Loading...
1/*
2 * Copyright (c) 2007-2014 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 DATAPATH_H
20#define DATAPATH_H 1
21
22#include <asm/page.h>
23#include <linux/kernel.h>
24#include <linux/mutex.h>
25#include <linux/netdevice.h>
26#include <linux/skbuff.h>
27#include <linux/u64_stats_sync.h>
28#include <net/ip_tunnels.h>
29
30#include "conntrack.h"
31#include "flow.h"
32#include "flow_table.h"
33#include "meter.h"
34#include "vport-internal_dev.h"
35
36#define DP_MAX_PORTS USHRT_MAX
37#define DP_VPORT_HASH_BUCKETS 1024
38
39/**
40 * struct dp_stats_percpu - per-cpu packet processing statistics for a given
41 * datapath.
42 * @n_hit: Number of received packets for which a matching flow was found in
43 * the flow table.
44 * @n_miss: Number of received packets that had no matching flow in the flow
45 * table. The sum of @n_hit and @n_miss is the number of packets that have
46 * been received by the datapath.
47 * @n_lost: Number of received packets that had no matching flow in the flow
48 * table that could not be sent to userspace (normally due to an overflow in
49 * one of the datapath's queues).
50 * @n_mask_hit: Number of masks looked up for flow match.
51 * @n_mask_hit / (@n_hit + @n_missed) will be the average masks looked
52 * up per packet.
53 */
54struct dp_stats_percpu {
55 u64 n_hit;
56 u64 n_missed;
57 u64 n_lost;
58 u64 n_mask_hit;
59 struct u64_stats_sync syncp;
60};
61
62/**
63 * struct datapath - datapath for flow-based packet switching
64 * @rcu: RCU callback head for deferred destruction.
65 * @list_node: Element in global 'dps' list.
66 * @table: flow table.
67 * @ports: Hash table for ports. %OVSP_LOCAL port always exists. Protected by
68 * ovs_mutex and RCU.
69 * @stats_percpu: Per-CPU datapath statistics.
70 * @net: Reference to net namespace.
71 * @max_headroom: the maximum headroom of all vports in this datapath; it will
72 * be used by all the internal vports in this dp.
73 *
74 * Context: See the comment on locking at the top of datapath.c for additional
75 * locking information.
76 */
77struct datapath {
78 struct rcu_head rcu;
79 struct list_head list_node;
80
81 /* Flow table. */
82 struct flow_table table;
83
84 /* Switch ports. */
85 struct hlist_head *ports;
86
87 /* Stats. */
88 struct dp_stats_percpu __percpu *stats_percpu;
89
90 /* Network namespace ref. */
91 possible_net_t net;
92
93 u32 user_features;
94
95 u32 max_headroom;
96
97 /* Switch meters. */
98 struct hlist_head *meters;
99};
100
101/**
102 * struct ovs_skb_cb - OVS data in skb CB
103 * @input_vport: The original vport packet came in on. This value is cached
104 * when a packet is received by OVS.
105 * @mru: The maximum received fragement size; 0 if the packet is not
106 * fragmented.
107 * @acts_origlen: The netlink size of the flow actions applied to this skb.
108 * @cutlen: The number of bytes from the packet end to be removed.
109 */
110struct ovs_skb_cb {
111 struct vport *input_vport;
112 u16 mru;
113 u16 acts_origlen;
114 u32 cutlen;
115};
116#define OVS_CB(skb) ((struct ovs_skb_cb *)(skb)->cb)
117
118/**
119 * struct dp_upcall - metadata to include with a packet to send to userspace
120 * @cmd: One of %OVS_PACKET_CMD_*.
121 * @userdata: If nonnull, its variable-length value is passed to userspace as
122 * %OVS_PACKET_ATTR_USERDATA.
123 * @portid: Netlink portid to which packet should be sent. If @portid is 0
124 * then no packet is sent and the packet is accounted in the datapath's @n_lost
125 * counter.
126 * @egress_tun_info: If nonnull, becomes %OVS_PACKET_ATTR_EGRESS_TUN_KEY.
127 * @mru: If not zero, Maximum received IP fragment size.
128 */
129struct dp_upcall_info {
130 struct ip_tunnel_info *egress_tun_info;
131 const struct nlattr *userdata;
132 const struct nlattr *actions;
133 int actions_len;
134 u32 portid;
135 u8 cmd;
136 u16 mru;
137};
138
139/**
140 * struct ovs_net - Per net-namespace data for ovs.
141 * @dps: List of datapaths to enable dumping them all out.
142 * Protected by genl_mutex.
143 */
144struct ovs_net {
145 struct list_head dps;
146 struct work_struct dp_notify_work;
147
148 /* Module reference for configuring conntrack. */
149 bool xt_label;
150};
151
152extern unsigned int ovs_net_id;
153void ovs_lock(void);
154void ovs_unlock(void);
155
156#ifdef CONFIG_LOCKDEP
157int lockdep_ovsl_is_held(void);
158#else
159#define lockdep_ovsl_is_held() 1
160#endif
161
162#define ASSERT_OVSL() WARN_ON(!lockdep_ovsl_is_held())
163#define ovsl_dereference(p) \
164 rcu_dereference_protected(p, lockdep_ovsl_is_held())
165#define rcu_dereference_ovsl(p) \
166 rcu_dereference_check(p, lockdep_ovsl_is_held())
167
168static inline struct net *ovs_dp_get_net(const struct datapath *dp)
169{
170 return read_pnet(&dp->net);
171}
172
173static inline void ovs_dp_set_net(struct datapath *dp, struct net *net)
174{
175 write_pnet(&dp->net, net);
176}
177
178struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no);
179
180static inline struct vport *ovs_vport_rcu(const struct datapath *dp, int port_no)
181{
182 WARN_ON_ONCE(!rcu_read_lock_held());
183 return ovs_lookup_vport(dp, port_no);
184}
185
186static inline struct vport *ovs_vport_ovsl_rcu(const struct datapath *dp, int port_no)
187{
188 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
189 return ovs_lookup_vport(dp, port_no);
190}
191
192static inline struct vport *ovs_vport_ovsl(const struct datapath *dp, int port_no)
193{
194 ASSERT_OVSL();
195 return ovs_lookup_vport(dp, port_no);
196}
197
198/* Must be called with rcu_read_lock. */
199static inline struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
200{
201 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
202
203 if (dev) {
204 struct vport *vport = ovs_internal_dev_get_vport(dev);
205
206 if (vport)
207 return vport->dp;
208 }
209
210 return NULL;
211}
212
213/* The caller must hold either ovs_mutex or rcu_read_lock to keep the
214 * returned dp pointer valid.
215 */
216static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
217{
218 struct datapath *dp;
219
220 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
221 rcu_read_lock();
222 dp = get_dp_rcu(net, dp_ifindex);
223 rcu_read_unlock();
224
225 return dp;
226}
227
228extern struct notifier_block ovs_dp_device_notifier;
229extern struct genl_family dp_vport_genl_family;
230
231void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key);
232void ovs_dp_detach_port(struct vport *);
233int ovs_dp_upcall(struct datapath *, struct sk_buff *,
234 const struct sw_flow_key *, const struct dp_upcall_info *,
235 uint32_t cutlen);
236
237const char *ovs_dp_name(const struct datapath *dp);
238struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
239 u32 portid, u32 seq, u8 cmd);
240
241int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
242 const struct sw_flow_actions *, struct sw_flow_key *);
243
244void ovs_dp_notify_wq(struct work_struct *work);
245
246int action_fifos_init(void);
247void action_fifos_exit(void);
248
249/* 'KEY' must not have any bits set outside of the 'MASK' */
250#define OVS_MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
251#define OVS_SET_MASKED(OLD, KEY, MASK) ((OLD) = OVS_MASKED(OLD, KEY, MASK))
252
253#define OVS_NLERR(logging_allowed, fmt, ...) \
254do { \
255 if (logging_allowed && net_ratelimit()) \
256 pr_info("netlink: " fmt "\n", ##__VA_ARGS__); \
257} while (0)
258#endif /* datapath.h */
1/*
2 * Copyright (c) 2007-2012 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 DATAPATH_H
20#define DATAPATH_H 1
21
22#include <asm/page.h>
23#include <linux/kernel.h>
24#include <linux/mutex.h>
25#include <linux/netdevice.h>
26#include <linux/skbuff.h>
27#include <linux/u64_stats_sync.h>
28
29#include "flow.h"
30#include "flow_table.h"
31#include "vport.h"
32
33#define DP_MAX_PORTS USHRT_MAX
34#define DP_VPORT_HASH_BUCKETS 1024
35
36#define SAMPLE_ACTION_DEPTH 3
37
38/**
39 * struct dp_stats_percpu - per-cpu packet processing statistics for a given
40 * datapath.
41 * @n_hit: Number of received packets for which a matching flow was found in
42 * the flow table.
43 * @n_miss: Number of received packets that had no matching flow in the flow
44 * table. The sum of @n_hit and @n_miss is the number of packets that have
45 * been received by the datapath.
46 * @n_lost: Number of received packets that had no matching flow in the flow
47 * table that could not be sent to userspace (normally due to an overflow in
48 * one of the datapath's queues).
49 * @n_mask_hit: Number of masks looked up for flow match.
50 * @n_mask_hit / (@n_hit + @n_missed) will be the average masks looked
51 * up per packet.
52 */
53struct dp_stats_percpu {
54 u64 n_hit;
55 u64 n_missed;
56 u64 n_lost;
57 u64 n_mask_hit;
58 struct u64_stats_sync syncp;
59};
60
61/**
62 * struct datapath - datapath for flow-based packet switching
63 * @rcu: RCU callback head for deferred destruction.
64 * @list_node: Element in global 'dps' list.
65 * @table: flow table.
66 * @ports: Hash table for ports. %OVSP_LOCAL port always exists. Protected by
67 * ovs_mutex and RCU.
68 * @stats_percpu: Per-CPU datapath statistics.
69 * @net: Reference to net namespace.
70 *
71 * Context: See the comment on locking at the top of datapath.c for additional
72 * locking information.
73 */
74struct datapath {
75 struct rcu_head rcu;
76 struct list_head list_node;
77
78 /* Flow table. */
79 struct flow_table table;
80
81 /* Switch ports. */
82 struct hlist_head *ports;
83
84 /* Stats. */
85 struct dp_stats_percpu __percpu *stats_percpu;
86
87#ifdef CONFIG_NET_NS
88 /* Network namespace ref. */
89 struct net *net;
90#endif
91
92 u32 user_features;
93};
94
95/**
96 * struct ovs_skb_cb - OVS data in skb CB
97 * @flow: The flow associated with this packet. May be %NULL if no flow.
98 * @pkt_key: The flow information extracted from the packet. Must be nonnull.
99 * @tun_key: Key for the tunnel that encapsulated this packet. NULL if the
100 * packet is not being tunneled.
101 */
102struct ovs_skb_cb {
103 struct sw_flow *flow;
104 struct sw_flow_key *pkt_key;
105 struct ovs_key_ipv4_tunnel *tun_key;
106};
107#define OVS_CB(skb) ((struct ovs_skb_cb *)(skb)->cb)
108
109/**
110 * struct dp_upcall - metadata to include with a packet to send to userspace
111 * @cmd: One of %OVS_PACKET_CMD_*.
112 * @key: Becomes %OVS_PACKET_ATTR_KEY. Must be nonnull.
113 * @userdata: If nonnull, its variable-length value is passed to userspace as
114 * %OVS_PACKET_ATTR_USERDATA.
115 * @pid: Netlink PID to which packet should be sent. If @pid is 0 then no
116 * packet is sent and the packet is accounted in the datapath's @n_lost
117 * counter.
118 */
119struct dp_upcall_info {
120 u8 cmd;
121 const struct sw_flow_key *key;
122 const struct nlattr *userdata;
123 u32 portid;
124};
125
126/**
127 * struct ovs_net - Per net-namespace data for ovs.
128 * @dps: List of datapaths to enable dumping them all out.
129 * Protected by genl_mutex.
130 */
131struct ovs_net {
132 struct list_head dps;
133 struct work_struct dp_notify_work;
134 struct vport_net vport_net;
135};
136
137extern int ovs_net_id;
138void ovs_lock(void);
139void ovs_unlock(void);
140
141#ifdef CONFIG_LOCKDEP
142int lockdep_ovsl_is_held(void);
143#else
144#define lockdep_ovsl_is_held() 1
145#endif
146
147#define ASSERT_OVSL() WARN_ON(unlikely(!lockdep_ovsl_is_held()))
148#define ovsl_dereference(p) \
149 rcu_dereference_protected(p, lockdep_ovsl_is_held())
150#define rcu_dereference_ovsl(p) \
151 rcu_dereference_check(p, lockdep_ovsl_is_held())
152
153static inline struct net *ovs_dp_get_net(struct datapath *dp)
154{
155 return read_pnet(&dp->net);
156}
157
158static inline void ovs_dp_set_net(struct datapath *dp, struct net *net)
159{
160 write_pnet(&dp->net, net);
161}
162
163struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no);
164
165static inline struct vport *ovs_vport_rcu(const struct datapath *dp, int port_no)
166{
167 WARN_ON_ONCE(!rcu_read_lock_held());
168 return ovs_lookup_vport(dp, port_no);
169}
170
171static inline struct vport *ovs_vport_ovsl_rcu(const struct datapath *dp, int port_no)
172{
173 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
174 return ovs_lookup_vport(dp, port_no);
175}
176
177static inline struct vport *ovs_vport_ovsl(const struct datapath *dp, int port_no)
178{
179 ASSERT_OVSL();
180 return ovs_lookup_vport(dp, port_no);
181}
182
183extern struct notifier_block ovs_dp_device_notifier;
184extern struct genl_family dp_vport_genl_family;
185
186void ovs_dp_process_received_packet(struct vport *, struct sk_buff *);
187void ovs_dp_detach_port(struct vport *);
188int ovs_dp_upcall(struct datapath *, struct sk_buff *,
189 const struct dp_upcall_info *);
190
191struct sk_buff *ovs_vport_cmd_build_info(struct vport *, u32 pid, u32 seq,
192 u8 cmd);
193
194int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb);
195void ovs_dp_notify_wq(struct work_struct *work);
196
197#define OVS_NLERR(fmt, ...) \
198 pr_info_once("netlink: " fmt, ##__VA_ARGS__)
199
200#endif /* datapath.h */