Don't need 'deleted' member of struct flow.
[openvswitch] / datapath / flow.c
1 /*
2  * Distributed under the terms of the GNU GPL version 2.
3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland 
4  * Stanford Junior University
5  */
6
7 #include "flow.h"
8 #include <linux/netdevice.h>
9 #include <linux/etherdevice.h>
10 #include <linux/if_ether.h>
11 #include <linux/if_vlan.h>
12 #include <net/llc_pdu.h>
13 #include <linux/ip.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/in.h>
19 #include <linux/rcupdate.h>
20
21 #include "openflow.h"
22 #include "compat.h"
23 #include "snap.h"
24
25 struct kmem_cache *flow_cache;
26
27 /* Internal function used to compare fields in flow. */
28 static inline
29 int flow_fields_match(const struct sw_flow_key *a, const struct sw_flow_key *b,
30                 uint16_t w)
31 {
32         return ((w & OFPFW_IN_PORT || a->in_port == b->in_port)
33                 && (w & OFPFW_DL_VLAN || a->dl_vlan == b->dl_vlan)
34                 && (w & OFPFW_DL_SRC || !memcmp(a->dl_src, b->dl_src, ETH_ALEN))
35                 && (w & OFPFW_DL_DST || !memcmp(a->dl_dst, b->dl_dst, ETH_ALEN))
36                 && (w & OFPFW_DL_TYPE || a->dl_type == b->dl_type)
37                 && (w & OFPFW_NW_SRC || a->nw_src == b->nw_src)
38                 && (w & OFPFW_NW_DST || a->nw_dst == b->nw_dst)
39                 && (w & OFPFW_NW_PROTO || a->nw_proto == b->nw_proto)
40                 && (w & OFPFW_TP_SRC || a->tp_src == b->tp_src)
41                 && (w & OFPFW_TP_DST || a->tp_dst == b->tp_dst));
42 }
43
44 /* Returns nonzero if 'a' and 'b' match, that is, if their fields are equal
45  * modulo wildcards, zero otherwise. */
46 int flow_matches(const struct sw_flow_key *a, const struct sw_flow_key *b)
47 {
48         return flow_fields_match(a, b, (a->wildcards | b->wildcards));
49 }
50 EXPORT_SYMBOL(flow_matches);
51
52 /* Returns nonzero if 't' (the table entry's key) and 'd' (the key 
53  * describing the deletion) match, that is, if their fields are 
54  * equal modulo wildcards, zero otherwise.  If 'strict' is nonzero, the
55  * wildcards must match in both 't_key' and 'd_key'.  Note that the
56  * table's wildcards are ignored unless 'strict' is set. */
57 int flow_del_matches(const struct sw_flow_key *t, const struct sw_flow_key *d, int strict)
58 {
59         if (strict && (t->wildcards != d->wildcards))
60                 return 0;
61
62         return flow_fields_match(t, d, d->wildcards);
63 }
64 EXPORT_SYMBOL(flow_del_matches);
65
66 void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
67 {
68         to->wildcards = ntohs(from->wildcards) & OFPFW_ALL;
69         memset(to->pad, '\0', sizeof(to->pad));
70         to->in_port = from->in_port;
71         to->dl_vlan = from->dl_vlan;
72         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
73         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
74         to->dl_type = from->dl_type;
75
76         to->nw_src = to->nw_dst = to->nw_proto = 0;
77         to->tp_src = to->tp_dst = 0;
78
79 #define OFPFW_TP (OFPFW_TP_SRC | OFPFW_TP_DST)
80 #define OFPFW_NW (OFPFW_NW_SRC | OFPFW_NW_DST | OFPFW_NW_PROTO)
81         if (to->wildcards & OFPFW_DL_TYPE) {
82                 /* Can't sensibly match on network or transport headers if the
83                  * data link type is unknown. */
84                 to->wildcards |= OFPFW_NW | OFPFW_TP;
85         } else if (from->dl_type == htons(ETH_P_IP)) {
86                 to->nw_src   = from->nw_src;
87                 to->nw_dst   = from->nw_dst;
88                 to->nw_proto = from->nw_proto;
89
90                 if (to->wildcards & OFPFW_NW_PROTO) {
91                         /* Can't sensibly match on transport headers if the
92                          * network protocol is unknown. */
93                         to->wildcards |= OFPFW_TP;
94                 } else if (from->nw_proto == IPPROTO_TCP
95                            || from->nw_proto == IPPROTO_UDP) {
96                         to->tp_src = from->tp_src;
97                         to->tp_dst = from->tp_dst;
98                 } else {
99                         /* Transport layer fields are undefined.  Mark them as
100                          * exact-match to allow such flows to reside in
101                          * table-hash, instead of falling into table-linear. */
102                         to->wildcards &= ~OFPFW_TP;
103                 }
104         } else {
105                 /* Network and transport layer fields are undefined.  Mark them
106                  * as exact-match to allow such flows to reside in table-hash,
107                  * instead of falling into table-linear. */
108                 to->wildcards &= ~(OFPFW_NW | OFPFW_TP);
109         }
110 }
111
112 void flow_fill_match(struct ofp_match* to, const struct sw_flow_key* from)
113 {
114         to->wildcards = htons(from->wildcards);
115         to->in_port   = from->in_port;
116         to->dl_vlan   = from->dl_vlan;
117         memcpy(to->dl_src, from->dl_src, ETH_ALEN);
118         memcpy(to->dl_dst, from->dl_dst, ETH_ALEN);
119         to->dl_type   = from->dl_type;
120         to->nw_src        = from->nw_src;
121         to->nw_dst        = from->nw_dst;
122         to->nw_proto  = from->nw_proto;
123         to->tp_src        = from->tp_src;
124         to->tp_dst        = from->tp_dst;
125         memset(to->pad, '\0', sizeof(to->pad));
126 }
127
128 /* Allocates and returns a new flow with 'n_actions' action, using allocation
129  * flags 'flags'.  Returns the new flow or a null pointer on failure. */
130 struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
131 {
132         struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
133         if (unlikely(!flow))
134                 return NULL;
135
136         flow->n_actions = n_actions;
137         flow->actions = kmalloc(n_actions * sizeof *flow->actions,
138                                 flags);
139         if (unlikely(!flow->actions) && n_actions > 0) {
140                 kmem_cache_free(flow_cache, flow);
141                 return NULL;
142         }
143         return flow;
144 }
145
146 /* Frees 'flow' immediately. */
147 void flow_free(struct sw_flow *flow)
148 {
149         if (unlikely(!flow))
150                 return;
151         kfree(flow->actions);
152         kmem_cache_free(flow_cache, flow);
153 }
154 EXPORT_SYMBOL(flow_free);
155
156 /* RCU callback used by flow_deferred_free. */
157 static void rcu_callback(struct rcu_head *rcu)
158 {
159         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
160         flow_free(flow);
161 }
162
163 /* Schedules 'flow' to be freed after the next RCU grace period.
164  * The caller must hold rcu_read_lock for this to be sensible. */
165 void flow_deferred_free(struct sw_flow *flow)
166 {
167         call_rcu(&flow->rcu, rcu_callback);
168 }
169 EXPORT_SYMBOL(flow_deferred_free);
170
171 /* Prints a representation of 'key' to the kernel log. */
172 void print_flow(const struct sw_flow_key *key)
173 {
174         printk("wild%04x port%04x:vlan%04x mac%02x:%02x:%02x:%02x:%02x:%02x"
175                         "->%02x:%02x:%02x:%02x:%02x:%02x "
176                         "proto%04x ip%u.%u.%u.%u->%u.%u.%u.%u port%d->%d\n",
177                         key->wildcards, ntohs(key->in_port), ntohs(key->dl_vlan),
178                         key->dl_src[0], key->dl_src[1], key->dl_src[2],
179                         key->dl_src[3], key->dl_src[4], key->dl_src[5],
180                         key->dl_dst[0], key->dl_dst[1], key->dl_dst[2],
181                         key->dl_dst[3], key->dl_dst[4], key->dl_dst[5],
182                         ntohs(key->dl_type),
183                         ((unsigned char *)&key->nw_src)[0],
184                         ((unsigned char *)&key->nw_src)[1],
185                         ((unsigned char *)&key->nw_src)[2],
186                         ((unsigned char *)&key->nw_src)[3],
187                         ((unsigned char *)&key->nw_dst)[0],
188                         ((unsigned char *)&key->nw_dst)[1],
189                         ((unsigned char *)&key->nw_dst)[2],
190                         ((unsigned char *)&key->nw_dst)[3],
191                         ntohs(key->tp_src), ntohs(key->tp_dst));
192 }
193 EXPORT_SYMBOL(print_flow);
194
195 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
196  * and initializes 'key' to match. */
197 void flow_extract(struct sk_buff *skb, uint16_t in_port,
198                   struct sw_flow_key *key)
199 {
200         struct ethhdr *mac;
201         struct udphdr *th;
202         int nh_ofs, th_ofs;
203
204         key->in_port = htons(in_port);
205         key->wildcards = 0;
206         memset(key->pad, '\0', sizeof(key->pad));
207
208         /* This code doesn't check that skb->len is long enough to contain the
209          * MAC or network header.  With a 46-byte minimum length frame this
210          * assumption is always correct. */
211
212         /* Doesn't verify checksums.  Should it? */
213
214         /* Data link layer.  We only support Ethernet. */
215         mac = eth_hdr(skb);
216         nh_ofs = sizeof(struct ethhdr);
217         if (likely(ntohs(mac->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF)) {
218                 /* This is an Ethernet II frame */
219                 key->dl_type = mac->h_proto;
220         } else {
221                 /* This is an 802.2 frame */
222                 if (snap_get_ethertype(skb, &key->dl_type) != -EINVAL) {
223                         nh_ofs += sizeof(struct snap_hdr);
224                 } else {
225                         key->dl_type = OFP_DL_TYPE_NOT_ETH_TYPE;
226                         nh_ofs += sizeof(struct llc_pdu_un);
227                 }
228         }
229
230         /* Check for a VLAN tag */
231         if (likely(key->dl_type != htons(ETH_P_8021Q))) {
232                 key->dl_vlan = htons(OFP_VLAN_NONE);
233         } else {
234                 struct vlan_hdr *vh = (struct vlan_hdr *)(skb_mac_header(skb) + nh_ofs);
235                 key->dl_type = vh->h_vlan_encapsulated_proto;
236                 key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
237                 nh_ofs += sizeof(*vh);
238         }
239         memcpy(key->dl_src, mac->h_source, ETH_ALEN);
240         memcpy(key->dl_dst, mac->h_dest, ETH_ALEN);
241         skb_set_network_header(skb, nh_ofs);
242
243         /* Network layer. */
244         if (likely(key->dl_type == htons(ETH_P_IP))) {
245                 struct iphdr *nh = ip_hdr(skb);
246                 key->nw_src = nh->saddr;
247                 key->nw_dst = nh->daddr;
248                 key->nw_proto = nh->protocol;
249                 th_ofs = nh_ofs + nh->ihl * 4;
250                 skb_set_transport_header(skb, th_ofs);
251
252                 /* Transport layer. */
253                 if ((key->nw_proto != IPPROTO_TCP && key->nw_proto != IPPROTO_UDP)
254                                 || skb->len < th_ofs + sizeof(struct udphdr)) {
255                         goto no_th;
256                 }
257                 th = udp_hdr(skb);
258                 key->tp_src = th->source;
259                 key->tp_dst = th->dest;
260
261                 return;
262         }
263
264         key->nw_src = 0;
265         key->nw_dst = 0;
266         key->nw_proto = 0;
267
268 no_th:
269         key->tp_src = 0;
270         key->tp_dst = 0;
271 }
272
273 /* Initializes the flow module.
274  * Returns zero if successful or a negative error code. */
275 int flow_init(void)
276 {
277         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
278                                         0, NULL);
279         if (flow_cache == NULL)
280                 return -ENOMEM;
281
282         return 0;
283 }
284
285 /* Uninitializes the flow module. */
286 void flow_exit(void)
287 {
288         kmem_cache_destroy(flow_cache);
289 }
290