xref: /openbmc/linux/net/openvswitch/flow.c (revision 206204a1)
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/smp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
44 #include <net/ip.h>
45 #include <net/ip_tunnels.h>
46 #include <net/ipv6.h>
47 #include <net/ndisc.h>
48 
49 u64 ovs_flow_used_time(unsigned long flow_jiffies)
50 {
51 	struct timespec cur_ts;
52 	u64 cur_ms, idle_ms;
53 
54 	ktime_get_ts(&cur_ts);
55 	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
56 	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
57 		 cur_ts.tv_nsec / NSEC_PER_MSEC;
58 
59 	return cur_ms - idle_ms;
60 }
61 
62 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
63 
64 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
65 {
66 	struct flow_stats *stats;
67 	__be16 tcp_flags = flow->key.tp.flags;
68 	int node = numa_node_id();
69 
70 	stats = rcu_dereference(flow->stats[node]);
71 
72 	/* Check if already have node-specific stats. */
73 	if (likely(stats)) {
74 		spin_lock(&stats->lock);
75 		/* Mark if we write on the pre-allocated stats. */
76 		if (node == 0 && unlikely(flow->stats_last_writer != node))
77 			flow->stats_last_writer = node;
78 	} else {
79 		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
80 		spin_lock(&stats->lock);
81 
82 		/* If the current NUMA-node is the only writer on the
83 		 * pre-allocated stats keep using them.
84 		 */
85 		if (unlikely(flow->stats_last_writer != node)) {
86 			/* A previous locker may have already allocated the
87 			 * stats, so we need to check again.  If node-specific
88 			 * stats were already allocated, we update the pre-
89 			 * allocated stats as we have already locked them.
90 			 */
91 			if (likely(flow->stats_last_writer != NUMA_NO_NODE)
92 			    && likely(!rcu_dereference(flow->stats[node]))) {
93 				/* Try to allocate node-specific stats. */
94 				struct flow_stats *new_stats;
95 
96 				new_stats =
97 					kmem_cache_alloc_node(flow_stats_cache,
98 							      GFP_THISNODE |
99 							      __GFP_NOMEMALLOC,
100 							      node);
101 				if (likely(new_stats)) {
102 					new_stats->used = jiffies;
103 					new_stats->packet_count = 1;
104 					new_stats->byte_count = skb->len;
105 					new_stats->tcp_flags = tcp_flags;
106 					spin_lock_init(&new_stats->lock);
107 
108 					rcu_assign_pointer(flow->stats[node],
109 							   new_stats);
110 					goto unlock;
111 				}
112 			}
113 			flow->stats_last_writer = node;
114 		}
115 	}
116 
117 	stats->used = jiffies;
118 	stats->packet_count++;
119 	stats->byte_count += skb->len;
120 	stats->tcp_flags |= tcp_flags;
121 unlock:
122 	spin_unlock(&stats->lock);
123 }
124 
125 /* Must be called with rcu_read_lock or ovs_mutex. */
126 void ovs_flow_stats_get(const struct sw_flow *flow,
127 			struct ovs_flow_stats *ovs_stats,
128 			unsigned long *used, __be16 *tcp_flags)
129 {
130 	int node;
131 
132 	*used = 0;
133 	*tcp_flags = 0;
134 	memset(ovs_stats, 0, sizeof(*ovs_stats));
135 
136 	for_each_node(node) {
137 		struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
138 
139 		if (stats) {
140 			/* Local CPU may write on non-local stats, so we must
141 			 * block bottom-halves here.
142 			 */
143 			spin_lock_bh(&stats->lock);
144 			if (!*used || time_after(stats->used, *used))
145 				*used = stats->used;
146 			*tcp_flags |= stats->tcp_flags;
147 			ovs_stats->n_packets += stats->packet_count;
148 			ovs_stats->n_bytes += stats->byte_count;
149 			spin_unlock_bh(&stats->lock);
150 		}
151 	}
152 }
153 
154 /* Called with ovs_mutex. */
155 void ovs_flow_stats_clear(struct sw_flow *flow)
156 {
157 	int node;
158 
159 	for_each_node(node) {
160 		struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
161 
162 		if (stats) {
163 			spin_lock_bh(&stats->lock);
164 			stats->used = 0;
165 			stats->packet_count = 0;
166 			stats->byte_count = 0;
167 			stats->tcp_flags = 0;
168 			spin_unlock_bh(&stats->lock);
169 		}
170 	}
171 }
172 
173 static int check_header(struct sk_buff *skb, int len)
174 {
175 	if (unlikely(skb->len < len))
176 		return -EINVAL;
177 	if (unlikely(!pskb_may_pull(skb, len)))
178 		return -ENOMEM;
179 	return 0;
180 }
181 
182 static bool arphdr_ok(struct sk_buff *skb)
183 {
184 	return pskb_may_pull(skb, skb_network_offset(skb) +
185 				  sizeof(struct arp_eth_header));
186 }
187 
188 static int check_iphdr(struct sk_buff *skb)
189 {
190 	unsigned int nh_ofs = skb_network_offset(skb);
191 	unsigned int ip_len;
192 	int err;
193 
194 	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
195 	if (unlikely(err))
196 		return err;
197 
198 	ip_len = ip_hdrlen(skb);
199 	if (unlikely(ip_len < sizeof(struct iphdr) ||
200 		     skb->len < nh_ofs + ip_len))
201 		return -EINVAL;
202 
203 	skb_set_transport_header(skb, nh_ofs + ip_len);
204 	return 0;
205 }
206 
207 static bool tcphdr_ok(struct sk_buff *skb)
208 {
209 	int th_ofs = skb_transport_offset(skb);
210 	int tcp_len;
211 
212 	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
213 		return false;
214 
215 	tcp_len = tcp_hdrlen(skb);
216 	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
217 		     skb->len < th_ofs + tcp_len))
218 		return false;
219 
220 	return true;
221 }
222 
223 static bool udphdr_ok(struct sk_buff *skb)
224 {
225 	return pskb_may_pull(skb, skb_transport_offset(skb) +
226 				  sizeof(struct udphdr));
227 }
228 
229 static bool sctphdr_ok(struct sk_buff *skb)
230 {
231 	return pskb_may_pull(skb, skb_transport_offset(skb) +
232 				  sizeof(struct sctphdr));
233 }
234 
235 static bool icmphdr_ok(struct sk_buff *skb)
236 {
237 	return pskb_may_pull(skb, skb_transport_offset(skb) +
238 				  sizeof(struct icmphdr));
239 }
240 
241 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
242 {
243 	unsigned int nh_ofs = skb_network_offset(skb);
244 	unsigned int nh_len;
245 	int payload_ofs;
246 	struct ipv6hdr *nh;
247 	uint8_t nexthdr;
248 	__be16 frag_off;
249 	int err;
250 
251 	err = check_header(skb, nh_ofs + sizeof(*nh));
252 	if (unlikely(err))
253 		return err;
254 
255 	nh = ipv6_hdr(skb);
256 	nexthdr = nh->nexthdr;
257 	payload_ofs = (u8 *)(nh + 1) - skb->data;
258 
259 	key->ip.proto = NEXTHDR_NONE;
260 	key->ip.tos = ipv6_get_dsfield(nh);
261 	key->ip.ttl = nh->hop_limit;
262 	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
263 	key->ipv6.addr.src = nh->saddr;
264 	key->ipv6.addr.dst = nh->daddr;
265 
266 	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
267 	if (unlikely(payload_ofs < 0))
268 		return -EINVAL;
269 
270 	if (frag_off) {
271 		if (frag_off & htons(~0x7))
272 			key->ip.frag = OVS_FRAG_TYPE_LATER;
273 		else
274 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
275 	}
276 
277 	nh_len = payload_ofs - nh_ofs;
278 	skb_set_transport_header(skb, nh_ofs + nh_len);
279 	key->ip.proto = nexthdr;
280 	return nh_len;
281 }
282 
283 static bool icmp6hdr_ok(struct sk_buff *skb)
284 {
285 	return pskb_may_pull(skb, skb_transport_offset(skb) +
286 				  sizeof(struct icmp6hdr));
287 }
288 
289 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
290 {
291 	struct qtag_prefix {
292 		__be16 eth_type; /* ETH_P_8021Q */
293 		__be16 tci;
294 	};
295 	struct qtag_prefix *qp;
296 
297 	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
298 		return 0;
299 
300 	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
301 					 sizeof(__be16))))
302 		return -ENOMEM;
303 
304 	qp = (struct qtag_prefix *) skb->data;
305 	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
306 	__skb_pull(skb, sizeof(struct qtag_prefix));
307 
308 	return 0;
309 }
310 
311 static __be16 parse_ethertype(struct sk_buff *skb)
312 {
313 	struct llc_snap_hdr {
314 		u8  dsap;  /* Always 0xAA */
315 		u8  ssap;  /* Always 0xAA */
316 		u8  ctrl;
317 		u8  oui[3];
318 		__be16 ethertype;
319 	};
320 	struct llc_snap_hdr *llc;
321 	__be16 proto;
322 
323 	proto = *(__be16 *) skb->data;
324 	__skb_pull(skb, sizeof(__be16));
325 
326 	if (ntohs(proto) >= ETH_P_802_3_MIN)
327 		return proto;
328 
329 	if (skb->len < sizeof(struct llc_snap_hdr))
330 		return htons(ETH_P_802_2);
331 
332 	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
333 		return htons(0);
334 
335 	llc = (struct llc_snap_hdr *) skb->data;
336 	if (llc->dsap != LLC_SAP_SNAP ||
337 	    llc->ssap != LLC_SAP_SNAP ||
338 	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
339 		return htons(ETH_P_802_2);
340 
341 	__skb_pull(skb, sizeof(struct llc_snap_hdr));
342 
343 	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
344 		return llc->ethertype;
345 
346 	return htons(ETH_P_802_2);
347 }
348 
349 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
350 			int nh_len)
351 {
352 	struct icmp6hdr *icmp = icmp6_hdr(skb);
353 
354 	/* The ICMPv6 type and code fields use the 16-bit transport port
355 	 * fields, so we need to store them in 16-bit network byte order.
356 	 */
357 	key->tp.src = htons(icmp->icmp6_type);
358 	key->tp.dst = htons(icmp->icmp6_code);
359 
360 	if (icmp->icmp6_code == 0 &&
361 	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
362 	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
363 		int icmp_len = skb->len - skb_transport_offset(skb);
364 		struct nd_msg *nd;
365 		int offset;
366 
367 		/* In order to process neighbor discovery options, we need the
368 		 * entire packet.
369 		 */
370 		if (unlikely(icmp_len < sizeof(*nd)))
371 			return 0;
372 
373 		if (unlikely(skb_linearize(skb)))
374 			return -ENOMEM;
375 
376 		nd = (struct nd_msg *)skb_transport_header(skb);
377 		key->ipv6.nd.target = nd->target;
378 
379 		icmp_len -= sizeof(*nd);
380 		offset = 0;
381 		while (icmp_len >= 8) {
382 			struct nd_opt_hdr *nd_opt =
383 				 (struct nd_opt_hdr *)(nd->opt + offset);
384 			int opt_len = nd_opt->nd_opt_len * 8;
385 
386 			if (unlikely(!opt_len || opt_len > icmp_len))
387 				return 0;
388 
389 			/* Store the link layer address if the appropriate
390 			 * option is provided.  It is considered an error if
391 			 * the same link layer option is specified twice.
392 			 */
393 			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
394 			    && opt_len == 8) {
395 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
396 					goto invalid;
397 				ether_addr_copy(key->ipv6.nd.sll,
398 						&nd->opt[offset+sizeof(*nd_opt)]);
399 			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
400 				   && opt_len == 8) {
401 				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
402 					goto invalid;
403 				ether_addr_copy(key->ipv6.nd.tll,
404 						&nd->opt[offset+sizeof(*nd_opt)]);
405 			}
406 
407 			icmp_len -= opt_len;
408 			offset += opt_len;
409 		}
410 	}
411 
412 	return 0;
413 
414 invalid:
415 	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
416 	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
417 	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
418 
419 	return 0;
420 }
421 
422 /**
423  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
424  * @skb: sk_buff that contains the frame, with skb->data pointing to the
425  * Ethernet header
426  * @in_port: port number on which @skb was received.
427  * @key: output flow key
428  *
429  * The caller must ensure that skb->len >= ETH_HLEN.
430  *
431  * Returns 0 if successful, otherwise a negative errno value.
432  *
433  * Initializes @skb header pointers as follows:
434  *
435  *    - skb->mac_header: the Ethernet header.
436  *
437  *    - skb->network_header: just past the Ethernet header, or just past the
438  *      VLAN header, to the first byte of the Ethernet payload.
439  *
440  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
441  *      on output, then just past the IP header, if one is present and
442  *      of a correct length, otherwise the same as skb->network_header.
443  *      For other key->eth.type values it is left untouched.
444  */
445 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
446 {
447 	int error;
448 	struct ethhdr *eth;
449 
450 	memset(key, 0, sizeof(*key));
451 
452 	key->phy.priority = skb->priority;
453 	if (OVS_CB(skb)->tun_key)
454 		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
455 	key->phy.in_port = in_port;
456 	key->phy.skb_mark = skb->mark;
457 
458 	skb_reset_mac_header(skb);
459 
460 	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
461 	 * header in the linear data area.
462 	 */
463 	eth = eth_hdr(skb);
464 	ether_addr_copy(key->eth.src, eth->h_source);
465 	ether_addr_copy(key->eth.dst, eth->h_dest);
466 
467 	__skb_pull(skb, 2 * ETH_ALEN);
468 	/* We are going to push all headers that we pull, so no need to
469 	 * update skb->csum here.
470 	 */
471 
472 	if (vlan_tx_tag_present(skb))
473 		key->eth.tci = htons(skb->vlan_tci);
474 	else if (eth->h_proto == htons(ETH_P_8021Q))
475 		if (unlikely(parse_vlan(skb, key)))
476 			return -ENOMEM;
477 
478 	key->eth.type = parse_ethertype(skb);
479 	if (unlikely(key->eth.type == htons(0)))
480 		return -ENOMEM;
481 
482 	skb_reset_network_header(skb);
483 	__skb_push(skb, skb->data - skb_mac_header(skb));
484 
485 	/* Network layer. */
486 	if (key->eth.type == htons(ETH_P_IP)) {
487 		struct iphdr *nh;
488 		__be16 offset;
489 
490 		error = check_iphdr(skb);
491 		if (unlikely(error)) {
492 			if (error == -EINVAL) {
493 				skb->transport_header = skb->network_header;
494 				error = 0;
495 			}
496 			return error;
497 		}
498 
499 		nh = ip_hdr(skb);
500 		key->ipv4.addr.src = nh->saddr;
501 		key->ipv4.addr.dst = nh->daddr;
502 
503 		key->ip.proto = nh->protocol;
504 		key->ip.tos = nh->tos;
505 		key->ip.ttl = nh->ttl;
506 
507 		offset = nh->frag_off & htons(IP_OFFSET);
508 		if (offset) {
509 			key->ip.frag = OVS_FRAG_TYPE_LATER;
510 			return 0;
511 		}
512 		if (nh->frag_off & htons(IP_MF) ||
513 			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
514 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
515 
516 		/* Transport layer. */
517 		if (key->ip.proto == IPPROTO_TCP) {
518 			if (tcphdr_ok(skb)) {
519 				struct tcphdr *tcp = tcp_hdr(skb);
520 				key->tp.src = tcp->source;
521 				key->tp.dst = tcp->dest;
522 				key->tp.flags = TCP_FLAGS_BE16(tcp);
523 			}
524 		} else if (key->ip.proto == IPPROTO_UDP) {
525 			if (udphdr_ok(skb)) {
526 				struct udphdr *udp = udp_hdr(skb);
527 				key->tp.src = udp->source;
528 				key->tp.dst = udp->dest;
529 			}
530 		} else if (key->ip.proto == IPPROTO_SCTP) {
531 			if (sctphdr_ok(skb)) {
532 				struct sctphdr *sctp = sctp_hdr(skb);
533 				key->tp.src = sctp->source;
534 				key->tp.dst = sctp->dest;
535 			}
536 		} else if (key->ip.proto == IPPROTO_ICMP) {
537 			if (icmphdr_ok(skb)) {
538 				struct icmphdr *icmp = icmp_hdr(skb);
539 				/* The ICMP type and code fields use the 16-bit
540 				 * transport port fields, so we need to store
541 				 * them in 16-bit network byte order. */
542 				key->tp.src = htons(icmp->type);
543 				key->tp.dst = htons(icmp->code);
544 			}
545 		}
546 
547 	} else if ((key->eth.type == htons(ETH_P_ARP) ||
548 		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
549 		struct arp_eth_header *arp;
550 
551 		arp = (struct arp_eth_header *)skb_network_header(skb);
552 
553 		if (arp->ar_hrd == htons(ARPHRD_ETHER)
554 				&& arp->ar_pro == htons(ETH_P_IP)
555 				&& arp->ar_hln == ETH_ALEN
556 				&& arp->ar_pln == 4) {
557 
558 			/* We only match on the lower 8 bits of the opcode. */
559 			if (ntohs(arp->ar_op) <= 0xff)
560 				key->ip.proto = ntohs(arp->ar_op);
561 			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
562 			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
563 			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
564 			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
565 		}
566 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
567 		int nh_len;             /* IPv6 Header + Extensions */
568 
569 		nh_len = parse_ipv6hdr(skb, key);
570 		if (unlikely(nh_len < 0)) {
571 			if (nh_len == -EINVAL) {
572 				skb->transport_header = skb->network_header;
573 				error = 0;
574 			} else {
575 				error = nh_len;
576 			}
577 			return error;
578 		}
579 
580 		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
581 			return 0;
582 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
583 			key->ip.frag = OVS_FRAG_TYPE_FIRST;
584 
585 		/* Transport layer. */
586 		if (key->ip.proto == NEXTHDR_TCP) {
587 			if (tcphdr_ok(skb)) {
588 				struct tcphdr *tcp = tcp_hdr(skb);
589 				key->tp.src = tcp->source;
590 				key->tp.dst = tcp->dest;
591 				key->tp.flags = TCP_FLAGS_BE16(tcp);
592 			}
593 		} else if (key->ip.proto == NEXTHDR_UDP) {
594 			if (udphdr_ok(skb)) {
595 				struct udphdr *udp = udp_hdr(skb);
596 				key->tp.src = udp->source;
597 				key->tp.dst = udp->dest;
598 			}
599 		} else if (key->ip.proto == NEXTHDR_SCTP) {
600 			if (sctphdr_ok(skb)) {
601 				struct sctphdr *sctp = sctp_hdr(skb);
602 				key->tp.src = sctp->source;
603 				key->tp.dst = sctp->dest;
604 			}
605 		} else if (key->ip.proto == NEXTHDR_ICMP) {
606 			if (icmp6hdr_ok(skb)) {
607 				error = parse_icmpv6(skb, key, nh_len);
608 				if (error)
609 					return error;
610 			}
611 		}
612 	}
613 
614 	return 0;
615 }
616