datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include "flow.h"
  10 #include "datapath.h"
  11 #include <linux/netdevice.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/if_ether.h>
  14 #include <linux/if_vlan.h>
  15 #include <net/llc_pdu.h>
  16 #include <linux/kernel.h>
  17 #include <linux/jhash.h>
  18 #include <linux/jiffies.h>
  19 #include <linux/llc.h>
  20 #include <linux/module.h>
  21 #include <linux/in.h>
  22 #include <linux/rcupdate.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/ip.h>
  26 #include <linux/tcp.h>
  27 #include <linux/udp.h>
  28 #include <linux/icmp.h>
  29 #include <net/inet_ecn.h>
  30 #include <net/ip.h>
  31
  32 #include "compat.h"
  33
  34 struct kmem_cache *flow_cache;
  35 static unsigned int hash_seed;
  36
  37 struct arp_eth_header
  38 {
  39         __be16      ar_hrd;     /* format of hardware address   */
  40         __be16      ar_pro;     /* format of protocol address   */
  41         unsigned char   ar_hln; /* length of hardware address   */
  42         unsigned char   ar_pln; /* length of protocol address   */
  43         __be16      ar_op;      /* ARP opcode (command)     */
  44
  45         /* Ethernet+IPv4 specific members. */
  46         unsigned char       ar_sha[ETH_ALEN];   /* sender hardware address  */
  47         unsigned char       ar_sip[4];          /* sender IP address        */
  48         unsigned char       ar_tha[ETH_ALEN];   /* target hardware address  */
  49         unsigned char       ar_tip[4];          /* target IP address        */
  50 } __attribute__((packed));
  51
  52 static inline int arphdr_ok(struct sk_buff *skb)
  53 {
  54         int nh_ofs = skb_network_offset(skb);
  55         return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
  56 }
  57
  58 static inline int iphdr_ok(struct sk_buff *skb)
  59 {
  60         int nh_ofs = skb_network_offset(skb);
  61         if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
  62                 int ip_len = ip_hdrlen(skb);
  63                 return (ip_len >= sizeof(struct iphdr)
  64                         && pskb_may_pull(skb, nh_ofs + ip_len));
  65         }
  66         return 0;
  67 }
  68
  69 static inline int tcphdr_ok(struct sk_buff *skb)
  70 {
  71         int th_ofs = skb_transport_offset(skb);
  72         if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
  73                 int tcp_len = tcp_hdrlen(skb);
  74                 return (tcp_len >= sizeof(struct tcphdr)
  75                         && skb->len >= th_ofs + tcp_len);
  76         }
  77         return 0;
  78 }
  79
  80 static inline int udphdr_ok(struct sk_buff *skb)
  81 {
  82         int th_ofs = skb_transport_offset(skb);
  83         return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
  84 }
  85
  86 static inline int icmphdr_ok(struct sk_buff *skb)
  87 {
  88         int th_ofs = skb_transport_offset(skb);
  89         return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
  90 }
  91
  92 #define TCP_FLAGS_OFFSET 13
  93 #define TCP_FLAG_MASK 0x3f
  94
  95 static inline struct ovs_tcphdr *ovs_tcp_hdr(const struct sk_buff *skb)
  96 {
  97         return (struct ovs_tcphdr *)skb_transport_header(skb);
  98 }
  99
 100 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 101 {
 102         u8 tcp_flags = 0;
 103
 104         if (flow->key.dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
 105                 struct iphdr *nh = ip_hdr(skb);
 106                 flow->ip_tos = nh->tos;
 107                 if (flow->key.nw_proto == IPPROTO_TCP && tcphdr_ok(skb)) {
 108                         u8 *tcp = (u8 *)tcp_hdr(skb);
 109                         tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 110                 }
 111         }
 112
 113         spin_lock_bh(&flow->lock);
 114         getnstimeofday(&flow->used);
 115         flow->packet_count++;
 116         flow->byte_count += skb->len;
 117         flow->tcp_flags |= tcp_flags;
 118         spin_unlock_bh(&flow->lock);
 119 }
 120
 121 struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
 122 {
 123         struct sw_flow_actions *sfa;
 124
 125         if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
 126                 return ERR_PTR(-EINVAL);
 127
 128         sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
 129                       GFP_KERNEL);
 130         if (!sfa)
 131                 return ERR_PTR(-ENOMEM);
 132
 133         sfa->n_actions = n_actions;
 134         return sfa;
 135 }
 136
 137
 138 /* Frees 'flow' immediately. */
 139 static void flow_free(struct sw_flow *flow)
 140 {
 141         if (unlikely(!flow))
 142                 return;
 143         kfree(flow->sf_acts);
 144         kmem_cache_free(flow_cache, flow);
 145 }
 146
 147 void flow_free_tbl(struct tbl_node *node)
 148 {
 149         struct sw_flow *flow = flow_cast(node);
 150         flow_free(flow);
 151 }
 152
 153 /* RCU callback used by flow_deferred_free. */
 154 static void rcu_free_flow_callback(struct rcu_head *rcu)
 155 {
 156         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 157         flow_free(flow);
 158 }
 159
 160 /* Schedules 'flow' to be freed after the next RCU grace period.
 161  * The caller must hold rcu_read_lock for this to be sensible. */
 162 void flow_deferred_free(struct sw_flow *flow)
 163 {
 164         call_rcu(&flow->rcu, rcu_free_flow_callback);
 165 }
 166
 167 /* RCU callback used by flow_deferred_free_acts. */
 168 static void rcu_free_acts_callback(struct rcu_head *rcu)
 169 {
 170         struct sw_flow_actions *sf_acts = container_of(rcu,
 171                         struct sw_flow_actions, rcu);
 172         kfree(sf_acts);
 173 }
 174
 175 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 176  * The caller must hold rcu_read_lock for this to be sensible. */
 177 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 178 {
 179         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 180 }
 181
 182 #define SNAP_OUI_LEN 3
 183
 184 struct eth_snap_hdr
 185 {
 186         struct ethhdr eth;
 187         u8  dsap;  /* Always 0xAA */
 188         u8  ssap;  /* Always 0xAA */
 189         u8  ctrl;
 190         u8  oui[SNAP_OUI_LEN];
 191         u16 ethertype;
 192 } __attribute__ ((packed));
 193
 194 static int is_snap(const struct eth_snap_hdr *esh)
 195 {
 196         return (esh->dsap == LLC_SAP_SNAP
 197                 && esh->ssap == LLC_SAP_SNAP
 198                 && !memcmp(esh->oui, "\0\0\0", 3));
 199 }
 200
 201 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
 202  * and initializes 'key' to match.  Returns 1 if 'skb' contains an IP
 203  * fragment, 0 otherwise. */
 204 int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
 205 {
 206         struct ethhdr *eth;
 207         struct eth_snap_hdr *esh;
 208         int retval = 0;
 209         int nh_ofs;
 210
 211         memset(key, 0, sizeof *key);
 212         key->tun_id = OVS_CB(skb)->tun_id;
 213         key->in_port = in_port;
 214         key->dl_vlan = htons(ODP_VLAN_NONE);
 215
 216         if (skb->len < sizeof *eth)
 217                 return 0;
 218         if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
 219                 return 0;
 220         }
 221
 222         skb_reset_mac_header(skb);
 223         eth = eth_hdr(skb);
 224         esh = (struct eth_snap_hdr *) eth;
 225         nh_ofs = sizeof *eth;
 226         if (likely(ntohs(eth->h_proto) >= ODP_DL_TYPE_ETH2_CUTOFF))
 227                 key->dl_type = eth->h_proto;
 228         else if (skb->len >= sizeof *esh && is_snap(esh)) {
 229                 key->dl_type = esh->ethertype;
 230                 nh_ofs = sizeof *esh;
 231         } else {
 232                 key->dl_type = htons(ODP_DL_TYPE_NOT_ETH_TYPE);
 233                 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
 234                         nh_ofs += sizeof(struct llc_pdu_un);
 235                 }
 236         }
 237
 238         /* Check for a VLAN tag */
 239         if (key->dl_type == htons(ETH_P_8021Q) &&
 240             skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
 241                 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
 242                 key->dl_type = vh->h_vlan_encapsulated_proto;
 243                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
 244                 key->dl_vlan_pcp = (ntohs(vh->h_vlan_TCI) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
 245                 nh_ofs += sizeof(struct vlan_hdr);
 246         }
 247         memcpy(key->dl_src, eth->h_source, ETH_ALEN);
 248         memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
 249         skb_set_network_header(skb, nh_ofs);
 250
 251         /* Network layer. */
 252         if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
 253                 struct iphdr *nh = ip_hdr(skb);
 254                 int th_ofs = nh_ofs + nh->ihl * 4;
 255                 key->nw_src = nh->saddr;
 256                 key->nw_dst = nh->daddr;
 257                 key->nw_tos = nh->tos & ~INET_ECN_MASK;
 258                 key->nw_proto = nh->protocol;
 259                 skb_set_transport_header(skb, th_ofs);
 260
 261                 /* Transport layer. */
 262                 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
 263                         if (key->nw_proto == IPPROTO_TCP) {
 264                                 if (tcphdr_ok(skb)) {
 265                                         struct tcphdr *tcp = tcp_hdr(skb);
 266                                         key->tp_src = tcp->source;
 267                                         key->tp_dst = tcp->dest;
 268                                 } else {
 269                                         /* Avoid tricking other code into
 270                                          * thinking that this packet has an L4
 271                                          * header. */
 272                                         key->nw_proto = 0;
 273                                 }
 274                         } else if (key->nw_proto == IPPROTO_UDP) {
 275                                 if (udphdr_ok(skb)) {
 276                                         struct udphdr *udp = udp_hdr(skb);
 277                                         key->tp_src = udp->source;
 278                                         key->tp_dst = udp->dest;
 279                                 } else {
 280                                         /* Avoid tricking other code into
 281                                          * thinking that this packet has an L4
 282                                          * header. */
 283                                         key->nw_proto = 0;
 284                                 }
 285                         } else if (key->nw_proto == IPPROTO_ICMP) {
 286                                 if (icmphdr_ok(skb)) {
 287                                         struct icmphdr *icmp = icmp_hdr(skb);
 288                                         /* The ICMP type and code fields use the 16-bit
 289                                          * transport port fields, so we need to store them
 290                                          * in 16-bit network byte order. */
 291                                         key->tp_src = htons(icmp->type);
 292                                         key->tp_dst = htons(icmp->code);
 293                                 } else {
 294                                         /* Avoid tricking other code into
 295                                          * thinking that this packet has an L4
 296                                          * header. */
 297                                         key->nw_proto = 0;
 298                                 }
 299                         }
 300                 } else {
 301                         retval = 1;
 302                 }
 303         } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 304                 struct arp_eth_header *arp;
 305
 306                 arp = (struct arp_eth_header *)skb_network_header(skb);
 307
 308                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 309                                 && arp->ar_pro == htons(ETH_P_IP)
 310                                 && arp->ar_hln == ETH_ALEN
 311                                 && arp->ar_pln == 4) {
 312
 313                         /* We only match on the lower 8 bits of the opcode. */
 314                         if (ntohs(arp->ar_op) <= 0xff) {
 315                                 key->nw_proto = ntohs(arp->ar_op);
 316                         }
 317
 318                         if (key->nw_proto == ARPOP_REQUEST
 319                                         || key->nw_proto == ARPOP_REPLY) {
 320                                 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
 321                                 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
 322                         }
 323                 }
 324         } else {
 325                 skb_reset_transport_header(skb);
 326         }
 327         return retval;
 328 }
 329
 330 struct sw_flow *flow_cast(const struct tbl_node *node)
 331 {
 332         return container_of(node, struct sw_flow, tbl_node);
 333 }
 334
 335 u32 flow_hash(const struct odp_flow_key *key)
 336 {
 337         return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
 338 }
 339
 340 int flow_cmp(const struct tbl_node *node, void *key2_)
 341 {
 342         const struct odp_flow_key *key1 = &flow_cast(node)->key;
 343         const struct odp_flow_key *key2 = key2_;
 344
 345         return !memcmp(key1, key2, sizeof(struct odp_flow_key));
 346 }
 347
 348 /* Initializes the flow module.
 349  * Returns zero if successful or a negative error code. */
 350 int flow_init(void)
 351 {
 352         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
 353                                         0, NULL);
 354         if (flow_cache == NULL)
 355                 return -ENOMEM;
 356
 357         get_random_bytes(&hash_seed, sizeof hash_seed);
 358
 359         return 0;
 360 }
 361
 362 /* Uninitializes the flow module. */
 363 void flow_exit(void)
 364 {
 365         kmem_cache_destroy(flow_cache);
 366 }