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