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 25
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 uint32_t dp_hash; /* Datapath computed hash value. The exact
102 computation is opaque to the user space.*/
103 uint32_t recirc_id; /* Must be exact match. */
106 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
107 ovs_be64 metadata; /* OpenFlow Metadata. */
108 uint32_t regs[FLOW_N_REGS]; /* Registers. */
109 uint32_t skb_priority; /* Packet priority for QoS. */
110 uint32_t pkt_mark; /* Packet mark. */
111 union flow_in_port in_port; /* Input port.*/
114 uint8_t dl_src[6]; /* Ethernet source address. */
115 uint8_t dl_dst[6]; /* Ethernet destination address. */
116 ovs_be16 dl_type; /* Ethernet frame type. */
117 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
118 ovs_be32 mpls_lse[FLOW_MAX_MPLS_LABELS]; /* MPLS label stack entry. */
121 struct in6_addr ipv6_src; /* IPv6 source address. */
122 struct in6_addr ipv6_dst; /* IPv6 destination address. */
123 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
124 ovs_be32 ipv6_label; /* IPv6 flow label. */
125 ovs_be32 nw_src; /* IPv4 source address. */
126 ovs_be32 nw_dst; /* IPv4 destination address. */
127 uint8_t nw_frag; /* FLOW_FRAG_* flags. */
128 uint8_t nw_tos; /* IP ToS (including DSCP and ECN). */
129 uint8_t nw_ttl; /* IP TTL/Hop Limit. */
130 uint8_t nw_proto; /* IP protocol or low 8 bits of ARP opcode. */
131 uint8_t arp_sha[6]; /* ARP/ND source hardware address. */
132 uint8_t arp_tha[6]; /* ARP/ND target hardware address. */
133 ovs_be16 tcp_flags; /* TCP flags. With L3 to avoid matching L4. */
134 ovs_be16 pad; /* Padding. */
137 ovs_be16 tp_src; /* TCP/UDP/SCTP source port. */
138 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port.
139 * Keep last for the BUILD_ASSERT_DECL below */
141 BUILD_ASSERT_DECL(sizeof(struct flow) % 4 == 0);
143 #define FLOW_U32S (sizeof(struct flow) / 4)
145 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
146 BUILD_ASSERT_DECL(offsetof(struct flow, tp_dst) + 2
147 == sizeof(struct flow_tnl) + 172
148 && FLOW_WC_SEQ == 25);
150 /* Incremental points at which flow classification may be performed in
152 * This is located here since this is dependent on the structure of the
153 * struct flow defined above:
154 * Each offset must be on a distint, successive U32 boundary srtictly
155 * within the struct flow. */
157 FLOW_SEGMENT_1_ENDS_AT = offsetof(struct flow, dl_src),
158 FLOW_SEGMENT_2_ENDS_AT = offsetof(struct flow, ipv6_src),
159 FLOW_SEGMENT_3_ENDS_AT = offsetof(struct flow, tp_src),
161 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT % 4 == 0);
162 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT % 4 == 0);
163 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT % 4 == 0);
164 BUILD_ASSERT_DECL( 0 < FLOW_SEGMENT_1_ENDS_AT);
165 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT < FLOW_SEGMENT_2_ENDS_AT);
166 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT < FLOW_SEGMENT_3_ENDS_AT);
167 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT < sizeof(struct flow));
169 extern const uint8_t flow_segment_u32s[];
171 /* Represents the metadata fields of struct flow. */
172 struct flow_metadata {
173 uint32_t dp_hash; /* Datapath computed hash field. */
174 uint32_t recirc_id; /* Recirculation ID. */
175 ovs_be64 tun_id; /* Encapsulating tunnel ID. */
176 ovs_be32 tun_src; /* Tunnel outer IPv4 src addr */
177 ovs_be32 tun_dst; /* Tunnel outer IPv4 dst addr */
178 ovs_be64 metadata; /* OpenFlow 1.1+ metadata field. */
179 uint32_t regs[FLOW_N_REGS]; /* Registers. */
180 uint32_t pkt_mark; /* Packet mark. */
181 ofp_port_t in_port; /* OpenFlow port or zero. */
184 void flow_extract(struct ofpbuf *, const struct pkt_metadata *md,
187 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
188 void flow_unwildcard_tp_ports(const struct flow *, struct flow_wildcards *);
189 void flow_get_metadata(const struct flow *, struct flow_metadata *);
191 char *flow_to_string(const struct flow *);
192 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
193 uint32_t flags, char del);
194 void format_flags_masked(struct ds *ds, const char *name,
195 const char *(*bit_to_string)(uint32_t),
196 uint32_t flags, uint32_t mask);
198 void flow_format(struct ds *, const struct flow *);
199 void flow_print(FILE *, const struct flow *);
200 static inline int flow_compare_3way(const struct flow *, const struct flow *);
201 static inline bool flow_equal(const struct flow *, const struct flow *);
202 static inline size_t flow_hash(const struct flow *, uint32_t basis);
204 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
205 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
206 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
208 int flow_count_mpls_labels(const struct flow *, struct flow_wildcards *);
209 int flow_count_common_mpls_labels(const struct flow *a, int an,
210 const struct flow *b, int bn,
211 struct flow_wildcards *wc);
212 void flow_push_mpls(struct flow *, int n, ovs_be16 mpls_eth_type,
213 struct flow_wildcards *);
214 bool flow_pop_mpls(struct flow *, int n, ovs_be16 eth_type,
215 struct flow_wildcards *);
216 void flow_set_mpls_label(struct flow *, int idx, ovs_be32 label);
217 void flow_set_mpls_ttl(struct flow *, int idx, uint8_t ttl);
218 void flow_set_mpls_tc(struct flow *, int idx, uint8_t tc);
219 void flow_set_mpls_bos(struct flow *, int idx, uint8_t stack);
220 void flow_set_mpls_lse(struct flow *, int idx, ovs_be32 lse);
222 void flow_compose(struct ofpbuf *, const struct flow *);
225 flow_compare_3way(const struct flow *a, const struct flow *b)
227 return memcmp(a, b, sizeof *a);
231 flow_equal(const struct flow *a, const struct flow *b)
233 return !flow_compare_3way(a, b);
237 flow_hash(const struct flow *flow, uint32_t basis)
239 return hash_words((const uint32_t *) flow, sizeof *flow / 4, basis);
242 static inline uint16_t
243 ofp_to_u16(ofp_port_t ofp_port)
245 return (OVS_FORCE uint16_t) ofp_port;
248 static inline uint32_t
249 odp_to_u32(odp_port_t odp_port)
251 return (OVS_FORCE uint32_t) odp_port;
254 static inline uint32_t
255 ofp11_to_u32(ofp11_port_t ofp11_port)
257 return (OVS_FORCE uint32_t) ofp11_port;
260 static inline ofp_port_t
261 u16_to_ofp(uint16_t port)
263 return OFP_PORT_C(port);
266 static inline odp_port_t
267 u32_to_odp(uint32_t port)
269 return ODP_PORT_C(port);
272 static inline ofp11_port_t
273 u32_to_ofp11(uint32_t port)
275 return OFP11_PORT_C(port);
278 static inline uint32_t
279 hash_ofp_port(ofp_port_t ofp_port)
281 return hash_int(ofp_to_u16(ofp_port), 0);
284 static inline uint32_t
285 hash_odp_port(odp_port_t odp_port)
287 return hash_int(odp_to_u32(odp_port), 0);
290 uint32_t flow_hash_in_minimask(const struct flow *, const struct minimask *,
292 uint32_t flow_hash_in_minimask_range(const struct flow *,
293 const struct minimask *,
294 uint8_t start, uint8_t end,
297 /* Wildcards for a flow.
299 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
300 * the flow is significant (must match). A 0-bit indicates that the
301 * corresponding bit of the flow is wildcarded (need not match). */
302 struct flow_wildcards {
306 void flow_wildcards_init_catchall(struct flow_wildcards *);
308 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
310 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
312 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
313 int idx, uint32_t mask);
315 void flow_wildcards_and(struct flow_wildcards *dst,
316 const struct flow_wildcards *src1,
317 const struct flow_wildcards *src2);
318 void flow_wildcards_or(struct flow_wildcards *dst,
319 const struct flow_wildcards *src1,
320 const struct flow_wildcards *src2);
321 bool flow_wildcards_has_extra(const struct flow_wildcards *,
322 const struct flow_wildcards *);
324 void flow_wildcards_fold_minimask(struct flow_wildcards *,
325 const struct minimask *);
326 void flow_wildcards_fold_minimask_range(struct flow_wildcards *,
327 const struct minimask *,
328 uint8_t start, uint8_t end);
330 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
331 bool flow_wildcards_equal(const struct flow_wildcards *,
332 const struct flow_wildcards *);
333 uint32_t flow_hash_5tuple(const struct flow *flow, uint32_t basis);
334 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
336 /* Initialize a flow with random fields that matter for nx_hash_fields. */
337 void flow_random_hash_fields(struct flow *);
338 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
339 enum nx_hash_fields);
340 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
342 const char *flow_hash_fields_to_str(enum nx_hash_fields);
343 bool flow_hash_fields_valid(enum nx_hash_fields);
345 uint32_t flow_hash_in_wildcards(const struct flow *,
346 const struct flow_wildcards *,
349 bool flow_equal_except(const struct flow *a, const struct flow *b,
350 const struct flow_wildcards *);
352 /* Compressed flow. */
354 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
355 BUILD_ASSERT_DECL(FLOW_U32S <= 64);
357 /* A sparse representation of a "struct flow".
359 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
360 * representation has two advantages. First, it saves memory. Second, it
361 * saves time when the goal is to iterate over only the nonzero parts of the
364 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
365 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
368 * 'values' points to the start of an array that has one element for each 1-bit
369 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
370 * values[1], and so on.
372 * 'values' may point to a few different locations:
374 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
375 * 'inline_values'. One hopes that this is the common case.
377 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
378 * allocated with malloc().
380 * - The caller could provide storage on the stack for situations where
381 * that makes sense. So far that's only proved useful for
382 * minimask_combine(), but the principle works elsewhere.
384 * Elements in 'values' are allowed to be zero. This is useful for "struct
385 * minimatch", for which ensuring that the miniflow and minimask members have
386 * same 'map' allows optimization. This allowance applies only to a miniflow
387 * that is not a mask. That is, a minimask may NOT have zero elements in
393 uint32_t inline_values[MINI_N_INLINE];
396 void miniflow_init(struct miniflow *, const struct flow *);
397 void miniflow_init_with_minimask(struct miniflow *, const struct flow *,
398 const struct minimask *);
399 void miniflow_clone(struct miniflow *, const struct miniflow *);
400 void miniflow_move(struct miniflow *dst, struct miniflow *);
401 void miniflow_destroy(struct miniflow *);
403 void miniflow_expand(const struct miniflow *, struct flow *);
405 uint32_t miniflow_get(const struct miniflow *, unsigned int u32_ofs);
406 uint16_t miniflow_get_vid(const struct miniflow *);
407 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
409 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
410 bool miniflow_equal_in_minimask(const struct miniflow *a,
411 const struct miniflow *b,
412 const struct minimask *);
413 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
414 const struct flow *b,
415 const struct minimask *);
416 uint32_t miniflow_hash(const struct miniflow *, uint32_t basis);
417 uint32_t miniflow_hash_in_minimask(const struct miniflow *,
418 const struct minimask *, uint32_t basis);
419 uint64_t miniflow_get_map_in_range(const struct miniflow *miniflow,
420 uint8_t start, uint8_t end,
421 unsigned int *offset);
424 /* Compressed flow wildcards. */
426 /* A sparse representation of a "struct flow_wildcards".
428 * See the large comment on struct miniflow for details.
430 * Note: While miniflow can have zero data for a 1-bit in the map,
431 * a minimask may not! We rely on this in the implementation. */
433 struct miniflow masks;
436 void minimask_init(struct minimask *, const struct flow_wildcards *);
437 void minimask_clone(struct minimask *, const struct minimask *);
438 void minimask_move(struct minimask *dst, struct minimask *src);
439 void minimask_combine(struct minimask *dst,
440 const struct minimask *a, const struct minimask *b,
441 uint32_t storage[FLOW_U32S]);
442 void minimask_destroy(struct minimask *);
444 void minimask_expand(const struct minimask *, struct flow_wildcards *);
446 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
447 uint16_t minimask_get_vid_mask(const struct minimask *);
448 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
450 bool minimask_equal(const struct minimask *a, const struct minimask *b);
451 uint32_t minimask_hash(const struct minimask *, uint32_t basis);
453 bool minimask_has_extra(const struct minimask *, const struct minimask *);
454 bool minimask_is_catchall(const struct minimask *);
456 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
457 static inline ovs_be64
458 miniflow_get_metadata(const struct miniflow *flow)
460 enum { MD_OFS = offsetof(struct flow, metadata) };
461 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
462 ovs_be32 hi = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4);
463 ovs_be32 lo = (OVS_FORCE ovs_be32) miniflow_get(flow, MD_OFS / 4 + 1);
465 return htonll(((uint64_t) ntohl(hi) << 32) | ntohl(lo));
468 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
470 * The return value is all-1-bits if 'mask' matches on the whole value of the
471 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
472 * or some other value if the metadata field is partially matched, partially
474 static inline ovs_be64
475 minimask_get_metadata_mask(const struct minimask *mask)
477 return miniflow_get_metadata(&mask->masks);