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;
36 /* This sequence number should be incremented whenever anything involving flows
37 * or the wildcarding of flows changes. This will cause build assertion
38 * failures in places which likely need to be updated. */
39 #define FLOW_WC_SEQ 23
42 BUILD_ASSERT_DECL(FLOW_N_REGS <= NXM_NX_MAX_REGS);
44 /* Used for struct flow's dl_type member for frames that have no Ethernet
45 * type, that is, pure 802.2 frames. */
46 #define FLOW_DL_TYPE_NONE 0x5ff
48 /* Fragment bits, used for IPv4 and IPv6, always zero for non-IP flows. */
49 #define FLOW_NW_FRAG_ANY (1 << 0) /* Set for any IP frag. */
50 #define FLOW_NW_FRAG_LATER (1 << 1) /* Set for IP frag with nonzero offset. */
51 #define FLOW_NW_FRAG_MASK (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
53 BUILD_ASSERT_DECL(FLOW_NW_FRAG_ANY == NX_IP_FRAG_ANY);
54 BUILD_ASSERT_DECL(FLOW_NW_FRAG_LATER == NX_IP_FRAG_LATER);
56 #define FLOW_TNL_F_DONT_FRAGMENT (1 << 0)
57 #define FLOW_TNL_F_CSUM (1 << 1)
58 #define FLOW_TNL_F_KEY (1 << 2)
60 const char *flow_tun_flag_to_string(uint32_t flags);
71 /* Unfortunately, a "struct flow" sometimes has to handle OpenFlow port
72 * numbers and other times datapath (dpif) port numbers. This union allows
80 * A flow in the network.
82 * Must be initialized to all zeros to make any compiler-induced padding
83 * zeroed. Helps also in keeping unused fields (such as mutually exclusive
84 * IPv4 and IPv6 addresses) zeroed out.
86 * The meaning of 'in_port' is context-dependent. In most cases, it is a
87 * 16-bit OpenFlow 1.0 port number. In the software datapath interface (dpif)
88 * layer and its implementations (e.g. dpif-linux, dpif-netdev), it is instead
89 * a 32-bit datapath port number.
91 * The fields are organized in four segments to facilitate staged lookup, where
92 * lower layer fields are first used to determine if the later fields need to
93 * be looked at. This enables better wildcarding for datapath flows.
97 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
98 ovs_be64 metadata; /* OpenFlow Metadata. */
99 uint32_t regs[FLOW_N_REGS]; /* Registers. */
100 uint32_t skb_priority; /* Packet priority for QoS. */
101 uint32_t pkt_mark; /* Packet mark. */
102 union flow_in_port in_port; /* Input port.*/
105 uint8_t dl_src[6]; /* Ethernet source address. */
106 uint8_t dl_dst[6]; /* Ethernet destination address. */
107 ovs_be16 dl_type; /* Ethernet frame type. */
108 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
111 ovs_be32 mpls_lse; /* MPLS label stack entry. */
112 struct in6_addr ipv6_src; /* IPv6 source address. */
113 struct in6_addr ipv6_dst; /* IPv6 destination address. */
114 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
115 ovs_be32 ipv6_label; /* IPv6 flow label. */
116 ovs_be32 nw_src; /* IPv4 source address. */
117 ovs_be32 nw_dst; /* IPv4 destination address. */
118 uint8_t nw_frag; /* FLOW_FRAG_* flags. */
119 uint8_t nw_tos; /* IP ToS (including DSCP and ECN). */
120 uint8_t nw_ttl; /* IP TTL/Hop Limit. */
121 uint8_t nw_proto; /* IP protocol or low 8 bits of ARP opcode. */
122 uint8_t arp_sha[6]; /* ARP/ND source hardware address. */
123 uint8_t arp_tha[6]; /* ARP/ND target hardware address. */
124 ovs_be16 tcp_flags; /* TCP flags. With L3 to avoid matching L4. */
125 ovs_be16 pad; /* Padding. */
127 ovs_be16 tp_src; /* TCP/UDP/SCTP source port. */
128 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port.
129 * Keep last for the BUILD_ASSERT_DECL below */
131 BUILD_ASSERT_DECL(sizeof(struct flow) % 4 == 0);
133 #define FLOW_U32S (sizeof(struct flow) / 4)
135 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
136 BUILD_ASSERT_DECL(offsetof(struct flow, tp_dst) + 2
137 == sizeof(struct flow_tnl) + 156
138 && FLOW_WC_SEQ == 23);
140 /* Incremental points at which flow classification may be performed in
142 * This is located here since this is dependent on the structure of the
143 * struct flow defined above:
144 * Each offset must be on a distint, successive U32 boundary srtictly
145 * within the struct flow. */
147 FLOW_SEGMENT_1_ENDS_AT = offsetof(struct flow, dl_src),
148 FLOW_SEGMENT_2_ENDS_AT = offsetof(struct flow, mpls_lse),
149 FLOW_SEGMENT_3_ENDS_AT = offsetof(struct flow, tp_src),
151 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT % 4 == 0);
152 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT % 4 == 0);
153 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT % 4 == 0);
154 BUILD_ASSERT_DECL( 0 < FLOW_SEGMENT_1_ENDS_AT);
155 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT < FLOW_SEGMENT_2_ENDS_AT);
156 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT < FLOW_SEGMENT_3_ENDS_AT);
157 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT < sizeof(struct flow));
159 extern const uint8_t flow_segment_u32s[];
161 /* Represents the metadata fields of struct flow. */
162 struct flow_metadata {
163 ovs_be64 tun_id; /* Encapsulating tunnel ID. */
164 ovs_be32 tun_src; /* Tunnel outer IPv4 src addr */
165 ovs_be32 tun_dst; /* Tunnel outer IPv4 dst addr */
166 ovs_be64 metadata; /* OpenFlow 1.1+ metadata field. */
167 uint32_t regs[FLOW_N_REGS]; /* Registers. */
168 uint32_t pkt_mark; /* Packet mark. */
169 ofp_port_t in_port; /* OpenFlow port or zero. */
172 void flow_extract(struct ofpbuf *, uint32_t priority, uint32_t mark,
173 const struct flow_tnl *, const union flow_in_port *in_port,
176 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
177 void flow_unwildcard_tp_ports(const struct flow *, struct flow_wildcards *);
178 void flow_get_metadata(const struct flow *, struct flow_metadata *);
180 char *flow_to_string(const struct flow *);
181 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
182 uint32_t flags, char del);
183 void format_flags_masked(struct ds *ds, const char *name,
184 const char *(*bit_to_string)(uint32_t),
185 uint32_t flags, uint32_t mask);
187 void flow_format(struct ds *, const struct flow *);
188 void flow_print(FILE *, const struct flow *);
189 static inline int flow_compare_3way(const struct flow *, const struct flow *);
190 static inline bool flow_equal(const struct flow *, const struct flow *);
191 static inline size_t flow_hash(const struct flow *, uint32_t basis);
193 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
194 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
195 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
197 void flow_set_mpls_label(struct flow *flow, ovs_be32 label);
198 void flow_set_mpls_ttl(struct flow *flow, uint8_t ttl);
199 void flow_set_mpls_tc(struct flow *flow, uint8_t tc);
200 void flow_set_mpls_bos(struct flow *flow, uint8_t stack);
202 void flow_compose(struct ofpbuf *, const struct flow *);
205 flow_compare_3way(const struct flow *a, const struct flow *b)
207 return memcmp(a, b, sizeof *a);
211 flow_equal(const struct flow *a, const struct flow *b)
213 return !flow_compare_3way(a, b);
217 flow_hash(const struct flow *flow, uint32_t basis)
219 return hash_words((const uint32_t *) flow, sizeof *flow / 4, basis);
222 static inline uint16_t
223 ofp_to_u16(ofp_port_t ofp_port)
225 return (OVS_FORCE uint16_t) ofp_port;
228 static inline uint32_t
229 odp_to_u32(odp_port_t odp_port)
231 return (OVS_FORCE uint32_t) odp_port;
234 static inline uint32_t
235 ofp11_to_u32(ofp11_port_t ofp11_port)
237 return (OVS_FORCE uint32_t) ofp11_port;
240 static inline ofp_port_t
241 u16_to_ofp(uint16_t port)
243 return OFP_PORT_C(port);
246 static inline odp_port_t
247 u32_to_odp(uint32_t port)
249 return ODP_PORT_C(port);
252 static inline ofp11_port_t
253 u32_to_ofp11(uint32_t port)
255 return OFP11_PORT_C(port);
258 static inline uint32_t
259 hash_ofp_port(ofp_port_t ofp_port)
261 return hash_int(ofp_to_u16(ofp_port), 0);
264 static inline uint32_t
265 hash_odp_port(odp_port_t odp_port)
267 return hash_int(odp_to_u32(odp_port), 0);
270 uint32_t flow_hash_in_minimask(const struct flow *, const struct minimask *,
272 uint32_t flow_hash_in_minimask_range(const struct flow *,
273 const struct minimask *,
274 uint8_t start, uint8_t end,
277 /* Wildcards for a flow.
279 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
280 * the flow is significant (must match). A 0-bit indicates that the
281 * corresponding bit of the flow is wildcarded (need not match). */
282 struct flow_wildcards {
286 void flow_wildcards_init_catchall(struct flow_wildcards *);
288 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
290 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
292 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
293 int idx, uint32_t mask);
295 void flow_wildcards_and(struct flow_wildcards *dst,
296 const struct flow_wildcards *src1,
297 const struct flow_wildcards *src2);
298 void flow_wildcards_or(struct flow_wildcards *dst,
299 const struct flow_wildcards *src1,
300 const struct flow_wildcards *src2);
301 bool flow_wildcards_has_extra(const struct flow_wildcards *,
302 const struct flow_wildcards *);
304 void flow_wildcards_fold_minimask(struct flow_wildcards *,
305 const struct minimask *);
306 void flow_wildcards_fold_minimask_range(struct flow_wildcards *,
307 const struct minimask *,
308 uint8_t start, uint8_t end);
310 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
311 bool flow_wildcards_equal(const struct flow_wildcards *,
312 const struct flow_wildcards *);
313 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
315 /* Initialize a flow with random fields that matter for nx_hash_fields. */
316 void flow_random_hash_fields(struct flow *);
317 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
318 enum nx_hash_fields);
319 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
321 const char *flow_hash_fields_to_str(enum nx_hash_fields);
322 bool flow_hash_fields_valid(enum nx_hash_fields);
324 uint32_t flow_hash_in_wildcards(const struct flow *,
325 const struct flow_wildcards *,
328 bool flow_equal_except(const struct flow *a, const struct flow *b,
329 const struct flow_wildcards *);
331 /* Compressed flow. */
333 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
334 BUILD_ASSERT_DECL(FLOW_U32S <= 64);
336 /* A sparse representation of a "struct flow".
338 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
339 * representation has two advantages. First, it saves memory. Second, it
340 * saves time when the goal is to iterate over only the nonzero parts of the
343 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
344 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
347 * 'values' points to the start of an array that has one element for each 1-bit
348 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
349 * values[1], and so on.
351 * 'values' may point to a few different locations:
353 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
354 * 'inline_values'. One hopes that this is the common case.
356 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
357 * allocated with malloc().
359 * - The caller could provide storage on the stack for situations where
360 * that makes sense. So far that's only proved useful for
361 * minimask_combine(), but the principle works elsewhere.
363 * Elements in 'values' are allowed to be zero. This is useful for "struct
364 * minimatch", for which ensuring that the miniflow and minimask members have
365 * same 'map' allows optimization. This allowance applies only to a miniflow
366 * that is not a mask. That is, a minimask may NOT have zero elements in
372 uint32_t inline_values[MINI_N_INLINE];
375 void miniflow_init(struct miniflow *, const struct flow *);
376 void miniflow_init_with_minimask(struct miniflow *, const struct flow *,
377 const struct minimask *);
378 void miniflow_clone(struct miniflow *, const struct miniflow *);
379 void miniflow_move(struct miniflow *dst, struct miniflow *);
380 void miniflow_destroy(struct miniflow *);
382 void miniflow_expand(const struct miniflow *, struct flow *);
384 uint32_t miniflow_get(const struct miniflow *, unsigned int u32_ofs);
385 uint16_t miniflow_get_vid(const struct miniflow *);
386 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
388 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
389 bool miniflow_equal_in_minimask(const struct miniflow *a,
390 const struct miniflow *b,
391 const struct minimask *);
392 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
393 const struct flow *b,
394 const struct minimask *);
395 uint32_t miniflow_hash(const struct miniflow *, uint32_t basis);
396 uint32_t miniflow_hash_in_minimask(const struct miniflow *,
397 const struct minimask *, uint32_t basis);
398 uint64_t miniflow_get_map_in_range(const struct miniflow *miniflow,
399 uint8_t start, uint8_t end,
400 unsigned int *offset);
403 /* Compressed flow wildcards. */
405 /* A sparse representation of a "struct flow_wildcards".
407 * See the large comment on struct miniflow for details.
409 * Note: While miniflow can have zero data for a 1-bit in the map,
410 * a minimask may not! We rely on this in the implementation. */
412 struct miniflow masks;
415 void minimask_init(struct minimask *, const struct flow_wildcards *);
416 void minimask_clone(struct minimask *, const struct minimask *);
417 void minimask_move(struct minimask *dst, struct minimask *src);
418 void minimask_combine(struct minimask *dst,
419 const struct minimask *a, const struct minimask *b,
420 uint32_t storage[FLOW_U32S]);
421 void minimask_destroy(struct minimask *);
423 void minimask_expand(const struct minimask *, struct flow_wildcards *);
425 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
426 uint16_t minimask_get_vid_mask(const struct minimask *);
427 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
429 bool minimask_equal(const struct minimask *a, const struct minimask *b);
430 uint32_t minimask_hash(const struct minimask *, uint32_t basis);
432 bool minimask_has_extra(const struct minimask *, const struct minimask *);
433 bool minimask_is_catchall(const struct minimask *);
435 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
436 static inline ovs_be64
437 miniflow_get_metadata(const struct miniflow *flow)
439 enum { MD_OFS = offsetof(struct flow, metadata) };
440 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
441 ovs_be32 hi = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4);
442 ovs_be32 lo = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4 + 1);
444 return htonll(((uint64_t) ntohl(hi) << 32) | ntohl(lo));
447 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
449 * The return value is all-1-bits if 'mask' matches on the whole value of the
450 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
451 * or some other value if the metadata field is partially matched, partially
453 static inline ovs_be64
454 minimask_get_metadata_mask(const struct minimask *mask)
456 return miniflow_get_metadata(&mask->masks);