xref: /linux/net/core/flow_dissector.c (revision 7f8998c7aef3ac9c5f3f2943e083dfa6302e90d0)
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 #include <scsi/fc/fc_fcoe.h>
17 
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 }
28 
29 /**
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)
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  */
40 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
41 			    void *data, int hlen)
42 {
43 	int poff = proto_ports_offset(ip_proto);
44 
45 	if (!data) {
46 		data = skb->data;
47 		hlen = skb_headlen(skb);
48 	}
49 
50 	if (poff >= 0) {
51 		__be32 *ports, _ports;
52 
53 		ports = __skb_header_pointer(skb, thoff + poff,
54 					     sizeof(_ports), data, hlen, &_ports);
55 		if (ports)
56 			return *ports;
57 	}
58 
59 	return 0;
60 }
61 EXPORT_SYMBOL(__skb_flow_get_ports);
62 
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)
76 {
77 	u8 ip_proto;
78 
79 	if (!data) {
80 		data = skb->data;
81 		proto = skb->protocol;
82 		nhoff = skb_network_offset(skb);
83 		hlen = skb_headlen(skb);
84 	}
85 
86 	memset(flow, 0, sizeof(*flow));
87 
88 again:
89 	switch (proto) {
90 	case htons(ETH_P_IP): {
91 		const struct iphdr *iph;
92 		struct iphdr _iph;
93 ip:
94 		iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
95 		if (!iph || iph->ihl < 5)
96 			return false;
97 		nhoff += iph->ihl * 4;
98 
99 		ip_proto = iph->protocol;
100 		if (ip_is_fragment(iph))
101 			ip_proto = 0;
102 
103 		iph_to_flow_copy_addrs(flow, iph);
104 		break;
105 	}
106 	case htons(ETH_P_IPV6): {
107 		const struct ipv6hdr *iph;
108 		struct ipv6hdr _iph;
109 		__be32 flow_label;
110 
111 ipv6:
112 		iph = __skb_header_pointer(skb, nhoff, sizeof(_iph), data, hlen, &_iph);
113 		if (!iph)
114 			return false;
115 
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);
120 
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.
124 		 */
125 		if (!skb)
126 			break;
127 
128 		flow_label = ip6_flowlabel(iph);
129 		if (flow_label) {
130 			/* Awesome, IPv6 packet has a flow label so we can
131 			 * use that to represent the ports without any
132 			 * further dissection.
133 			 */
134 			flow->n_proto = proto;
135 			flow->ip_proto = ip_proto;
136 			flow->ports = flow_label;
137 			flow->thoff = (u16)nhoff;
138 
139 			return true;
140 		}
141 
142 		break;
143 	}
144 	case htons(ETH_P_8021AD):
145 	case htons(ETH_P_8021Q): {
146 		const struct vlan_hdr *vlan;
147 		struct vlan_hdr _vlan;
148 
149 		vlan = __skb_header_pointer(skb, nhoff, sizeof(_vlan), data, hlen, &_vlan);
150 		if (!vlan)
151 			return false;
152 
153 		proto = vlan->h_vlan_encapsulated_proto;
154 		nhoff += sizeof(*vlan);
155 		goto again;
156 	}
157 	case htons(ETH_P_PPP_SES): {
158 		struct {
159 			struct pppoe_hdr hdr;
160 			__be16 proto;
161 		} *hdr, _hdr;
162 		hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
163 		if (!hdr)
164 			return false;
165 		proto = hdr->proto;
166 		nhoff += PPPOE_SES_HLEN;
167 		switch (proto) {
168 		case htons(PPP_IP):
169 			goto ip;
170 		case htons(PPP_IPV6):
171 			goto ipv6;
172 		default:
173 			return false;
174 		}
175 	}
176 	case htons(ETH_P_FCOE):
177 		flow->thoff = (u16)(nhoff + FCOE_HEADER_LEN);
178 		/* fall through */
179 	default:
180 		return false;
181 	}
182 
183 	switch (ip_proto) {
184 	case IPPROTO_GRE: {
185 		struct gre_hdr {
186 			__be16 flags;
187 			__be16 proto;
188 		} *hdr, _hdr;
189 
190 		hdr = __skb_header_pointer(skb, nhoff, sizeof(_hdr), data, hlen, &_hdr);
191 		if (!hdr)
192 			return false;
193 		/*
194 		 * Only look inside GRE if version zero and no
195 		 * routing
196 		 */
197 		if (!(hdr->flags & (GRE_VERSION|GRE_ROUTING))) {
198 			proto = hdr->proto;
199 			nhoff += 4;
200 			if (hdr->flags & GRE_CSUM)
201 				nhoff += 4;
202 			if (hdr->flags & GRE_KEY)
203 				nhoff += 4;
204 			if (hdr->flags & GRE_SEQ)
205 				nhoff += 4;
206 			if (proto == htons(ETH_P_TEB)) {
207 				const struct ethhdr *eth;
208 				struct ethhdr _eth;
209 
210 				eth = __skb_header_pointer(skb, nhoff,
211 							   sizeof(_eth),
212 							   data, hlen, &_eth);
213 				if (!eth)
214 					return false;
215 				proto = eth->h_proto;
216 				nhoff += sizeof(*eth);
217 			}
218 			goto again;
219 		}
220 		break;
221 	}
222 	case IPPROTO_IPIP:
223 		proto = htons(ETH_P_IP);
224 		goto ip;
225 	case IPPROTO_IPV6:
226 		proto = htons(ETH_P_IPV6);
227 		goto ipv6;
228 	default:
229 		break;
230 	}
231 
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;
236 
237 	return true;
238 }
239 EXPORT_SYMBOL(__skb_flow_dissect);
240 
241 static u32 hashrnd __read_mostly;
242 static __always_inline void __flow_hash_secret_init(void)
243 {
244 	net_get_random_once(&hashrnd, sizeof(hashrnd));
245 }
246 
247 static __always_inline u32 __flow_hash_3words(u32 a, u32 b, u32 c)
248 {
249 	__flow_hash_secret_init();
250 	return jhash_3words(a, b, c, hashrnd);
251 }
252 
253 static inline u32 __flow_hash_from_keys(struct flow_keys *keys)
254 {
255 	u32 hash;
256 
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]);
263 	}
264 
265 	hash = __flow_hash_3words((__force u32)keys->dst,
266 				  (__force u32)keys->src,
267 				  (__force u32)keys->ports);
268 	if (!hash)
269 		hash = 1;
270 
271 	return hash;
272 }
273 
274 u32 flow_hash_from_keys(struct flow_keys *keys)
275 {
276 	return __flow_hash_from_keys(keys);
277 }
278 EXPORT_SYMBOL(flow_hash_from_keys);
279 
280 /*
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.
285  */
286 void __skb_get_hash(struct sk_buff *skb)
287 {
288 	struct flow_keys keys;
289 
290 	if (!skb_flow_dissect(skb, &keys))
291 		return;
292 
293 	if (keys.ports)
294 		skb->l4_hash = 1;
295 
296 	skb->sw_hash = 1;
297 
298 	skb->hash = __flow_hash_from_keys(&keys);
299 }
300 EXPORT_SYMBOL(__skb_get_hash);
301 
302 /*
303  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
304  * to be used as a distribution range.
305  */
306 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
307 		  unsigned int num_tx_queues)
308 {
309 	u32 hash;
310 	u16 qoffset = 0;
311 	u16 qcount = num_tx_queues;
312 
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;
317 		return hash;
318 	}
319 
320 	if (dev->num_tc) {
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;
324 	}
325 
326 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
327 }
328 EXPORT_SYMBOL(__skb_tx_hash);
329 
330 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
331 		   const struct flow_keys *keys, int hlen)
332 {
333 	u32 poff = keys->thoff;
334 
335 	switch (keys->ip_proto) {
336 	case IPPROTO_TCP: {
337 		const struct tcphdr *tcph;
338 		struct tcphdr _tcph;
339 
340 		tcph = __skb_header_pointer(skb, poff, sizeof(_tcph),
341 					    data, hlen, &_tcph);
342 		if (!tcph)
343 			return poff;
344 
345 		poff += max_t(u32, sizeof(struct tcphdr), tcph->doff * 4);
346 		break;
347 	}
348 	case IPPROTO_UDP:
349 	case IPPROTO_UDPLITE:
350 		poff += sizeof(struct udphdr);
351 		break;
352 	/* For the rest, we do not really care about header
353 	 * extensions at this point for now.
354 	 */
355 	case IPPROTO_ICMP:
356 		poff += sizeof(struct icmphdr);
357 		break;
358 	case IPPROTO_ICMPV6:
359 		poff += sizeof(struct icmp6hdr);
360 		break;
361 	case IPPROTO_IGMP:
362 		poff += sizeof(struct igmphdr);
363 		break;
364 	case IPPROTO_DCCP:
365 		poff += sizeof(struct dccp_hdr);
366 		break;
367 	case IPPROTO_SCTP:
368 		poff += sizeof(struct sctphdr);
369 		break;
370 	}
371 
372 	return poff;
373 }
374 
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.
379  */
380 u32 skb_get_poff(const struct sk_buff *skb)
381 {
382 	struct flow_keys keys;
383 
384 	if (!skb_flow_dissect(skb, &keys))
385 		return 0;
386 
387 	return __skb_get_poff(skb, skb->data, &keys, skb_headlen(skb));
388 }
389 
390 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
391 {
392 #ifdef CONFIG_XPS
393 	struct xps_dev_maps *dev_maps;
394 	struct xps_map *map;
395 	int queue_index = -1;
396 
397 	rcu_read_lock();
398 	dev_maps = rcu_dereference(dev->xps_maps);
399 	if (dev_maps) {
400 		map = rcu_dereference(
401 		    dev_maps->cpu_map[raw_smp_processor_id()]);
402 		if (map) {
403 			if (map->len == 1)
404 				queue_index = map->queues[0];
405 			else
406 				queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
407 									   map->len)];
408 			if (unlikely(queue_index >= dev->real_num_tx_queues))
409 				queue_index = -1;
410 		}
411 	}
412 	rcu_read_unlock();
413 
414 	return queue_index;
415 #else
416 	return -1;
417 #endif
418 }
419 
420 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
421 {
422 	struct sock *sk = skb->sk;
423 	int queue_index = sk_tx_queue_get(sk);
424 
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);
428 		if (new_index < 0)
429 			new_index = skb_tx_hash(dev, skb);
430 
431 		if (queue_index != new_index && sk &&
432 		    rcu_access_pointer(sk->sk_dst_cache))
433 			sk_tx_queue_set(sk, new_index);
434 
435 		queue_index = new_index;
436 	}
437 
438 	return queue_index;
439 }
440 
441 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
442 				    struct sk_buff *skb,
443 				    void *accel_priv)
444 {
445 	int queue_index = 0;
446 
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,
451 							    __netdev_pick_tx);
452 		else
453 			queue_index = __netdev_pick_tx(dev, skb);
454 
455 		if (!accel_priv)
456 			queue_index = netdev_cap_txqueue(dev, queue_index);
457 	}
458 
459 	skb_set_queue_mapping(skb, queue_index);
460 	return netdev_get_tx_queue(dev, queue_index);
461 }
462