xref: /openbmc/linux/net/openvswitch/flow.c (revision e6445719)
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_OFFSET 13
62 #define TCP_FLAG_MASK 0x3f
63 
64 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
65 {
66 	u8 tcp_flags = 0;
67 
68 	if ((flow->key.eth.type == htons(ETH_P_IP) ||
69 	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
70 	    flow->key.ip.proto == IPPROTO_TCP &&
71 	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
72 		u8 *tcp = (u8 *)tcp_hdr(skb);
73 		tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
74 	}
75 
76 	spin_lock(&flow->lock);
77 	flow->used = jiffies;
78 	flow->packet_count++;
79 	flow->byte_count += skb->len;
80 	flow->tcp_flags |= tcp_flags;
81 	spin_unlock(&flow->lock);
82 }
83 
84 static int check_header(struct sk_buff *skb, int len)
85 {
86 	if (unlikely(skb->len < len))
87 		return -EINVAL;
88 	if (unlikely(!pskb_may_pull(skb, len)))
89 		return -ENOMEM;
90 	return 0;
91 }
92 
93 static bool arphdr_ok(struct sk_buff *skb)
94 {
95 	return pskb_may_pull(skb, skb_network_offset(skb) +
96 				  sizeof(struct arp_eth_header));
97 }
98 
99 static int check_iphdr(struct sk_buff *skb)
100 {
101 	unsigned int nh_ofs = skb_network_offset(skb);
102 	unsigned int ip_len;
103 	int err;
104 
105 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
106 	if (unlikely(err))
107 		return err;
108 
109 	ip_len = ip_hdrlen(skb);
110 	if (unlikely(ip_len < sizeof(struct iphdr) ||
111 		     skb->len < nh_ofs + ip_len))
112 		return -EINVAL;
113 
114 	skb_set_transport_header(skb, nh_ofs + ip_len);
115 	return 0;
116 }
117 
118 static bool tcphdr_ok(struct sk_buff *skb)
119 {
120 	int th_ofs = skb_transport_offset(skb);
121 	int tcp_len;
122 
123 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
124 		return false;
125 
126 	tcp_len = tcp_hdrlen(skb);
127 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
128 		     skb->len < th_ofs + tcp_len))
129 		return false;
130 
131 	return true;
132 }
133 
134 static bool udphdr_ok(struct sk_buff *skb)
135 {
136 	return pskb_may_pull(skb, skb_transport_offset(skb) +
137 				  sizeof(struct udphdr));
138 }
139 
140 static bool sctphdr_ok(struct sk_buff *skb)
141 {
142 	return pskb_may_pull(skb, skb_transport_offset(skb) +
143 				  sizeof(struct sctphdr));
144 }
145 
146 static bool icmphdr_ok(struct sk_buff *skb)
147 {
148 	return pskb_may_pull(skb, skb_transport_offset(skb) +
149 				  sizeof(struct icmphdr));
150 }
151 
152 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
153 {
154 	unsigned int nh_ofs = skb_network_offset(skb);
155 	unsigned int nh_len;
156 	int payload_ofs;
157 	struct ipv6hdr *nh;
158 	uint8_t nexthdr;
159 	__be16 frag_off;
160 	int err;
161 
162 	err = check_header(skb, nh_ofs + sizeof(*nh));
163 	if (unlikely(err))
164 		return err;
165 
166 	nh = ipv6_hdr(skb);
167 	nexthdr = nh->nexthdr;
168 	payload_ofs = (u8 *)(nh + 1) - skb->data;
169 
170 	key->ip.proto = NEXTHDR_NONE;
171 	key->ip.tos = ipv6_get_dsfield(nh);
172 	key->ip.ttl = nh->hop_limit;
173 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
174 	key->ipv6.addr.src = nh->saddr;
175 	key->ipv6.addr.dst = nh->daddr;
176 
177 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
178 	if (unlikely(payload_ofs < 0))
179 		return -EINVAL;
180 
181 	if (frag_off) {
182 		if (frag_off & htons(~0x7))
183 			key->ip.frag = OVS_FRAG_TYPE_LATER;
184 		else
185 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
186 	}
187 
188 	nh_len = payload_ofs - nh_ofs;
189 	skb_set_transport_header(skb, nh_ofs + nh_len);
190 	key->ip.proto = nexthdr;
191 	return nh_len;
192 }
193 
194 static bool icmp6hdr_ok(struct sk_buff *skb)
195 {
196 	return pskb_may_pull(skb, skb_transport_offset(skb) +
197 				  sizeof(struct icmp6hdr));
198 }
199 
200 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
201 {
202 	struct qtag_prefix {
203 		__be16 eth_type; /* ETH_P_8021Q */
204 		__be16 tci;
205 	};
206 	struct qtag_prefix *qp;
207 
208 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
209 		return 0;
210 
211 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
212 					 sizeof(__be16))))
213 		return -ENOMEM;
214 
215 	qp = (struct qtag_prefix *) skb->data;
216 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
217 	__skb_pull(skb, sizeof(struct qtag_prefix));
218 
219 	return 0;
220 }
221 
222 static __be16 parse_ethertype(struct sk_buff *skb)
223 {
224 	struct llc_snap_hdr {
225 		u8  dsap;  /* Always 0xAA */
226 		u8  ssap;  /* Always 0xAA */
227 		u8  ctrl;
228 		u8  oui[3];
229 		__be16 ethertype;
230 	};
231 	struct llc_snap_hdr *llc;
232 	__be16 proto;
233 
234 	proto = *(__be16 *) skb->data;
235 	__skb_pull(skb, sizeof(__be16));
236 
237 	if (ntohs(proto) >= ETH_P_802_3_MIN)
238 		return proto;
239 
240 	if (skb->len < sizeof(struct llc_snap_hdr))
241 		return htons(ETH_P_802_2);
242 
243 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
244 		return htons(0);
245 
246 	llc = (struct llc_snap_hdr *) skb->data;
247 	if (llc->dsap != LLC_SAP_SNAP ||
248 	    llc->ssap != LLC_SAP_SNAP ||
249 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
250 		return htons(ETH_P_802_2);
251 
252 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
253 
254 	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
255 		return llc->ethertype;
256 
257 	return htons(ETH_P_802_2);
258 }
259 
260 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
261 			int nh_len)
262 {
263 	struct icmp6hdr *icmp = icmp6_hdr(skb);
264 
265 	/* The ICMPv6 type and code fields use the 16-bit transport port
266 	 * fields, so we need to store them in 16-bit network byte order.
267 	 */
268 	key->ipv6.tp.src = htons(icmp->icmp6_type);
269 	key->ipv6.tp.dst = htons(icmp->icmp6_code);
270 
271 	if (icmp->icmp6_code == 0 &&
272 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
273 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
274 		int icmp_len = skb->len - skb_transport_offset(skb);
275 		struct nd_msg *nd;
276 		int offset;
277 
278 		/* In order to process neighbor discovery options, we need the
279 		 * entire packet.
280 		 */
281 		if (unlikely(icmp_len < sizeof(*nd)))
282 			return 0;
283 
284 		if (unlikely(skb_linearize(skb)))
285 			return -ENOMEM;
286 
287 		nd = (struct nd_msg *)skb_transport_header(skb);
288 		key->ipv6.nd.target = nd->target;
289 
290 		icmp_len -= sizeof(*nd);
291 		offset = 0;
292 		while (icmp_len >= 8) {
293 			struct nd_opt_hdr *nd_opt =
294 				 (struct nd_opt_hdr *)(nd->opt + offset);
295 			int opt_len = nd_opt->nd_opt_len * 8;
296 
297 			if (unlikely(!opt_len || opt_len > icmp_len))
298 				return 0;
299 
300 			/* Store the link layer address if the appropriate
301 			 * option is provided.  It is considered an error if
302 			 * the same link layer option is specified twice.
303 			 */
304 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
305 			    && opt_len == 8) {
306 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
307 					goto invalid;
308 				memcpy(key->ipv6.nd.sll,
309 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
310 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
311 				   && opt_len == 8) {
312 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
313 					goto invalid;
314 				memcpy(key->ipv6.nd.tll,
315 				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
316 			}
317 
318 			icmp_len -= opt_len;
319 			offset += opt_len;
320 		}
321 	}
322 
323 	return 0;
324 
325 invalid:
326 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
327 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
328 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
329 
330 	return 0;
331 }
332 
333 /**
334  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
335  * @skb: sk_buff that contains the frame, with skb->data pointing to the
336  * Ethernet header
337  * @in_port: port number on which @skb was received.
338  * @key: output flow key
339  *
340  * The caller must ensure that skb->len >= ETH_HLEN.
341  *
342  * Returns 0 if successful, otherwise a negative errno value.
343  *
344  * Initializes @skb header pointers as follows:
345  *
346  *    - skb->mac_header: the Ethernet header.
347  *
348  *    - skb->network_header: just past the Ethernet header, or just past the
349  *      VLAN header, to the first byte of the Ethernet payload.
350  *
351  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
352  *      on output, then just past the IP header, if one is present and
353  *      of a correct length, otherwise the same as skb->network_header.
354  *      For other key->eth.type values it is left untouched.
355  */
356 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
357 {
358 	int error;
359 	struct ethhdr *eth;
360 
361 	memset(key, 0, sizeof(*key));
362 
363 	key->phy.priority = skb->priority;
364 	if (OVS_CB(skb)->tun_key)
365 		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
366 	key->phy.in_port = in_port;
367 	key->phy.skb_mark = skb->mark;
368 
369 	skb_reset_mac_header(skb);
370 
371 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
372 	 * header in the linear data area.
373 	 */
374 	eth = eth_hdr(skb);
375 	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
376 	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
377 
378 	__skb_pull(skb, 2 * ETH_ALEN);
379 	/* We are going to push all headers that we pull, so no need to
380 	 * update skb->csum here.
381 	 */
382 
383 	if (vlan_tx_tag_present(skb))
384 		key->eth.tci = htons(skb->vlan_tci);
385 	else if (eth->h_proto == htons(ETH_P_8021Q))
386 		if (unlikely(parse_vlan(skb, key)))
387 			return -ENOMEM;
388 
389 	key->eth.type = parse_ethertype(skb);
390 	if (unlikely(key->eth.type == htons(0)))
391 		return -ENOMEM;
392 
393 	skb_reset_network_header(skb);
394 	__skb_push(skb, skb->data - skb_mac_header(skb));
395 
396 	/* Network layer. */
397 	if (key->eth.type == htons(ETH_P_IP)) {
398 		struct iphdr *nh;
399 		__be16 offset;
400 
401 		error = check_iphdr(skb);
402 		if (unlikely(error)) {
403 			if (error == -EINVAL) {
404 				skb->transport_header = skb->network_header;
405 				error = 0;
406 			}
407 			return error;
408 		}
409 
410 		nh = ip_hdr(skb);
411 		key->ipv4.addr.src = nh->saddr;
412 		key->ipv4.addr.dst = nh->daddr;
413 
414 		key->ip.proto = nh->protocol;
415 		key->ip.tos = nh->tos;
416 		key->ip.ttl = nh->ttl;
417 
418 		offset = nh->frag_off & htons(IP_OFFSET);
419 		if (offset) {
420 			key->ip.frag = OVS_FRAG_TYPE_LATER;
421 			return 0;
422 		}
423 		if (nh->frag_off & htons(IP_MF) ||
424 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
425 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
426 
427 		/* Transport layer. */
428 		if (key->ip.proto == IPPROTO_TCP) {
429 			if (tcphdr_ok(skb)) {
430 				struct tcphdr *tcp = tcp_hdr(skb);
431 				key->ipv4.tp.src = tcp->source;
432 				key->ipv4.tp.dst = tcp->dest;
433 			}
434 		} else if (key->ip.proto == IPPROTO_UDP) {
435 			if (udphdr_ok(skb)) {
436 				struct udphdr *udp = udp_hdr(skb);
437 				key->ipv4.tp.src = udp->source;
438 				key->ipv4.tp.dst = udp->dest;
439 			}
440 		} else if (key->ip.proto == IPPROTO_SCTP) {
441 			if (sctphdr_ok(skb)) {
442 				struct sctphdr *sctp = sctp_hdr(skb);
443 				key->ipv4.tp.src = sctp->source;
444 				key->ipv4.tp.dst = sctp->dest;
445 			}
446 		} else if (key->ip.proto == IPPROTO_ICMP) {
447 			if (icmphdr_ok(skb)) {
448 				struct icmphdr *icmp = icmp_hdr(skb);
449 				/* The ICMP type and code fields use the 16-bit
450 				 * transport port fields, so we need to store
451 				 * them in 16-bit network byte order. */
452 				key->ipv4.tp.src = htons(icmp->type);
453 				key->ipv4.tp.dst = htons(icmp->code);
454 			}
455 		}
456 
457 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
458 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
459 		struct arp_eth_header *arp;
460 
461 		arp = (struct arp_eth_header *)skb_network_header(skb);
462 
463 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
464 				&& arp->ar_pro == htons(ETH_P_IP)
465 				&& arp->ar_hln == ETH_ALEN
466 				&& arp->ar_pln == 4) {
467 
468 			/* We only match on the lower 8 bits of the opcode. */
469 			if (ntohs(arp->ar_op) <= 0xff)
470 				key->ip.proto = ntohs(arp->ar_op);
471 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
472 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
473 			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
474 			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
475 		}
476 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
477 		int nh_len;             /* IPv6 Header + Extensions */
478 
479 		nh_len = parse_ipv6hdr(skb, key);
480 		if (unlikely(nh_len < 0)) {
481 			if (nh_len == -EINVAL) {
482 				skb->transport_header = skb->network_header;
483 				error = 0;
484 			} else {
485 				error = nh_len;
486 			}
487 			return error;
488 		}
489 
490 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
491 			return 0;
492 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
493 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
494 
495 		/* Transport layer. */
496 		if (key->ip.proto == NEXTHDR_TCP) {
497 			if (tcphdr_ok(skb)) {
498 				struct tcphdr *tcp = tcp_hdr(skb);
499 				key->ipv6.tp.src = tcp->source;
500 				key->ipv6.tp.dst = tcp->dest;
501 			}
502 		} else if (key->ip.proto == NEXTHDR_UDP) {
503 			if (udphdr_ok(skb)) {
504 				struct udphdr *udp = udp_hdr(skb);
505 				key->ipv6.tp.src = udp->source;
506 				key->ipv6.tp.dst = udp->dest;
507 			}
508 		} else if (key->ip.proto == NEXTHDR_SCTP) {
509 			if (sctphdr_ok(skb)) {
510 				struct sctphdr *sctp = sctp_hdr(skb);
511 				key->ipv6.tp.src = sctp->source;
512 				key->ipv6.tp.dst = sctp->dest;
513 			}
514 		} else if (key->ip.proto == NEXTHDR_ICMP) {
515 			if (icmp6hdr_ok(skb)) {
516 				error = parse_icmpv6(skb, key, nh_len);
517 				if (error)
518 					return error;
519 			}
520 		}
521 	}
522 
523 	return 0;
524 }
525