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/igmp.h>
   9 #include <linux/icmp.h>
  10 #include <linux/sctp.h>
  11 #include <linux/dccp.h>
  12 #include <linux/if_tunnel.h>
  13 #include <linux/if_pppox.h>
  14 #include <linux/ppp_defs.h>
  15 #include <net/flow_keys.h>
  16
  17 /* copy saddr & daddr, possibly using 64bit load/store
  18  * Equivalent to :      flow->src = iph->saddr;
  19  *                      flow->dst = iph->daddr;
  20  */
  21 static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *iph)
  22 {
  23         BUILD_BUG_ON(offsetof(typeof(*flow), dst) !=
  24                      offsetof(typeof(*flow), src) + sizeof(flow->src));
  25         memcpy(&flow->src, &iph->saddr, sizeof(flow->src) + sizeof(flow->dst));
  26 }
  27
  28 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow)
  29 {
  30         int poff, nhoff = skb_network_offset(skb);
  31         u8 ip_proto;
  32         __be16 proto = skb->protocol;
  33
  34         memset(flow, 0, sizeof(*flow));
  35
  36 again:
  37         switch (proto) {
  38         case __constant_htons(ETH_P_IP): {
  39                 const struct iphdr *iph;
  40                 struct iphdr _iph;
  41 ip:
  42                 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  43                 if (!iph || iph->ihl < 5)
  44                         return false;
  45
  46                 if (ip_is_fragment(iph))
  47                         ip_proto = 0;
  48                 else
  49                         ip_proto = iph->protocol;
  50                 iph_to_flow_copy_addrs(flow, iph);
  51                 nhoff += iph->ihl * 4;
  52                 break;
  53         }
  54         case __constant_htons(ETH_P_IPV6): {
  55                 const struct ipv6hdr *iph;
  56                 struct ipv6hdr _iph;
  57 ipv6:
  58                 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  59                 if (!iph)
  60                         return false;
  61
  62                 ip_proto = iph->nexthdr;
  63                 flow->src = (__force __be32)ipv6_addr_hash(&iph->saddr);
  64                 flow->dst = (__force __be32)ipv6_addr_hash(&iph->daddr);
  65                 nhoff += sizeof(struct ipv6hdr);
  66                 break;
  67         }
  68         case __constant_htons(ETH_P_8021Q): {
  69                 const struct vlan_hdr *vlan;
  70                 struct vlan_hdr _vlan;
  71
  72                 vlan = skb_header_pointer(skb, nhoff, sizeof(_vlan), &_vlan);
  73                 if (!vlan)
  74                         return false;
  75
  76                 proto = vlan->h_vlan_encapsulated_proto;
  77                 nhoff += sizeof(*vlan);
  78                 goto again;
  79         }
  80         case __constant_htons(ETH_P_PPP_SES): {
  81                 struct {
  82                         struct pppoe_hdr hdr;
  83                         __be16 proto;
  84                 } *hdr, _hdr;
  85                 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
  86                 if (!hdr)
  87                         return false;
  88                 proto = hdr->proto;
  89                 nhoff += PPPOE_SES_HLEN;
  90                 switch (proto) {
  91                 case __constant_htons(PPP_IP):
  92                         goto ip;
  93                 case __constant_htons(PPP_IPV6):
  94                         goto ipv6;
  95                 default:
  96                         return false;
  97                 }
  98         }
  99         default:
 100                 return false;
 101         }
 102
 103         switch (ip_proto) {
 104         case IPPROTO_GRE: {
 105                 struct gre_hdr {
 106                         __be16 flags;
 107                         __be16 proto;
 108                 } *hdr, _hdr;
 109
 110                 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
 111                 if (!hdr)
 112                         return false;
 113                 /*
 114                  * Only look inside GRE if version zero and no
 115                  * routing
 116                  */
 117                 if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
 118                         proto = hdr->proto;
 119                         nhoff += 4;
 120                         if (hdr->flags & GRE_CSUM)
 121                                 nhoff += 4;
 122                         if (hdr->flags & GRE_KEY)
 123                                 nhoff += 4;
 124                         if (hdr->flags & GRE_SEQ)
 125                                 nhoff += 4;
 126                         if (proto == htons(ETH_P_TEB)) {
 127                                 const struct ethhdr *eth;
 128                                 struct ethhdr _eth;
 129
 130                                 eth = skb_header_pointer(skb, nhoff,
 131                                                          sizeof(_eth), &_eth);
 132                                 if (!eth)
 133                                         return false;
 134                                 proto = eth->h_proto;
 135                                 nhoff += sizeof(*eth);
 136                         }
 137                         goto again;
 138                 }
 139                 break;
 140         }
 141         case IPPROTO_IPIP:
 142                 goto again;
 143         default:
 144                 break;
 145         }
 146
 147         flow->ip_proto = ip_proto;
 148         poff = proto_ports_offset(ip_proto);
 149         if (poff >= 0) {
 150                 __be32 *ports, _ports;
 151
 152                 ports = skb_header_pointer(skb, nhoff + poff,
 153                                            sizeof(_ports), &_ports);
 154                 if (ports)
 155                         flow->ports = *ports;
 156         }
 157
 158         flow->thoff = (u16) nhoff;
 159
 160         return true;
 161 }
 162 EXPORT_SYMBOL(skb_flow_dissect);
 163
 164 static u32 hashrnd __read_mostly;
 165
 166 /*
 167  * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
 168  * and src/dst port numbers.  Sets rxhash in skb to non-zero hash value
 169  * on success, zero indicates no valid hash.  Also, sets l4_rxhash in skb
 170  * if hash is a canonical 4-tuple hash over transport ports.
 171  */
 172 void __skb_get_rxhash(struct sk_buff *skb)
 173 {
 174         struct flow_keys keys;
 175         u32 hash;
 176
 177         if (!skb_flow_dissect(skb, &keys))
 178                 return;
 179
 180         if (keys.ports)
 181                 skb->l4_rxhash = 1;
 182
 183         /* get a consistent hash (same value on both flow directions) */
 184         if (((__force u32)keys.dst < (__force u32)keys.src) ||
 185             (((__force u32)keys.dst == (__force u32)keys.src) &&
 186              ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
 187                 swap(keys.dst, keys.src);
 188                 swap(keys.port16[0], keys.port16[1]);
 189         }
 190
 191         hash = jhash_3words((__force u32)keys.dst,
 192                             (__force u32)keys.src,
 193                             (__force u32)keys.ports, hashrnd);
 194         if (!hash)
 195                 hash = 1;
 196
 197         skb->rxhash = hash;
 198 }
 199 EXPORT_SYMBOL(__skb_get_rxhash);
 200
 201 /*
 202  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
 203  * to be used as a distribution range.
 204  */
 205 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
 206                   unsigned int num_tx_queues)
 207 {
 208         u32 hash;
 209         u16 qoffset = 0;
 210         u16 qcount = num_tx_queues;
 211
 212         if (skb_rx_queue_recorded(skb)) {
 213                 hash = skb_get_rx_queue(skb);
 214                 while (unlikely(hash >= num_tx_queues))
 215                         hash -= num_tx_queues;
 216                 return hash;
 217         }
 218
 219         if (dev->num_tc) {
 220                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
 221                 qoffset = dev->tc_to_txq[tc].offset;
 222                 qcount = dev->tc_to_txq[tc].count;
 223         }
 224
 225         if (skb->sk && skb->sk->sk_hash)
 226                 hash = skb->sk->sk_hash;
 227         else
 228                 hash = (__force u16) skb->protocol;
 229         hash = jhash_1word(hash, hashrnd);
 230
 231         return (u16) (((u64) hash * qcount) >> 32) + qoffset;
 232 }
 233 EXPORT_SYMBOL(__skb_tx_hash);
 234
 235 /* __skb_get_poff() returns the offset to the payload as far as it could
 236  * be dissected. The main user is currently BPF, so that we can dynamically
 237  * truncate packets without needing to push actual payload to the user
 238  * space and can analyze headers only, instead.
 239  */
 240 u32 __skb_get_poff(const struct sk_buff *skb)
 241 {
 242         struct flow_keys keys;
 243         u32 poff = 0;
 244
 245         if (!skb_flow_dissect(skb, &keys))
 246                 return 0;
 247
 248         poff += keys.thoff;
 249         switch (keys.ip_proto) {
 250         case IPPROTO_TCP: {
 251                 const struct tcphdr *tcph;
 252                 struct tcphdr _tcph;
 253
 254                 tcph = skb_header_pointer(skb, poff, sizeof(_tcph), &_tcph);
 255                 if (!tcph)
 256                         return poff;
 257
 258                 poff += max_t(u32, sizeof(struct tcphdr), tcph->doff * 4);
 259                 break;
 260         }
 261         case IPPROTO_UDP:
 262         case IPPROTO_UDPLITE:
 263                 poff += sizeof(struct udphdr);
 264                 break;
 265         /* For the rest, we do not really care about header
 266          * extensions at this point for now.
 267          */
 268         case IPPROTO_ICMP:
 269                 poff += sizeof(struct icmphdr);
 270                 break;
 271         case IPPROTO_ICMPV6:
 272                 poff += sizeof(struct icmp6hdr);
 273                 break;
 274         case IPPROTO_IGMP:
 275                 poff += sizeof(struct igmphdr);
 276                 break;
 277         case IPPROTO_DCCP:
 278                 poff += sizeof(struct dccp_hdr);
 279                 break;
 280         case IPPROTO_SCTP:
 281                 poff += sizeof(struct sctphdr);
 282                 break;
 283         }
 284
 285         return poff;
 286 }
 287
 288 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
 289 {
 290         if (unlikely(queue_index >= dev->real_num_tx_queues)) {
 291                 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
 292                                      dev->name, queue_index,
 293                                      dev->real_num_tx_queues);
 294                 return 0;
 295         }
 296         return queue_index;
 297 }
 298
 299 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
 300 {
 301 #ifdef CONFIG_XPS
 302         struct xps_dev_maps *dev_maps;
 303         struct xps_map *map;
 304         int queue_index = -1;
 305
 306         rcu_read_lock();
 307         dev_maps = rcu_dereference(dev->xps_maps);
 308         if (dev_maps) {
 309                 map = rcu_dereference(
 310                     dev_maps->cpu_map[raw_smp_processor_id()]);
 311                 if (map) {
 312                         if (map->len == 1)
 313                                 queue_index = map->queues[0];
 314                         else {
 315                                 u32 hash;
 316                                 if (skb->sk && skb->sk->sk_hash)
 317                                         hash = skb->sk->sk_hash;
 318                                 else
 319                                         hash = (__force u16) skb->protocol ^
 320                                             skb->rxhash;
 321                                 hash = jhash_1word(hash, hashrnd);
 322                                 queue_index = map->queues[
 323                                     ((u64)hash * map->len) >> 32];
 324                         }
 325                         if (unlikely(queue_index >= dev->real_num_tx_queues))
 326                                 queue_index = -1;
 327                 }
 328         }
 329         rcu_read_unlock();
 330
 331         return queue_index;
 332 #else
 333         return -1;
 334 #endif
 335 }
 336
 337 u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
 338 {
 339         struct sock *sk = skb->sk;
 340         int queue_index = sk_tx_queue_get(sk);
 341
 342         if (queue_index < 0 || skb->ooo_okay ||
 343             queue_index >= dev->real_num_tx_queues) {
 344                 int new_index = get_xps_queue(dev, skb);
 345                 if (new_index < 0)
 346                         new_index = skb_tx_hash(dev, skb);
 347
 348                 if (queue_index != new_index && sk &&
 349                     rcu_access_pointer(sk->sk_dst_cache))
 350                         sk_tx_queue_set(sk, new_index);
 351
 352                 queue_index = new_index;
 353         }
 354
 355         return queue_index;
 356 }
 357 EXPORT_SYMBOL(__netdev_pick_tx);
 358
 359 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
 360                                     struct sk_buff *skb)
 361 {
 362         int queue_index = 0;
 363
 364         if (dev->real_num_tx_queues != 1) {
 365                 const struct net_device_ops *ops = dev->netdev_ops;
 366                 if (ops->ndo_select_queue)
 367                         queue_index = ops->ndo_select_queue(dev, skb);
 368                 else
 369                         queue_index = __netdev_pick_tx(dev, skb);
 370                 queue_index = dev_cap_txqueue(dev, queue_index);
 371         }
 372
 373         skb_set_queue_mapping(skb, queue_index);
 374         return netdev_get_tx_queue(dev, queue_index);
 375 }
 376
 377 static int __init initialize_hashrnd(void)
 378 {
 379         get_random_bytes(&hashrnd, sizeof(hashrnd));
 380         return 0;
 381 }
 382
 383 late_initcall_sync(initialize_hashrnd);