lib/odp-util.h

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #ifndef ODP_UTIL_H
  18 #define ODP_UTIL_H 1
  19
  20 #include <stdbool.h>
  21 #include <stddef.h>
  22 #include <stdint.h>
  23 #include <string.h>
  24 #include "flow.h"
  25 #include "hash.h"
  26 #include "openvswitch/hmap.h"
  27 #include "openvswitch/ofp-actions.h"
  28 #include "odp-netlink.h"
  29 #include "openflow/openflow.h"
  30 #include "util.h"
  31
  32 struct ds;
  33 struct nlattr;
  34 struct ofpbuf;
  35 struct simap;
  36 struct pkt_metadata;
  37
  38 #define SLOW_PATH_REASONS                                               \
  39     /* These reasons are mutually exclusive. */                         \
  40     SPR(SLOW_CFM,        "cfm",        "Consists of CFM packets")       \
  41     SPR(SLOW_BFD,        "bfd",        "Consists of BFD packets")       \
  42     SPR(SLOW_LACP,       "lacp",       "Consists of LACP packets")      \
  43     SPR(SLOW_STP,        "stp",        "Consists of STP packets")       \
  44     SPR(SLOW_LLDP,       "lldp",       "Consists of LLDP packets")    \
  45     SPR(SLOW_CONTROLLER, "controller",                                  \
  46         "Sends \"packet-in\" messages to the OpenFlow controller")      \
  47     SPR(SLOW_ACTION,     "action",                                      \
  48         "Uses action(s) not supported by datapath")
  49
  50 /* Indexes for slow-path reasons.  Client code uses "enum slow_path_reason"
  51  * values instead of these, these are just a way to construct those. */
  52 enum {
  53 #define SPR(ENUM, STRING, EXPLANATION) ENUM##_INDEX,
  54     SLOW_PATH_REASONS
  55 #undef SPR
  56 };
  57
  58 /* Reasons why a subfacet might not be fast-pathable.
  59  *
  60  * Each reason is a separate bit to allow reasons to be combined. */
  61 enum slow_path_reason {
  62 #define SPR(ENUM, STRING, EXPLANATION) ENUM = 1 << ENUM##_INDEX,
  63     SLOW_PATH_REASONS
  64 #undef SPR
  65 };
  66
  67 /* Mask of all slow_path_reasons. */
  68 enum {
  69     SLOW_PATH_REASON_MASK = 0
  70 #define SPR(ENUM, STRING, EXPLANATION) | 1 << ENUM##_INDEX
  71     SLOW_PATH_REASONS
  72 #undef SPR
  73 };
  74
  75 const char *slow_path_reason_to_explanation(enum slow_path_reason);
  76
  77 #define ODPP_LOCAL ODP_PORT_C(OVSP_LOCAL)
  78 #define ODPP_NONE  ODP_PORT_C(UINT32_MAX)
  79
  80 void format_odp_actions(struct ds *, const struct nlattr *odp_actions,
  81                         size_t actions_len);
  82 int odp_actions_from_string(const char *, const struct simap *port_names,
  83                             struct ofpbuf *odp_actions);
  84
  85 /* A map from odp port number to its name. */
  86 struct odp_portno_names {
  87     struct hmap_node hmap_node; /* A node in a port number to name hmap. */
  88     odp_port_t port_no;         /* Port number in the datapath. */
  89     char *name;                 /* Name associated with the above 'port_no'. */
  90 };
  91
  92 void odp_portno_names_set(struct hmap *portno_names, odp_port_t port_no,
  93                           char *port_name);
  94 void odp_portno_names_destroy(struct hmap *portno_names);
  95 /* The maximum number of bytes that odp_flow_key_from_flow() appends to a
  96  * buffer.  This is the upper bound on the length of a nlattr-formatted flow
  97  * key that ovs-vswitchd fully understands.
  98  *
  99  * OVS doesn't insist that ovs-vswitchd and the datapath have exactly the same
 100  * idea of a flow, so therefore this value isn't necessarily an upper bound on
 101  * the length of a flow key that the datapath can pass to ovs-vswitchd.
 102  *
 103  * The longest nlattr-formatted flow key appended by odp_flow_key_from_flow()
 104  * would be:
 105  *
 106  *                                     struct  pad  nl hdr  total
 107  *                                     ------  ---  ------  -----
 108  *  OVS_KEY_ATTR_PRIORITY                4    --     4      8
 109  *  OVS_KEY_ATTR_TUNNEL                  0    --     4      4
 110  *  - OVS_TUNNEL_KEY_ATTR_ID             8    --     4     12
 111  *  - OVS_TUNNEL_KEY_ATTR_IPV4_SRC       4    --     4      8
 112  *  - OVS_TUNNEL_KEY_ATTR_IPV4_DST       4    --     4      8
 113  *  - OVS_TUNNEL_KEY_ATTR_IPV6_SRC       16   --     4     20
 114  *  - OVS_TUNNEL_KEY_ATTR_IPV6_DST       16   --     4     20
 115  *  - OVS_TUNNEL_KEY_ATTR_TOS            1    3      4      8
 116  *  - OVS_TUNNEL_KEY_ATTR_TTL            1    3      4      8
 117  *  - OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT  0    --     4      4
 118  *  - OVS_TUNNEL_KEY_ATTR_CSUM           0    --     4      4
 119  *  - OVS_TUNNEL_KEY_ATTR_OAM            0    --     4      4
 120  *  - OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS    256  --     4      260
 121  *  - OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS     -    --     -      - (shared with _GENEVE_OPTS)
 122  *  OVS_KEY_ATTR_IN_PORT                 4    --     4      8
 123  *  OVS_KEY_ATTR_SKB_MARK                4    --     4      8
 124  *  OVS_KEY_ATTR_DP_HASH                 4    --     4      8
 125  *  OVS_KEY_ATTR_RECIRC_ID               4    --     4      8
 126  *  OVS_KEY_ATTR_CT_STATE                4    --     4      8
 127  *  OVS_KEY_ATTR_CT_ZONE                 2     2     4      8
 128  *  OVS_KEY_ATTR_CT_MARK                 4    --     4      8
 129  *  OVS_KEY_ATTR_CT_LABEL               16    --     4     20
 130  *  OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6     40    --     4     44
 131  *  OVS_KEY_ATTR_ETHERNET               12    --     4     16
 132  *  OVS_KEY_ATTR_ETHERTYPE               2     2     4      8  (outer VLAN ethertype)
 133  *  OVS_KEY_ATTR_VLAN                    2     2     4      8
 134  *  OVS_KEY_ATTR_ENCAP                   0    --     4      4  (VLAN encapsulation)
 135  *  OVS_KEY_ATTR_ETHERTYPE               2     2     4      8  (inner VLAN ethertype)
 136  *  OVS_KEY_ATTR_IPV6                   40    --     4     44
 137  *  OVS_KEY_ATTR_ICMPV6                  2     2     4      8
 138  *  OVS_KEY_ATTR_ND                     28    --     4     32
 139  *  ----------------------------------------------------------
 140  *  total                                                 616
 141  *
 142  * We include some slack space in case the calculation isn't quite right or we
 143  * add another field and forget to adjust this value.
 144  */
 145 #define ODPUTIL_FLOW_KEY_BYTES 640
 146 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 38);
 147
 148 /* A buffer with sufficient size and alignment to hold an nlattr-formatted flow
 149  * key.  An array of "struct nlattr" might not, in theory, be sufficiently
 150  * aligned because it only contains 16-bit types. */
 151 struct odputil_keybuf {
 152     uint32_t keybuf[DIV_ROUND_UP(ODPUTIL_FLOW_KEY_BYTES, 4)];
 153 };
 154
 155 enum odp_key_fitness odp_tun_key_from_attr(const struct nlattr *,
 156                                            struct flow_tnl *);
 157
 158 int odp_ufid_from_string(const char *s_, ovs_u128 *ufid);
 159 void odp_format_ufid(const ovs_u128 *ufid, struct ds *);
 160
 161 void odp_flow_format(const struct nlattr *key, size_t key_len,
 162                      const struct nlattr *mask, size_t mask_len,
 163                      const struct hmap *portno_names, struct ds *,
 164                      bool verbose);
 165 void odp_flow_key_format(const struct nlattr *, size_t, struct ds *);
 166 int odp_flow_from_string(const char *s,
 167                          const struct simap *port_names,
 168                          struct ofpbuf *, struct ofpbuf *);
 169
 170 /* ODP_SUPPORT_FIELD(TYPE, FIELD_NAME, FIELD_DESCRIPTION)
 171  *
 172  * Each 'ODP_SUPPORT_FIELD' defines a member in 'struct odp_support',
 173  * and represents support for related OVS_KEY_ATTR_* fields.
 174  * They are defined as macros to keep 'dpif_show_support()' in sync
 175  * as new fields are added.   */
 176 #define ODP_SUPPORT_FIELDS                                                   \
 177     /* Maximum number of 802.1q VLAN headers to serialize in a mask. */      \
 178     ODP_SUPPORT_FIELD(size_t, max_vlan_headers, "Max VLAN headers")          \
 179     /* Maximum number of MPLS label stack entries to serialise in a mask. */ \
 180     ODP_SUPPORT_FIELD(size_t, max_mpls_depth, "Max MPLS depth")              \
 181     /* If this is true, then recirculation fields will always be             \
 182      * serialised. */                                                        \
 183     ODP_SUPPORT_FIELD(bool, recirc, "Recirc")                                \
 184     /* If true, serialise the corresponding OVS_KEY_ATTR_CONN_* field. */    \
 185     ODP_SUPPORT_FIELD(bool, ct_state, "CT state")                            \
 186     ODP_SUPPORT_FIELD(bool, ct_zone, "CT zone")                              \
 187     ODP_SUPPORT_FIELD(bool, ct_mark, "CT mark")                              \
 188     ODP_SUPPORT_FIELD(bool, ct_label, "CT label")                            \
 189                                                                              \
 190     /* If true, it means that the datapath supports the NAT bits in          \
 191      * 'ct_state'.  The above 'ct_state' member must be true for this        \
 192      * to make sense */                                                      \
 193     ODP_SUPPORT_FIELD(bool, ct_state_nat, "CT state NAT")                    \
 194                                                                              \
 195     /* Conntrack original direction tuple matching * supported. */           \
 196     ODP_SUPPORT_FIELD(bool, ct_orig_tuple, "CT orig tuple")
 197
 198 /* Indicates support for various fields. This defines how flows will be
 199  * serialised. */
 200 struct odp_support {
 201 #define ODP_SUPPORT_FIELD(TYPE, NAME, TITLE) TYPE NAME;
 202     ODP_SUPPORT_FIELDS
 203 #undef ODP_SUPPORT_FIELD
 204 };
 205
 206 struct odp_flow_key_parms {
 207     /* The flow and mask to be serialized. In the case of masks, 'flow'
 208      * is used as a template to determine how to interpret 'mask'.  For
 209      * example, the 'dl_type' of 'mask' describes the mask, but it doesn't
 210      * indicate whether the other fields should be interpreted as ARP, IPv4,
 211      * IPv6, etc. */
 212     const struct flow *flow;
 213     const struct flow *mask;
 214
 215     /* Indicates support for various fields. If the datapath supports a field,
 216      * then it will always be serialised. */
 217     struct odp_support support;
 218
 219     /* Indicates if we are probing datapath capability. If true, ignore the
 220      * configured flow limits. */
 221     bool probe;
 222
 223     /* The netlink formatted version of the flow. It is used in cases where
 224      * the mask cannot be constructed from the OVS internal representation
 225      * and needs to see the original form. */
 226     const struct ofpbuf *key_buf;
 227 };
 228
 229 void odp_flow_key_from_flow(const struct odp_flow_key_parms *, struct ofpbuf *);
 230 void odp_flow_key_from_mask(const struct odp_flow_key_parms *, struct ofpbuf *);
 231
 232 uint32_t odp_flow_key_hash(const struct nlattr *, size_t);
 233
 234 /* Estimated space needed for metadata. */
 235 enum { ODP_KEY_METADATA_SIZE = 9 * 8 };
 236 void odp_key_from_pkt_metadata(struct ofpbuf *, const struct pkt_metadata *);
 237 void odp_key_to_pkt_metadata(const struct nlattr *key, size_t key_len,
 238                               struct pkt_metadata *md);
 239
 240 /* How well a kernel-provided flow key (a sequence of OVS_KEY_ATTR_*
 241  * attributes) matches OVS userspace expectations.
 242  *
 243  * These values are arranged so that greater values are "more important" than
 244  * lesser ones.  In particular, a single flow key can fit the descriptions for
 245  * both ODP_FIT_TOO_LITTLE and ODP_FIT_TOO_MUCH.  Such a key is treated as
 246  * ODP_FIT_TOO_LITTLE. */
 247 enum odp_key_fitness {
 248     ODP_FIT_PERFECT,            /* The key had exactly the fields we expect. */
 249     ODP_FIT_TOO_MUCH,           /* The key had fields we don't understand. */
 250     ODP_FIT_TOO_LITTLE,         /* The key lacked fields we expected to see. */
 251     ODP_FIT_ERROR,              /* The key was invalid. */
 252 };
 253 enum odp_key_fitness odp_flow_key_to_flow(const struct nlattr *, size_t,
 254                                           struct flow *);
 255 enum odp_key_fitness odp_flow_key_to_mask(const struct nlattr *mask_key,
 256                                           size_t mask_key_len,
 257                                           struct flow_wildcards *mask,
 258                                           const struct flow *flow);
 259
 260 const char *odp_key_fitness_to_string(enum odp_key_fitness);
 261
 262 void commit_odp_tunnel_action(const struct flow *, struct flow *base,
 263                               struct ofpbuf *odp_actions);
 264 void commit_masked_set_action(struct ofpbuf *odp_actions,
 265                               enum ovs_key_attr key_type, const void *key,
 266                               const void *mask, size_t key_size);
 267 enum slow_path_reason commit_odp_actions(const struct flow *,
 268                                          struct flow *base,
 269                                          struct ofpbuf *odp_actions,
 270                                          struct flow_wildcards *wc,
 271                                          bool use_masked);
 272 \f
 273 /* ofproto-dpif interface.
 274  *
 275  * The following types and functions are logically part of ofproto-dpif.
 276  * ofproto-dpif puts values of these types into the flows that it installs in
 277  * the kernel datapath, though, so ovs-dpctl needs to interpret them so that
 278  * it can print flows in a more human-readable manner. */
 279
 280 enum user_action_cookie_type {
 281     USER_ACTION_COOKIE_UNSPEC,
 282     USER_ACTION_COOKIE_SFLOW,        /* Packet for per-bridge sFlow sampling. */
 283     USER_ACTION_COOKIE_SLOW_PATH,    /* Userspace must process this flow. */
 284     USER_ACTION_COOKIE_FLOW_SAMPLE,  /* Packet for per-flow sampling. */
 285     USER_ACTION_COOKIE_IPFIX,        /* Packet for per-bridge IPFIX sampling. */
 286 };
 287
 288 /* user_action_cookie is passed as argument to OVS_ACTION_ATTR_USERSPACE. */
 289 union user_action_cookie {
 290     uint16_t type;              /* enum user_action_cookie_type. */
 291
 292     struct {
 293         uint16_t type;          /* USER_ACTION_COOKIE_SFLOW. */
 294         ovs_be16 vlan_tci;      /* Destination VLAN TCI. */
 295         uint32_t output;        /* SFL_FLOW_SAMPLE_TYPE 'output' value. */
 296     } sflow;
 297
 298     struct {
 299         uint16_t type;          /* USER_ACTION_COOKIE_SLOW_PATH. */
 300         uint16_t unused;
 301         uint32_t reason;        /* enum slow_path_reason. */
 302     } slow_path;
 303
 304     struct {
 305         uint16_t type;          /* USER_ACTION_COOKIE_FLOW_SAMPLE. */
 306         uint16_t probability;   /* Sampling probability. */
 307         uint32_t collector_set_id; /* ID of IPFIX collector set. */
 308         uint32_t obs_domain_id; /* Observation Domain ID. */
 309         uint32_t obs_point_id;  /* Observation Point ID. */
 310         odp_port_t output_odp_port; /* The output odp port. */
 311         enum nx_action_sample_direction direction;
 312     } flow_sample;
 313
 314     struct {
 315         uint16_t   type;            /* USER_ACTION_COOKIE_IPFIX. */
 316         odp_port_t output_odp_port; /* The output odp port. */
 317     } ipfix;
 318 };
 319 BUILD_ASSERT_DECL(sizeof(union user_action_cookie) == 24);
 320
 321 size_t odp_put_userspace_action(uint32_t pid,
 322                                 const void *userdata, size_t userdata_size,
 323                                 odp_port_t tunnel_out_port,
 324                                 bool include_actions,
 325                                 struct ofpbuf *odp_actions);
 326 void odp_put_tunnel_action(const struct flow_tnl *tunnel,
 327                            struct ofpbuf *odp_actions);
 328
 329 void odp_put_tnl_push_action(struct ofpbuf *odp_actions,
 330                              struct ovs_action_push_tnl *data);
 331 #endif /* odp-util.h */