Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 1 | #include <linux/skbuff.h> |
Jesper Dangaard Brouer | c452ed7 | 2012-01-24 16:03:33 -0500 | [diff] [blame] | 2 | #include <linux/export.h> |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 3 | #include <linux/ip.h> |
| 4 | #include <linux/ipv6.h> |
| 5 | #include <linux/if_vlan.h> |
| 6 | #include <net/ip.h> |
Eric Dumazet | ddbe503 | 2012-07-18 08:11:12 +0000 | [diff] [blame] | 7 | #include <net/ipv6.h> |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 8 | #include <linux/igmp.h> |
| 9 | #include <linux/icmp.h> |
| 10 | #include <linux/sctp.h> |
| 11 | #include <linux/dccp.h> |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 12 | #include <linux/if_tunnel.h> |
| 13 | #include <linux/if_pppox.h> |
| 14 | #include <linux/ppp_defs.h> |
| 15 | #include <net/flow_keys.h> |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 16 | #include <scsi/fc/fc_fcoe.h> |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 17 | |
Eric Dumazet | 4d77d2b | 2011-11-28 20:30:35 +0000 | [diff] [blame] | 18 | /* copy saddr & daddr, possibly using 64bit load/store |
| 19 | * Equivalent to : flow->src = iph->saddr; |
| 20 | * flow->dst = iph->daddr; |
| 21 | */ |
| 22 | static void iph_to_flow_copy_addrs(struct flow_keys *flow, const struct iphdr *iph) |
| 23 | { |
| 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)); |
| 27 | } |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 28 | |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 29 | /** |
WANG Cong | 6451b3f | 2014-08-25 17:03:46 -0700 | [diff] [blame] | 30 | * __skb_flow_get_ports - extract the upper layer ports and return them |
| 31 | * @skb: sk_buff to extract the ports from |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 32 | * @thoff: transport header offset |
| 33 | * @ip_proto: protocol for which to get port offset |
WANG Cong | 6451b3f | 2014-08-25 17:03:46 -0700 | [diff] [blame] | 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) |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 36 | * |
| 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 |
| 39 | */ |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 40 | __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, |
| 41 | void *data, int hlen) |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 42 | { |
| 43 | int poff = proto_ports_offset(ip_proto); |
| 44 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 45 | if (!data) { |
| 46 | data = skb->data; |
| 47 | hlen = skb_headlen(skb); |
| 48 | } |
| 49 | |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 50 | if (poff >= 0) { |
| 51 | __be32 *ports, _ports; |
| 52 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 53 | ports = __skb_header_pointer(skb, thoff + poff, |
| 54 | sizeof(_ports), data, hlen, &_ports); |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 55 | if (ports) |
| 56 | return *ports; |
| 57 | } |
| 58 | |
| 59 | return 0; |
| 60 | } |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 61 | EXPORT_SYMBOL(__skb_flow_get_ports); |
Nikolay Aleksandrov | 357afe9 | 2013-10-02 13:39:24 +0200 | [diff] [blame] | 62 | |
WANG Cong | 453a940 | 2014-08-25 17:03:47 -0700 | [diff] [blame] | 63 | /** |
| 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) |
| 70 | * |
| 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 |
| 73 | */ |
| 74 | bool __skb_flow_dissect(const struct sk_buff *skb, struct flow_keys *flow, |
| 75 | void *data, __be16 proto, int nhoff, int hlen) |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 76 | { |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 77 | u8 ip_proto; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 78 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 79 | if (!data) { |
| 80 | data = skb->data; |
WANG Cong | 453a940 | 2014-08-25 17:03:47 -0700 | [diff] [blame] | 81 | proto = skb->protocol; |
| 82 | nhoff = skb_network_offset(skb); |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 83 | hlen = skb_headlen(skb); |
| 84 | } |
| 85 | |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 86 | memset(flow, 0, sizeof(*flow)); |
| 87 | |
| 88 | again: |
| 89 | switch (proto) { |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 90 | case htons(ETH_P_IP): { |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 91 | const struct iphdr *iph; |
| 92 | struct iphdr _iph; |
| 93 | ip: |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 94 | iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph); |
Jason Wang | 6f09234 | 2013-11-01 15:01:10 +0800 | [diff] [blame] | 95 | if (!iph || iph->ihl < 5) |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 96 | return false; |
Eric Dumazet | 3797d3e | 2013-11-07 08:37:28 -0800 | [diff] [blame] | 97 | nhoff += iph->ihl * 4; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 98 | |
Eric Dumazet | 3797d3e | 2013-11-07 08:37:28 -0800 | [diff] [blame] | 99 | ip_proto = iph->protocol; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 100 | if (ip_is_fragment(iph)) |
| 101 | ip_proto = 0; |
Eric Dumazet | 3797d3e | 2013-11-07 08:37:28 -0800 | [diff] [blame] | 102 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 103 | /* skip the address processing if skb is NULL. The assumption |
| 104 | * here is that if there is no skb we are not looking for flow |
| 105 | * info but lengths and protocols. |
| 106 | */ |
| 107 | if (!skb) |
| 108 | break; |
| 109 | |
Eric Dumazet | 4d77d2b | 2011-11-28 20:30:35 +0000 | [diff] [blame] | 110 | iph_to_flow_copy_addrs(flow, iph); |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 111 | break; |
| 112 | } |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 113 | case htons(ETH_P_IPV6): { |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 114 | const struct ipv6hdr *iph; |
| 115 | struct ipv6hdr _iph; |
Tom Herbert | 19469a8 | 2014-07-01 21:33:01 -0700 | [diff] [blame] | 116 | __be32 flow_label; |
| 117 | |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 118 | ipv6: |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 119 | iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph); |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 120 | if (!iph) |
| 121 | return false; |
| 122 | |
| 123 | ip_proto = iph->nexthdr; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 124 | nhoff += sizeof(struct ipv6hdr); |
Tom Herbert | 19469a8 | 2014-07-01 21:33:01 -0700 | [diff] [blame] | 125 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 126 | /* see comment above in IPv4 section */ |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 127 | if (!skb) |
| 128 | break; |
| 129 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 130 | flow->src = (__force __be32)ipv6_addr_hash(&iph->saddr); |
| 131 | flow->dst = (__force __be32)ipv6_addr_hash(&iph->daddr); |
| 132 | |
Tom Herbert | 19469a8 | 2014-07-01 21:33:01 -0700 | [diff] [blame] | 133 | flow_label = ip6_flowlabel(iph); |
| 134 | if (flow_label) { |
| 135 | /* Awesome, IPv6 packet has a flow label so we can |
| 136 | * use that to represent the ports without any |
| 137 | * further dissection. |
| 138 | */ |
| 139 | flow->n_proto = proto; |
| 140 | flow->ip_proto = ip_proto; |
| 141 | flow->ports = flow_label; |
| 142 | flow->thoff = (u16)nhoff; |
| 143 | |
| 144 | return true; |
| 145 | } |
| 146 | |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 147 | break; |
| 148 | } |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 149 | case htons(ETH_P_8021AD): |
| 150 | case htons(ETH_P_8021Q): { |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 151 | const struct vlan_hdr *vlan; |
| 152 | struct vlan_hdr _vlan; |
| 153 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 154 | vlan = __skb_header_pointer(skb, nhoff, sizeof(_vlan), data, hlen, &_vlan); |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 155 | if (!vlan) |
| 156 | return false; |
| 157 | |
| 158 | proto = vlan->h_vlan_encapsulated_proto; |
| 159 | nhoff += sizeof(*vlan); |
| 160 | goto again; |
| 161 | } |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 162 | case htons(ETH_P_PPP_SES): { |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 163 | struct { |
| 164 | struct pppoe_hdr hdr; |
| 165 | __be16 proto; |
| 166 | } *hdr, _hdr; |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 167 | hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr); |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 168 | if (!hdr) |
| 169 | return false; |
| 170 | proto = hdr->proto; |
| 171 | nhoff += PPPOE_SES_HLEN; |
| 172 | switch (proto) { |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 173 | case htons(PPP_IP): |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 174 | goto ip; |
Joe Perches | 2b8837a | 2014-03-12 10:04:17 -0700 | [diff] [blame] | 175 | case htons(PPP_IPV6): |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 176 | goto ipv6; |
| 177 | default: |
| 178 | return false; |
| 179 | } |
| 180 | } |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 181 | case htons(ETH_P_FCOE): |
| 182 | flow->thoff = (u16)(nhoff + FCOE_HEADER_LEN); |
| 183 | /* fall through */ |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 184 | default: |
| 185 | return false; |
| 186 | } |
| 187 | |
| 188 | switch (ip_proto) { |
| 189 | case IPPROTO_GRE: { |
| 190 | struct gre_hdr { |
| 191 | __be16 flags; |
| 192 | __be16 proto; |
| 193 | } *hdr, _hdr; |
| 194 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 195 | hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr); |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 196 | if (!hdr) |
| 197 | return false; |
| 198 | /* |
| 199 | * Only look inside GRE if version zero and no |
| 200 | * routing |
| 201 | */ |
| 202 | if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) { |
| 203 | proto = hdr->proto; |
| 204 | nhoff += 4; |
| 205 | if (hdr->flags & GRE_CSUM) |
| 206 | nhoff += 4; |
| 207 | if (hdr->flags & GRE_KEY) |
| 208 | nhoff += 4; |
| 209 | if (hdr->flags & GRE_SEQ) |
| 210 | nhoff += 4; |
Michael Dalton | e1733de | 2013-03-11 06:52:28 +0000 | [diff] [blame] | 211 | if (proto == htons(ETH_P_TEB)) { |
| 212 | const struct ethhdr *eth; |
| 213 | struct ethhdr _eth; |
| 214 | |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 215 | eth = __skb_header_pointer(skb, nhoff, |
| 216 | sizeof(_eth), |
| 217 | data, hlen, &_eth); |
Michael Dalton | e1733de | 2013-03-11 06:52:28 +0000 | [diff] [blame] | 218 | if (!eth) |
| 219 | return false; |
| 220 | proto = eth->h_proto; |
| 221 | nhoff += sizeof(*eth); |
| 222 | } |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 223 | goto again; |
| 224 | } |
| 225 | break; |
| 226 | } |
| 227 | case IPPROTO_IPIP: |
Tom Herbert | fca4189 | 2013-07-29 11:07:36 -0700 | [diff] [blame] | 228 | proto = htons(ETH_P_IP); |
| 229 | goto ip; |
Tom Herbert | b438f94 | 2013-07-29 11:07:42 -0700 | [diff] [blame] | 230 | case IPPROTO_IPV6: |
| 231 | proto = htons(ETH_P_IPV6); |
| 232 | goto ipv6; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 233 | default: |
| 234 | break; |
| 235 | } |
| 236 | |
Govindarajulu Varadarajan | e0f31d8 | 2014-06-23 16:07:58 +0530 | [diff] [blame] | 237 | flow->n_proto = proto; |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 238 | flow->ip_proto = ip_proto; |
Daniel Borkmann | 8ed7816 | 2013-03-19 06:39:29 +0000 | [diff] [blame] | 239 | flow->thoff = (u16) nhoff; |
| 240 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 241 | /* unless skb is set we don't need to record port info */ |
| 242 | if (skb) |
| 243 | flow->ports = __skb_flow_get_ports(skb, nhoff, ip_proto, |
| 244 | data, hlen); |
| 245 | |
Eric Dumazet | 0744dd0 | 2011-11-28 05:22:18 +0000 | [diff] [blame] | 246 | return true; |
| 247 | } |
David S. Miller | 690e36e | 2014-08-23 12:13:41 -0700 | [diff] [blame] | 248 | EXPORT_SYMBOL(__skb_flow_dissect); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 249 | |
| 250 | static u32 hashrnd __read_mostly; |
Hannes Frederic Sowa | 66415cf | 2013-10-23 20:06:00 +0200 | [diff] [blame] | 251 | static __always_inline void __flow_hash_secret_init(void) |
| 252 | { |
| 253 | net_get_random_once(&hashrnd, sizeof(hashrnd)); |
| 254 | } |
| 255 | |
| 256 | static __always_inline u32 __flow_hash_3words(u32 a, u32 b, u32 c) |
| 257 | { |
| 258 | __flow_hash_secret_init(); |
| 259 | return jhash_3words(a, b, c, hashrnd); |
| 260 | } |
| 261 | |
Tom Herbert | 5ed20a6 | 2014-07-01 21:32:05 -0700 | [diff] [blame] | 262 | static inline u32 __flow_hash_from_keys(struct flow_keys *keys) |
| 263 | { |
| 264 | u32 hash; |
| 265 | |
| 266 | /* get a consistent hash (same value on both flow directions) */ |
| 267 | if (((__force u32)keys->dst < (__force u32)keys->src) || |
| 268 | (((__force u32)keys->dst == (__force u32)keys->src) && |
| 269 | ((__force u16)keys->port16[1] < (__force u16)keys->port16[0]))) { |
| 270 | swap(keys->dst, keys->src); |
| 271 | swap(keys->port16[0], keys->port16[1]); |
| 272 | } |
| 273 | |
| 274 | hash = __flow_hash_3words((__force u32)keys->dst, |
| 275 | (__force u32)keys->src, |
| 276 | (__force u32)keys->ports); |
| 277 | if (!hash) |
| 278 | hash = 1; |
| 279 | |
| 280 | return hash; |
| 281 | } |
| 282 | |
| 283 | u32 flow_hash_from_keys(struct flow_keys *keys) |
| 284 | { |
| 285 | return __flow_hash_from_keys(keys); |
| 286 | } |
| 287 | EXPORT_SYMBOL(flow_hash_from_keys); |
| 288 | |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 289 | /* |
Tom Herbert | 3958afa1b | 2013-12-15 22:12:06 -0800 | [diff] [blame] | 290 | * __skb_get_hash: calculate a flow hash based on src/dst addresses |
Tom Herbert | 61b905d | 2014-03-24 15:34:47 -0700 | [diff] [blame] | 291 | * and src/dst port numbers. Sets hash in skb to non-zero hash value |
| 292 | * on success, zero indicates no valid hash. Also, sets l4_hash in skb |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 293 | * if hash is a canonical 4-tuple hash over transport ports. |
| 294 | */ |
Tom Herbert | 3958afa1b | 2013-12-15 22:12:06 -0800 | [diff] [blame] | 295 | void __skb_get_hash(struct sk_buff *skb) |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 296 | { |
| 297 | struct flow_keys keys; |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 298 | |
| 299 | if (!skb_flow_dissect(skb, &keys)) |
| 300 | return; |
| 301 | |
| 302 | if (keys.ports) |
Tom Herbert | 61b905d | 2014-03-24 15:34:47 -0700 | [diff] [blame] | 303 | skb->l4_hash = 1; |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 304 | |
Tom Herbert | a3b18dd | 2014-07-01 21:33:17 -0700 | [diff] [blame] | 305 | skb->sw_hash = 1; |
| 306 | |
Tom Herbert | 5ed20a6 | 2014-07-01 21:32:05 -0700 | [diff] [blame] | 307 | skb->hash = __flow_hash_from_keys(&keys); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 308 | } |
Tom Herbert | 3958afa1b | 2013-12-15 22:12:06 -0800 | [diff] [blame] | 309 | EXPORT_SYMBOL(__skb_get_hash); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 310 | |
| 311 | /* |
| 312 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number |
| 313 | * to be used as a distribution range. |
| 314 | */ |
Tom Herbert | 0e00161 | 2014-07-01 21:32:27 -0700 | [diff] [blame] | 315 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 316 | unsigned int num_tx_queues) |
| 317 | { |
| 318 | u32 hash; |
| 319 | u16 qoffset = 0; |
| 320 | u16 qcount = num_tx_queues; |
| 321 | |
| 322 | if (skb_rx_queue_recorded(skb)) { |
| 323 | hash = skb_get_rx_queue(skb); |
| 324 | while (unlikely(hash >= num_tx_queues)) |
| 325 | hash -= num_tx_queues; |
| 326 | return hash; |
| 327 | } |
| 328 | |
| 329 | if (dev->num_tc) { |
| 330 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); |
| 331 | qoffset = dev->tc_to_txq[tc].offset; |
| 332 | qcount = dev->tc_to_txq[tc].count; |
| 333 | } |
| 334 | |
Daniel Borkmann | 8fc54f6 | 2014-08-23 20:58:54 +0200 | [diff] [blame] | 335 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 336 | } |
| 337 | EXPORT_SYMBOL(__skb_tx_hash); |
| 338 | |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 339 | u32 __skb_get_poff(const struct sk_buff *skb, void *data, |
| 340 | const struct flow_keys *keys, int hlen) |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 341 | { |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 342 | u32 poff = keys->thoff; |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 343 | |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 344 | switch (keys->ip_proto) { |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 345 | case IPPROTO_TCP: { |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 346 | /* access doff as u8 to avoid unaligned access */ |
| 347 | const u8 *doff; |
| 348 | u8 _doff; |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 349 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 350 | doff = __skb_header_pointer(skb, poff + 12, sizeof(_doff), |
| 351 | data, hlen, &_doff); |
| 352 | if (!doff) |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 353 | return poff; |
| 354 | |
Alexander Duyck | 5af7fb6 | 2014-10-10 12:09:12 -0700 | [diff] [blame] | 355 | poff += max_t(u32, sizeof(struct tcphdr), (*doff & 0xF0) >> 2); |
Daniel Borkmann | f77668d | 2013-03-19 06:39:30 +0000 | [diff] [blame] | 356 | break; |
| 357 | } |
| 358 | case IPPROTO_UDP: |
| 359 | case IPPROTO_UDPLITE: |
| 360 | poff += sizeof(struct udphdr); |
| 361 | break; |
| 362 | /* For the rest, we do not really care about header |
| 363 | * extensions at this point for now. |
| 364 | */ |
| 365 | case IPPROTO_ICMP: |
| 366 | poff += sizeof(struct icmphdr); |
| 367 | break; |
| 368 | case IPPROTO_ICMPV6: |
| 369 | poff += sizeof(struct icmp6hdr); |
| 370 | break; |
| 371 | case IPPROTO_IGMP: |
| 372 | poff += sizeof(struct igmphdr); |
| 373 | break; |
| 374 | case IPPROTO_DCCP: |
| 375 | poff += sizeof(struct dccp_hdr); |
| 376 | break; |
| 377 | case IPPROTO_SCTP: |
| 378 | poff += sizeof(struct sctphdr); |
| 379 | break; |
| 380 | } |
| 381 | |
| 382 | return poff; |
| 383 | } |
| 384 | |
Alexander Duyck | 56193d1 | 2014-09-05 19:20:26 -0400 | [diff] [blame] | 385 | /* skb_get_poff() returns the offset to the payload as far as it could |
| 386 | * be dissected. The main user is currently BPF, so that we can dynamically |
| 387 | * truncate packets without needing to push actual payload to the user |
| 388 | * space and can analyze headers only, instead. |
| 389 | */ |
| 390 | u32 skb_get_poff(const struct sk_buff *skb) |
| 391 | { |
| 392 | struct flow_keys keys; |
| 393 | |
| 394 | if (!skb_flow_dissect(skb, &keys)) |
| 395 | return 0; |
| 396 | |
| 397 | return __skb_get_poff(skb, skb->data, &keys, skb_headlen(skb)); |
| 398 | } |
| 399 | |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 400 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
| 401 | { |
| 402 | #ifdef CONFIG_XPS |
| 403 | struct xps_dev_maps *dev_maps; |
| 404 | struct xps_map *map; |
| 405 | int queue_index = -1; |
| 406 | |
| 407 | rcu_read_lock(); |
| 408 | dev_maps = rcu_dereference(dev->xps_maps); |
| 409 | if (dev_maps) { |
| 410 | map = rcu_dereference( |
| 411 | dev_maps->cpu_map[raw_smp_processor_id()]); |
| 412 | if (map) { |
| 413 | if (map->len == 1) |
| 414 | queue_index = map->queues[0]; |
Tom Herbert | 0e00161 | 2014-07-01 21:32:27 -0700 | [diff] [blame] | 415 | else |
Daniel Borkmann | 8fc54f6 | 2014-08-23 20:58:54 +0200 | [diff] [blame] | 416 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), |
| 417 | map->len)]; |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 418 | if (unlikely(queue_index >= dev->real_num_tx_queues)) |
| 419 | queue_index = -1; |
| 420 | } |
| 421 | } |
| 422 | rcu_read_unlock(); |
| 423 | |
| 424 | return queue_index; |
| 425 | #else |
| 426 | return -1; |
| 427 | #endif |
| 428 | } |
| 429 | |
Daniel Borkmann | 99932d4 | 2014-02-16 15:55:20 +0100 | [diff] [blame] | 430 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 431 | { |
| 432 | struct sock *sk = skb->sk; |
| 433 | int queue_index = sk_tx_queue_get(sk); |
| 434 | |
| 435 | if (queue_index < 0 || skb->ooo_okay || |
| 436 | queue_index >= dev->real_num_tx_queues) { |
| 437 | int new_index = get_xps_queue(dev, skb); |
| 438 | if (new_index < 0) |
| 439 | new_index = skb_tx_hash(dev, skb); |
| 440 | |
Eric Dumazet | 702821f | 2013-08-28 18:10:43 -0700 | [diff] [blame] | 441 | if (queue_index != new_index && sk && |
| 442 | rcu_access_pointer(sk->sk_dst_cache)) |
Eric Dumazet | 50d1784 | 2013-09-07 12:02:57 -0700 | [diff] [blame] | 443 | sk_tx_queue_set(sk, new_index); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 444 | |
| 445 | queue_index = new_index; |
| 446 | } |
| 447 | |
| 448 | return queue_index; |
| 449 | } |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 450 | |
| 451 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, |
Jason Wang | f663dd9 | 2014-01-10 16:18:26 +0800 | [diff] [blame] | 452 | struct sk_buff *skb, |
| 453 | void *accel_priv) |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 454 | { |
| 455 | int queue_index = 0; |
| 456 | |
| 457 | if (dev->real_num_tx_queues != 1) { |
| 458 | const struct net_device_ops *ops = dev->netdev_ops; |
| 459 | if (ops->ndo_select_queue) |
Daniel Borkmann | 99932d4 | 2014-02-16 15:55:20 +0100 | [diff] [blame] | 460 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, |
| 461 | __netdev_pick_tx); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 462 | else |
| 463 | queue_index = __netdev_pick_tx(dev, skb); |
Jason Wang | f663dd9 | 2014-01-10 16:18:26 +0800 | [diff] [blame] | 464 | |
| 465 | if (!accel_priv) |
Daniel Borkmann | b9507bd | 2014-02-16 15:55:21 +0100 | [diff] [blame] | 466 | queue_index = netdev_cap_txqueue(dev, queue_index); |
Cong Wang | 441d9d3 | 2013-01-21 00:39:24 +0000 | [diff] [blame] | 467 | } |
| 468 | |
| 469 | skb_set_queue_mapping(skb, queue_index); |
| 470 | return netdev_get_tx_queue(dev, queue_index); |
| 471 | } |