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));
29 * skb_flow_get_ports - extract the upper layer ports and return them
30 * @skb: buffer to extract the ports from
31 * @thoff: transport header offset
32 * @ip_proto: protocol for which to get port offset
34 * The function will try to retrieve the ports at offset thoff + poff where poff
35 * is the protocol port offset returned from proto_ports_offset
37 __be32 skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto)
39 int poff = proto_ports_offset(ip_proto);
42 __be32 *ports, _ports;
44 ports = skb_header_pointer(skb, thoff + poff,
45 sizeof(_ports), &_ports);
52 EXPORT_SYMBOL(skb_flow_get_ports);
54 bool skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow)
56 int nhoff = skb_network_offset(skb);
58 __be16 proto = skb->protocol;
60 memset(flow, 0, sizeof(*flow));
64 case htons(ETH_P_IP): {
65 const struct iphdr *iph;
68 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
69 if (!iph || iph->ihl < 5)
71 nhoff += iph->ihl * 4;
73 ip_proto = iph->protocol;
74 if (ip_is_fragment(iph))
77 iph_to_flow_copy_addrs(flow, iph);
80 case htons(ETH_P_IPV6): {
81 const struct ipv6hdr *iph;
86 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
90 ip_proto = iph->nexthdr;
91 flow->src = (__force __be32)ipv6_addr_hash(&iph->saddr);
92 flow->dst = (__force __be32)ipv6_addr_hash(&iph->daddr);
93 nhoff += sizeof(struct ipv6hdr);
95 flow_label = ip6_flowlabel(iph);
97 /* Awesome, IPv6 packet has a flow label so we can
98 * use that to represent the ports without any
101 flow->n_proto = proto;
102 flow->ip_proto = ip_proto;
103 flow->ports = flow_label;
104 flow->thoff = (u16)nhoff;
111 case htons(ETH_P_8021AD):
112 case htons(ETH_P_8021Q): {
113 const struct vlan_hdr *vlan;
114 struct vlan_hdr _vlan;
116 vlan = skb_header_pointer(skb, nhoff, sizeof(_vlan), &_vlan);
120 proto = vlan->h_vlan_encapsulated_proto;
121 nhoff += sizeof(*vlan);
124 case htons(ETH_P_PPP_SES): {
126 struct pppoe_hdr hdr;
129 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
133 nhoff += PPPOE_SES_HLEN;
137 case htons(PPP_IPV6):
154 hdr = skb_header_pointer(skb, nhoff, sizeof(_hdr), &_hdr);
158 * Only look inside GRE if version zero and no
161 if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
164 if (hdr->flags & GRE_CSUM)
166 if (hdr->flags & GRE_KEY)
168 if (hdr->flags & GRE_SEQ)
170 if (proto == htons(ETH_P_TEB)) {
171 const struct ethhdr *eth;
174 eth = skb_header_pointer(skb, nhoff,
175 sizeof(_eth), &_eth);
178 proto = eth->h_proto;
179 nhoff += sizeof(*eth);
186 proto = htons(ETH_P_IP);
189 proto = htons(ETH_P_IPV6);
195 flow->n_proto = proto;
196 flow->ip_proto = ip_proto;
197 flow->ports = skb_flow_get_ports(skb, nhoff, ip_proto);
198 flow->thoff = (u16) nhoff;
202 EXPORT_SYMBOL(skb_flow_dissect);
204 static u32 hashrnd __read_mostly;
205 static __always_inline void __flow_hash_secret_init(void)
207 net_get_random_once(&hashrnd, sizeof(hashrnd));
210 static __always_inline u32 __flow_hash_3words(u32 a, u32 b, u32 c)
212 __flow_hash_secret_init();
213 return jhash_3words(a, b, c, hashrnd);
216 static inline u32 __flow_hash_from_keys(struct flow_keys *keys)
220 /* get a consistent hash (same value on both flow directions) */
221 if (((__force u32)keys->dst < (__force u32)keys->src) ||
222 (((__force u32)keys->dst == (__force u32)keys->src) &&
223 ((__force u16)keys->port16[1] < (__force u16)keys->port16[0]))) {
224 swap(keys->dst, keys->src);
225 swap(keys->port16[0], keys->port16[1]);
228 hash = __flow_hash_3words((__force u32)keys->dst,
229 (__force u32)keys->src,
230 (__force u32)keys->ports);
237 u32 flow_hash_from_keys(struct flow_keys *keys)
239 return __flow_hash_from_keys(keys);
241 EXPORT_SYMBOL(flow_hash_from_keys);
244 * __skb_get_hash: calculate a flow hash based on src/dst addresses
245 * and src/dst port numbers. Sets hash in skb to non-zero hash value
246 * on success, zero indicates no valid hash. Also, sets l4_hash in skb
247 * if hash is a canonical 4-tuple hash over transport ports.
249 void __skb_get_hash(struct sk_buff *skb)
251 struct flow_keys keys;
253 if (!skb_flow_dissect(skb, &keys))
261 skb->hash = __flow_hash_from_keys(&keys);
263 EXPORT_SYMBOL(__skb_get_hash);
266 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
267 * to be used as a distribution range.
269 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
270 unsigned int num_tx_queues)
274 u16 qcount = num_tx_queues;
276 if (skb_rx_queue_recorded(skb)) {
277 hash = skb_get_rx_queue(skb);
278 while (unlikely(hash >= num_tx_queues))
279 hash -= num_tx_queues;
284 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
285 qoffset = dev->tc_to_txq[tc].offset;
286 qcount = dev->tc_to_txq[tc].count;
289 return (u16) (((u64)skb_get_hash(skb) * qcount) >> 32) + qoffset;
291 EXPORT_SYMBOL(__skb_tx_hash);
293 /* __skb_get_poff() returns the offset to the payload as far as it could
294 * be dissected. The main user is currently BPF, so that we can dynamically
295 * truncate packets without needing to push actual payload to the user
296 * space and can analyze headers only, instead.
298 u32 __skb_get_poff(const struct sk_buff *skb)
300 struct flow_keys keys;
303 if (!skb_flow_dissect(skb, &keys))
307 switch (keys.ip_proto) {
309 const struct tcphdr *tcph;
312 tcph = skb_header_pointer(skb, poff, sizeof(_tcph), &_tcph);
316 poff += max_t(u32, sizeof(struct tcphdr), tcph->doff * 4);
320 case IPPROTO_UDPLITE:
321 poff += sizeof(struct udphdr);
323 /* For the rest, we do not really care about header
324 * extensions at this point for now.
327 poff += sizeof(struct icmphdr);
330 poff += sizeof(struct icmp6hdr);
333 poff += sizeof(struct igmphdr);
336 poff += sizeof(struct dccp_hdr);
339 poff += sizeof(struct sctphdr);
346 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
349 struct xps_dev_maps *dev_maps;
351 int queue_index = -1;
354 dev_maps = rcu_dereference(dev->xps_maps);
356 map = rcu_dereference(
357 dev_maps->cpu_map[raw_smp_processor_id()]);
360 queue_index = map->queues[0];
362 queue_index = map->queues[
363 ((u64)skb_get_hash(skb) * map->len) >> 32];
365 if (unlikely(queue_index >= dev->real_num_tx_queues))
377 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
379 struct sock *sk = skb->sk;
380 int queue_index = sk_tx_queue_get(sk);
382 if (queue_index < 0 || skb->ooo_okay ||
383 queue_index >= dev->real_num_tx_queues) {
384 int new_index = get_xps_queue(dev, skb);
386 new_index = skb_tx_hash(dev, skb);
388 if (queue_index != new_index && sk &&
389 rcu_access_pointer(sk->sk_dst_cache))
390 sk_tx_queue_set(sk, new_index);
392 queue_index = new_index;
398 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
404 if (dev->real_num_tx_queues != 1) {
405 const struct net_device_ops *ops = dev->netdev_ops;
406 if (ops->ndo_select_queue)
407 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
410 queue_index = __netdev_pick_tx(dev, skb);
413 queue_index = netdev_cap_txqueue(dev, queue_index);
416 skb_set_queue_mapping(skb, queue_index);
417 return netdev_get_tx_queue(dev, queue_index);