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>
16 #include <scsi/fc/fc_fcoe.h>
18 /* copy saddr & daddr, possibly using 64bit load/store
19 * Equivalent to : flow->src = iph->saddr;
20 * flow->dst = iph->daddr;
22 static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *iph)
24 BUILD_BUG_ON(offsetof(typeof(*flow), dst) !=
25 offsetof(typeof(*flow), src) + sizeof(flow->src));
26 memcpy(&flow->src, &iph->saddr, sizeof(flow->src) + sizeof(flow->dst));
30 * __skb_flow_get_ports - extract the upper layer ports and return them
31 * @skb: sk_buff to extract the ports from
32 * @thoff: transport header offset
33 * @ip_proto: protocol for which to get port offset
34 * @data: raw buffer pointer to the packet, if NULL use skb->data
35 * @hlen: packet header length, if @data is NULL use skb_headlen(skb)
37 * The function will try to retrieve the ports at offset thoff + poff where poff
38 * is the protocol port offset returned from proto_ports_offset
40 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
43 int poff = proto_ports_offset(ip_proto);
47 hlen = skb_headlen(skb);
51 __be32 *ports, _ports;
53 ports = __skb_header_pointer(skb, thoff + poff,
54 sizeof(_ports), data, hlen, &_ports);
61 EXPORT_SYMBOL(__skb_flow_get_ports);
64 * __skb_flow_dissect - extract the flow_keys struct and return it
65 * @skb: sk_buff to extract the flow from, can be NULL if the rest are specified
66 * @data: raw buffer pointer to the packet, if NULL use skb->data
67 * @proto: protocol for which to get the flow, if @data is NULL use skb->protocol
68 * @nhoff: network header offset, if @data is NULL use skb_network_offset(skb)
69 * @hlen: packet header length, if @data is NULL use skb_headlen(skb)
71 * The function will try to retrieve the struct flow_keys from either the skbuff
72 * or a raw buffer specified by the rest parameters
74 bool __skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow,
75 void *data, __be16 proto, int nhoff, int hlen)
81 proto = skb->protocol;
82 nhoff = skb_network_offset(skb);
83 hlen = skb_headlen(skb);
86 memset(flow, 0, sizeof(*flow));
90 case htons(ETH_P_IP): {
91 const struct iphdr *iph;
94 iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
95 if (!iph || iph->ihl < 5)
97 nhoff += iph->ihl * 4;
99 ip_proto = iph->protocol;
100 if (ip_is_fragment(iph))
103 iph_to_flow_copy_addrs(flow, iph);
106 case htons(ETH_P_IPV6): {
107 const struct ipv6hdr *iph;
112 iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
116 ip_proto = iph->nexthdr;
117 flow->src = (__force __be32)ipv6_addr_hash(&iph->saddr);
118 flow->dst = (__force __be32)ipv6_addr_hash(&iph->daddr);
119 nhoff += sizeof(struct ipv6hdr);
121 /* skip the flow label processing if skb is NULL. The
122 * assumption here is that if there is no skb we are not
123 * looking for flow info as much as we are length.
128 flow_label = ip6_flowlabel(iph);
130 /* Awesome, IPv6 packet has a flow label so we can
131 * use that to represent the ports without any
132 * further dissection.
134 flow->n_proto = proto;
135 flow->ip_proto = ip_proto;
136 flow->ports = flow_label;
137 flow->thoff = (u16)nhoff;
144 case htons(ETH_P_8021AD):
145 case htons(ETH_P_8021Q): {
146 const struct vlan_hdr *vlan;
147 struct vlan_hdr _vlan;
149 vlan = __skb_header_pointer(skb, nhoff, sizeof(_vlan), data, hlen, &_vlan);
153 proto = vlan->h_vlan_encapsulated_proto;
154 nhoff += sizeof(*vlan);
157 case htons(ETH_P_PPP_SES): {
159 struct pppoe_hdr hdr;
162 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
166 nhoff += PPPOE_SES_HLEN;
170 case htons(PPP_IPV6):
176 case htons(ETH_P_FCOE):
177 flow->thoff = (u16)(nhoff + FCOE_HEADER_LEN);
190 hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
194 * Only look inside GRE if version zero and no
197 if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
200 if (hdr->flags & GRE_CSUM)
202 if (hdr->flags & GRE_KEY)
204 if (hdr->flags & GRE_SEQ)
206 if (proto == htons(ETH_P_TEB)) {
207 const struct ethhdr *eth;
210 eth = __skb_header_pointer(skb, nhoff,
215 proto = eth->h_proto;
216 nhoff += sizeof(*eth);
223 proto = htons(ETH_P_IP);
226 proto = htons(ETH_P_IPV6);
232 flow->n_proto = proto;
233 flow->ip_proto = ip_proto;
234 flow->ports = __skb_flow_get_ports(skb, nhoff, ip_proto, data, hlen);
235 flow->thoff = (u16) nhoff;
239 EXPORT_SYMBOL(__skb_flow_dissect);
241 static u32 hashrnd __read_mostly;
242 static __always_inline void __flow_hash_secret_init(void)
244 net_get_random_once(&hashrnd, sizeof(hashrnd));
247 static __always_inline u32 __flow_hash_3words(u32 a, u32 b, u32 c)
249 __flow_hash_secret_init();
250 return jhash_3words(a, b, c, hashrnd);
253 static inline u32 __flow_hash_from_keys(struct flow_keys *keys)
257 /* get a consistent hash (same value on both flow directions) */
258 if (((__force u32)keys->dst < (__force u32)keys->src) ||
259 (((__force u32)keys->dst == (__force u32)keys->src) &&
260 ((__force u16)keys->port16[1] < (__force u16)keys->port16[0]))) {
261 swap(keys->dst, keys->src);
262 swap(keys->port16[0], keys->port16[1]);
265 hash = __flow_hash_3words((__force u32)keys->dst,
266 (__force u32)keys->src,
267 (__force u32)keys->ports);
274 u32 flow_hash_from_keys(struct flow_keys *keys)
276 return __flow_hash_from_keys(keys);
278 EXPORT_SYMBOL(flow_hash_from_keys);
281 * __skb_get_hash: calculate a flow hash based on src/dst addresses
282 * and src/dst port numbers. Sets hash in skb to non-zero hash value
283 * on success, zero indicates no valid hash. Also, sets l4_hash in skb
284 * if hash is a canonical 4-tuple hash over transport ports.
286 void __skb_get_hash(struct sk_buff *skb)
288 struct flow_keys keys;
290 if (!skb_flow_dissect(skb, &keys))
298 skb->hash = __flow_hash_from_keys(&keys);
300 EXPORT_SYMBOL(__skb_get_hash);
303 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
304 * to be used as a distribution range.
306 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
307 unsigned int num_tx_queues)
311 u16 qcount = num_tx_queues;
313 if (skb_rx_queue_recorded(skb)) {
314 hash = skb_get_rx_queue(skb);
315 while (unlikely(hash >= num_tx_queues))
316 hash -= num_tx_queues;
321 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
322 qoffset = dev->tc_to_txq[tc].offset;
323 qcount = dev->tc_to_txq[tc].count;
326 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
328 EXPORT_SYMBOL(__skb_tx_hash);
330 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
331 const struct flow_keys *keys, int hlen)
333 u32 poff = keys->thoff;
335 switch (keys->ip_proto) {
337 const struct tcphdr *tcph;
340 tcph = __skb_header_pointer(skb, poff, sizeof(_tcph),
345 poff += max_t(u32, sizeof(struct tcphdr), tcph->doff * 4);
349 case IPPROTO_UDPLITE:
350 poff += sizeof(struct udphdr);
352 /* For the rest, we do not really care about header
353 * extensions at this point for now.
356 poff += sizeof(struct icmphdr);
359 poff += sizeof(struct icmp6hdr);
362 poff += sizeof(struct igmphdr);
365 poff += sizeof(struct dccp_hdr);
368 poff += sizeof(struct sctphdr);
375 /* skb_get_poff() returns the offset to the payload as far as it could
376 * be dissected. The main user is currently BPF, so that we can dynamically
377 * truncate packets without needing to push actual payload to the user
378 * space and can analyze headers only, instead.
380 u32 skb_get_poff(const struct sk_buff *skb)
382 struct flow_keys keys;
384 if (!skb_flow_dissect(skb, &keys))
387 return __skb_get_poff(skb, skb->data, &keys, skb_headlen(skb));
390 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
393 struct xps_dev_maps *dev_maps;
395 int queue_index = -1;
398 dev_maps = rcu_dereference(dev->xps_maps);
400 map = rcu_dereference(
401 dev_maps->cpu_map[raw_smp_processor_id()]);
404 queue_index = map->queues[0];
406 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
408 if (unlikely(queue_index >= dev->real_num_tx_queues))
420 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
422 struct sock *sk = skb->sk;
423 int queue_index = sk_tx_queue_get(sk);
425 if (queue_index < 0 || skb->ooo_okay ||
426 queue_index >= dev->real_num_tx_queues) {
427 int new_index = get_xps_queue(dev, skb);
429 new_index = skb_tx_hash(dev, skb);
431 if (queue_index != new_index && sk &&
432 rcu_access_pointer(sk->sk_dst_cache))
433 sk_tx_queue_set(sk, new_index);
435 queue_index = new_index;
441 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
447 if (dev->real_num_tx_queues != 1) {
448 const struct net_device_ops *ops = dev->netdev_ops;
449 if (ops->ndo_select_queue)
450 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
453 queue_index = __netdev_pick_tx(dev, skb);
456 queue_index = netdev_cap_txqueue(dev, queue_index);
459 skb_set_queue_mapping(skb, queue_index);
460 return netdev_get_tx_queue(dev, queue_index);