net/core/flow_dissector.c

   1 #include <linux/skbuff.h>
   2 #include <linux/export.h>
   3 #include <linux/ip.h>
   4 #include <linux/ipv6.h>
   5 #include <linux/if_vlan.h>
   6 #include <net/ip.h>
   7 #include <net/ipv6.h>
   8 #include <linux/if_tunnel.h>
   9 #include <linux/if_pppox.h>
  10 #include <linux/ppp_defs.h>
  11 #include <net/flow_keys.h>
  12
  13 /* copy saddr & daddr, possibly using 64bit load/store
  14  * Equivalent to :      flow->src = iph->saddr;
  15  *                      flow->dst = iph->daddr;
  16  */
  17 static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *iph)
  18 {
  19         BUILD_BUG_ON(offsetof(typeof(*flow), dst) !=
  20                      offsetof(typeof(*flow), src) + sizeof(flow->src));
  21         memcpy(&flow->src, &iph->saddr, sizeof(flow->src) + sizeof(flow->dst));
  22 }
  23
  24 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow)
  25 {
  26         int poff, nhoff = skb_network_offset(skb);
  27         u8 ip_proto;
  28         __be16 proto = skb->protocol;
  29
  30         memset(flow, 0, sizeof(*flow));
  31
  32 again:
  33         switch (proto) {
  34         case __constant_htons(ETH_P_IP): {
  35                 const struct iphdr *iph;
  36                 struct iphdr _iph;
  37 ip:
  38                 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  39                 if (!iph)
  40                         return false;
  41
  42                 if (ip_is_fragment(iph))
  43                         ip_proto = 0;
  44                 else
  45                         ip_proto = iph->protocol;
  46                 iph_to_flow_copy_addrs(flow, iph);
  47                 nhoff += iph->ihl * 4;
  48                 break;
  49         }
  50         case __constant_htons(ETH_P_IPV6): {
  51                 const struct ipv6hdr *iph;
  52                 struct ipv6hdr _iph;
  53 ipv6:
  54                 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  55                 if (!iph)
  56                         return false;
  57
  58                 ip_proto = iph->nexthdr;
  59                 flow->src = (__force __be32)ipv6_addr_hash(&iph->saddr);
  60                 flow->dst = (__force __be32)ipv6_addr_hash(&iph->daddr);
  61                 nhoff += sizeof(struct ipv6hdr);
  62                 break;
  63         }
  64         case __constant_htons(ETH_P_8021Q): {
  65                 const struct vlan_hdr *vlan;
  66                 struct vlan_hdr _vlan;
  67
  68                 vlan = skb_header_pointer(skb, nhoff, sizeof(_vlan), &_vlan);
  69                 if (!vlan)
  70                         return false;
  71
  72                 proto = vlan->h_vlan_encapsulated_proto;
  73                 nhoff += sizeof(*vlan);
  74                 goto again;
  75         }
  76         case __constant_htons(ETH_P_PPP_SES): {
  77                 struct {
  78                         struct pppoe_hdr hdr;
  79                         __be16 proto;
  80                 } *hdr, _hdr;
  81                 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
  82                 if (!hdr)
  83                         return false;
  84                 proto = hdr->proto;
  85                 nhoff += PPPOE_SES_HLEN;
  86                 switch (proto) {
  87                 case __constant_htons(PPP_IP):
  88                         goto ip;
  89                 case __constant_htons(PPP_IPV6):
  90                         goto ipv6;
  91                 default:
  92                         return false;
  93                 }
  94         }
  95         default:
  96                 return false;
  97         }
  98
  99         switch (ip_proto) {
 100         case IPPROTO_GRE: {
 101                 struct gre_hdr {
 102                         __be16 flags;
 103                         __be16 proto;
 104                 } *hdr, _hdr;
 105
 106                 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
 107                 if (!hdr)
 108                         return false;
 109                 /*
 110                  * Only look inside GRE if version zero and no
 111                  * routing
 112                  */
 113                 if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
 114                         proto = hdr->proto;
 115                         nhoff += 4;
 116                         if (hdr->flags & GRE_CSUM)
 117                                 nhoff += 4;
 118                         if (hdr->flags & GRE_KEY)
 119                                 nhoff += 4;
 120                         if (hdr->flags & GRE_SEQ)
 121                                 nhoff += 4;
 122                         if (proto == htons(ETH_P_TEB)) {
 123                                 const struct ethhdr *eth;
 124                                 struct ethhdr _eth;
 125
 126                                 eth = skb_header_pointer(skb, nhoff,
 127                                                          sizeof(_eth), &_eth);
 128                                 if (!eth)
 129                                         return false;
 130                                 proto = eth->h_proto;
 131                                 nhoff += sizeof(*eth);
 132                         }
 133                         goto again;
 134                 }
 135                 break;
 136         }
 137         case IPPROTO_IPIP:
 138                 goto again;
 139         default:
 140                 break;
 141         }
 142
 143         flow->ip_proto = ip_proto;
 144         poff = proto_ports_offset(ip_proto);
 145         if (poff >= 0) {
 146                 __be32 *ports, _ports;
 147
 148                 nhoff += poff;
 149                 ports = skb_header_pointer(skb, nhoff, sizeof(_ports), &_ports);
 150                 if (ports)
 151                         flow->ports = *ports;
 152         }
 153
 154         flow->thoff = (u16) nhoff;
 155
 156         return true;
 157 }
 158 EXPORT_SYMBOL(skb_flow_dissect);
 159
 160 static u32 hashrnd __read_mostly;
 161
 162 /*
 163  * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
 164  * and src/dst port numbers.  Sets rxhash in skb to non-zero hash value
 165  * on success, zero indicates no valid hash.  Also, sets l4_rxhash in skb
 166  * if hash is a canonical 4-tuple hash over transport ports.
 167  */
 168 void __skb_get_rxhash(struct sk_buff *skb)
 169 {
 170         struct flow_keys keys;
 171         u32 hash;
 172
 173         if (!skb_flow_dissect(skb, &keys))
 174                 return;
 175
 176         if (keys.ports)
 177                 skb->l4_rxhash = 1;
 178
 179         /* get a consistent hash (same value on both flow directions) */
 180         if (((__force u32)keys.dst < (__force u32)keys.src) ||
 181             (((__force u32)keys.dst == (__force u32)keys.src) &&
 182              ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
 183                 swap(keys.dst, keys.src);
 184                 swap(keys.port16[0], keys.port16[1]);
 185         }
 186
 187         hash = jhash_3words((__force u32)keys.dst,
 188                             (__force u32)keys.src,
 189                             (__force u32)keys.ports, hashrnd);
 190         if (!hash)
 191                 hash = 1;
 192
 193         skb->rxhash = hash;
 194 }
 195 EXPORT_SYMBOL(__skb_get_rxhash);
 196
 197 /*
 198  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
 199  * to be used as a distribution range.
 200  */
 201 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
 202                   unsigned int num_tx_queues)
 203 {
 204         u32 hash;
 205         u16 qoffset = 0;
 206         u16 qcount = num_tx_queues;
 207
 208         if (skb_rx_queue_recorded(skb)) {
 209                 hash = skb_get_rx_queue(skb);
 210                 while (unlikely(hash >= num_tx_queues))
 211                         hash -= num_tx_queues;
 212                 return hash;
 213         }
 214
 215         if (dev->num_tc) {
 216                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
 217                 qoffset = dev->tc_to_txq[tc].offset;
 218                 qcount = dev->tc_to_txq[tc].count;
 219         }
 220
 221         if (skb->sk && skb->sk->sk_hash)
 222                 hash = skb->sk->sk_hash;
 223         else
 224                 hash = (__force u16) skb->protocol;
 225         hash = jhash_1word(hash, hashrnd);
 226
 227         return (u16) (((u64) hash * qcount) >> 32) + qoffset;
 228 }
 229 EXPORT_SYMBOL(__skb_tx_hash);
 230
 231 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
 232 {
 233         if (unlikely(queue_index >= dev->real_num_tx_queues)) {
 234                 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
 235                                      dev->name, queue_index,
 236                                      dev->real_num_tx_queues);
 237                 return 0;
 238         }
 239         return queue_index;
 240 }
 241
 242 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
 243 {
 244 #ifdef CONFIG_XPS
 245         struct xps_dev_maps *dev_maps;
 246         struct xps_map *map;
 247         int queue_index = -1;
 248
 249         rcu_read_lock();
 250         dev_maps = rcu_dereference(dev->xps_maps);
 251         if (dev_maps) {
 252                 map = rcu_dereference(
 253                     dev_maps->cpu_map[raw_smp_processor_id()]);
 254                 if (map) {
 255                         if (map->len == 1)
 256                                 queue_index = map->queues[0];
 257                         else {
 258                                 u32 hash;
 259                                 if (skb->sk && skb->sk->sk_hash)
 260                                         hash = skb->sk->sk_hash;
 261                                 else
 262                                         hash = (__force u16) skb->protocol ^
 263                                             skb->rxhash;
 264                                 hash = jhash_1word(hash, hashrnd);
 265                                 queue_index = map->queues[
 266                                     ((u64)hash * map->len) >> 32];
 267                         }
 268                         if (unlikely(queue_index >= dev->real_num_tx_queues))
 269                                 queue_index = -1;
 270                 }
 271         }
 272         rcu_read_unlock();
 273
 274         return queue_index;
 275 #else
 276         return -1;
 277 #endif
 278 }
 279
 280 u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
 281 {
 282         struct sock *sk = skb->sk;
 283         int queue_index = sk_tx_queue_get(sk);
 284
 285         if (queue_index < 0 || skb->ooo_okay ||
 286             queue_index >= dev->real_num_tx_queues) {
 287                 int new_index = get_xps_queue(dev, skb);
 288                 if (new_index < 0)
 289                         new_index = skb_tx_hash(dev, skb);
 290
 291                 if (queue_index != new_index && sk) {
 292                         struct dst_entry *dst =
 293                                     rcu_dereference_check(sk->sk_dst_cache, 1);
 294
 295                         if (dst && skb_dst(skb) == dst)
 296                                 sk_tx_queue_set(sk, queue_index);
 297
 298                 }
 299
 300                 queue_index = new_index;
 301         }
 302
 303         return queue_index;
 304 }
 305 EXPORT_SYMBOL(__netdev_pick_tx);
 306
 307 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
 308                                     struct sk_buff *skb)
 309 {
 310         int queue_index = 0;
 311
 312         if (dev->real_num_tx_queues != 1) {
 313                 const struct net_device_ops *ops = dev->netdev_ops;
 314                 if (ops->ndo_select_queue)
 315                         queue_index = ops->ndo_select_queue(dev, skb);
 316                 else
 317                         queue_index = __netdev_pick_tx(dev, skb);
 318                 queue_index = dev_cap_txqueue(dev, queue_index);
 319         }
 320
 321         skb_set_queue_mapping(skb, queue_index);
 322         return netdev_get_tx_queue(dev, queue_index);
 323 }
 324
 325 static int __init initialize_hashrnd(void)
 326 {
 327         get_random_bytes(&hashrnd, sizeof(hashrnd));
 328         return 0;
 329 }
 330
 331 late_initcall_sync(initialize_hashrnd);