]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - net/openvswitch/flow.c
openvswitch: Per NUMA node flow stats.
[mirror_ubuntu-zesty-kernel.git] / net / openvswitch / flow.c
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 = 0;
68 int node = numa_node_id();
69
70 stats = rcu_dereference(flow->stats[node]);
71
72 if ((flow->key.eth.type == htons(ETH_P_IP) ||
73 flow->key.eth.type == htons(ETH_P_IPV6)) &&
74 flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
75 flow->key.ip.proto == IPPROTO_TCP &&
76 likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
77 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
78 }
79
80 /* Check if already have node-specific stats. */
81 if (likely(stats)) {
82 spin_lock(&stats->lock);
83 /* Mark if we write on the pre-allocated stats. */
84 if (node == 0 && unlikely(flow->stats_last_writer != node))
85 flow->stats_last_writer = node;
86 } else {
87 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
88 spin_lock(&stats->lock);
89
90 /* If the current NUMA-node is the only writer on the
91 * pre-allocated stats keep using them.
92 */
93 if (unlikely(flow->stats_last_writer != node)) {
94 /* A previous locker may have already allocated the
95 * stats, so we need to check again. If node-specific
96 * stats were already allocated, we update the pre-
97 * allocated stats as we have already locked them.
98 */
99 if (likely(flow->stats_last_writer != NUMA_NO_NODE)
100 && likely(!rcu_dereference(flow->stats[node]))) {
101 /* Try to allocate node-specific stats. */
102 struct flow_stats *new_stats;
103
104 new_stats =
105 kmem_cache_alloc_node(flow_stats_cache,
106 GFP_THISNODE |
107 __GFP_NOMEMALLOC,
108 node);
109 if (likely(new_stats)) {
110 new_stats->used = jiffies;
111 new_stats->packet_count = 1;
112 new_stats->byte_count = skb->len;
113 new_stats->tcp_flags = tcp_flags;
114 spin_lock_init(&new_stats->lock);
115
116 rcu_assign_pointer(flow->stats[node],
117 new_stats);
118 goto unlock;
119 }
120 }
121 flow->stats_last_writer = node;
122 }
123 }
124
125 stats->used = jiffies;
126 stats->packet_count++;
127 stats->byte_count += skb->len;
128 stats->tcp_flags |= tcp_flags;
129 unlock:
130 spin_unlock(&stats->lock);
131 }
132
133 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
134 unsigned long *used, __be16 *tcp_flags)
135 {
136 int node;
137
138 *used = 0;
139 *tcp_flags = 0;
140 memset(ovs_stats, 0, sizeof(*ovs_stats));
141
142 for_each_node(node) {
143 struct flow_stats *stats = rcu_dereference(flow->stats[node]);
144
145 if (stats) {
146 /* Local CPU may write on non-local stats, so we must
147 * block bottom-halves here.
148 */
149 spin_lock_bh(&stats->lock);
150 if (!*used || time_after(stats->used, *used))
151 *used = stats->used;
152 *tcp_flags |= stats->tcp_flags;
153 ovs_stats->n_packets += stats->packet_count;
154 ovs_stats->n_bytes += stats->byte_count;
155 spin_unlock_bh(&stats->lock);
156 }
157 }
158 }
159
160 void ovs_flow_stats_clear(struct sw_flow *flow)
161 {
162 int node;
163
164 for_each_node(node) {
165 struct flow_stats *stats = rcu_dereference(flow->stats[node]);
166
167 if (stats) {
168 spin_lock_bh(&stats->lock);
169 stats->used = 0;
170 stats->packet_count = 0;
171 stats->byte_count = 0;
172 stats->tcp_flags = 0;
173 spin_unlock_bh(&stats->lock);
174 }
175 }
176 }
177
178 static int check_header(struct sk_buff *skb, int len)
179 {
180 if (unlikely(skb->len < len))
181 return -EINVAL;
182 if (unlikely(!pskb_may_pull(skb, len)))
183 return -ENOMEM;
184 return 0;
185 }
186
187 static bool arphdr_ok(struct sk_buff *skb)
188 {
189 return pskb_may_pull(skb, skb_network_offset(skb) +
190 sizeof(struct arp_eth_header));
191 }
192
193 static int check_iphdr(struct sk_buff *skb)
194 {
195 unsigned int nh_ofs = skb_network_offset(skb);
196 unsigned int ip_len;
197 int err;
198
199 err = check_header(skb, nh_ofs + sizeof(struct iphdr));
200 if (unlikely(err))
201 return err;
202
203 ip_len = ip_hdrlen(skb);
204 if (unlikely(ip_len < sizeof(struct iphdr) ||
205 skb->len < nh_ofs + ip_len))
206 return -EINVAL;
207
208 skb_set_transport_header(skb, nh_ofs + ip_len);
209 return 0;
210 }
211
212 static bool tcphdr_ok(struct sk_buff *skb)
213 {
214 int th_ofs = skb_transport_offset(skb);
215 int tcp_len;
216
217 if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
218 return false;
219
220 tcp_len = tcp_hdrlen(skb);
221 if (unlikely(tcp_len < sizeof(struct tcphdr) ||
222 skb->len < th_ofs + tcp_len))
223 return false;
224
225 return true;
226 }
227
228 static bool udphdr_ok(struct sk_buff *skb)
229 {
230 return pskb_may_pull(skb, skb_transport_offset(skb) +
231 sizeof(struct udphdr));
232 }
233
234 static bool sctphdr_ok(struct sk_buff *skb)
235 {
236 return pskb_may_pull(skb, skb_transport_offset(skb) +
237 sizeof(struct sctphdr));
238 }
239
240 static bool icmphdr_ok(struct sk_buff *skb)
241 {
242 return pskb_may_pull(skb, skb_transport_offset(skb) +
243 sizeof(struct icmphdr));
244 }
245
246 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
247 {
248 unsigned int nh_ofs = skb_network_offset(skb);
249 unsigned int nh_len;
250 int payload_ofs;
251 struct ipv6hdr *nh;
252 uint8_t nexthdr;
253 __be16 frag_off;
254 int err;
255
256 err = check_header(skb, nh_ofs + sizeof(*nh));
257 if (unlikely(err))
258 return err;
259
260 nh = ipv6_hdr(skb);
261 nexthdr = nh->nexthdr;
262 payload_ofs = (u8 *)(nh + 1) - skb->data;
263
264 key->ip.proto = NEXTHDR_NONE;
265 key->ip.tos = ipv6_get_dsfield(nh);
266 key->ip.ttl = nh->hop_limit;
267 key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
268 key->ipv6.addr.src = nh->saddr;
269 key->ipv6.addr.dst = nh->daddr;
270
271 payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
272 if (unlikely(payload_ofs < 0))
273 return -EINVAL;
274
275 if (frag_off) {
276 if (frag_off & htons(~0x7))
277 key->ip.frag = OVS_FRAG_TYPE_LATER;
278 else
279 key->ip.frag = OVS_FRAG_TYPE_FIRST;
280 }
281
282 nh_len = payload_ofs - nh_ofs;
283 skb_set_transport_header(skb, nh_ofs + nh_len);
284 key->ip.proto = nexthdr;
285 return nh_len;
286 }
287
288 static bool icmp6hdr_ok(struct sk_buff *skb)
289 {
290 return pskb_may_pull(skb, skb_transport_offset(skb) +
291 sizeof(struct icmp6hdr));
292 }
293
294 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
295 {
296 struct qtag_prefix {
297 __be16 eth_type; /* ETH_P_8021Q */
298 __be16 tci;
299 };
300 struct qtag_prefix *qp;
301
302 if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
303 return 0;
304
305 if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
306 sizeof(__be16))))
307 return -ENOMEM;
308
309 qp = (struct qtag_prefix *) skb->data;
310 key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
311 __skb_pull(skb, sizeof(struct qtag_prefix));
312
313 return 0;
314 }
315
316 static __be16 parse_ethertype(struct sk_buff *skb)
317 {
318 struct llc_snap_hdr {
319 u8 dsap; /* Always 0xAA */
320 u8 ssap; /* Always 0xAA */
321 u8 ctrl;
322 u8 oui[3];
323 __be16 ethertype;
324 };
325 struct llc_snap_hdr *llc;
326 __be16 proto;
327
328 proto = *(__be16 *) skb->data;
329 __skb_pull(skb, sizeof(__be16));
330
331 if (ntohs(proto) >= ETH_P_802_3_MIN)
332 return proto;
333
334 if (skb->len < sizeof(struct llc_snap_hdr))
335 return htons(ETH_P_802_2);
336
337 if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
338 return htons(0);
339
340 llc = (struct llc_snap_hdr *) skb->data;
341 if (llc->dsap != LLC_SAP_SNAP ||
342 llc->ssap != LLC_SAP_SNAP ||
343 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
344 return htons(ETH_P_802_2);
345
346 __skb_pull(skb, sizeof(struct llc_snap_hdr));
347
348 if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
349 return llc->ethertype;
350
351 return htons(ETH_P_802_2);
352 }
353
354 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
355 int nh_len)
356 {
357 struct icmp6hdr *icmp = icmp6_hdr(skb);
358
359 /* The ICMPv6 type and code fields use the 16-bit transport port
360 * fields, so we need to store them in 16-bit network byte order.
361 */
362 key->ipv6.tp.src = htons(icmp->icmp6_type);
363 key->ipv6.tp.dst = htons(icmp->icmp6_code);
364
365 if (icmp->icmp6_code == 0 &&
366 (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
367 icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
368 int icmp_len = skb->len - skb_transport_offset(skb);
369 struct nd_msg *nd;
370 int offset;
371
372 /* In order to process neighbor discovery options, we need the
373 * entire packet.
374 */
375 if (unlikely(icmp_len < sizeof(*nd)))
376 return 0;
377
378 if (unlikely(skb_linearize(skb)))
379 return -ENOMEM;
380
381 nd = (struct nd_msg *)skb_transport_header(skb);
382 key->ipv6.nd.target = nd->target;
383
384 icmp_len -= sizeof(*nd);
385 offset = 0;
386 while (icmp_len >= 8) {
387 struct nd_opt_hdr *nd_opt =
388 (struct nd_opt_hdr *)(nd->opt + offset);
389 int opt_len = nd_opt->nd_opt_len * 8;
390
391 if (unlikely(!opt_len || opt_len > icmp_len))
392 return 0;
393
394 /* Store the link layer address if the appropriate
395 * option is provided. It is considered an error if
396 * the same link layer option is specified twice.
397 */
398 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
399 && opt_len == 8) {
400 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
401 goto invalid;
402 ether_addr_copy(key->ipv6.nd.sll,
403 &nd->opt[offset+sizeof(*nd_opt)]);
404 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
405 && opt_len == 8) {
406 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
407 goto invalid;
408 ether_addr_copy(key->ipv6.nd.tll,
409 &nd->opt[offset+sizeof(*nd_opt)]);
410 }
411
412 icmp_len -= opt_len;
413 offset += opt_len;
414 }
415 }
416
417 return 0;
418
419 invalid:
420 memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
421 memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
422 memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
423
424 return 0;
425 }
426
427 /**
428 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
429 * @skb: sk_buff that contains the frame, with skb->data pointing to the
430 * Ethernet header
431 * @in_port: port number on which @skb was received.
432 * @key: output flow key
433 *
434 * The caller must ensure that skb->len >= ETH_HLEN.
435 *
436 * Returns 0 if successful, otherwise a negative errno value.
437 *
438 * Initializes @skb header pointers as follows:
439 *
440 * - skb->mac_header: the Ethernet header.
441 *
442 * - skb->network_header: just past the Ethernet header, or just past the
443 * VLAN header, to the first byte of the Ethernet payload.
444 *
445 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
446 * on output, then just past the IP header, if one is present and
447 * of a correct length, otherwise the same as skb->network_header.
448 * For other key->eth.type values it is left untouched.
449 */
450 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
451 {
452 int error;
453 struct ethhdr *eth;
454
455 memset(key, 0, sizeof(*key));
456
457 key->phy.priority = skb->priority;
458 if (OVS_CB(skb)->tun_key)
459 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
460 key->phy.in_port = in_port;
461 key->phy.skb_mark = skb->mark;
462
463 skb_reset_mac_header(skb);
464
465 /* Link layer. We are guaranteed to have at least the 14 byte Ethernet
466 * header in the linear data area.
467 */
468 eth = eth_hdr(skb);
469 ether_addr_copy(key->eth.src, eth->h_source);
470 ether_addr_copy(key->eth.dst, eth->h_dest);
471
472 __skb_pull(skb, 2 * ETH_ALEN);
473 /* We are going to push all headers that we pull, so no need to
474 * update skb->csum here.
475 */
476
477 if (vlan_tx_tag_present(skb))
478 key->eth.tci = htons(skb->vlan_tci);
479 else if (eth->h_proto == htons(ETH_P_8021Q))
480 if (unlikely(parse_vlan(skb, key)))
481 return -ENOMEM;
482
483 key->eth.type = parse_ethertype(skb);
484 if (unlikely(key->eth.type == htons(0)))
485 return -ENOMEM;
486
487 skb_reset_network_header(skb);
488 __skb_push(skb, skb->data - skb_mac_header(skb));
489
490 /* Network layer. */
491 if (key->eth.type == htons(ETH_P_IP)) {
492 struct iphdr *nh;
493 __be16 offset;
494
495 error = check_iphdr(skb);
496 if (unlikely(error)) {
497 if (error == -EINVAL) {
498 skb->transport_header = skb->network_header;
499 error = 0;
500 }
501 return error;
502 }
503
504 nh = ip_hdr(skb);
505 key->ipv4.addr.src = nh->saddr;
506 key->ipv4.addr.dst = nh->daddr;
507
508 key->ip.proto = nh->protocol;
509 key->ip.tos = nh->tos;
510 key->ip.ttl = nh->ttl;
511
512 offset = nh->frag_off & htons(IP_OFFSET);
513 if (offset) {
514 key->ip.frag = OVS_FRAG_TYPE_LATER;
515 return 0;
516 }
517 if (nh->frag_off & htons(IP_MF) ||
518 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
519 key->ip.frag = OVS_FRAG_TYPE_FIRST;
520
521 /* Transport layer. */
522 if (key->ip.proto == IPPROTO_TCP) {
523 if (tcphdr_ok(skb)) {
524 struct tcphdr *tcp = tcp_hdr(skb);
525 key->ipv4.tp.src = tcp->source;
526 key->ipv4.tp.dst = tcp->dest;
527 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
528 }
529 } else if (key->ip.proto == IPPROTO_UDP) {
530 if (udphdr_ok(skb)) {
531 struct udphdr *udp = udp_hdr(skb);
532 key->ipv4.tp.src = udp->source;
533 key->ipv4.tp.dst = udp->dest;
534 }
535 } else if (key->ip.proto == IPPROTO_SCTP) {
536 if (sctphdr_ok(skb)) {
537 struct sctphdr *sctp = sctp_hdr(skb);
538 key->ipv4.tp.src = sctp->source;
539 key->ipv4.tp.dst = sctp->dest;
540 }
541 } else if (key->ip.proto == IPPROTO_ICMP) {
542 if (icmphdr_ok(skb)) {
543 struct icmphdr *icmp = icmp_hdr(skb);
544 /* The ICMP type and code fields use the 16-bit
545 * transport port fields, so we need to store
546 * them in 16-bit network byte order. */
547 key->ipv4.tp.src = htons(icmp->type);
548 key->ipv4.tp.dst = htons(icmp->code);
549 }
550 }
551
552 } else if ((key->eth.type == htons(ETH_P_ARP) ||
553 key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
554 struct arp_eth_header *arp;
555
556 arp = (struct arp_eth_header *)skb_network_header(skb);
557
558 if (arp->ar_hrd == htons(ARPHRD_ETHER)
559 && arp->ar_pro == htons(ETH_P_IP)
560 && arp->ar_hln == ETH_ALEN
561 && arp->ar_pln == 4) {
562
563 /* We only match on the lower 8 bits of the opcode. */
564 if (ntohs(arp->ar_op) <= 0xff)
565 key->ip.proto = ntohs(arp->ar_op);
566 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
567 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
568 ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
569 ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
570 }
571 } else if (key->eth.type == htons(ETH_P_IPV6)) {
572 int nh_len; /* IPv6 Header + Extensions */
573
574 nh_len = parse_ipv6hdr(skb, key);
575 if (unlikely(nh_len < 0)) {
576 if (nh_len == -EINVAL) {
577 skb->transport_header = skb->network_header;
578 error = 0;
579 } else {
580 error = nh_len;
581 }
582 return error;
583 }
584
585 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
586 return 0;
587 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
588 key->ip.frag = OVS_FRAG_TYPE_FIRST;
589
590 /* Transport layer. */
591 if (key->ip.proto == NEXTHDR_TCP) {
592 if (tcphdr_ok(skb)) {
593 struct tcphdr *tcp = tcp_hdr(skb);
594 key->ipv6.tp.src = tcp->source;
595 key->ipv6.tp.dst = tcp->dest;
596 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
597 }
598 } else if (key->ip.proto == NEXTHDR_UDP) {
599 if (udphdr_ok(skb)) {
600 struct udphdr *udp = udp_hdr(skb);
601 key->ipv6.tp.src = udp->source;
602 key->ipv6.tp.dst = udp->dest;
603 }
604 } else if (key->ip.proto == NEXTHDR_SCTP) {
605 if (sctphdr_ok(skb)) {
606 struct sctphdr *sctp = sctp_hdr(skb);
607 key->ipv6.tp.src = sctp->source;
608 key->ipv6.tp.dst = sctp->dest;
609 }
610 } else if (key->ip.proto == NEXTHDR_ICMP) {
611 if (icmp6hdr_ok(skb)) {
612 error = parse_icmpv6(skb, key, nh_len);
613 if (error)
614 return error;
615 }
616 }
617 }
618
619 return 0;
620 }