xref: /openbmc/linux/net/openvswitch/flow.c (revision df23e9f6)
1 /*
2  * Copyright (c) 2007-2013 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #include "flow.h"
20 #include "datapath.h"
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
32 #include <linux/in.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
43 #include <net/ip.h>
44 #include <net/ip_tunnels.h>
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 
48 u64 ovs_flow_used_time(unsigned long flow_jiffies)
49 {
50 	struct timespec cur_ts;
51 	u64 cur_ms, idle_ms;
52 
53 	ktime_get_ts(&cur_ts);
54 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
55 	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
56 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
57 
58 	return cur_ms - idle_ms;
59 }
60 
61 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
62 
63 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
64 {
65 	__be16 tcp_flags = 0;
66 
67 	if ((flow->key.eth.type == htons(ETH_P_IP) ||
68 	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
69 	    flow->key.ip.proto == IPPROTO_TCP &&
70 	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
71 		tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
72 	}
73 
74 	spin_lock(&flow->lock);
75 	flow->used = jiffies;
76 	flow->packet_count++;
77 	flow->byte_count += skb->len;
78 	flow->tcp_flags |= tcp_flags;
79 	spin_unlock(&flow->lock);
80 }
81 
82 static int check_header(struct sk_buff *skb, int len)
83 {
84 	if (unlikely(skb->len < len))
85 		return -EINVAL;
86 	if (unlikely(!pskb_may_pull(skb, len)))
87 		return -ENOMEM;
88 	return 0;
89 }
90 
91 static bool arphdr_ok(struct sk_buff *skb)
92 {
93 	return pskb_may_pull(skb, skb_network_offset(skb) +
94 				  sizeof(struct arp_eth_header));
95 }
96 
97 static int check_iphdr(struct sk_buff *skb)
98 {
99 	unsigned int nh_ofs = skb_network_offset(skb);
100 	unsigned int ip_len;
101 	int err;
102 
103 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
104 	if (unlikely(err))
105 		return err;
106 
107 	ip_len = ip_hdrlen(skb);
108 	if (unlikely(ip_len < sizeof(struct iphdr) ||
109 		     skb->len < nh_ofs + ip_len))
110 		return -EINVAL;
111 
112 	skb_set_transport_header(skb, nh_ofs + ip_len);
113 	return 0;
114 }
115 
116 static bool tcphdr_ok(struct sk_buff *skb)
117 {
118 	int th_ofs = skb_transport_offset(skb);
119 	int tcp_len;
120 
121 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
122 		return false;
123 
124 	tcp_len = tcp_hdrlen(skb);
125 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
126 		     skb->len < th_ofs + tcp_len))
127 		return false;
128 
129 	return true;
130 }
131 
132 static bool udphdr_ok(struct sk_buff *skb)
133 {
134 	return pskb_may_pull(skb, skb_transport_offset(skb) +
135 				  sizeof(struct udphdr));
136 }
137 
138 static bool sctphdr_ok(struct sk_buff *skb)
139 {
140 	return pskb_may_pull(skb, skb_transport_offset(skb) +
141 				  sizeof(struct sctphdr));
142 }
143 
144 static bool icmphdr_ok(struct sk_buff *skb)
145 {
146 	return pskb_may_pull(skb, skb_transport_offset(skb) +
147 				  sizeof(struct icmphdr));
148 }
149 
150 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
151 {
152 	unsigned int nh_ofs = skb_network_offset(skb);
153 	unsigned int nh_len;
154 	int payload_ofs;
155 	struct ipv6hdr *nh;
156 	uint8_t nexthdr;
157 	__be16 frag_off;
158 	int err;
159 
160 	err = check_header(skb, nh_ofs + sizeof(*nh));
161 	if (unlikely(err))
162 		return err;
163 
164 	nh = ipv6_hdr(skb);
165 	nexthdr = nh->nexthdr;
166 	payload_ofs = (u8 *)(nh + 1) - skb->data;
167 
168 	key->ip.proto = NEXTHDR_NONE;
169 	key->ip.tos = ipv6_get_dsfield(nh);
170 	key->ip.ttl = nh->hop_limit;
171 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
172 	key->ipv6.addr.src = nh->saddr;
173 	key->ipv6.addr.dst = nh->daddr;
174 
175 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
176 	if (unlikely(payload_ofs < 0))
177 		return -EINVAL;
178 
179 	if (frag_off) {
180 		if (frag_off & htons(~0x7))
181 			key->ip.frag = OVS_FRAG_TYPE_LATER;
182 		else
183 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
184 	}
185 
186 	nh_len = payload_ofs - nh_ofs;
187 	skb_set_transport_header(skb, nh_ofs + nh_len);
188 	key->ip.proto = nexthdr;
189 	return nh_len;
190 }
191 
192 static bool icmp6hdr_ok(struct sk_buff *skb)
193 {
194 	return pskb_may_pull(skb, skb_transport_offset(skb) +
195 				  sizeof(struct icmp6hdr));
196 }
197 
198 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
199 {
200 	struct qtag_prefix {
201 		__be16 eth_type; /* ETH_P_8021Q */
202 		__be16 tci;
203 	};
204 	struct qtag_prefix *qp;
205 
206 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
207 		return 0;
208 
209 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
210 					 sizeof(__be16))))
211 		return -ENOMEM;
212 
213 	qp = (struct qtag_prefix *) skb->data;
214 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
215 	__skb_pull(skb, sizeof(struct qtag_prefix));
216 
217 	return 0;
218 }
219 
220 static __be16 parse_ethertype(struct sk_buff *skb)
221 {
222 	struct llc_snap_hdr {
223 		u8  dsap;  /* Always 0xAA */
224 		u8  ssap;  /* Always 0xAA */
225 		u8  ctrl;
226 		u8  oui[3];
227 		__be16 ethertype;
228 	};
229 	struct llc_snap_hdr *llc;
230 	__be16 proto;
231 
232 	proto = *(__be16 *) skb->data;
233 	__skb_pull(skb, sizeof(__be16));
234 
235 	if (ntohs(proto) >= ETH_P_802_3_MIN)
236 		return proto;
237 
238 	if (skb->len < sizeof(struct llc_snap_hdr))
239 		return htons(ETH_P_802_2);
240 
241 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
242 		return htons(0);
243 
244 	llc = (struct llc_snap_hdr *) skb->data;
245 	if (llc->dsap != LLC_SAP_SNAP ||
246 	    llc->ssap != LLC_SAP_SNAP ||
247 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
248 		return htons(ETH_P_802_2);
249 
250 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
251 
252 	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
253 		return llc->ethertype;
254 
255 	return htons(ETH_P_802_2);
256 }
257 
258 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
259 			int nh_len)
260 {
261 	struct icmp6hdr *icmp = icmp6_hdr(skb);
262 
263 	/* The ICMPv6 type and code fields use the 16-bit transport port
264 	 * fields, so we need to store them in 16-bit network byte order.
265 	 */
266 	key->ipv6.tp.src = htons(icmp->icmp6_type);
267 	key->ipv6.tp.dst = htons(icmp->icmp6_code);
268 
269 	if (icmp->icmp6_code == 0 &&
270 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
271 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
272 		int icmp_len = skb->len - skb_transport_offset(skb);
273 		struct nd_msg *nd;
274 		int offset;
275 
276 		/* In order to process neighbor discovery options, we need the
277 		 * entire packet.
278 		 */
279 		if (unlikely(icmp_len < sizeof(*nd)))
280 			return 0;
281 
282 		if (unlikely(skb_linearize(skb)))
283 			return -ENOMEM;
284 
285 		nd = (struct nd_msg *)skb_transport_header(skb);
286 		key->ipv6.nd.target = nd->target;
287 
288 		icmp_len -= sizeof(*nd);
289 		offset = 0;
290 		while (icmp_len >= 8) {
291 			struct nd_opt_hdr *nd_opt =
292 				 (struct nd_opt_hdr *)(nd->opt + offset);
293 			int opt_len = nd_opt->nd_opt_len * 8;
294 
295 			if (unlikely(!opt_len || opt_len > icmp_len))
296 				return 0;
297 
298 			/* Store the link layer address if the appropriate
299 			 * option is provided.  It is considered an error if
300 			 * the same link layer option is specified twice.
301 			 */
302 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
303 			    && opt_len == 8) {
304 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
305 					goto invalid;
306 				memcpy(key->ipv6.nd.sll,
307 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
308 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
309 				   && opt_len == 8) {
310 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
311 					goto invalid;
312 				memcpy(key->ipv6.nd.tll,
313 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
314 			}
315 
316 			icmp_len -= opt_len;
317 			offset += opt_len;
318 		}
319 	}
320 
321 	return 0;
322 
323 invalid:
324 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
325 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
326 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
327 
328 	return 0;
329 }
330 
331 /**
332  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
333  * @skb: sk_buff that contains the frame, with skb->data pointing to the
334  * Ethernet header
335  * @in_port: port number on which @skb was received.
336  * @key: output flow key
337  *
338  * The caller must ensure that skb->len >= ETH_HLEN.
339  *
340  * Returns 0 if successful, otherwise a negative errno value.
341  *
342  * Initializes @skb header pointers as follows:
343  *
344  *    - skb->mac_header: the Ethernet header.
345  *
346  *    - skb->network_header: just past the Ethernet header, or just past the
347  *      VLAN header, to the first byte of the Ethernet payload.
348  *
349  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
350  *      on output, then just past the IP header, if one is present and
351  *      of a correct length, otherwise the same as skb->network_header.
352  *      For other key->eth.type values it is left untouched.
353  */
354 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
355 {
356 	int error;
357 	struct ethhdr *eth;
358 
359 	memset(key, 0, sizeof(*key));
360 
361 	key->phy.priority = skb->priority;
362 	if (OVS_CB(skb)->tun_key)
363 		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
364 	key->phy.in_port = in_port;
365 	key->phy.skb_mark = skb->mark;
366 
367 	skb_reset_mac_header(skb);
368 
369 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
370 	 * header in the linear data area.
371 	 */
372 	eth = eth_hdr(skb);
373 	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
374 	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
375 
376 	__skb_pull(skb, 2 * ETH_ALEN);
377 	/* We are going to push all headers that we pull, so no need to
378 	 * update skb->csum here.
379 	 */
380 
381 	if (vlan_tx_tag_present(skb))
382 		key->eth.tci = htons(skb->vlan_tci);
383 	else if (eth->h_proto == htons(ETH_P_8021Q))
384 		if (unlikely(parse_vlan(skb, key)))
385 			return -ENOMEM;
386 
387 	key->eth.type = parse_ethertype(skb);
388 	if (unlikely(key->eth.type == htons(0)))
389 		return -ENOMEM;
390 
391 	skb_reset_network_header(skb);
392 	__skb_push(skb, skb->data - skb_mac_header(skb));
393 
394 	/* Network layer. */
395 	if (key->eth.type == htons(ETH_P_IP)) {
396 		struct iphdr *nh;
397 		__be16 offset;
398 
399 		error = check_iphdr(skb);
400 		if (unlikely(error)) {
401 			if (error == -EINVAL) {
402 				skb->transport_header = skb->network_header;
403 				error = 0;
404 			}
405 			return error;
406 		}
407 
408 		nh = ip_hdr(skb);
409 		key->ipv4.addr.src = nh->saddr;
410 		key->ipv4.addr.dst = nh->daddr;
411 
412 		key->ip.proto = nh->protocol;
413 		key->ip.tos = nh->tos;
414 		key->ip.ttl = nh->ttl;
415 
416 		offset = nh->frag_off & htons(IP_OFFSET);
417 		if (offset) {
418 			key->ip.frag = OVS_FRAG_TYPE_LATER;
419 			return 0;
420 		}
421 		if (nh->frag_off & htons(IP_MF) ||
422 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
423 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
424 
425 		/* Transport layer. */
426 		if (key->ip.proto == IPPROTO_TCP) {
427 			if (tcphdr_ok(skb)) {
428 				struct tcphdr *tcp = tcp_hdr(skb);
429 				key->ipv4.tp.src = tcp->source;
430 				key->ipv4.tp.dst = tcp->dest;
431 			}
432 		} else if (key->ip.proto == IPPROTO_UDP) {
433 			if (udphdr_ok(skb)) {
434 				struct udphdr *udp = udp_hdr(skb);
435 				key->ipv4.tp.src = udp->source;
436 				key->ipv4.tp.dst = udp->dest;
437 			}
438 		} else if (key->ip.proto == IPPROTO_SCTP) {
439 			if (sctphdr_ok(skb)) {
440 				struct sctphdr *sctp = sctp_hdr(skb);
441 				key->ipv4.tp.src = sctp->source;
442 				key->ipv4.tp.dst = sctp->dest;
443 			}
444 		} else if (key->ip.proto == IPPROTO_ICMP) {
445 			if (icmphdr_ok(skb)) {
446 				struct icmphdr *icmp = icmp_hdr(skb);
447 				/* The ICMP type and code fields use the 16-bit
448 				 * transport port fields, so we need to store
449 				 * them in 16-bit network byte order. */
450 				key->ipv4.tp.src = htons(icmp->type);
451 				key->ipv4.tp.dst = htons(icmp->code);
452 			}
453 		}
454 
455 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
456 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
457 		struct arp_eth_header *arp;
458 
459 		arp = (struct arp_eth_header *)skb_network_header(skb);
460 
461 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
462 				&& arp->ar_pro == htons(ETH_P_IP)
463 				&& arp->ar_hln == ETH_ALEN
464 				&& arp->ar_pln == 4) {
465 
466 			/* We only match on the lower 8 bits of the opcode. */
467 			if (ntohs(arp->ar_op) <= 0xff)
468 				key->ip.proto = ntohs(arp->ar_op);
469 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
470 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
471 			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
472 			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
473 		}
474 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
475 		int nh_len;             /* IPv6 Header + Extensions */
476 
477 		nh_len = parse_ipv6hdr(skb, key);
478 		if (unlikely(nh_len < 0)) {
479 			if (nh_len == -EINVAL) {
480 				skb->transport_header = skb->network_header;
481 				error = 0;
482 			} else {
483 				error = nh_len;
484 			}
485 			return error;
486 		}
487 
488 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
489 			return 0;
490 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
491 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
492 
493 		/* Transport layer. */
494 		if (key->ip.proto == NEXTHDR_TCP) {
495 			if (tcphdr_ok(skb)) {
496 				struct tcphdr *tcp = tcp_hdr(skb);
497 				key->ipv6.tp.src = tcp->source;
498 				key->ipv6.tp.dst = tcp->dest;
499 			}
500 		} else if (key->ip.proto == NEXTHDR_UDP) {
501 			if (udphdr_ok(skb)) {
502 				struct udphdr *udp = udp_hdr(skb);
503 				key->ipv6.tp.src = udp->source;
504 				key->ipv6.tp.dst = udp->dest;
505 			}
506 		} else if (key->ip.proto == NEXTHDR_SCTP) {
507 			if (sctphdr_ok(skb)) {
508 				struct sctphdr *sctp = sctp_hdr(skb);
509 				key->ipv6.tp.src = sctp->source;
510 				key->ipv6.tp.dst = sctp->dest;
511 			}
512 		} else if (key->ip.proto == NEXTHDR_ICMP) {
513 			if (icmp6hdr_ok(skb)) {
514 				error = parse_icmpv6(skb, key, nh_len);
515 				if (error)
516 					return error;
517 			}
518 		}
519 	}
520 
521 	return 0;
522 }
523