2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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 24
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
80 /* Maximum number of supported MPLS labels. */
81 #define FLOW_MAX_MPLS_LABELS 3
84 * A flow in the network.
86 * Must be initialized to all zeros to make any compiler-induced padding
87 * zeroed. Helps also in keeping unused fields (such as mutually exclusive
88 * IPv4 and IPv6 addresses) zeroed out.
90 * The meaning of 'in_port' is context-dependent. In most cases, it is a
91 * 16-bit OpenFlow 1.0 port number. In the software datapath interface (dpif)
92 * layer and its implementations (e.g. dpif-linux, dpif-netdev), it is instead
93 * a 32-bit datapath port number.
95 * The fields are organized in four segments to facilitate staged lookup, where
96 * lower layer fields are first used to determine if the later fields need to
97 * be looked at. This enables better wildcarding for datapath flows.
101 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
102 ovs_be64 metadata; /* OpenFlow Metadata. */
103 uint32_t regs[FLOW_N_REGS]; /* Registers. */
104 uint32_t skb_priority; /* Packet priority for QoS. */
105 uint32_t pkt_mark; /* Packet mark. */
106 union flow_in_port in_port; /* Input port.*/
109 uint8_t dl_src[6]; /* Ethernet source address. */
110 uint8_t dl_dst[6]; /* Ethernet destination address. */
111 ovs_be16 dl_type; /* Ethernet frame type. */
112 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
113 ovs_be32 mpls_lse[FLOW_MAX_MPLS_LABELS]; /* MPLS label stack entry. */
116 struct in6_addr ipv6_src; /* IPv6 source address. */
117 struct in6_addr ipv6_dst; /* IPv6 destination address. */
118 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
119 ovs_be32 ipv6_label; /* IPv6 flow label. */
120 ovs_be32 nw_src; /* IPv4 source address. */
121 ovs_be32 nw_dst; /* IPv4 destination address. */
122 uint8_t nw_frag; /* FLOW_FRAG_* flags. */
123 uint8_t nw_tos; /* IP ToS (including DSCP and ECN). */
124 uint8_t nw_ttl; /* IP TTL/Hop Limit. */
125 uint8_t nw_proto; /* IP protocol or low 8 bits of ARP opcode. */
126 uint8_t arp_sha[6]; /* ARP/ND source hardware address. */
127 uint8_t arp_tha[6]; /* ARP/ND target hardware address. */
128 ovs_be16 tcp_flags; /* TCP flags. With L3 to avoid matching L4. */
129 ovs_be16 pad; /* Padding. */
132 ovs_be16 tp_src; /* TCP/UDP/SCTP source port. */
133 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port.
134 * Keep last for the BUILD_ASSERT_DECL below */
136 BUILD_ASSERT_DECL(sizeof(struct flow) % 4 == 0);
138 #define FLOW_U32S (sizeof(struct flow) / 4)
140 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
141 BUILD_ASSERT_DECL(offsetof(struct flow, tp_dst) + 2
142 == sizeof(struct flow_tnl) + 164
143 && FLOW_WC_SEQ == 24);
145 /* Incremental points at which flow classification may be performed in
147 * This is located here since this is dependent on the structure of the
148 * struct flow defined above:
149 * Each offset must be on a distint, successive U32 boundary srtictly
150 * within the struct flow. */
152 FLOW_SEGMENT_1_ENDS_AT = offsetof(struct flow, dl_src),
153 FLOW_SEGMENT_2_ENDS_AT = offsetof(struct flow, ipv6_src),
154 FLOW_SEGMENT_3_ENDS_AT = offsetof(struct flow, tp_src),
156 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT % 4 == 0);
157 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT % 4 == 0);
158 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT % 4 == 0);
159 BUILD_ASSERT_DECL( 0 < FLOW_SEGMENT_1_ENDS_AT);
160 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT < FLOW_SEGMENT_2_ENDS_AT);
161 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT < FLOW_SEGMENT_3_ENDS_AT);
162 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT < sizeof(struct flow));
164 extern const uint8_t flow_segment_u32s[];
166 /* Represents the metadata fields of struct flow. */
167 struct flow_metadata {
168 ovs_be64 tun_id; /* Encapsulating tunnel ID. */
169 ovs_be32 tun_src; /* Tunnel outer IPv4 src addr */
170 ovs_be32 tun_dst; /* Tunnel outer IPv4 dst addr */
171 ovs_be64 metadata; /* OpenFlow 1.1+ metadata field. */
172 uint32_t regs[FLOW_N_REGS]; /* Registers. */
173 uint32_t pkt_mark; /* Packet mark. */
174 ofp_port_t in_port; /* OpenFlow port or zero. */
177 void flow_extract(struct ofpbuf *, const struct pkt_metadata *md,
180 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
181 void flow_unwildcard_tp_ports(const struct flow *, struct flow_wildcards *);
182 void flow_get_metadata(const struct flow *, struct flow_metadata *);
184 char *flow_to_string(const struct flow *);
185 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
186 uint32_t flags, char del);
187 void format_flags_masked(struct ds *ds, const char *name,
188 const char *(*bit_to_string)(uint32_t),
189 uint32_t flags, uint32_t mask);
191 void flow_format(struct ds *, const struct flow *);
192 void flow_print(FILE *, const struct flow *);
193 static inline int flow_compare_3way(const struct flow *, const struct flow *);
194 static inline bool flow_equal(const struct flow *, const struct flow *);
195 static inline size_t flow_hash(const struct flow *, uint32_t basis);
197 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
198 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
199 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
201 int flow_count_mpls_labels(const struct flow *, struct flow_wildcards *);
202 int flow_count_common_mpls_labels(const struct flow *a, int an,
203 const struct flow *b, int bn,
204 struct flow_wildcards *wc);
205 void flow_push_mpls(struct flow *, int n, ovs_be16 mpls_eth_type,
206 struct flow_wildcards *);
207 bool flow_pop_mpls(struct flow *, int n, ovs_be16 eth_type,
208 struct flow_wildcards *);
209 void flow_set_mpls_label(struct flow *, int idx, ovs_be32 label);
210 void flow_set_mpls_ttl(struct flow *, int idx, uint8_t ttl);
211 void flow_set_mpls_tc(struct flow *, int idx, uint8_t tc);
212 void flow_set_mpls_bos(struct flow *, int idx, uint8_t stack);
213 void flow_set_mpls_lse(struct flow *, int idx, ovs_be32 lse);
215 void flow_compose(struct ofpbuf *, const struct flow *);
218 flow_compare_3way(const struct flow *a, const struct flow *b)
220 return memcmp(a, b, sizeof *a);
224 flow_equal(const struct flow *a, const struct flow *b)
226 return !flow_compare_3way(a, b);
230 flow_hash(const struct flow *flow, uint32_t basis)
232 return hash_words((const uint32_t *) flow, sizeof *flow / 4, basis);
235 static inline uint16_t
236 ofp_to_u16(ofp_port_t ofp_port)
238 return (OVS_FORCE uint16_t) ofp_port;
241 static inline uint32_t
242 odp_to_u32(odp_port_t odp_port)
244 return (OVS_FORCE uint32_t) odp_port;
247 static inline uint32_t
248 ofp11_to_u32(ofp11_port_t ofp11_port)
250 return (OVS_FORCE uint32_t) ofp11_port;
253 static inline ofp_port_t
254 u16_to_ofp(uint16_t port)
256 return OFP_PORT_C(port);
259 static inline odp_port_t
260 u32_to_odp(uint32_t port)
262 return ODP_PORT_C(port);
265 static inline ofp11_port_t
266 u32_to_ofp11(uint32_t port)
268 return OFP11_PORT_C(port);
271 static inline uint32_t
272 hash_ofp_port(ofp_port_t ofp_port)
274 return hash_int(ofp_to_u16(ofp_port), 0);
277 static inline uint32_t
278 hash_odp_port(odp_port_t odp_port)
280 return hash_int(odp_to_u32(odp_port), 0);
283 uint32_t flow_hash_in_minimask(const struct flow *, const struct minimask *,
285 uint32_t flow_hash_in_minimask_range(const struct flow *,
286 const struct minimask *,
287 uint8_t start, uint8_t end,
290 /* Wildcards for a flow.
292 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
293 * the flow is significant (must match). A 0-bit indicates that the
294 * corresponding bit of the flow is wildcarded (need not match). */
295 struct flow_wildcards {
299 void flow_wildcards_init_catchall(struct flow_wildcards *);
301 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
303 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
305 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
306 int idx, uint32_t mask);
308 void flow_wildcards_and(struct flow_wildcards *dst,
309 const struct flow_wildcards *src1,
310 const struct flow_wildcards *src2);
311 void flow_wildcards_or(struct flow_wildcards *dst,
312 const struct flow_wildcards *src1,
313 const struct flow_wildcards *src2);
314 bool flow_wildcards_has_extra(const struct flow_wildcards *,
315 const struct flow_wildcards *);
317 void flow_wildcards_fold_minimask(struct flow_wildcards *,
318 const struct minimask *);
319 void flow_wildcards_fold_minimask_range(struct flow_wildcards *,
320 const struct minimask *,
321 uint8_t start, uint8_t end);
323 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
324 bool flow_wildcards_equal(const struct flow_wildcards *,
325 const struct flow_wildcards *);
326 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
328 /* Initialize a flow with random fields that matter for nx_hash_fields. */
329 void flow_random_hash_fields(struct flow *);
330 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
331 enum nx_hash_fields);
332 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
334 const char *flow_hash_fields_to_str(enum nx_hash_fields);
335 bool flow_hash_fields_valid(enum nx_hash_fields);
337 uint32_t flow_hash_in_wildcards(const struct flow *,
338 const struct flow_wildcards *,
341 bool flow_equal_except(const struct flow *a, const struct flow *b,
342 const struct flow_wildcards *);
344 /* Compressed flow. */
346 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
347 BUILD_ASSERT_DECL(FLOW_U32S <= 64);
349 /* A sparse representation of a "struct flow".
351 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
352 * representation has two advantages. First, it saves memory. Second, it
353 * saves time when the goal is to iterate over only the nonzero parts of the
356 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
357 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
360 * 'values' points to the start of an array that has one element for each 1-bit
361 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
362 * values[1], and so on.
364 * 'values' may point to a few different locations:
366 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
367 * 'inline_values'. One hopes that this is the common case.
369 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
370 * allocated with malloc().
372 * - The caller could provide storage on the stack for situations where
373 * that makes sense. So far that's only proved useful for
374 * minimask_combine(), but the principle works elsewhere.
376 * Elements in 'values' are allowed to be zero. This is useful for "struct
377 * minimatch", for which ensuring that the miniflow and minimask members have
378 * same 'map' allows optimization. This allowance applies only to a miniflow
379 * that is not a mask. That is, a minimask may NOT have zero elements in
385 uint32_t inline_values[MINI_N_INLINE];
388 void miniflow_init(struct miniflow *, const struct flow *);
389 void miniflow_init_with_minimask(struct miniflow *, const struct flow *,
390 const struct minimask *);
391 void miniflow_clone(struct miniflow *, const struct miniflow *);
392 void miniflow_move(struct miniflow *dst, struct miniflow *);
393 void miniflow_destroy(struct miniflow *);
395 void miniflow_expand(const struct miniflow *, struct flow *);
397 uint32_t miniflow_get(const struct miniflow *, unsigned int u32_ofs);
398 uint16_t miniflow_get_vid(const struct miniflow *);
399 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
401 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
402 bool miniflow_equal_in_minimask(const struct miniflow *a,
403 const struct miniflow *b,
404 const struct minimask *);
405 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
406 const struct flow *b,
407 const struct minimask *);
408 uint32_t miniflow_hash(const struct miniflow *, uint32_t basis);
409 uint32_t miniflow_hash_in_minimask(const struct miniflow *,
410 const struct minimask *, uint32_t basis);
411 uint64_t miniflow_get_map_in_range(const struct miniflow *miniflow,
412 uint8_t start, uint8_t end,
413 unsigned int *offset);
416 /* Compressed flow wildcards. */
418 /* A sparse representation of a "struct flow_wildcards".
420 * See the large comment on struct miniflow for details.
422 * Note: While miniflow can have zero data for a 1-bit in the map,
423 * a minimask may not! We rely on this in the implementation. */
425 struct miniflow masks;
428 void minimask_init(struct minimask *, const struct flow_wildcards *);
429 void minimask_clone(struct minimask *, const struct minimask *);
430 void minimask_move(struct minimask *dst, struct minimask *src);
431 void minimask_combine(struct minimask *dst,
432 const struct minimask *a, const struct minimask *b,
433 uint32_t storage[FLOW_U32S]);
434 void minimask_destroy(struct minimask *);
436 void minimask_expand(const struct minimask *, struct flow_wildcards *);
438 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
439 uint16_t minimask_get_vid_mask(const struct minimask *);
440 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
442 bool minimask_equal(const struct minimask *a, const struct minimask *b);
443 uint32_t minimask_hash(const struct minimask *, uint32_t basis);
445 bool minimask_has_extra(const struct minimask *, const struct minimask *);
446 bool minimask_is_catchall(const struct minimask *);
448 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
449 static inline ovs_be64
450 miniflow_get_metadata(const struct miniflow *flow)
452 enum { MD_OFS = offsetof(struct flow, metadata) };
453 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
454 ovs_be32 hi = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4);
455 ovs_be32 lo = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4 + 1);
457 return htonll(((uint64_t) ntohl(hi) << 32) | ntohl(lo));
460 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
462 * The return value is all-1-bits if 'mask' matches on the whole value of the
463 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
464 * or some other value if the metadata field is partially matched, partially
466 static inline ovs_be64
467 minimask_get_metadata_mask(const struct minimask *mask)
469 return miniflow_get_metadata(&mask->masks);