]> git.proxmox.com Git - ovs.git/blame - datapath/flow.c
bridge: Don't learn from inadmissible flows when revising learning table.
[ovs.git] / datapath / flow.c
CommitLineData
064af421
BP
1/*
2 * Distributed under the terms of the GNU GPL version 2.
834377ea 3 * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
a14bc59f
BP
4 *
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
064af421
BP
7 */
8
9#include "flow.h"
f5e86186 10#include "datapath.h"
064af421
BP
11#include <linux/netdevice.h>
12#include <linux/etherdevice.h>
13#include <linux/if_ether.h>
14#include <linux/if_vlan.h>
15#include <net/llc_pdu.h>
16#include <linux/kernel.h>
17#include <linux/jiffies.h>
18#include <linux/llc.h>
19#include <linux/module.h>
20#include <linux/in.h>
21#include <linux/rcupdate.h>
a26ef517 22#include <linux/if_arp.h>
064af421
BP
23#include <linux/if_ether.h>
24#include <linux/ip.h>
25#include <linux/tcp.h>
26#include <linux/udp.h>
27#include <linux/icmp.h>
3c5f6de3 28#include <net/inet_ecn.h>
064af421
BP
29#include <net/ip.h>
30
31#include "compat.h"
32
33struct kmem_cache *flow_cache;
34
a26ef517
JP
35struct arp_eth_header
36{
37 __be16 ar_hrd; /* format of hardware address */
38 __be16 ar_pro; /* format of protocol address */
39 unsigned char ar_hln; /* length of hardware address */
40 unsigned char ar_pln; /* length of protocol address */
41 __be16 ar_op; /* ARP opcode (command) */
42
43 /* Ethernet+IPv4 specific members. */
44 unsigned char ar_sha[ETH_ALEN]; /* sender hardware address */
45 unsigned char ar_sip[4]; /* sender IP address */
46 unsigned char ar_tha[ETH_ALEN]; /* target hardware address */
47 unsigned char ar_tip[4]; /* target IP address */
48} __attribute__((packed));
49
50static inline int arphdr_ok(struct sk_buff *skb)
51{
52 int nh_ofs = skb_network_offset(skb);
53 return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
54}
55
064af421
BP
56static inline int iphdr_ok(struct sk_buff *skb)
57{
58 int nh_ofs = skb_network_offset(skb);
59 if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
60 int ip_len = ip_hdrlen(skb);
61 return (ip_len >= sizeof(struct iphdr)
62 && pskb_may_pull(skb, nh_ofs + ip_len));
63 }
64 return 0;
65}
66
67static inline int tcphdr_ok(struct sk_buff *skb)
68{
69 int th_ofs = skb_transport_offset(skb);
70 if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
71 int tcp_len = tcp_hdrlen(skb);
72 return (tcp_len >= sizeof(struct tcphdr)
73 && skb->len >= th_ofs + tcp_len);
74 }
75 return 0;
76}
77
78static inline int udphdr_ok(struct sk_buff *skb)
79{
80 int th_ofs = skb_transport_offset(skb);
81 return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
82}
83
84static inline int icmphdr_ok(struct sk_buff *skb)
85{
86 int th_ofs = skb_transport_offset(skb);
87 return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
88}
89
90#define TCP_FLAGS_OFFSET 13
91#define TCP_FLAG_MASK 0x3f
92
93static inline struct ovs_tcphdr *ovs_tcp_hdr(const struct sk_buff *skb)
94{
95 return (struct ovs_tcphdr *)skb_transport_header(skb);
96}
97
98void flow_used(struct sw_flow *flow, struct sk_buff *skb)
99{
100 unsigned long flags;
101 u8 tcp_flags = 0;
102
103 if (flow->key.dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
104 struct iphdr *nh = ip_hdr(skb);
105 flow->ip_tos = nh->tos;
106 if (flow->key.nw_proto == IPPROTO_TCP && tcphdr_ok(skb)) {
107 u8 *tcp = (u8 *)tcp_hdr(skb);
108 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
109 }
110 }
111
112 spin_lock_irqsave(&flow->lock, flags);
113 getnstimeofday(&flow->used);
114 flow->packet_count++;
115 flow->byte_count += skb->len;
116 flow->tcp_flags |= tcp_flags;
117 spin_unlock_irqrestore(&flow->lock, flags);
118}
119
120struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
121{
122 struct sw_flow_actions *sfa;
123
124 if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
125 return ERR_PTR(-EINVAL);
126
127 sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
128 GFP_KERNEL);
129 if (!sfa)
130 return ERR_PTR(-ENOMEM);
131
132 sfa->n_actions = n_actions;
133 return sfa;
134}
135
136
137/* Frees 'flow' immediately. */
138void flow_free(struct sw_flow *flow)
139{
140 if (unlikely(!flow))
141 return;
142 kfree(flow->sf_acts);
143 kmem_cache_free(flow_cache, flow);
144}
145
146/* RCU callback used by flow_deferred_free. */
147static void rcu_free_flow_callback(struct rcu_head *rcu)
148{
149 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
150 flow_free(flow);
151}
152
153/* Schedules 'flow' to be freed after the next RCU grace period.
154 * The caller must hold rcu_read_lock for this to be sensible. */
155void flow_deferred_free(struct sw_flow *flow)
156{
157 call_rcu(&flow->rcu, rcu_free_flow_callback);
158}
159
160/* RCU callback used by flow_deferred_free_acts. */
161static void rcu_free_acts_callback(struct rcu_head *rcu)
162{
163 struct sw_flow_actions *sf_acts = container_of(rcu,
164 struct sw_flow_actions, rcu);
165 kfree(sf_acts);
166}
167
168/* Schedules 'sf_acts' to be freed after the next RCU grace period.
169 * The caller must hold rcu_read_lock for this to be sensible. */
170void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
171{
172 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
173}
174
175#define SNAP_OUI_LEN 3
176
177struct eth_snap_hdr
178{
179 struct ethhdr eth;
180 u8 dsap; /* Always 0xAA */
181 u8 ssap; /* Always 0xAA */
182 u8 ctrl;
183 u8 oui[SNAP_OUI_LEN];
184 u16 ethertype;
185} __attribute__ ((packed));
186
187static int is_snap(const struct eth_snap_hdr *esh)
188{
189 return (esh->dsap == LLC_SAP_SNAP
190 && esh->ssap == LLC_SAP_SNAP
191 && !memcmp(esh->oui, "\0\0\0", 3));
192}
193
194/* Parses the Ethernet frame in 'skb', which was received on 'in_port',
195 * and initializes 'key' to match. Returns 1 if 'skb' contains an IP
196 * fragment, 0 otherwise. */
197int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
198{
199 struct ethhdr *eth;
200 struct eth_snap_hdr *esh;
201 int retval = 0;
202 int nh_ofs;
203
204 memset(key, 0, sizeof *key);
205 key->dl_vlan = htons(ODP_VLAN_NONE);
206 key->in_port = in_port;
207
208 if (skb->len < sizeof *eth)
209 return 0;
210 if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
211 return 0;
212 }
213
214 skb_reset_mac_header(skb);
215 eth = eth_hdr(skb);
216 esh = (struct eth_snap_hdr *) eth;
217 nh_ofs = sizeof *eth;
218 if (likely(ntohs(eth->h_proto) >= ODP_DL_TYPE_ETH2_CUTOFF))
219 key->dl_type = eth->h_proto;
220 else if (skb->len >= sizeof *esh && is_snap(esh)) {
221 key->dl_type = esh->ethertype;
222 nh_ofs = sizeof *esh;
223 } else {
224 key->dl_type = htons(ODP_DL_TYPE_NOT_ETH_TYPE);
225 if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
226 nh_ofs += sizeof(struct llc_pdu_un);
227 }
228 }
229
230 /* Check for a VLAN tag */
231 if (key->dl_type == htons(ETH_P_8021Q) &&
232 skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
233 struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
234 key->dl_type = vh->h_vlan_encapsulated_proto;
235 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
f5e86186 236 key->dl_vlan_pcp = (ntohs(vh->h_vlan_TCI) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
064af421
BP
237 nh_ofs += sizeof(struct vlan_hdr);
238 }
239 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
240 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
241 skb_set_network_header(skb, nh_ofs);
242
243 /* Network layer. */
244 if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
245 struct iphdr *nh = ip_hdr(skb);
246 int th_ofs = nh_ofs + nh->ihl * 4;
247 key->nw_src = nh->saddr;
248 key->nw_dst = nh->daddr;
f5e86186 249 key->nw_tos = nh->tos & ~INET_ECN_MASK;
064af421
BP
250 key->nw_proto = nh->protocol;
251 skb_set_transport_header(skb, th_ofs);
252
253 /* Transport layer. */
254 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
255 if (key->nw_proto == IPPROTO_TCP) {
256 if (tcphdr_ok(skb)) {
257 struct tcphdr *tcp = tcp_hdr(skb);
258 key->tp_src = tcp->source;
259 key->tp_dst = tcp->dest;
260 } else {
261 /* Avoid tricking other code into
262 * thinking that this packet has an L4
263 * header. */
264 key->nw_proto = 0;
265 }
266 } else if (key->nw_proto == IPPROTO_UDP) {
267 if (udphdr_ok(skb)) {
268 struct udphdr *udp = udp_hdr(skb);
269 key->tp_src = udp->source;
270 key->tp_dst = udp->dest;
271 } else {
272 /* Avoid tricking other code into
273 * thinking that this packet has an L4
274 * header. */
275 key->nw_proto = 0;
276 }
277 } else if (key->nw_proto == IPPROTO_ICMP) {
278 if (icmphdr_ok(skb)) {
279 struct icmphdr *icmp = icmp_hdr(skb);
280 /* The ICMP type and code fields use the 16-bit
281 * transport port fields, so we need to store them
282 * in 16-bit network byte order. */
283 key->tp_src = htons(icmp->type);
284 key->tp_dst = htons(icmp->code);
285 } else {
286 /* Avoid tricking other code into
287 * thinking that this packet has an L4
288 * header. */
289 key->nw_proto = 0;
290 }
291 }
292 } else {
293 retval = 1;
294 }
a26ef517
JP
295 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
296 struct arp_eth_header *arp;
297
298 arp = (struct arp_eth_header *)skb_network_header(skb);
299
f5e86186 300 if (arp->ar_hrd == htons(ARPHRD_ETHER)
de3f65ea
JP
301 && arp->ar_pro == htons(ETH_P_IP)
302 && arp->ar_hln == ETH_ALEN
303 && arp->ar_pln == 4) {
304
305 /* We only match on the lower 8 bits of the opcode. */
306 if (ntohs(arp->ar_op) <= 0xff) {
307 key->nw_proto = ntohs(arp->ar_op);
308 }
309
310 if (key->nw_proto == ARPOP_REQUEST
311 || key->nw_proto == ARPOP_REPLY) {
312 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
313 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
314 }
315 }
064af421
BP
316 } else {
317 skb_reset_transport_header(skb);
318 }
319 return retval;
320}
321
322/* Initializes the flow module.
323 * Returns zero if successful or a negative error code. */
324int flow_init(void)
325{
326 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
327 0, NULL);
328 if (flow_cache == NULL)
329 return -ENOMEM;
330
331 return 0;
332}
333
334/* Uninitializes the flow module. */
335void flow_exit(void)
336{
337 kmem_cache_destroy(flow_cache);
338}