blob: ab85c1cae255b8fd373a0d5e0e1d05841b5bd8d0 [file] [log] [blame]
Jesse Grossccb13522011-10-25 19:26:31 -07001/*
Andy Zhou971427f32014-09-15 19:37:25 -07002 * Copyright (c) 2007-2014 Nicira, Inc.
Jesse Grossccb13522011-10-25 19:26:31 -07003 *
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>
Thomas Graf1d8fff92015-07-21 10:43:54 +020028#include <net/ip_tunnels.h>
Jesse Grossccb13522011-10-25 19:26:31 -070029
Joe Stringer7f8a4362015-08-26 11:31:48 -070030#include "conntrack.h"
Jesse Grossccb13522011-10-25 19:26:31 -070031#include "flow.h"
Pravin B Shelare6445712013-10-03 18:16:47 -070032#include "flow_table.h"
Jesse Grossccb13522011-10-25 19:26:31 -070033
Pravin B Shelar15eac2a2012-08-23 12:40:54 -070034#define DP_MAX_PORTS USHRT_MAX
35#define DP_VPORT_HASH_BUCKETS 1024
36
Jesse Grossccb13522011-10-25 19:26:31 -070037#define SAMPLE_ACTION_DEPTH 3
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).
Andy Zhou1bd71162013-10-22 10:42:46 -070050 * @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.
Jesse Grossccb13522011-10-25 19:26:31 -070053 */
54struct dp_stats_percpu {
55 u64 n_hit;
56 u64 n_missed;
57 u64 n_lost;
Andy Zhou1bd71162013-10-22 10:42:46 -070058 u64 n_mask_hit;
WANG Congdf9d9fd2014-02-14 15:10:46 -080059 struct u64_stats_sync syncp;
Jesse Grossccb13522011-10-25 19:26:31 -070060};
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.
Pravin B Shelarb637e492013-10-04 00:14:23 -070066 * @table: flow table.
Pravin B Shelar15eac2a2012-08-23 12:40:54 -070067 * @ports: Hash table for ports. %OVSP_LOCAL port always exists. Protected by
Pravin B Shelar8e4e1712013-04-15 13:23:03 -070068 * ovs_mutex and RCU.
Jesse Grossccb13522011-10-25 19:26:31 -070069 * @stats_percpu: Per-CPU datapath statistics.
Pravin B Shelar46df7b82012-02-22 19:58:59 -080070 * @net: Reference to net namespace.
Paolo Abeni3a927bc2016-02-26 10:45:39 +010071 * @max_headroom: the maximum headroom of all vports in this datapath; it will
72 * be used by all the internal vports in this dp.
Jesse Grossccb13522011-10-25 19:26:31 -070073 *
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. */
Pravin B Shelarb637e492013-10-04 00:14:23 -070082 struct flow_table table;
Jesse Grossccb13522011-10-25 19:26:31 -070083
84 /* Switch ports. */
Pravin B Shelar15eac2a2012-08-23 12:40:54 -070085 struct hlist_head *ports;
Jesse Grossccb13522011-10-25 19:26:31 -070086
87 /* Stats. */
88 struct dp_stats_percpu __percpu *stats_percpu;
Pravin B Shelar46df7b82012-02-22 19:58:59 -080089
Pravin B Shelar46df7b82012-02-22 19:58:59 -080090 /* Network namespace ref. */
Eric W. Biederman0c5c9fb2015-03-11 23:06:44 -050091 possible_net_t net;
Thomas Graf43d4be92013-12-13 15:22:18 +010092
93 u32 user_features;
Paolo Abeni3a927bc2016-02-26 10:45:39 +010094
95 u32 max_headroom;
Jesse Grossccb13522011-10-25 19:26:31 -070096};
97
98/**
99 * struct ovs_skb_cb - OVS data in skb CB
Pravin B Shelar83c8df22014-09-15 19:20:31 -0700100 * @input_vport: The original vport packet came in on. This value is cached
101 * when a packet is received by OVS.
Joe Stringer7f8a4362015-08-26 11:31:48 -0700102 * @mru: The maximum received fragement size; 0 if the packet is not
William Tuf2a4d082016-06-10 11:49:33 -0700103 * @cutlen: The number of bytes from the packet end to be removed.
Joe Stringer7f8a4362015-08-26 11:31:48 -0700104 * fragmented.
Jesse Grossccb13522011-10-25 19:26:31 -0700105 */
106struct ovs_skb_cb {
Pravin B Shelar83c8df22014-09-15 19:20:31 -0700107 struct vport *input_vport;
Joe Stringer7f8a4362015-08-26 11:31:48 -0700108 u16 mru;
William Tuf2a4d082016-06-10 11:49:33 -0700109 u32 cutlen;
Jesse Grossccb13522011-10-25 19:26:31 -0700110};
111#define OVS_CB(skb) ((struct ovs_skb_cb *)(skb)->cb)
112
113/**
114 * struct dp_upcall - metadata to include with a packet to send to userspace
115 * @cmd: One of %OVS_PACKET_CMD_*.
Ben Pfaff44901082013-02-15 17:29:22 -0800116 * @userdata: If nonnull, its variable-length value is passed to userspace as
Jesse Grossccb13522011-10-25 19:26:31 -0700117 * %OVS_PACKET_ATTR_USERDATA.
Pravin B Shelare8eedb82014-11-06 06:57:27 -0800118 * @portid: Netlink portid to which packet should be sent. If @portid is 0
119 * then no packet is sent and the packet is accounted in the datapath's @n_lost
Jesse Grossccb13522011-10-25 19:26:31 -0700120 * counter.
Wenyu Zhang8f0aad62014-11-06 06:51:24 -0800121 * @egress_tun_info: If nonnull, becomes %OVS_PACKET_ATTR_EGRESS_TUN_KEY.
Joe Stringer7f8a4362015-08-26 11:31:48 -0700122 * @mru: If not zero, Maximum received IP fragment size.
Jesse Grossccb13522011-10-25 19:26:31 -0700123 */
124struct dp_upcall_info {
Pravin B Shelar4c222792015-08-30 18:09:38 -0700125 struct ip_tunnel_info *egress_tun_info;
Jesse Grossccb13522011-10-25 19:26:31 -0700126 const struct nlattr *userdata;
Neil McKeeccea7442015-05-26 20:59:43 -0700127 const struct nlattr *actions;
128 int actions_len;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000129 u32 portid;
Pravin B Shelare8eedb82014-11-06 06:57:27 -0800130 u8 cmd;
Joe Stringer7f8a4362015-08-26 11:31:48 -0700131 u16 mru;
Jesse Grossccb13522011-10-25 19:26:31 -0700132};
133
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700134/**
135 * struct ovs_net - Per net-namespace data for ovs.
136 * @dps: List of datapaths to enable dumping them all out.
137 * Protected by genl_mutex.
138 */
139struct ovs_net {
140 struct list_head dps;
141 struct work_struct dp_notify_work;
Joe Stringerc2ac6672015-08-26 11:31:52 -0700142
143 /* Module reference for configuring conntrack. */
144 bool xt_label;
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700145};
146
147extern int ovs_net_id;
148void ovs_lock(void);
149void ovs_unlock(void);
150
151#ifdef CONFIG_LOCKDEP
152int lockdep_ovsl_is_held(void);
153#else
154#define lockdep_ovsl_is_held() 1
155#endif
156
Thomas Graf80019d32014-07-30 02:31:08 +0200157#define ASSERT_OVSL() WARN_ON(!lockdep_ovsl_is_held())
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700158#define ovsl_dereference(p) \
159 rcu_dereference_protected(p, lockdep_ovsl_is_held())
Jesse Gross663efa32013-12-03 10:58:53 -0800160#define rcu_dereference_ovsl(p) \
161 rcu_dereference_check(p, lockdep_ovsl_is_held())
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700162
Thomas Graf12eb18f2014-11-06 06:58:52 -0800163static inline struct net *ovs_dp_get_net(const struct datapath *dp)
Pravin B Shelar46df7b82012-02-22 19:58:59 -0800164{
165 return read_pnet(&dp->net);
166}
167
168static inline void ovs_dp_set_net(struct datapath *dp, struct net *net)
169{
170 write_pnet(&dp->net, net);
171}
172
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700173struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no);
174
175static inline struct vport *ovs_vport_rcu(const struct datapath *dp, int port_no)
176{
177 WARN_ON_ONCE(!rcu_read_lock_held());
178 return ovs_lookup_vport(dp, port_no);
179}
180
181static inline struct vport *ovs_vport_ovsl_rcu(const struct datapath *dp, int port_no)
182{
183 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
184 return ovs_lookup_vport(dp, port_no);
185}
186
187static inline struct vport *ovs_vport_ovsl(const struct datapath *dp, int port_no)
188{
189 ASSERT_OVSL();
190 return ovs_lookup_vport(dp, port_no);
191}
192
Jesse Grossccb13522011-10-25 19:26:31 -0700193extern struct notifier_block ovs_dp_device_notifier;
Johannes Berg68eb5502013-11-19 15:19:38 +0100194extern struct genl_family dp_vport_genl_family;
Jesse Grossccb13522011-10-25 19:26:31 -0700195
Pravin B Shelar8c8b1b82014-09-15 19:28:44 -0700196void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key);
Jesse Grossccb13522011-10-25 19:26:31 -0700197void ovs_dp_detach_port(struct vport *);
198int ovs_dp_upcall(struct datapath *, struct sk_buff *,
William Tuf2a4d082016-06-10 11:49:33 -0700199 const struct sw_flow_key *, const struct dp_upcall_info *,
200 uint32_t cutlen);
Jesse Grossccb13522011-10-25 19:26:31 -0700201
Andy Zhou971427f32014-09-15 19:37:25 -0700202const char *ovs_dp_name(const struct datapath *dp);
Jesse Grossccb13522011-10-25 19:26:31 -0700203struct sk_buff *ovs_vport_cmd_build_info(struct vport *, u32 pid, u32 seq,
204 u8 cmd);
205
Pravin B Shelar2ff3e4e2014-09-15 19:15:28 -0700206int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
Thomas Graf12eb18f2014-11-06 06:58:52 -0800207 const struct sw_flow_actions *, struct sw_flow_key *);
Andy Zhou971427f32014-09-15 19:37:25 -0700208
Pravin B Shelar8e4e1712013-04-15 13:23:03 -0700209void ovs_dp_notify_wq(struct work_struct *work);
Andy Zhou03f0d912013-08-07 20:01:00 -0700210
Andy Zhou971427f32014-09-15 19:37:25 -0700211int action_fifos_init(void);
212void action_fifos_exit(void);
213
Joe Stringerbe26b9a2015-08-26 11:31:45 -0700214/* 'KEY' must not have any bits set outside of the 'MASK' */
215#define OVS_MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
216#define OVS_SET_MASKED(OLD, KEY, MASK) ((OLD) = OVS_MASKED(OLD, KEY, MASK))
217
Jarno Rajahalme05da5892014-11-06 07:03:05 -0800218#define OVS_NLERR(logging_allowed, fmt, ...) \
Joe Perches1815a882014-02-03 17:06:46 -0800219do { \
Jarno Rajahalme05da5892014-11-06 07:03:05 -0800220 if (logging_allowed && net_ratelimit()) \
221 pr_info("netlink: " fmt "\n", ##__VA_ARGS__); \
Joe Perches1815a882014-02-03 17:06:46 -0800222} while (0)
Jesse Grossccb13522011-10-25 19:26:31 -0700223#endif /* datapath.h */