]> git.proxmox.com Git - ovs.git/blob - datapath/flow.h
datapath: Mega flow implementation
[ovs.git] / datapath / flow.h
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 #ifndef FLOW_H
20 #define FLOW_H 1
21
22 #include <linux/kernel.h>
23 #include <linux/netlink.h>
24 #include <linux/openvswitch.h>
25 #include <linux/spinlock.h>
26 #include <linux/types.h>
27 #include <linux/rcupdate.h>
28 #include <linux/if_ether.h>
29 #include <linux/in6.h>
30 #include <linux/jiffies.h>
31 #include <linux/time.h>
32 #include <linux/flex_array.h>
33 #include <net/inet_ecn.h>
34
35 struct sk_buff;
36 struct sw_flow_mask;
37 struct flow_table;
38
39 struct sw_flow_actions {
40 struct rcu_head rcu;
41 u32 actions_len;
42 struct nlattr actions[];
43 };
44
45 /* Tunnel flow flags. */
46 #define OVS_TNL_F_DONT_FRAGMENT (1 << 0)
47 #define OVS_TNL_F_CSUM (1 << 1)
48 #define OVS_TNL_F_KEY (1 << 2)
49
50 /* Used to memset ovs_key_ipv4_tunnel padding. */
51 #define OVS_TUNNEL_KEY_SIZE \
52 (offsetof(struct ovs_key_ipv4_tunnel, ipv4_ttl) + \
53 FIELD_SIZEOF(struct ovs_key_ipv4_tunnel, ipv4_ttl))
54
55 struct ovs_key_ipv4_tunnel {
56 __be64 tun_id;
57 __be32 ipv4_src;
58 __be32 ipv4_dst;
59 u16 tun_flags;
60 u8 ipv4_tos;
61 u8 ipv4_ttl;
62 };
63
64 struct sw_flow_key {
65 struct ovs_key_ipv4_tunnel tun_key; /* Encapsulating tunnel key. */
66 struct {
67 u32 priority; /* Packet QoS priority. */
68 u32 skb_mark; /* SKB mark. */
69 u16 in_port; /* Input switch port (or DP_MAX_PORTS). */
70 } phy;
71 struct {
72 u8 src[ETH_ALEN]; /* Ethernet source address. */
73 u8 dst[ETH_ALEN]; /* Ethernet destination address. */
74 __be16 tci; /* 0 if no VLAN, VLAN_TAG_PRESENT set otherwise. */
75 __be16 type; /* Ethernet frame type. */
76 } eth;
77 struct {
78 u8 proto; /* IP protocol or lower 8 bits of ARP opcode. */
79 u8 tos; /* IP ToS. */
80 u8 ttl; /* IP TTL/hop limit. */
81 u8 frag; /* One of OVS_FRAG_TYPE_*. */
82 } ip;
83 union {
84 struct {
85 struct {
86 __be32 src; /* IP source address. */
87 __be32 dst; /* IP destination address. */
88 } addr;
89 union {
90 struct {
91 __be16 src; /* TCP/UDP source port. */
92 __be16 dst; /* TCP/UDP destination port. */
93 } tp;
94 struct {
95 u8 sha[ETH_ALEN]; /* ARP source hardware address. */
96 u8 tha[ETH_ALEN]; /* ARP target hardware address. */
97 } arp;
98 };
99 } ipv4;
100 struct {
101 struct {
102 struct in6_addr src; /* IPv6 source address. */
103 struct in6_addr dst; /* IPv6 destination address. */
104 } addr;
105 __be32 label; /* IPv6 flow label. */
106 struct {
107 __be16 src; /* TCP/UDP source port. */
108 __be16 dst; /* TCP/UDP destination port. */
109 } tp;
110 struct {
111 struct in6_addr target; /* ND target address. */
112 u8 sll[ETH_ALEN]; /* ND source link layer address. */
113 u8 tll[ETH_ALEN]; /* ND target link layer address. */
114 } nd;
115 } ipv6;
116 };
117 };
118
119 struct sw_flow {
120 struct rcu_head rcu;
121 struct hlist_node hash_node[2];
122 u32 hash;
123
124 struct sw_flow_key key;
125 struct sw_flow_key unmasked_key;
126 struct sw_flow_mask __rcu *mask;
127 struct sw_flow_actions __rcu *sf_acts;
128
129 spinlock_t lock; /* Lock for values below. */
130 unsigned long used; /* Last used time (in jiffies). */
131 u64 packet_count; /* Number of packets matched. */
132 u64 byte_count; /* Number of bytes matched. */
133 u8 tcp_flags; /* Union of seen TCP flags. */
134 };
135
136 struct sw_flow_key_range {
137 size_t start;
138 size_t end;
139 };
140
141 static inline u16 ovs_sw_flow_key_range_actual_size(const struct sw_flow_key_range *range)
142 {
143 return range->end - range->start;
144 }
145
146 struct sw_flow_match {
147 struct sw_flow_key *key;
148 struct sw_flow_key_range range;
149 struct sw_flow_mask *mask;
150 };
151
152 void ovs_match_init(struct sw_flow_match *match,
153 struct sw_flow_key *key, struct sw_flow_mask *mask);
154
155 struct arp_eth_header {
156 __be16 ar_hrd; /* format of hardware address */
157 __be16 ar_pro; /* format of protocol address */
158 unsigned char ar_hln; /* length of hardware address */
159 unsigned char ar_pln; /* length of protocol address */
160 __be16 ar_op; /* ARP opcode (command) */
161
162 /* Ethernet+IPv4 specific members. */
163 unsigned char ar_sha[ETH_ALEN]; /* sender hardware address */
164 unsigned char ar_sip[4]; /* sender IP address */
165 unsigned char ar_tha[ETH_ALEN]; /* target hardware address */
166 unsigned char ar_tip[4]; /* target IP address */
167 } __packed;
168
169 int ovs_flow_init(void);
170 void ovs_flow_exit(void);
171
172 struct sw_flow *ovs_flow_alloc(void);
173 void ovs_flow_deferred_free(struct sw_flow *);
174 void ovs_flow_free(struct sw_flow *, bool deferred);
175
176 struct sw_flow_actions *ovs_flow_actions_alloc(int actions_len);
177 void ovs_flow_deferred_free_acts(struct sw_flow_actions *);
178
179 int ovs_flow_extract(struct sk_buff *, u16 in_port, struct sw_flow_key *);
180 void ovs_flow_used(struct sw_flow *, struct sk_buff *);
181 u64 ovs_flow_used_time(unsigned long flow_jiffies);
182 int ovs_flow_to_nlattrs(const struct sw_flow_key *,
183 const struct sw_flow_key *, struct sk_buff *);
184 int ovs_match_from_nlattrs(struct sw_flow_match *match,
185 const struct nlattr *,
186 const struct nlattr *);
187 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
188 const struct nlattr *attr);
189
190 #define MAX_ACTIONS_BUFSIZE (32 * 1024)
191 #define TBL_MIN_BUCKETS 1024
192
193 struct flow_table {
194 struct flex_array *buckets;
195 unsigned int count, n_buckets;
196 struct rcu_head rcu;
197 struct list_head mask_list;
198 int node_ver;
199 u32 hash_seed;
200 bool keep_flows;
201 };
202
203 static inline int ovs_flow_tbl_count(struct flow_table *table)
204 {
205 return table->count;
206 }
207
208 static inline int ovs_flow_tbl_need_to_expand(struct flow_table *table)
209 {
210 return (table->count > table->n_buckets);
211 }
212
213 struct sw_flow *ovs_flow_lookup(struct flow_table *,
214 const struct sw_flow_key *);
215 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
216 struct sw_flow_match *match);
217
218 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred);
219 struct flow_table *ovs_flow_tbl_alloc(int new_size);
220 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table);
221 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table);
222 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow,
223 const struct sw_flow_key *key, int key_len);
224
225 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow);
226
227 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *idx);
228 extern const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1];
229 int ipv4_tun_from_nlattr(const struct nlattr *attr,
230 struct sw_flow_match *match, bool is_mask);
231 int ipv4_tun_to_nlattr(struct sk_buff *skb,
232 const struct ovs_key_ipv4_tunnel *tun_key,
233 const struct ovs_key_ipv4_tunnel *output);
234
235 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
236 const struct sw_flow_key *key, int key_len);
237
238 struct sw_flow_mask {
239 int ref_count;
240 struct rcu_head rcu;
241 struct list_head list;
242 struct sw_flow_key_range range;
243 struct sw_flow_key key;
244 };
245
246 static inline u16
247 ovs_sw_flow_mask_actual_size(const struct sw_flow_mask *mask)
248 {
249 return ovs_sw_flow_key_range_actual_size(&mask->range);
250 }
251
252 static inline u16
253 ovs_sw_flow_mask_size_roundup(const struct sw_flow_mask *mask)
254 {
255 return roundup(ovs_sw_flow_mask_actual_size(mask), sizeof(u32));
256 }
257
258 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void);
259 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *);
260 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *, bool deferred);
261 void ovs_sw_flow_mask_insert(struct flow_table *, struct sw_flow_mask *);
262 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *,
263 const struct sw_flow_mask *);
264 #endif /* flow.h */