datapath/datapath.h

   1 /*
   2  * Copyright (c) 2009, 2010 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Interface exported by openvswitch_mod. */
  10
  11 #ifndef DATAPATH_H
  12 #define DATAPATH_H 1
  13
  14 #include <asm/page.h>
  15 #include <linux/kernel.h>
  16 #include <linux/mutex.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/workqueue.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/version.h>
  21 #include "flow.h"
  22 #include "dp_sysfs.h"
  23
  24 /* Mask for the priority bits in a vlan header.  If we ever merge upstream
  25  * then this should go into include/linux/if_vlan.h. */
  26 #define VLAN_PCP_MASK 0xe000
  27 #define VLAN_PCP_SHIFT 13
  28
  29 #define DP_MAX_PORTS 1024
  30 #define DP_MAX_GROUPS 16
  31
  32 #define DP_L2_BITS (PAGE_SHIFT - ilog2(sizeof(struct dp_bucket*)))
  33 #define DP_L2_SIZE (1 << DP_L2_BITS)
  34 #define DP_L2_SHIFT 0
  35
  36 #define DP_L1_BITS (PAGE_SHIFT - ilog2(sizeof(struct dp_bucket**)))
  37 #define DP_L1_SIZE (1 << DP_L1_BITS)
  38 #define DP_L1_SHIFT DP_L2_BITS
  39
  40 /* For 4 kB pages, this is 1,048,576 on 32-bit or 262,144 on 64-bit. */
  41 #define DP_MAX_BUCKETS (DP_L1_SIZE * DP_L2_SIZE)
  42
  43 /**
  44  * struct dp_table - flow table
  45  * @n_buckets: number of buckets (a power of 2 between %DP_L1_SIZE and
  46  * %DP_MAX_BUCKETS)
  47  * @buckets: pointer to @n_buckets/%DP_L1_SIZE pointers to %DP_L1_SIZE pointers
  48  * to buckets
  49  * @hash_seed: random number used for flow hashing, to make the hash
  50  * distribution harder to predict
  51  * @rcu: RCU callback structure
  52  *
  53  * The @buckets array is logically an array of pointers to buckets.  It is
  54  * broken into two levels to avoid the need to kmalloc() any object larger than
  55  * a single page or to use vmalloc().  @buckets is always nonnull, as is each
  56  * @buckets[i], but each @buckets[i][j] is nonnull only if the specified hash
  57  * bucket is nonempty (for 0 <= i < @n_buckets/%DP_L1_SIZE, 0 <= j <
  58  * %DP_L1_SIZE).
  59  */
  60 struct dp_table {
  61         unsigned int n_buckets;
  62         struct dp_bucket ***buckets;
  63         unsigned int hash_seed;
  64         struct rcu_head rcu;
  65 };
  66
  67 /**
  68  * struct dp_bucket - single bucket within datapath flow table
  69  * @rcu: RCU callback structure
  70  * @n_flows: number of flows in @flows[] array
  71  * @flows: array of @n_flows pointers to flows
  72  *
  73  * The expected number of flows per bucket is 1, but this allows for an
  74  * arbitrary number of collisions.
  75  */
  76 struct dp_bucket {
  77         struct rcu_head rcu;
  78         unsigned int n_flows;
  79         struct sw_flow *flows[];
  80 };
  81
  82 #define DP_N_QUEUES 3
  83 #define DP_MAX_QUEUE_LEN 100
  84
  85 /**
  86  * struct dp_stats_percpu - per-cpu packet processing statistics for a given
  87  * datapath.
  88  * @n_frags: Number of IP fragments processed by datapath.
  89  * @n_hit: Number of received packets for which a matching flow was found in
  90  * the flow table.
  91  * @n_miss: Number of received packets that had no matching flow in the flow
  92  * table.  The sum of @n_hit and @n_miss is the number of packets that have
  93  * been received by the datapath.
  94  * @n_lost: Number of received packets that had no matching flow in the flow
  95  * table that could not be sent to userspace (normally due to an overflow in
  96  * one of the datapath's queues).
  97  */
  98 struct dp_stats_percpu {
  99         u64 n_frags;
 100         u64 n_hit;
 101         u64 n_missed;
 102         u64 n_lost;
 103 };
 104
 105 struct dp_port_group {
 106         struct rcu_head rcu;
 107         int n_ports;
 108         u16 ports[];
 109 };
 110
 111 /**
 112  * struct datapath - datapath for flow-based packet switching
 113  * @mutex: Mutual exclusion for ioctls.
 114  * @dp_idx: Datapath number (index into the dps[] array in datapath.c).
 115  * @ifobj: Represents /sys/class/net/<devname>/brif.
 116  * @drop_frags: Drop all IP fragments if nonzero.
 117  * @queues: %DP_N_QUEUES sets of queued packets for userspace to handle.
 118  * @waitqueue: Waitqueue, for waiting for new packets in @queues.
 119  * @n_flows: Number of flows currently in flow table.
 120  * @table: Current flow table (RCU protected).
 121  * @groups: Port groups, used by ODPAT_OUTPUT_GROUP action (RCU protected).
 122  * @n_ports: Number of ports currently in @ports.
 123  * @ports: Map from port number to &struct net_bridge_port.  %ODPP_LOCAL port
 124  * always exists, other ports may be %NULL.
 125  * @port_list: List of all ports in @ports in arbitrary order.
 126  * @stats_percpu: Per-CPU datapath statistics.
 127  * @sflow_probability: Number of packets out of UINT_MAX to sample to the
 128  * %ODPL_SFLOW queue, e.g. (@sflow_probability/UINT_MAX) is the probability of
 129  * sampling a given packet.
 130  */
 131 struct datapath {
 132         struct mutex mutex;
 133         int dp_idx;
 134         struct kobject ifobj;
 135
 136         int drop_frags;
 137
 138         /* Queued data. */
 139         struct sk_buff_head queues[DP_N_QUEUES];
 140         wait_queue_head_t waitqueue;
 141
 142         /* Flow table. */
 143         unsigned int n_flows;
 144         struct dp_table *table;
 145
 146         /* Port groups. */
 147         struct dp_port_group *groups[DP_MAX_GROUPS];
 148
 149         /* Switch ports. */
 150         unsigned int n_ports;
 151         struct net_bridge_port *ports[DP_MAX_PORTS];
 152         struct list_head port_list;
 153
 154         /* Stats. */
 155         struct dp_stats_percpu *stats_percpu;
 156
 157         /* sFlow Sampling */
 158         unsigned int sflow_probability;
 159 };
 160
 161 /**
 162  * struct net_bridge_port - one port within a datapath
 163  * @port_no: Index into @dp's @ports array.
 164  * @dp: Datapath to which this port belongs.
 165  * @dev: The network device attached to this port.  The @br_port member in @dev
 166  * points back to this &struct net_bridge_port.
 167  * @kobj: Represents /sys/class/net/<devname>/brport.
 168  * @linkname: The name of the link from /sys/class/net/<datapath>/brif to this
 169  * &struct net_bridge_port.  (We keep this around so that we can delete it
 170  * if @dev gets renamed.)  Set to the null string when no link exists.
 171  * @node: Element in @dp's @port_list.
 172  * @sflow_pool: Number of packets that were candidates for sFlow sampling,
 173  * regardless of whether they were actually chosen and sent down to userspace.
 174  */
 175 struct net_bridge_port {
 176         u16 port_no;
 177         struct datapath *dp;
 178         struct net_device *dev;
 179         struct kobject kobj;
 180         char linkname[IFNAMSIZ];
 181         struct list_head node;
 182         atomic_t sflow_pool;
 183 };
 184
 185 enum csum_type {
 186         OVS_CSUM_NONE = 0,
 187         OVS_CSUM_UNNECESSARY = 1,
 188         OVS_CSUM_COMPLETE = 2,
 189         OVS_CSUM_PARTIAL = 3,
 190 };
 191
 192 /**
 193  * struct ovs_skb_cb - OVS data in skb CB
 194  * @ip_summed: Consistently stores L4 checksumming status across different
 195  * kernel versions.
 196  */
 197 struct ovs_skb_cb {
 198         enum csum_type          ip_summed;
 199         __be32                  tun_id;
 200 };
 201 #define OVS_CB(skb) ((struct ovs_skb_cb *)(skb)->cb)
 202
 203 extern struct notifier_block dp_device_notifier;
 204 extern int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
 205
 206 /* Flow table. */
 207 struct dp_table *dp_table_create(unsigned int n_buckets);
 208 void dp_table_destroy(struct dp_table *, int free_flows);
 209 struct sw_flow *dp_table_lookup(struct dp_table *, const struct odp_flow_key *);
 210 int dp_table_insert(struct dp_table *, struct sw_flow *);
 211 int dp_table_delete(struct dp_table *, struct sw_flow *);
 212 int dp_table_expand(struct datapath *);
 213 int dp_table_flush(struct datapath *);
 214 int dp_table_foreach(struct dp_table *table,
 215                      int (*callback)(struct sw_flow *flow, void *aux),
 216                      void *aux);
 217
 218 void dp_process_received_packet(struct sk_buff *, struct net_bridge_port *);
 219 int dp_del_port(struct net_bridge_port *);
 220 int dp_output_control(struct datapath *, struct sk_buff *, int, u32 arg);
 221 int dp_min_mtu(const struct datapath *dp);
 222 void set_dp_devs_mtu(const struct datapath *dp, struct net_device *dev);
 223
 224 struct datapath *get_dp(int dp_idx);
 225
 226 static inline const char *dp_name(const struct datapath *dp)
 227 {
 228         return dp->ports[ODPP_LOCAL]->dev->name;
 229 }
 230
 231 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
 232 int vswitch_skb_checksum_setup(struct sk_buff *skb);
 233 #else
 234 static inline int vswitch_skb_checksum_setup(struct sk_buff *skb)
 235 {
 236         return 0;
 237 }
 238 #endif
 239
 240 void compute_ip_summed(struct sk_buff *skb, bool xmit);
 241 void forward_ip_summed(struct sk_buff *skb);
 242
 243 #endif /* datapath.h */