPFW_NW_SRC_SHIFT);
 135         to->nw_dst_mask = make_nw_mask(to->wildcards >> OFPFW_NW_DST_SHIFT);
 136 }
 137
 138 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
 139 {
 140         to->wildcards = htonl(from->wildcards);
 141         to->in_port   = from->in_port;
 142         to->dl_vlan   = from->dl_vlan;
 143         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
 144         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
 145         to->dl_type   = from->dl_type;
 146         to->nw_src    = from->nw_src;
 147         to->nw_dst    = from->nw_dst;
 148         to->nw_proto  = from->nw_proto;
 149         to->tp_src    = from->tp_src;
 150         to->tp_dst    = from->tp_dst;
 151         to->pad       = 0;
 152 }
 153
 154 int flow_timeout(struct sw_flow *flow)
 155 {
 156         if (flow->idle_timeout != OFP_FLOW_PERMANENT
 157             && time_after(jiffies, flow->used + flow->idle_timeout * HZ))
 158                 return OFPER_IDLE_TIMEOUT;
 159         else if (flow->hard_timeout != OFP_FLOW_PERMANENT
 160                  && time_after(jiffies,
 161                                flow->init_time + flow->hard_timeout * HZ))
 162                 return OFPER_HARD_TIMEOUT;
 163         else
 164                 return -1;
 165 }
 166 EXPORT_SYMBOL(flow_timeout);
 167
 168 /* Allocates and returns a new flow with 'n_actions' action, using allocation
 169  * flags 'flags'.  Returns the new flow or a null pointer on failure. */
 170 struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
 171 {
 172         struct sw_flow_actions *sfa;
 173         int size = sizeof *sfa + (n_actions * sizeof sfa->actions[0]);
 174         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
 175         if (unlikely(!flow))
 176                 return NULL;
 177
 178         sfa = kmalloc(size, flags);
 179         if (unlikely(!sfa)) {
 180                 kmem_cache_free(flow_cache, flow);
 181                 return NULL;
 182         }
 183         sfa->n_actions = n_actions;
 184         flow->sf_acts = sfa;
 185
 186         return flow;
 187 }
 188
 189 /* Frees 'flow' immediately. */
 190 void flow_free(struct sw_flow *flow)
 191 {
 192         if (unlikely(!flow))
 193                 return;
 194         kfree(flow->sf_acts);
 195         kmem_cache_free(flow_cache, flow);
 196 }
 197 EXPORT_SYMBOL(flow_free);
 198
 199 /* RCU callback used by flow_deferred_free. */
 200 static void rcu_free_flow_callback(struct rcu_head *rcu)
 201 {
 202         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 203         flow_free(flow);
 204 }
 205
 206 /* Schedules 'flow' to be freed after the next RCU grace period.
 207  * The caller must hold rcu_read_lock for this to be sensible. */
 208 void flow_deferred_free(struct sw_flow *flow)
 209 {
 210         call_rcu(&flow->rcu, rcu_free_flow_callback);
 211 }
 212 EXPORT_SYMBOL(flow_deferred_free);
 213
 214 /* RCU callback used by flow_deferred_free_acts. */
 215 static void rcu_free_acts_callback(struct rcu_head *rcu)
 216 {
 217         struct sw_flow_actions *sf_acts = container_of(rcu,
 218                         struct sw_flow_actions, rcu);
 219         kfree(sf_acts);
 220 }
 221
 222 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 223  * The caller must hold rcu_read_lock for this to be sensible. */
 224 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 225 {
 226         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 227 }
 228 EXPORT_SYMBOL(flow_deferred_free_acts);
 229
 230 /* Copies 'actions' into a newly allocated structure for use by 'flow'
 231  * and safely frees the structure that defined the previous actions. */
 232 void flow_replace_acts(struct sw_flow *flow, const struct ofp_action *actions,
 233                         int n_actions)
 234 {
 235         struct sw_flow_actions *sfa;
 236         struct sw_flow_actions *orig_sfa = flow->sf_acts;
 237         int size = sizeof *sfa + (n_actions * sizeof sfa->actions[0]);
 238
 239         sfa = kmalloc(size, GFP_ATOMIC);
 240         if (unlikely(!sfa))
 241                 return;
 242
 243         sfa->n_actions = n_actions;
 244         memcpy(sfa->actions, actions, n_actions * sizeof sfa->actions[0]);
 245
 246         rcu_assign_pointer(flow->sf_acts, sfa);
 247         flow_deferred_free_acts(orig_sfa);
 248
 249         return;
 250 }
 251 EXPORT_SYMBOL(flow_replace_acts);
 252
 253 /* Prints a representation of 'key' to the kernel log. */
 254 void print_flow(const struct sw_flow_key *key)
 255 {
 256         printk("wild%08x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
 257                         "->%02x:%02x:%02x:%02x:%02x:%02x "
 258                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
 259                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
 260                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
 261                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
 262                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
 263                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
 264                         ntohs(key->dl_type),
 265                         ((unsigned char *)&key->nw_src)[0],
 266                         ((unsigned char *)&key->nw_src)[1],
 267                         ((unsigned char *)&key->nw_src)[2],
 268                         ((unsigned char *)&key->nw_src)[3],
 269                         ((unsigned char *)&key->nw_dst)[0],
 270                         ((unsigned char *)&key->nw_dst)[1],
 271                         ((unsigned char *)&key->nw_dst)[2],
 272                         ((unsigned char *)&key->nw_dst)[3],
 273                         ntohs(key->tp_src), ntohs(key->tp_dst));
 274 }
 275 EXPORT_SYMBOL(print_flow);
 276
 277 static int tcphdr_ok(struct sk_buff *skb)
 278 {
 279         int th_ofs = skb_transport_offset(skb);
 280         if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
 281                 int tcp_len = tcp_hdrlen(skb);
 282                 return (tcp_len >= sizeof(struct tcphdr)
 283                         && skb->len >= th_ofs + tcp_len);
 284         }
 285         return 0;
 286 }
 287
 288 static int udphdr_ok(struct sk_buff *skb)
 289 {
 290         int th_ofs = skb_transport_offset(skb);
 291         return skb->len >= th_ofs + sizeof(struct udphdr);
 292 }
 293
 294 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 295  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 296  * fragment, 0 otherwise. */
 297 int flow_extract(struct sk_buff *skb, uint16_t in_port,
 298                  struct sw_flow_key *key)
 299 {
 300         struct ethhdr *mac;
 301         int nh_ofs, th_ofs;
 302         int retval = 0;
 303
 304         key->in_port = htons(in_port);
 305         key->pad = 0;
 306         key->wildcards = 0;
 307         key->nw_src_mask = 0;
 308         key->nw_dst_mask = 0;
 309
 310         /* This code doesn't check that skb->len is long enough to contain the
 311          * MAC or network header.  With a 46-byte minimum length frame this
 312          * assumption is always correct. */
 313
 314         /* Doesn't verify checksums.  Should it? */
 315
 316         /* Data link layer.  We only support Ethernet. */
 317         mac = eth_hdr(skb);
 318         nh_ofs = sizeof(struct ethhdr);
 319         if (likely(ntohs(mac->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF)) {
 320                 /* This is an Ethernet II frame */
 321                 key->dl_type = mac->h_proto;
 322         } else {
 323                 /* This is an 802.2 frame */
 324                 if (snap_get_ethertype(skb, &key->dl_type) != -EINVAL) {
 325                         nh_ofs += sizeof(struct snap_hdr);
 326                 } else {
 327                         key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
 328                         nh_ofs += sizeof(struct llc_pdu_un);
 329                 }
 330         }
 331
 332         /* Check for a VLAN tag */
 333         if (likely(key->dl_type != htons(ETH_P_8021Q))) {
 334                 key->dl_vlan = htons(OFP_VLAN_NONE);
 335         } else {
 336                 struct vlan_hdr *vh = (struct vlan_hdr *)(skb_mac_header(skb) + nh_ofs);
 337                 key->dl_type = vh->h_vlan_encapsulated_proto;
 338                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 339                 nh_ofs += sizeof(*vh);
 340         }
 341         memcpy(key->dl_src, mac->h_source, ETH_ALEN);
 342         memcpy(key->dl_dst, mac->h_dest, ETH_ALEN);
 343         skb_set_network_header(skb, nh_ofs);
 344
 345         /* Network layer. */
 346         if (likely(key->dl_type == htons(ETH_P_IP))) {
 347                 struct iphdr *nh = ip_hdr(skb);
 348                 key->nw_src = nh->saddr;
 349                 key->nw_dst = nh->daddr;
 350                 key->nw_proto = nh->protocol;
 351                 th_ofs = nh_ofs + nh->ihl * 4;
 352                 skb_set_transport_header(skb, th_ofs);
 353
 354                 /* Transport layer. */
 355                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 356                         if (key->nw_proto == IPPROTO_TCP) {
 357                                 if (tcphdr_ok(skb)) {
 358                                         struct tcphdr *tcp = tcp_hdr(skb);
 359                                         key->tp_src = tcp->source;
 360                                         key->tp_dst = tcp->dest;
 361                                 } else {
 362                                         /* Avoid tricking other code into
 363                                          * thinking that this packet has an L4
 364                                          * header. */
 365                                         goto no_proto;
 366                                 }
 367                         } else if (key->nw_proto == IPPROTO_UDP) {
 368                                 if (udphdr_ok(skb)) {
 369                                         struct udphdr *udp = udp_hdr(skb);
 370                                         key->tp_src = udp->source;
 371                                         key->tp_dst = udp->dest;
 372                                 } else {
 373                                         /* Avoid tricking other code into
 374                                          * thinking that this packet has an L4
 375                                          * header. */
 376                                         goto no_proto;
 377                                 }
 378                         } else {
 379                                 goto no_th;
 380                         }
 381                 } else {
 382                         retval = 1;
 383                         goto no_th;
 384                 }
 385
 386                 return 0;
 387         }
 388
 389         key->nw_src = 0;
 390         key->nw_dst = 0;
 391
 392 no_proto:
 393         key->nw_proto = 0;
 394
 395 no_th:
 396         key->tp_src = 0;
 397         key->tp_dst = 0;
 398         return retval;
 399 }
 400
 401 /* Initializes the flow module.
 402  * Returns zero if successful or a negative error code. */
 403 int flow_init(void)
 404 {
 405         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 406                                         0, NULL);
 407         if (flow_cache == NULL)
 408                 return -ENOMEM;
 409
 410         return 0;
 411 }
 412
 413 /* Uninitializes the flow module. */
 414 void flow_exit(void)
 415 {
 416         kmem_cache_destroy(flow_cache);
 417 }
 418