2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
26 #include "openvswitch/hmap.h"
27 #include "openvswitch/ofp-actions.h"
28 #include "openvswitch/uuid.h"
29 #include "odp-netlink.h"
30 #include "openflow/openflow.h"
39 #define SLOW_PATH_REASONS \
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_ACTION, "action", \
46 "Uses action(s) not supported by datapath") \
47 SPR(SLOW_MATCH, "match", \
48 "Datapath can't match specifically enough")
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. */
53 #define SPR(ENUM, STRING, EXPLANATION) ENUM##_INDEX,
58 /* Reasons why a flow might not be fast-pathable.
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,
67 /* Mask of all slow_path_reasons. */
69 SLOW_PATH_REASON_MASK
= 0
70 #define SPR(ENUM, STRING, EXPLANATION) | 1 << ENUM##_INDEX
75 const char *slow_path_reason_to_explanation(enum slow_path_reason
);
77 #define ODPP_LOCAL ODP_PORT_C(OVSP_LOCAL)
78 #define ODPP_NONE ODP_PORT_C(UINT32_MAX)
80 void format_odp_actions(struct ds
*, const struct nlattr
*odp_actions
,
81 size_t actions_len
, const struct hmap
*portno_names
);
82 int odp_actions_from_string(const char *, const struct simap
*port_names
,
83 struct ofpbuf
*odp_actions
);
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'. */
92 void odp_portno_names_set(struct hmap
*portno_names
, odp_port_t port_no
,
94 void odp_portno_names_destroy(struct hmap
*portno_names
);
95 void odp_portno_name_format(const struct hmap
*portno_names
,
96 odp_port_t
, struct ds
*);
98 /* The maximum number of bytes that odp_flow_key_from_flow() appends to a
99 * buffer. This is the upper bound on the length of a nlattr-formatted flow
100 * key that ovs-vswitchd fully understands.
102 * OVS doesn't insist that ovs-vswitchd and the datapath have exactly the same
103 * idea of a flow, so therefore this value isn't necessarily an upper bound on
104 * the length of a flow key that the datapath can pass to ovs-vswitchd.
106 * The longest nlattr-formatted flow key appended by odp_flow_key_from_flow()
109 * struct pad nl hdr total
110 * ------ --- ------ -----
111 * OVS_KEY_ATTR_PRIORITY 4 -- 4 8
112 * OVS_KEY_ATTR_TUNNEL 0 -- 4 4
113 * - OVS_TUNNEL_KEY_ATTR_ID 8 -- 4 12
114 * - OVS_TUNNEL_KEY_ATTR_IPV4_SRC 4 -- 4 8
115 * - OVS_TUNNEL_KEY_ATTR_IPV4_DST 4 -- 4 8
116 * - OVS_TUNNEL_KEY_ATTR_IPV6_SRC 16 -- 4 20
117 * - OVS_TUNNEL_KEY_ATTR_IPV6_DST 16 -- 4 20
118 * - OVS_TUNNEL_KEY_ATTR_TOS 1 3 4 8
119 * - OVS_TUNNEL_KEY_ATTR_TTL 1 3 4 8
120 * - OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT 0 -- 4 4
121 * - OVS_TUNNEL_KEY_ATTR_CSUM 0 -- 4 4
122 * - OVS_TUNNEL_KEY_ATTR_OAM 0 -- 4 4
123 * - OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS 256 -- 4 260
124 * - OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS - -- - - (shared with _GENEVE_OPTS)
125 * OVS_KEY_ATTR_IN_PORT 4 -- 4 8
126 * OVS_KEY_ATTR_SKB_MARK 4 -- 4 8
127 * OVS_KEY_ATTR_DP_HASH 4 -- 4 8
128 * OVS_KEY_ATTR_RECIRC_ID 4 -- 4 8
129 * OVS_KEY_ATTR_CT_STATE 4 -- 4 8
130 * OVS_KEY_ATTR_CT_ZONE 2 2 4 8
131 * OVS_KEY_ATTR_CT_MARK 4 -- 4 8
132 * OVS_KEY_ATTR_CT_LABEL 16 -- 4 20
133 * OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6 40 -- 4 44
134 * OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4 - -- - - (exclusive of_CT_ORIG_TUPLE_IPV6)
135 * OVS_KEY_ATTR_ETHERNET 12 -- 4 16
136 * OVS_KEY_ATTR_ETHERTYPE 2 2 4 8 (outer VLAN ethertype)
137 * OVS_KEY_ATTR_VLAN 2 2 4 8
138 * OVS_KEY_ATTR_ENCAP 0 -- 4 4 (VLAN encapsulation)
139 * OVS_KEY_ATTR_ETHERTYPE 2 2 4 8 (inner VLAN ethertype)
140 * OVS_KEY_ATTR_IPV6 40 -- 4 44
141 * OVS_KEY_ATTR_ICMPV6 2 2 4 8
142 * OVS_KEY_ATTR_ND 28 -- 4 32
143 * ----------------------------------------------------------
146 * We include some slack space in case the calculation isn't quite right or we
147 * add another field and forget to adjust this value.
149 #define ODPUTIL_FLOW_KEY_BYTES 640
150 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 41);
152 /* A buffer with sufficient size and alignment to hold an nlattr-formatted flow
153 * key. An array of "struct nlattr" might not, in theory, be sufficiently
154 * aligned because it only contains 16-bit types. */
155 struct odputil_keybuf
{
156 uint32_t keybuf
[DIV_ROUND_UP(ODPUTIL_FLOW_KEY_BYTES
, 4)];
159 enum odp_key_fitness
odp_tun_key_from_attr(const struct nlattr
*,
160 struct flow_tnl
*, char **errorp
);
161 enum odp_key_fitness
odp_nsh_key_from_attr(const struct nlattr
*,
162 struct ovs_key_nsh
*,
163 struct ovs_key_nsh
*,
165 enum odp_key_fitness
odp_nsh_hdr_from_attr(const struct nlattr
*,
166 struct nsh_hdr
*, size_t);
168 int odp_ufid_from_string(const char *s_
, ovs_u128
*ufid
);
169 void odp_format_ufid(const ovs_u128
*ufid
, struct ds
*);
171 void odp_flow_format(const struct nlattr
*key
, size_t key_len
,
172 const struct nlattr
*mask
, size_t mask_len
,
173 const struct hmap
*portno_names
, struct ds
*,
175 void odp_flow_key_format(const struct nlattr
*, size_t, struct ds
*);
176 int odp_flow_from_string(const char *s
, const struct simap
*port_names
,
177 struct ofpbuf
*, struct ofpbuf
*, char **errorp
);
179 /* ODP_SUPPORT_FIELD(TYPE, FIELD_NAME, FIELD_DESCRIPTION)
181 * Each 'ODP_SUPPORT_FIELD' defines a member in 'struct odp_support',
182 * and represents support for related OVS_KEY_ATTR_* fields.
183 * They are defined as macros to keep 'dpif_show_support()' in sync
184 * as new fields are added. */
185 #define ODP_SUPPORT_FIELDS \
186 /* Maximum number of 802.1q VLAN headers to serialize in a mask. */ \
187 ODP_SUPPORT_FIELD(size_t, max_vlan_headers, "Max VLAN headers") \
188 /* Maximum number of MPLS label stack entries to serialise in a mask. */ \
189 ODP_SUPPORT_FIELD(size_t, max_mpls_depth, "Max MPLS depth") \
190 /* If this is true, then recirculation fields will always be \
192 ODP_SUPPORT_FIELD(bool, recirc, "Recirc") \
193 /* If true, serialise the corresponding OVS_KEY_ATTR_CONN_* field. */ \
194 ODP_SUPPORT_FIELD(bool, ct_state, "CT state") \
195 ODP_SUPPORT_FIELD(bool, ct_zone, "CT zone") \
196 ODP_SUPPORT_FIELD(bool, ct_mark, "CT mark") \
197 ODP_SUPPORT_FIELD(bool, ct_label, "CT label") \
199 /* If true, it means that the datapath supports the NAT bits in \
200 * 'ct_state'. The above 'ct_state' member must be true for this \
202 ODP_SUPPORT_FIELD(bool, ct_state_nat, "CT state NAT") \
204 /* Conntrack original direction tuple matching * supported. */ \
205 ODP_SUPPORT_FIELD(bool, ct_orig_tuple, "CT orig tuple") \
206 ODP_SUPPORT_FIELD(bool, ct_orig_tuple6, "CT orig tuple for IPv6") \
208 /* If true, it means that the datapath supports the IPv6 Neigh \
209 * Discovery Extension bits. */ \
210 ODP_SUPPORT_FIELD(bool, nd_ext, "IPv6 ND Extension")
212 /* Indicates support for various fields. This defines how flows will be
215 #define ODP_SUPPORT_FIELD(TYPE, NAME, TITLE) TYPE NAME;
217 #undef ODP_SUPPORT_FIELD
220 struct odp_flow_key_parms
{
221 /* The flow and mask to be serialized. In the case of masks, 'flow'
222 * is used as a template to determine how to interpret 'mask'. For
223 * example, the 'dl_type' of 'mask' describes the mask, but it doesn't
224 * indicate whether the other fields should be interpreted as ARP, IPv4,
226 const struct flow
*flow
;
227 const struct flow
*mask
;
229 /* Indicates support for various fields. If the datapath supports a field,
230 * then it will always be serialised. */
231 struct odp_support support
;
233 /* Indicates if we are probing datapath capability. If true, ignore the
234 * configured flow limits. */
237 /* The netlink formatted version of the flow. It is used in cases where
238 * the mask cannot be constructed from the OVS internal representation
239 * and needs to see the original form. */
240 const struct ofpbuf
*key_buf
;
243 void odp_flow_key_from_flow(const struct odp_flow_key_parms
*, struct ofpbuf
*);
244 void odp_flow_key_from_mask(const struct odp_flow_key_parms
*, struct ofpbuf
*);
246 void odp_flow_key_hash(const void *key
, size_t key_len
, ovs_u128
*hash
);
248 /* Estimated space needed for metadata. */
249 enum { ODP_KEY_METADATA_SIZE
= 9 * 8 };
250 void odp_key_from_dp_packet(struct ofpbuf
*, const struct dp_packet
*);
251 void odp_key_to_dp_packet(const struct nlattr
*key
, size_t key_len
,
252 struct dp_packet
*md
);
254 /* How well a kernel-provided flow key (a sequence of OVS_KEY_ATTR_*
255 * attributes) matches OVS userspace expectations.
257 * These values are arranged so that greater values are "more important" than
258 * lesser ones. In particular, a single flow key can fit the descriptions for
259 * both ODP_FIT_TOO_LITTLE and ODP_FIT_TOO_MUCH. Such a key is treated as
260 * ODP_FIT_TOO_LITTLE. */
261 enum odp_key_fitness
{
262 ODP_FIT_PERFECT
, /* The key had exactly the fields we expect. */
263 ODP_FIT_TOO_MUCH
, /* The key had fields we don't understand. */
264 ODP_FIT_TOO_LITTLE
, /* The key lacked fields we expected to see. */
265 ODP_FIT_ERROR
, /* The key was invalid. */
267 enum odp_key_fitness
odp_flow_key_to_flow(const struct nlattr
*, size_t,
268 struct flow
*, char **errorp
);
269 enum odp_key_fitness
odp_flow_key_to_mask(const struct nlattr
*mask_key
,
271 struct flow_wildcards
*mask
,
272 const struct flow
*flow
,
274 int parse_key_and_mask_to_match(const struct nlattr
*key
, size_t key_len
,
275 const struct nlattr
*mask
, size_t mask_len
,
276 struct match
*match
);
278 const char *odp_key_fitness_to_string(enum odp_key_fitness
);
280 void commit_odp_tunnel_action(const struct flow
*, struct flow
*base
,
281 struct ofpbuf
*odp_actions
,
282 const char *tnl_type
);
283 void commit_masked_set_action(struct ofpbuf
*odp_actions
,
284 enum ovs_key_attr key_type
, const void *key
,
285 const void *mask
, size_t key_size
);
286 enum slow_path_reason
commit_odp_actions(const struct flow
*,
288 struct ofpbuf
*odp_actions
,
289 struct flow_wildcards
*wc
,
293 struct ofpbuf
*encap_data
);
295 /* ofproto-dpif interface.
297 * The following types and functions are logically part of ofproto-dpif.
298 * ofproto-dpif puts values of these types into the flows that it installs in
299 * the kernel datapath, though, so ovs-dpctl needs to interpret them so that
300 * it can print flows in a more human-readable manner. */
302 enum user_action_cookie_type
{
303 USER_ACTION_COOKIE_UNSPEC
,
304 USER_ACTION_COOKIE_SFLOW
, /* Packet for per-bridge sFlow sampling. */
305 USER_ACTION_COOKIE_SLOW_PATH
, /* Userspace must process this flow. */
306 USER_ACTION_COOKIE_FLOW_SAMPLE
, /* Packet for per-flow sampling. */
307 USER_ACTION_COOKIE_IPFIX
, /* Packet for per-bridge IPFIX sampling. */
308 USER_ACTION_COOKIE_CONTROLLER
, /* Forward packet to controller. */
311 /* user_action_cookie is passed as argument to OVS_ACTION_ATTR_USERSPACE. */
312 struct user_action_cookie
{
313 uint16_t type
; /* enum user_action_cookie_type. */
314 ofp_port_t ofp_in_port
; /* OpenFlow in port, or OFPP_NONE. */
315 struct uuid ofproto_uuid
; /* UUID of ofproto-dpif. */
319 /* USER_ACTION_COOKIE_SFLOW. */
320 ovs_be16 vlan_tci
; /* Destination VLAN TCI. */
321 uint32_t output
; /* SFL_FLOW_SAMPLE_TYPE 'output' value. */
325 /* USER_ACTION_COOKIE_SLOW_PATH. */
327 uint32_t reason
; /* enum slow_path_reason. */
331 /* USER_ACTION_COOKIE_FLOW_SAMPLE. */
332 uint16_t probability
; /* Sampling probability. */
333 uint32_t collector_set_id
; /* ID of IPFIX collector set. */
334 uint32_t obs_domain_id
; /* Observation Domain ID. */
335 uint32_t obs_point_id
; /* Observation Point ID. */
336 odp_port_t output_odp_port
; /* The output odp port. */
337 enum nx_action_sample_direction direction
;
341 /* USER_ACTION_COOKIE_IPFIX. */
342 odp_port_t output_odp_port
; /* The output odp port. */
346 /* USER_ACTION_COOKIE_CONTROLLER. */
347 uint8_t dont_send
; /* Don't send the packet to controller. */
348 uint8_t continuation
; /* Send packet-in as a continuation. */
351 ovs_32aligned_be64 rule_cookie
;
352 uint16_t controller_id
;
357 BUILD_ASSERT_DECL(sizeof(struct user_action_cookie
) == 48);
359 size_t odp_put_userspace_action(uint32_t pid
,
360 const void *userdata
, size_t userdata_size
,
361 odp_port_t tunnel_out_port
,
362 bool include_actions
,
363 struct ofpbuf
*odp_actions
);
364 void odp_put_tunnel_action(const struct flow_tnl
*tunnel
,
365 struct ofpbuf
*odp_actions
,
366 const char *tnl_type
);
368 void odp_put_tnl_push_action(struct ofpbuf
*odp_actions
,
369 struct ovs_action_push_tnl
*data
);
371 void odp_put_pop_eth_action(struct ofpbuf
*odp_actions
);
372 void odp_put_push_eth_action(struct ofpbuf
*odp_actions
,
373 const struct eth_addr
*eth_src
,
374 const struct eth_addr
*eth_dst
);
376 struct attr_len_tbl
{
378 const struct attr_len_tbl
*next
;
382 #define ATTR_LEN_INVALID -1
383 #define ATTR_LEN_VARIABLE -2
384 #define ATTR_LEN_NESTED -3
386 extern const struct attr_len_tbl ovs_flow_key_attr_lens
[OVS_KEY_ATTR_MAX
+ 1];
387 #endif /* odp-util.h */