Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 1 | /* |
Andy Zhou | 971427f3 | 2014-09-15 19:37:25 -0700 | [diff] [blame] | 2 | * Copyright (c) 2007-2014 Nicira, Inc. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 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> |
Thomas Graf | 1d8fff9 | 2015-07-21 10:43:54 +0200 | [diff] [blame] | 28 | #include <net/ip_tunnels.h> |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 29 | |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 30 | #include "conntrack.h" |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 31 | #include "flow.h" |
Pravin B Shelar | e644571 | 2013-10-03 18:16:47 -0700 | [diff] [blame] | 32 | #include "flow_table.h" |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 33 | |
Pravin B Shelar | 15eac2a | 2012-08-23 12:40:54 -0700 | [diff] [blame] | 34 | #define DP_MAX_PORTS USHRT_MAX |
| 35 | #define DP_VPORT_HASH_BUCKETS 1024 |
| 36 | |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 37 | #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 Zhou | 1bd7116 | 2013-10-22 10:42:46 -0700 | [diff] [blame] | 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. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 53 | */ |
| 54 | struct dp_stats_percpu { |
| 55 | u64 n_hit; |
| 56 | u64 n_missed; |
| 57 | u64 n_lost; |
Andy Zhou | 1bd7116 | 2013-10-22 10:42:46 -0700 | [diff] [blame] | 58 | u64 n_mask_hit; |
WANG Cong | df9d9fd | 2014-02-14 15:10:46 -0800 | [diff] [blame] | 59 | struct u64_stats_sync syncp; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 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. |
Pravin B Shelar | b637e49 | 2013-10-04 00:14:23 -0700 | [diff] [blame] | 66 | * @table: flow table. |
Pravin B Shelar | 15eac2a | 2012-08-23 12:40:54 -0700 | [diff] [blame] | 67 | * @ports: Hash table for ports. %OVSP_LOCAL port always exists. Protected by |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 68 | * ovs_mutex and RCU. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 69 | * @stats_percpu: Per-CPU datapath statistics. |
Pravin B Shelar | 46df7b8 | 2012-02-22 19:58:59 -0800 | [diff] [blame] | 70 | * @net: Reference to net namespace. |
Paolo Abeni | 3a927bc | 2016-02-26 10:45:39 +0100 | [diff] [blame] | 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. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 73 | * |
| 74 | * Context: See the comment on locking at the top of datapath.c for additional |
| 75 | * locking information. |
| 76 | */ |
| 77 | struct datapath { |
| 78 | struct rcu_head rcu; |
| 79 | struct list_head list_node; |
| 80 | |
| 81 | /* Flow table. */ |
Pravin B Shelar | b637e49 | 2013-10-04 00:14:23 -0700 | [diff] [blame] | 82 | struct flow_table table; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 83 | |
| 84 | /* Switch ports. */ |
Pravin B Shelar | 15eac2a | 2012-08-23 12:40:54 -0700 | [diff] [blame] | 85 | struct hlist_head *ports; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 86 | |
| 87 | /* Stats. */ |
| 88 | struct dp_stats_percpu __percpu *stats_percpu; |
Pravin B Shelar | 46df7b8 | 2012-02-22 19:58:59 -0800 | [diff] [blame] | 89 | |
Pravin B Shelar | 46df7b8 | 2012-02-22 19:58:59 -0800 | [diff] [blame] | 90 | /* Network namespace ref. */ |
Eric W. Biederman | 0c5c9fb | 2015-03-11 23:06:44 -0500 | [diff] [blame] | 91 | possible_net_t net; |
Thomas Graf | 43d4be9 | 2013-12-13 15:22:18 +0100 | [diff] [blame] | 92 | |
| 93 | u32 user_features; |
Paolo Abeni | 3a927bc | 2016-02-26 10:45:39 +0100 | [diff] [blame] | 94 | |
| 95 | u32 max_headroom; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 96 | }; |
| 97 | |
| 98 | /** |
| 99 | * struct ovs_skb_cb - OVS data in skb CB |
Pravin B Shelar | 83c8df2 | 2014-09-15 19:20:31 -0700 | [diff] [blame] | 100 | * @input_vport: The original vport packet came in on. This value is cached |
| 101 | * when a packet is received by OVS. |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 102 | * @mru: The maximum received fragement size; 0 if the packet is not |
William Tu | f2a4d08 | 2016-06-10 11:49:33 -0700 | [diff] [blame] | 103 | * @cutlen: The number of bytes from the packet end to be removed. |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 104 | * fragmented. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 105 | */ |
| 106 | struct ovs_skb_cb { |
Pravin B Shelar | 83c8df2 | 2014-09-15 19:20:31 -0700 | [diff] [blame] | 107 | struct vport *input_vport; |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 108 | u16 mru; |
William Tu | f2a4d08 | 2016-06-10 11:49:33 -0700 | [diff] [blame] | 109 | u32 cutlen; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 110 | }; |
| 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 Pfaff | 4490108 | 2013-02-15 17:29:22 -0800 | [diff] [blame] | 116 | * @userdata: If nonnull, its variable-length value is passed to userspace as |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 117 | * %OVS_PACKET_ATTR_USERDATA. |
Pravin B Shelar | e8eedb8 | 2014-11-06 06:57:27 -0800 | [diff] [blame] | 118 | * @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 Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 120 | * counter. |
Wenyu Zhang | 8f0aad6 | 2014-11-06 06:51:24 -0800 | [diff] [blame] | 121 | * @egress_tun_info: If nonnull, becomes %OVS_PACKET_ATTR_EGRESS_TUN_KEY. |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 122 | * @mru: If not zero, Maximum received IP fragment size. |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 123 | */ |
| 124 | struct dp_upcall_info { |
Pravin B Shelar | 4c22279 | 2015-08-30 18:09:38 -0700 | [diff] [blame] | 125 | struct ip_tunnel_info *egress_tun_info; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 126 | const struct nlattr *userdata; |
Neil McKee | ccea744 | 2015-05-26 20:59:43 -0700 | [diff] [blame] | 127 | const struct nlattr *actions; |
| 128 | int actions_len; |
Eric W. Biederman | 15e4730 | 2012-09-07 20:12:54 +0000 | [diff] [blame] | 129 | u32 portid; |
Pravin B Shelar | e8eedb8 | 2014-11-06 06:57:27 -0800 | [diff] [blame] | 130 | u8 cmd; |
Joe Stringer | 7f8a436 | 2015-08-26 11:31:48 -0700 | [diff] [blame] | 131 | u16 mru; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 132 | }; |
| 133 | |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 134 | /** |
| 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 | */ |
| 139 | struct ovs_net { |
| 140 | struct list_head dps; |
| 141 | struct work_struct dp_notify_work; |
Joe Stringer | c2ac667 | 2015-08-26 11:31:52 -0700 | [diff] [blame] | 142 | |
| 143 | /* Module reference for configuring conntrack. */ |
| 144 | bool xt_label; |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 145 | }; |
| 146 | |
| 147 | extern int ovs_net_id; |
| 148 | void ovs_lock(void); |
| 149 | void ovs_unlock(void); |
| 150 | |
| 151 | #ifdef CONFIG_LOCKDEP |
| 152 | int lockdep_ovsl_is_held(void); |
| 153 | #else |
| 154 | #define lockdep_ovsl_is_held() 1 |
| 155 | #endif |
| 156 | |
Thomas Graf | 80019d3 | 2014-07-30 02:31:08 +0200 | [diff] [blame] | 157 | #define ASSERT_OVSL() WARN_ON(!lockdep_ovsl_is_held()) |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 158 | #define ovsl_dereference(p) \ |
| 159 | rcu_dereference_protected(p, lockdep_ovsl_is_held()) |
Jesse Gross | 663efa3 | 2013-12-03 10:58:53 -0800 | [diff] [blame] | 160 | #define rcu_dereference_ovsl(p) \ |
| 161 | rcu_dereference_check(p, lockdep_ovsl_is_held()) |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 162 | |
Thomas Graf | 12eb18f | 2014-11-06 06:58:52 -0800 | [diff] [blame] | 163 | static inline struct net *ovs_dp_get_net(const struct datapath *dp) |
Pravin B Shelar | 46df7b8 | 2012-02-22 19:58:59 -0800 | [diff] [blame] | 164 | { |
| 165 | return read_pnet(&dp->net); |
| 166 | } |
| 167 | |
| 168 | static inline void ovs_dp_set_net(struct datapath *dp, struct net *net) |
| 169 | { |
| 170 | write_pnet(&dp->net, net); |
| 171 | } |
| 172 | |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 173 | struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no); |
| 174 | |
| 175 | static 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 | |
| 181 | static 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 | |
| 187 | static 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 Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 193 | extern struct notifier_block ovs_dp_device_notifier; |
Johannes Berg | 68eb550 | 2013-11-19 15:19:38 +0100 | [diff] [blame] | 194 | extern struct genl_family dp_vport_genl_family; |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 195 | |
Pravin B Shelar | 8c8b1b8 | 2014-09-15 19:28:44 -0700 | [diff] [blame] | 196 | void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key); |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 197 | void ovs_dp_detach_port(struct vport *); |
| 198 | int ovs_dp_upcall(struct datapath *, struct sk_buff *, |
William Tu | f2a4d08 | 2016-06-10 11:49:33 -0700 | [diff] [blame] | 199 | const struct sw_flow_key *, const struct dp_upcall_info *, |
| 200 | uint32_t cutlen); |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 201 | |
Andy Zhou | 971427f3 | 2014-09-15 19:37:25 -0700 | [diff] [blame] | 202 | const char *ovs_dp_name(const struct datapath *dp); |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 203 | struct sk_buff *ovs_vport_cmd_build_info(struct vport *, u32 pid, u32 seq, |
| 204 | u8 cmd); |
| 205 | |
Pravin B Shelar | 2ff3e4e | 2014-09-15 19:15:28 -0700 | [diff] [blame] | 206 | int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, |
Thomas Graf | 12eb18f | 2014-11-06 06:58:52 -0800 | [diff] [blame] | 207 | const struct sw_flow_actions *, struct sw_flow_key *); |
Andy Zhou | 971427f3 | 2014-09-15 19:37:25 -0700 | [diff] [blame] | 208 | |
Pravin B Shelar | 8e4e171 | 2013-04-15 13:23:03 -0700 | [diff] [blame] | 209 | void ovs_dp_notify_wq(struct work_struct *work); |
Andy Zhou | 03f0d91 | 2013-08-07 20:01:00 -0700 | [diff] [blame] | 210 | |
Andy Zhou | 971427f3 | 2014-09-15 19:37:25 -0700 | [diff] [blame] | 211 | int action_fifos_init(void); |
| 212 | void action_fifos_exit(void); |
| 213 | |
Joe Stringer | be26b9a | 2015-08-26 11:31:45 -0700 | [diff] [blame] | 214 | /* '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 Rajahalme | 05da589 | 2014-11-06 07:03:05 -0800 | [diff] [blame] | 218 | #define OVS_NLERR(logging_allowed, fmt, ...) \ |
Joe Perches | 1815a88 | 2014-02-03 17:06:46 -0800 | [diff] [blame] | 219 | do { \ |
Jarno Rajahalme | 05da589 | 2014-11-06 07:03:05 -0800 | [diff] [blame] | 220 | if (logging_allowed && net_ratelimit()) \ |
| 221 | pr_info("netlink: " fmt "\n", ##__VA_ARGS__); \ |
Joe Perches | 1815a88 | 2014-02-03 17:06:46 -0800 | [diff] [blame] | 222 | } while (0) |
Jesse Gross | ccb1352 | 2011-10-25 19:26:31 -0700 | [diff] [blame] | 223 | #endif /* datapath.h */ |