1 #include <linux/skbuff.h>
2 #include <linux/export.h>
4 #include <linux/ipv6.h>
5 #include <linux/if_vlan.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>
17 /* copy saddr & daddr, possibly using 64bit load/store
18 * Equivalent to : flow->src = iph->saddr;
19 * flow->dst = iph->daddr;
21 static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *iph)
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));
28 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow)
30 int poff, nhoff = skb_network_offset(skb);
32 __be16 proto = skb->protocol;
34 memset(flow, 0, sizeof(*flow));
38 case __constant_htons(ETH_P_IP): {
39 const struct iphdr *iph;
42 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
43 if (!iph || iph->ihl < 5)
46 if (ip_is_fragment(iph))
49 ip_proto = iph->protocol;
50 iph_to_flow_copy_addrs(flow, iph);
51 nhoff += iph->ihl * 4;
54 case __constant_htons(ETH_P_IPV6): {
55 const struct ipv6hdr *iph;
58 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
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);
68 case __constant_htons(ETH_P_8021Q): {
69 const struct vlan_hdr *vlan;
70 struct vlan_hdr _vlan;
72 vlan = skb_header_pointer(skb, nhoff, sizeof(_vlan), &_vlan);
76 proto = vlan->h_vlan_encapsulated_proto;
77 nhoff += sizeof(*vlan);
80 case __constant_htons(ETH_P_PPP_SES): {
85 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
89 nhoff += PPPOE_SES_HLEN;
91 case __constant_htons(PPP_IP):
93 case __constant_htons(PPP_IPV6):
110 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
114 * Only look inside GRE if version zero and no
117 if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
120 if (hdr->flags & GRE_CSUM)
122 if (hdr->flags & GRE_KEY)
124 if (hdr->flags & GRE_SEQ)
126 if (proto == htons(ETH_P_TEB)) {
127 const struct ethhdr *eth;
130 eth = skb_header_pointer(skb, nhoff,
131 sizeof(_eth), &_eth);
134 proto = eth->h_proto;
135 nhoff += sizeof(*eth);
147 flow->ip_proto = ip_proto;
148 poff = proto_ports_offset(ip_proto);
150 __be32 *ports, _ports;
152 ports = skb_header_pointer(skb, nhoff + poff,
153 sizeof(_ports), &_ports);
155 flow->ports = *ports;
158 flow->thoff = (u16) nhoff;
162 EXPORT_SYMBOL(skb_flow_dissect);
164 static u32 hashrnd __read_mostly;
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.
172 void __skb_get_rxhash(struct sk_buff *skb)
174 struct flow_keys keys;
177 if (!skb_flow_dissect(skb, &keys))
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]);
191 hash = jhash_3words((__force u32)keys.dst,
192 (__force u32)keys.src,
193 (__force u32)keys.ports, hashrnd);
199 EXPORT_SYMBOL(__skb_get_rxhash);
202 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
203 * to be used as a distribution range.
205 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
206 unsigned int num_tx_queues)
210 u16 qcount = num_tx_queues;
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;
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;
225 if (skb->sk && skb->sk->sk_hash)
226 hash = skb->sk->sk_hash;
228 hash = (__force u16) skb->protocol;
229 hash = jhash_1word(hash, hashrnd);
231 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
233 EXPORT_SYMBOL(__skb_tx_hash);
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.
240 u32 __skb_get_poff(const struct sk_buff *skb)
242 struct flow_keys keys;
245 if (!skb_flow_dissect(skb, &keys))
249 switch (keys.ip_proto) {
251 const struct tcphdr *tcph;
254 tcph = skb_header_pointer(skb, poff, sizeof(_tcph), &_tcph);
258 poff += max_t(u32, sizeof(struct tcphdr), tcph->doff * 4);
262 case IPPROTO_UDPLITE:
263 poff += sizeof(struct udphdr);
265 /* For the rest, we do not really care about header
266 * extensions at this point for now.
269 poff += sizeof(struct icmphdr);
272 poff += sizeof(struct icmp6hdr);
275 poff += sizeof(struct igmphdr);
278 poff += sizeof(struct dccp_hdr);
281 poff += sizeof(struct sctphdr);
288 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
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);
299 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
302 struct xps_dev_maps *dev_maps;
304 int queue_index = -1;
307 dev_maps = rcu_dereference(dev->xps_maps);
309 map = rcu_dereference(
310 dev_maps->cpu_map[raw_smp_processor_id()]);
313 queue_index = map->queues[0];
316 if (skb->sk && skb->sk->sk_hash)
317 hash = skb->sk->sk_hash;
319 hash = (__force u16) skb->protocol ^
321 hash = jhash_1word(hash, hashrnd);
322 queue_index = map->queues[
323 ((u64)hash * map->len) >> 32];
325 if (unlikely(queue_index >= dev->real_num_tx_queues))
337 u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
339 struct sock *sk = skb->sk;
340 int queue_index = sk_tx_queue_get(sk);
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);
346 new_index = skb_tx_hash(dev, skb);
348 if (queue_index != new_index && sk &&
349 rcu_access_pointer(sk->sk_dst_cache))
350 sk_tx_queue_set(sk, new_index);
352 queue_index = new_index;
357 EXPORT_SYMBOL(__netdev_pick_tx);
359 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
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);
369 queue_index = __netdev_pick_tx(dev, skb);
370 queue_index = dev_cap_txqueue(dev, queue_index);
373 skb_set_queue_mapping(skb, queue_index);
374 return netdev_get_tx_queue(dev, queue_index);
377 static int __init initialize_hashrnd(void)
379 get_random_bytes(&hashrnd, sizeof(hashrnd));
383 late_initcall_sync(initialize_hashrnd);