2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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.
19 #include <sys/types.h>
20 #include <netinet/in.h>
24 #include "byte-order.h"
25 #include "openflow/nicira-ext.h"
26 #include "openflow/openflow.h"
30 struct dpif_flow_stats;
32 struct flow_wildcards;
37 /* This sequence number should be incremented whenever anything involving flows
38 * or the wildcarding of flows changes. This will cause build assertion
39 * failures in places which likely need to be updated. */
40 #define FLOW_WC_SEQ 21
43 BUILD_ASSERT_DECL(FLOW_N_REGS <= NXM_NX_MAX_REGS);
45 /* Used for struct flow's dl_type member for frames that have no Ethernet
46 * type, that is, pure 802.2 frames. */
47 #define FLOW_DL_TYPE_NONE 0x5ff
49 /* Fragment bits, used for IPv4 and IPv6, always zero for non-IP flows. */
50 #define FLOW_NW_FRAG_ANY (1 << 0) /* Set for any IP frag. */
51 #define FLOW_NW_FRAG_LATER (1 << 1) /* Set for IP frag with nonzero offset. */
52 #define FLOW_NW_FRAG_MASK (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
54 BUILD_ASSERT_DECL(FLOW_NW_FRAG_ANY == NX_IP_FRAG_ANY);
55 BUILD_ASSERT_DECL(FLOW_NW_FRAG_LATER == NX_IP_FRAG_LATER);
57 #define FLOW_TNL_F_DONT_FRAGMENT (1 << 0)
58 #define FLOW_TNL_F_CSUM (1 << 1)
59 #define FLOW_TNL_F_KEY (1 << 2)
61 const char *flow_tun_flag_to_string(uint32_t flags);
72 /* Unfortunately, a "struct flow" sometimes has to handle OpenFlow port
73 * numbers and other times datapath (dpif) port numbers. This union allows
81 * A flow in the network.
83 * Must be initialized to all zeros to make any compiler-induced padding
84 * zeroed. Helps also in keeping unused fields (such as mutually exclusive
85 * IPv4 and IPv6 addresses) zeroed out.
87 * The meaning of 'in_port' is context-dependent. In most cases, it is a
88 * 16-bit OpenFlow 1.0 port number. In the software datapath interface (dpif)
89 * layer and its implementations (e.g. dpif-linux, dpif-netdev), it is instead
90 * a 32-bit datapath port number.
93 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
94 ovs_be64 metadata; /* OpenFlow Metadata. */
95 struct in6_addr ipv6_src; /* IPv6 source address. */
96 struct in6_addr ipv6_dst; /* IPv6 destination address. */
97 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
98 uint32_t skb_priority; /* Packet priority for QoS. */
99 uint32_t regs[FLOW_N_REGS]; /* Registers. */
100 ovs_be32 nw_src; /* IPv4 source address. */
101 ovs_be32 nw_dst; /* IPv4 destination address. */
102 ovs_be32 ipv6_label; /* IPv6 flow label. */
103 union flow_in_port in_port; /* Input port.*/
104 uint32_t pkt_mark; /* Packet mark. */
105 ovs_be32 mpls_lse; /* MPLS label stack entry. */
106 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
107 ovs_be16 dl_type; /* Ethernet frame type. */
108 ovs_be16 tp_src; /* TCP/UDP/SCTP source port. */
109 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port. */
110 uint8_t dl_src[6]; /* Ethernet source address. */
111 uint8_t dl_dst[6]; /* Ethernet destination address. */
112 uint8_t nw_proto; /* IP protocol or low 8 bits of ARP opcode. */
113 uint8_t nw_tos; /* IP ToS (including DSCP and ECN). */
114 uint8_t arp_sha[6]; /* ARP/ND source hardware address. */
115 uint8_t arp_tha[6]; /* ARP/ND target hardware address. */
116 uint8_t nw_ttl; /* IP TTL/Hop Limit. */
117 uint8_t nw_frag; /* FLOW_FRAG_* flags. Keep last for the
118 BUILD_ASSERT_DECL below */
120 BUILD_ASSERT_DECL(sizeof(struct flow) % 4 == 0);
122 #define FLOW_U32S (sizeof(struct flow) / 4)
124 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
125 BUILD_ASSERT_DECL(offsetof(struct flow, nw_frag) + 1
126 == sizeof(struct flow_tnl) + 152
127 && FLOW_WC_SEQ == 21);
129 /* Represents the metadata fields of struct flow. */
130 struct flow_metadata {
131 ovs_be64 tun_id; /* Encapsulating tunnel ID. */
132 ovs_be32 tun_src; /* Tunnel outer IPv4 src addr */
133 ovs_be32 tun_dst; /* Tunnel outer IPv4 dst addr */
134 ovs_be64 metadata; /* OpenFlow 1.1+ metadata field. */
135 uint32_t regs[FLOW_N_REGS]; /* Registers. */
136 uint32_t pkt_mark; /* Packet mark. */
137 ofp_port_t in_port; /* OpenFlow port or zero. */
140 void flow_extract(struct ofpbuf *, uint32_t priority, uint32_t mark,
141 const struct flow_tnl *, const union flow_in_port *in_port,
144 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
145 void flow_get_metadata(const struct flow *, struct flow_metadata *);
147 char *flow_to_string(const struct flow *);
148 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
149 uint32_t flags, char del);
151 void flow_format(struct ds *, const struct flow *);
152 void flow_print(FILE *, const struct flow *);
153 static inline int flow_compare_3way(const struct flow *, const struct flow *);
154 static inline bool flow_equal(const struct flow *, const struct flow *);
155 static inline size_t flow_hash(const struct flow *, uint32_t basis);
157 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
158 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
159 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
161 void flow_set_mpls_label(struct flow *flow, ovs_be32 label);
162 void flow_set_mpls_ttl(struct flow *flow, uint8_t ttl);
163 void flow_set_mpls_tc(struct flow *flow, uint8_t tc);
164 void flow_set_mpls_bos(struct flow *flow, uint8_t stack);
166 void flow_compose(struct ofpbuf *, const struct flow *);
169 flow_compare_3way(const struct flow *a, const struct flow *b)
171 return memcmp(a, b, sizeof *a);
175 flow_equal(const struct flow *a, const struct flow *b)
177 return !flow_compare_3way(a, b);
181 flow_hash(const struct flow *flow, uint32_t basis)
183 return hash_words((const uint32_t *) flow, sizeof *flow / 4, basis);
186 static inline uint16_t
187 ofp_to_u16(ofp_port_t ofp_port)
189 return (OVS_FORCE uint16_t) ofp_port;
192 static inline uint32_t
193 odp_to_u32(odp_port_t odp_port)
195 return (OVS_FORCE uint32_t) odp_port;
198 static inline uint32_t
199 ofp11_to_u32(ofp11_port_t ofp11_port)
201 return (OVS_FORCE uint32_t) ofp11_port;
204 static inline ofp_port_t
205 u16_to_ofp(uint16_t port)
207 return OFP_PORT_C(port);
210 static inline odp_port_t
211 u32_to_odp(uint32_t port)
213 return ODP_PORT_C(port);
216 static inline ofp11_port_t
217 u32_to_ofp11(uint32_t port)
219 return OFP11_PORT_C(port);
222 static inline uint32_t
223 hash_ofp_port(ofp_port_t ofp_port)
225 return hash_int(ofp_to_u16(ofp_port), 0);
228 static inline uint32_t
229 hash_odp_port(odp_port_t odp_port)
231 return hash_int(odp_to_u32(odp_port), 0);
234 uint32_t flow_hash_in_minimask(const struct flow *, const struct minimask *,
237 /* Wildcards for a flow.
239 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
240 * the flow is significant (must match). A 0-bit indicates that the
241 * corresponding bit of the flow is wildcarded (need not match). */
242 struct flow_wildcards {
246 void flow_wildcards_init_catchall(struct flow_wildcards *);
248 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
250 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
251 int idx, uint32_t mask);
253 void flow_wildcards_and(struct flow_wildcards *dst,
254 const struct flow_wildcards *src1,
255 const struct flow_wildcards *src2);
256 void flow_wildcards_or(struct flow_wildcards *dst,
257 const struct flow_wildcards *src1,
258 const struct flow_wildcards *src2);
259 bool flow_wildcards_has_extra(const struct flow_wildcards *,
260 const struct flow_wildcards *);
262 void flow_wildcards_fold_minimask(struct flow_wildcards *,
263 const struct minimask *);
265 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
266 bool flow_wildcards_equal(const struct flow_wildcards *,
267 const struct flow_wildcards *);
268 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
270 /* Initialize a flow with random fields that matter for nx_hash_fields. */
271 void flow_random_hash_fields(struct flow *);
272 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
273 enum nx_hash_fields);
274 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
276 const char *flow_hash_fields_to_str(enum nx_hash_fields);
277 bool flow_hash_fields_valid(enum nx_hash_fields);
279 uint32_t flow_hash_in_wildcards(const struct flow *,
280 const struct flow_wildcards *,
283 bool flow_equal_except(const struct flow *a, const struct flow *b,
284 const struct flow_wildcards *);
286 /* Compressed flow. */
288 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
289 #define MINI_N_MAPS DIV_ROUND_UP(FLOW_U32S, 32)
291 /* A sparse representation of a "struct flow".
293 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
294 * representation has two advantages. First, it saves memory. Second, it
295 * saves time when the goal is to iterate over only the nonzero parts of the
298 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
299 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
302 * 'values' points to the start of an array that has one element for each 1-bit
303 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
304 * values[1], and so on.
306 * 'values' may point to a few different locations:
308 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
309 * 'inline_values'. One hopes that this is the common case.
311 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
312 * allocated with malloc().
314 * - The caller could provide storage on the stack for situations where
315 * that makes sense. So far that's only proved useful for
316 * minimask_combine(), but the principle works elsewhere.
318 * Elements in 'values' are allowed to be zero. This is useful for "struct
319 * minimatch", for which ensuring that the miniflow and minimask members have
320 * same 'map' allows optimization .
324 uint32_t inline_values[MINI_N_INLINE];
325 uint32_t map[MINI_N_MAPS];
328 void miniflow_init(struct miniflow *, const struct flow *);
329 void miniflow_init_with_minimask(struct miniflow *, const struct flow *,
330 const struct minimask *);
331 void miniflow_clone(struct miniflow *, const struct miniflow *);
332 void miniflow_move(struct miniflow *dst, struct miniflow *);
333 void miniflow_destroy(struct miniflow *);
335 void miniflow_expand(const struct miniflow *, struct flow *);
337 uint32_t miniflow_get(const struct miniflow *, unsigned int u32_ofs);
338 uint16_t miniflow_get_vid(const struct miniflow *);
339 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
341 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
342 bool miniflow_equal_in_minimask(const struct miniflow *a,
343 const struct miniflow *b,
344 const struct minimask *);
345 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
346 const struct flow *b,
347 const struct minimask *);
348 uint32_t miniflow_hash(const struct miniflow *, uint32_t basis);
349 uint32_t miniflow_hash_in_minimask(const struct miniflow *,
350 const struct minimask *, uint32_t basis);
352 /* Compressed flow wildcards. */
354 /* A sparse representation of a "struct flow_wildcards".
356 * See the large comment on struct miniflow for details. */
358 struct miniflow masks;
361 void minimask_init(struct minimask *, const struct flow_wildcards *);
362 void minimask_clone(struct minimask *, const struct minimask *);
363 void minimask_move(struct minimask *dst, struct minimask *src);
364 void minimask_combine(struct minimask *dst,
365 const struct minimask *a, const struct minimask *b,
366 uint32_t storage[FLOW_U32S]);
367 void minimask_destroy(struct minimask *);
369 void minimask_expand(const struct minimask *, struct flow_wildcards *);
371 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
372 uint16_t minimask_get_vid_mask(const struct minimask *);
373 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
375 bool minimask_equal(const struct minimask *a, const struct minimask *b);
376 uint32_t minimask_hash(const struct minimask *, uint32_t basis);
378 bool minimask_has_extra(const struct minimask *, const struct minimask *);
379 bool minimask_is_catchall(const struct minimask *);
381 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
382 static inline ovs_be64
383 miniflow_get_metadata(const struct miniflow *flow)
385 enum { MD_OFS = offsetof(struct flow, metadata) };
386 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
387 ovs_be32 hi = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4);
388 ovs_be32 lo = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4 + 1);
390 return htonll(((uint64_t) ntohl(hi) << 32) | ntohl(lo));
393 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
395 * The return value is all-1-bits if 'mask' matches on the whole value of the
396 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
397 * or some other value if the metadata field is partially matched, partially
399 static inline ovs_be64
400 minimask_get_metadata_mask(const struct minimask *mask)
402 return miniflow_get_metadata(&mask->masks);