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"
31 struct dpif_flow_stats;
33 struct flow_wildcards;
38 /* This sequence number should be incremented whenever anything involving flows
39 * or the wildcarding of flows changes. This will cause build assertion
40 * failures in places which likely need to be updated. */
41 #define FLOW_WC_SEQ 26
44 BUILD_ASSERT_DECL(FLOW_N_REGS <= NXM_NX_MAX_REGS);
46 /* Used for struct flow's dl_type member for frames that have no Ethernet
47 * type, that is, pure 802.2 frames. */
48 #define FLOW_DL_TYPE_NONE 0x5ff
50 /* Fragment bits, used for IPv4 and IPv6, always zero for non-IP flows. */
51 #define FLOW_NW_FRAG_ANY (1 << 0) /* Set for any IP frag. */
52 #define FLOW_NW_FRAG_LATER (1 << 1) /* Set for IP frag with nonzero offset. */
53 #define FLOW_NW_FRAG_MASK (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
55 BUILD_ASSERT_DECL(FLOW_NW_FRAG_ANY == NX_IP_FRAG_ANY);
56 BUILD_ASSERT_DECL(FLOW_NW_FRAG_LATER == NX_IP_FRAG_LATER);
58 #define FLOW_TNL_F_DONT_FRAGMENT (1 << 0)
59 #define FLOW_TNL_F_CSUM (1 << 1)
60 #define FLOW_TNL_F_KEY (1 << 2)
62 const char *flow_tun_flag_to_string(uint32_t flags);
64 /* Maximum number of supported MPLS labels. */
65 #define FLOW_MAX_MPLS_LABELS 3
68 * A flow in the network.
70 * Must be initialized to all zeros to make any compiler-induced padding
71 * zeroed. Helps also in keeping unused fields (such as mutually exclusive
72 * IPv4 and IPv6 addresses) zeroed out.
74 * The meaning of 'in_port' is context-dependent. In most cases, it is a
75 * 16-bit OpenFlow 1.0 port number. In the software datapath interface (dpif)
76 * layer and its implementations (e.g. dpif-linux, dpif-netdev), it is instead
77 * a 32-bit datapath port number.
79 * The fields are organized in four segments to facilitate staged lookup, where
80 * lower layer fields are first used to determine if the later fields need to
81 * be looked at. This enables better wildcarding for datapath flows.
83 * NOTE: Order of the fields is significant, any change in the order must be
84 * reflected in miniflow_extract()!
88 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
89 ovs_be64 metadata; /* OpenFlow Metadata. */
90 uint32_t regs[FLOW_N_REGS]; /* Registers. */
91 uint32_t skb_priority; /* Packet priority for QoS. */
92 uint32_t pkt_mark; /* Packet mark. */
93 uint32_t recirc_id; /* Must be exact match. */
94 union flow_in_port in_port; /* Input port.*/
96 /* L2, Order the same as in the Ethernet header! */
97 uint8_t dl_dst[6]; /* Ethernet destination address. */
98 uint8_t dl_src[6]; /* Ethernet source address. */
99 ovs_be16 dl_type; /* Ethernet frame type. */
100 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
101 ovs_be32 mpls_lse[FLOW_MAX_MPLS_LABELS]; /* MPLS label stack entry. */
104 struct in6_addr ipv6_src; /* IPv6 source address. */
105 struct in6_addr ipv6_dst; /* IPv6 destination address. */
106 ovs_be32 ipv6_label; /* IPv6 flow label. */
107 ovs_be32 nw_src; /* IPv4 source address. */
108 ovs_be32 nw_dst; /* IPv4 destination address. */
109 uint8_t nw_frag; /* FLOW_FRAG_* flags. */
110 uint8_t nw_tos; /* IP ToS (including DSCP and ECN). */
111 uint8_t nw_ttl; /* IP TTL/Hop Limit. */
112 uint8_t nw_proto; /* IP protocol or low 8 bits of ARP opcode. */
113 uint8_t arp_sha[6]; /* ARP/ND source hardware address. */
114 uint8_t arp_tha[6]; /* ARP/ND target hardware address. */
115 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
116 ovs_be16 tcp_flags; /* TCP flags. With L3 to avoid matching L4. */
117 ovs_be16 pad; /* Padding. */
120 ovs_be16 tp_src; /* TCP/UDP/SCTP source port. */
121 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port.
122 * Keep last for the BUILD_ASSERT_DECL below */
123 uint32_t dp_hash; /* Datapath computed hash value. The exact
124 computation is opaque to the user space.*/
126 BUILD_ASSERT_DECL(sizeof(struct flow) % 4 == 0);
128 #define FLOW_U32S (sizeof(struct flow) / 4)
130 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
131 BUILD_ASSERT_DECL(offsetof(struct flow, dp_hash) + sizeof(uint32_t)
132 == sizeof(struct flow_tnl) + 172
133 && FLOW_WC_SEQ == 26);
135 /* Incremental points at which flow classification may be performed in
137 * This is located here since this is dependent on the structure of the
138 * struct flow defined above:
139 * Each offset must be on a distinct, successive U32 boundary strictly
140 * within the struct flow. */
142 FLOW_SEGMENT_1_ENDS_AT = offsetof(struct flow, dl_dst),
143 FLOW_SEGMENT_2_ENDS_AT = offsetof(struct flow, ipv6_src),
144 FLOW_SEGMENT_3_ENDS_AT = offsetof(struct flow, tp_src),
146 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT % 4 == 0);
147 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT % 4 == 0);
148 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT % 4 == 0);
149 BUILD_ASSERT_DECL( 0 < FLOW_SEGMENT_1_ENDS_AT);
150 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT < FLOW_SEGMENT_2_ENDS_AT);
151 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT < FLOW_SEGMENT_3_ENDS_AT);
152 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT < sizeof(struct flow));
154 extern const uint8_t flow_segment_u32s[];
156 /* Represents the metadata fields of struct flow. */
157 struct flow_metadata {
158 uint32_t dp_hash; /* Datapath computed hash field. */
159 uint32_t recirc_id; /* Recirculation ID. */
160 ovs_be64 tun_id; /* Encapsulating tunnel ID. */
161 ovs_be32 tun_src; /* Tunnel outer IPv4 src addr */
162 ovs_be32 tun_dst; /* Tunnel outer IPv4 dst addr */
163 ovs_be64 metadata; /* OpenFlow 1.1+ metadata field. */
164 uint32_t regs[FLOW_N_REGS]; /* Registers. */
165 uint32_t pkt_mark; /* Packet mark. */
166 ofp_port_t in_port; /* OpenFlow port or zero. */
169 void flow_extract(struct ofpbuf *, const struct pkt_metadata *md,
172 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
173 void flow_unwildcard_tp_ports(const struct flow *, struct flow_wildcards *);
174 void flow_get_metadata(const struct flow *, struct flow_metadata *);
176 char *flow_to_string(const struct flow *);
177 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
178 uint32_t flags, char del);
179 void format_flags_masked(struct ds *ds, const char *name,
180 const char *(*bit_to_string)(uint32_t),
181 uint32_t flags, uint32_t mask);
183 void flow_format(struct ds *, const struct flow *);
184 void flow_print(FILE *, const struct flow *);
185 static inline int flow_compare_3way(const struct flow *, const struct flow *);
186 static inline bool flow_equal(const struct flow *, const struct flow *);
187 static inline size_t flow_hash(const struct flow *, uint32_t basis);
189 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
190 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
191 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
193 int flow_count_mpls_labels(const struct flow *, struct flow_wildcards *);
194 int flow_count_common_mpls_labels(const struct flow *a, int an,
195 const struct flow *b, int bn,
196 struct flow_wildcards *wc);
197 void flow_push_mpls(struct flow *, int n, ovs_be16 mpls_eth_type,
198 struct flow_wildcards *);
199 bool flow_pop_mpls(struct flow *, int n, ovs_be16 eth_type,
200 struct flow_wildcards *);
201 void flow_set_mpls_label(struct flow *, int idx, ovs_be32 label);
202 void flow_set_mpls_ttl(struct flow *, int idx, uint8_t ttl);
203 void flow_set_mpls_tc(struct flow *, int idx, uint8_t tc);
204 void flow_set_mpls_bos(struct flow *, int idx, uint8_t stack);
205 void flow_set_mpls_lse(struct flow *, int idx, ovs_be32 lse);
207 void flow_compose(struct ofpbuf *, const struct flow *);
210 flow_compare_3way(const struct flow *a, const struct flow *b)
212 return memcmp(a, b, sizeof *a);
216 flow_equal(const struct flow *a, const struct flow *b)
218 return !flow_compare_3way(a, b);
222 flow_hash(const struct flow *flow, uint32_t basis)
224 return hash_words((const uint32_t *) flow, sizeof *flow / 4, basis);
227 static inline uint16_t
228 ofp_to_u16(ofp_port_t ofp_port)
230 return (OVS_FORCE uint16_t) ofp_port;
233 static inline uint32_t
234 odp_to_u32(odp_port_t odp_port)
236 return (OVS_FORCE uint32_t) odp_port;
239 static inline uint32_t
240 ofp11_to_u32(ofp11_port_t ofp11_port)
242 return (OVS_FORCE uint32_t) ofp11_port;
245 static inline ofp_port_t
246 u16_to_ofp(uint16_t port)
248 return OFP_PORT_C(port);
251 static inline odp_port_t
252 u32_to_odp(uint32_t port)
254 return ODP_PORT_C(port);
257 static inline ofp11_port_t
258 u32_to_ofp11(uint32_t port)
260 return OFP11_PORT_C(port);
263 static inline uint32_t
264 hash_ofp_port(ofp_port_t ofp_port)
266 return hash_int(ofp_to_u16(ofp_port), 0);
269 static inline uint32_t
270 hash_odp_port(odp_port_t odp_port)
272 return hash_int(odp_to_u32(odp_port), 0);
275 uint32_t flow_hash_in_minimask(const struct flow *, const struct minimask *,
277 uint32_t flow_hash_in_minimask_range(const struct flow *,
278 const struct minimask *,
279 uint8_t start, uint8_t end,
282 /* Wildcards for a flow.
284 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
285 * the flow is significant (must match). A 0-bit indicates that the
286 * corresponding bit of the flow is wildcarded (need not match). */
287 struct flow_wildcards {
291 void flow_wildcards_init_catchall(struct flow_wildcards *);
293 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
295 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
297 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
298 int idx, uint32_t mask);
300 void flow_wildcards_and(struct flow_wildcards *dst,
301 const struct flow_wildcards *src1,
302 const struct flow_wildcards *src2);
303 void flow_wildcards_or(struct flow_wildcards *dst,
304 const struct flow_wildcards *src1,
305 const struct flow_wildcards *src2);
306 bool flow_wildcards_has_extra(const struct flow_wildcards *,
307 const struct flow_wildcards *);
309 void flow_wildcards_fold_minimask(struct flow_wildcards *,
310 const struct minimask *);
311 void flow_wildcards_fold_minimask_range(struct flow_wildcards *,
312 const struct minimask *,
313 uint8_t start, uint8_t end);
315 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
316 bool flow_wildcards_equal(const struct flow_wildcards *,
317 const struct flow_wildcards *);
318 uint32_t flow_hash_5tuple(const struct flow *flow, uint32_t basis);
319 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
321 /* Initialize a flow with random fields that matter for nx_hash_fields. */
322 void flow_random_hash_fields(struct flow *);
323 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
324 enum nx_hash_fields);
325 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
327 const char *flow_hash_fields_to_str(enum nx_hash_fields);
328 bool flow_hash_fields_valid(enum nx_hash_fields);
330 uint32_t flow_hash_in_wildcards(const struct flow *,
331 const struct flow_wildcards *,
334 bool flow_equal_except(const struct flow *a, const struct flow *b,
335 const struct flow_wildcards *);
337 /* Compressed flow. */
339 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
340 BUILD_ASSERT_DECL(FLOW_U32S <= 64);
342 /* A sparse representation of a "struct flow".
344 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
345 * representation has two advantages. First, it saves memory. Second, it
346 * saves time when the goal is to iterate over only the nonzero parts of the
349 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
350 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
353 * 'values' points to the start of an array that has one element for each 1-bit
354 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
355 * values[1], and so on.
357 * 'values' may point to a few different locations:
359 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
360 * 'inline_values'. One hopes that this is the common case.
362 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
363 * allocated with malloc().
365 * - The caller could provide storage on the stack for situations where
366 * that makes sense. So far that's only proved useful for
367 * minimask_combine(), but the principle works elsewhere.
369 * Elements in 'values' are allowed to be zero. This is useful for "struct
370 * minimatch", for which ensuring that the miniflow and minimask members have
371 * same 'map' allows optimization. This allowance applies only to a miniflow
372 * that is not a mask. That is, a minimask may NOT have zero elements in
378 uint32_t inline_values[MINI_N_INLINE];
381 /* This is useful for initializing a miniflow for a miniflow_extract() call. */
382 static inline void miniflow_initialize(struct miniflow *mf,
383 uint32_t buf[FLOW_U32S])
391 /* The 'dst->values' must be initialized with a buffer with space for
392 * FLOW_U32S. 'dst->map' is ignored on input and set on output to
393 * indicate which fields were extracted. */
394 void miniflow_extract(struct ofpbuf *packet, const struct pkt_metadata *,
395 struct miniflow *dst);
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 /* These accessors use byte offsets, which are assumed to be compile-time
407 static inline uint8_t miniflow_get_u8(const struct miniflow *,
409 static inline uint16_t miniflow_get_u16(const struct miniflow *,
411 static inline ovs_be16 miniflow_get_be16(const struct miniflow *,
413 static inline uint32_t miniflow_get_u32(const struct miniflow *,
415 static inline ovs_be32 miniflow_get_be32(const struct miniflow *,
418 static inline uint16_t miniflow_get_vid(const struct miniflow *);
419 static inline uint16_t miniflow_get_tcp_flags(const struct miniflow *);
420 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
422 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
423 bool miniflow_equal_in_minimask(const struct miniflow *a,
424 const struct miniflow *b,
425 const struct minimask *);
426 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
427 const struct flow *b,
428 const struct minimask *);
429 uint32_t miniflow_hash(const struct miniflow *, uint32_t basis);
430 uint32_t miniflow_hash_in_minimask(const struct miniflow *,
431 const struct minimask *, uint32_t basis);
432 uint64_t miniflow_get_map_in_range(const struct miniflow *miniflow,
433 uint8_t start, uint8_t end,
434 unsigned int *offset);
437 /* Compressed flow wildcards. */
439 /* A sparse representation of a "struct flow_wildcards".
441 * See the large comment on struct miniflow for details.
443 * Note: While miniflow can have zero data for a 1-bit in the map,
444 * a minimask may not! We rely on this in the implementation. */
446 struct miniflow masks;
449 void minimask_init(struct minimask *, const struct flow_wildcards *);
450 void minimask_clone(struct minimask *, const struct minimask *);
451 void minimask_move(struct minimask *dst, struct minimask *src);
452 void minimask_combine(struct minimask *dst,
453 const struct minimask *a, const struct minimask *b,
454 uint32_t storage[FLOW_U32S]);
455 void minimask_destroy(struct minimask *);
457 void minimask_expand(const struct minimask *, struct flow_wildcards *);
459 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
460 static inline uint16_t minimask_get_vid_mask(const struct minimask *);
461 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
463 bool minimask_equal(const struct minimask *a, const struct minimask *b);
464 uint32_t minimask_hash(const struct minimask *, uint32_t basis);
466 bool minimask_has_extra(const struct minimask *, const struct minimask *);
467 bool minimask_is_catchall(const struct minimask *);
470 /* 'OFS' is a compile-time constant. */
471 #define MINIFLOW_GET_TYPE(MF, TYPE, OFS) \
472 (MF->map & UINT64_C(1) << OFS / 4) \
473 ? ((OVS_FORCE const TYPE *) \
474 (MF->values + count_1bits(MF->map & ((UINT64_C(1) << OFS / 4) - 1)))) \
475 [OFS % 4 / sizeof(TYPE)] \
478 static inline uint8_t
479 miniflow_get_u8(const struct miniflow *flow, unsigned int ofs)
481 return MINIFLOW_GET_TYPE(flow, uint8_t, ofs);
484 static inline uint16_t
485 miniflow_get_u16(const struct miniflow *flow, unsigned int ofs)
487 return MINIFLOW_GET_TYPE(flow, uint16_t, ofs);
490 /* Returns the ovs_be16 that would be at byte offset 'u8_ofs' if 'flow' were
491 * expanded into a "struct flow". */
492 static inline ovs_be16
493 miniflow_get_be16(const struct miniflow *flow, unsigned int ofs)
495 return MINIFLOW_GET_TYPE(flow, ovs_be16, ofs);
498 static inline uint32_t
499 miniflow_get_u32(const struct miniflow *flow, unsigned int ofs)
501 return MINIFLOW_GET_TYPE(flow, uint32_t, ofs);
504 static inline ovs_be32
505 miniflow_get_be32(const struct miniflow *flow, unsigned int ofs)
507 return MINIFLOW_GET_TYPE(flow, ovs_be32, ofs);
510 #undef MINIFLOW_GET_TYPE
512 /* Returns the VID within the vlan_tci member of the "struct flow" represented
514 static inline uint16_t
515 miniflow_get_vid(const struct miniflow *flow)
517 ovs_be16 tci = miniflow_get_be16(flow, offsetof(struct flow, vlan_tci));
518 return vlan_tci_to_vid(tci);
521 /* Returns the VID mask within the vlan_tci member of the "struct
522 * flow_wildcards" represented by 'mask'. */
523 static inline uint16_t
524 minimask_get_vid_mask(const struct minimask *mask)
526 return miniflow_get_vid(&mask->masks);
529 /* Returns the value of the "tcp_flags" field in 'flow'. */
530 static inline uint16_t
531 miniflow_get_tcp_flags(const struct miniflow *flow)
533 return ntohs(miniflow_get_be16(flow, offsetof(struct flow, tcp_flags)));
536 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
537 static inline ovs_be64
538 miniflow_get_metadata(const struct miniflow *flow)
540 enum { MD_OFS = offsetof(struct flow, metadata) };
541 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
542 ovs_be32 hi = miniflow_get_be32(flow, MD_OFS);
543 ovs_be32 lo = miniflow_get_be32(flow, MD_OFS + 4);
545 return htonll(((uint64_t) ntohl(hi) << 32) | ntohl(lo));
548 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
550 * The return value is all-1-bits if 'mask' matches on the whole value of the
551 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
552 * or some other value if the metadata field is partially matched, partially
554 static inline ovs_be64
555 minimask_get_metadata_mask(const struct minimask *mask)
557 return miniflow_get_metadata(&mask->masks);
560 static inline struct pkt_metadata
561 pkt_metadata_from_flow(const struct flow *flow)
563 struct pkt_metadata md;
565 md.recirc_id = flow->recirc_id;
566 md.dp_hash = flow->dp_hash;
567 md.tunnel = flow->tunnel;
568 md.skb_priority = flow->skb_priority;
569 md.pkt_mark = flow->pkt_mark;
570 md.in_port = flow->in_port;
575 static inline bool is_ip_any(const struct flow *flow)
577 return dl_type_is_ip_any(flow->dl_type);
580 static inline bool is_icmpv4(const struct flow *flow)
582 return (flow->dl_type == htons(ETH_TYPE_IP)
583 && flow->nw_proto == IPPROTO_ICMP);
586 static inline bool is_icmpv6(const struct flow *flow)
588 return (flow->dl_type == htons(ETH_TYPE_IPV6)
589 && flow->nw_proto == IPPROTO_ICMPV6);