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 /* Wildcards for a flow.
277 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
278 * the flow is significant (must match). A 0-bit indicates that the
279 * corresponding bit of the flow is wildcarded (need not match). */
280 struct flow_wildcards {
284 void flow_wildcards_init_catchall(struct flow_wildcards *);
286 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
288 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
290 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
291 int idx, uint32_t mask);
293 void flow_wildcards_and(struct flow_wildcards *dst,
294 const struct flow_wildcards *src1,
295 const struct flow_wildcards *src2);
296 void flow_wildcards_or(struct flow_wildcards *dst,
297 const struct flow_wildcards *src1,
298 const struct flow_wildcards *src2);
299 bool flow_wildcards_has_extra(const struct flow_wildcards *,
300 const struct flow_wildcards *);
301 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
302 bool flow_wildcards_equal(const struct flow_wildcards *,
303 const struct flow_wildcards *);
304 uint32_t flow_hash_5tuple(const struct flow *flow, uint32_t basis);
305 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
307 /* Initialize a flow with random fields that matter for nx_hash_fields. */
308 void flow_random_hash_fields(struct flow *);
309 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
310 enum nx_hash_fields);
311 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
313 const char *flow_hash_fields_to_str(enum nx_hash_fields);
314 bool flow_hash_fields_valid(enum nx_hash_fields);
316 uint32_t flow_hash_in_wildcards(const struct flow *,
317 const struct flow_wildcards *,
320 bool flow_equal_except(const struct flow *a, const struct flow *b,
321 const struct flow_wildcards *);
323 /* Compressed flow. */
325 #define MINI_N_INLINE (sizeof(void *) == 4 ? 7 : 8)
326 BUILD_ASSERT_DECL(FLOW_U32S <= 64);
328 /* A sparse representation of a "struct flow".
330 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
331 * representation has two advantages. First, it saves memory. Second, it
332 * saves time when the goal is to iterate over only the nonzero parts of the
335 * The 'map' member holds one bit for each uint32_t in a "struct flow". Each
336 * 0-bit indicates that the corresponding uint32_t is zero, each 1-bit that it
339 * 'values' points to the start of an array that has one element for each 1-bit
340 * in 'map'. The least-numbered 1-bit is in values[0], the next 1-bit is in
341 * values[1], and so on.
343 * 'values' may point to a few different locations:
345 * - If 'map' has MINI_N_INLINE or fewer 1-bits, it may point to
346 * 'inline_values'. One hopes that this is the common case.
348 * - If 'map' has more than MINI_N_INLINE 1-bits, it may point to memory
349 * allocated with malloc().
351 * - The caller could provide storage on the stack for situations where
352 * that makes sense. So far that's only proved useful for
353 * minimask_combine(), but the principle works elsewhere.
355 * Elements in 'values' are allowed to be zero. This is useful for "struct
356 * minimatch", for which ensuring that the miniflow and minimask members have
357 * same 'map' allows optimization. This allowance applies only to a miniflow
358 * that is not a mask. That is, a minimask may NOT have zero elements in
364 uint32_t inline_values[MINI_N_INLINE];
367 /* This is useful for initializing a miniflow for a miniflow_extract() call. */
368 static inline void miniflow_initialize(struct miniflow *mf,
369 uint32_t buf[FLOW_U32S])
377 /* The 'dst->values' must be initialized with a buffer with space for
378 * FLOW_U32S. 'dst->map' is ignored on input and set on output to
379 * indicate which fields were extracted. */
380 void miniflow_extract(struct ofpbuf *packet, const struct pkt_metadata *,
381 struct miniflow *dst);
382 void miniflow_init(struct miniflow *, const struct flow *);
383 void miniflow_init_with_minimask(struct miniflow *, const struct flow *,
384 const struct minimask *);
385 void miniflow_clone(struct miniflow *, const struct miniflow *);
386 void miniflow_move(struct miniflow *dst, struct miniflow *);
387 void miniflow_destroy(struct miniflow *);
389 void miniflow_expand(const struct miniflow *, struct flow *);
391 #define FLOW_U32_SIZE(FIELD) \
392 DIV_ROUND_UP(sizeof(((struct flow *)0)->FIELD), sizeof(uint32_t))
394 #define MINIFLOW_MAP(FIELD) \
395 (((UINT64_C(1) << FLOW_U32_SIZE(FIELD)) - 1) \
396 << (offsetof(struct flow, FIELD) / 4))
398 static inline uint32_t
399 mf_get_next_in_map(uint64_t *fmap, uint64_t rm1bit, const uint32_t **fp,
403 if (*fmap & rm1bit) {
404 uint64_t trash = *fmap & (rm1bit - 1);
408 *fp += count_1bits(trash);
415 /* Iterate through all miniflow u32 values specified by the 'MAP'.
416 * This works as the first statement in a block.*/
417 #define MINIFLOW_FOR_EACH_IN_MAP(VALUE, FLOW, MAP) \
418 const uint32_t *fp_ = (FLOW)->values; \
419 uint64_t rm1bit_, fmap_, map_; \
420 for (fmap_ = (FLOW)->map, map_ = (MAP), rm1bit_ = rightmost_1bit(map_); \
421 mf_get_next_in_map(&fmap_, rm1bit_, &fp_, &(VALUE)); \
422 map_ -= rm1bit_, rm1bit_ = rightmost_1bit(map_))
424 /* Get the value of 'FIELD' of an up to 4 byte wide integer type 'TYPE' of
426 #define MINIFLOW_GET_TYPE(MF, TYPE, OFS) \
427 (((MF)->map & (UINT64_C(1) << (OFS) / 4)) \
428 ? ((OVS_FORCE const TYPE *) \
430 + count_1bits((MF)->map & ((UINT64_C(1) << (OFS) / 4) - 1)))) \
431 [(OFS) % 4 / sizeof(TYPE)] \
434 #define MINIFLOW_GET_U8(FLOW, FIELD) \
435 MINIFLOW_GET_TYPE(FLOW, uint8_t, offsetof(struct flow, FIELD))
436 #define MINIFLOW_GET_U16(FLOW, FIELD) \
437 MINIFLOW_GET_TYPE(FLOW, uint16_t, offsetof(struct flow, FIELD))
438 #define MINIFLOW_GET_BE16(FLOW, FIELD) \
439 MINIFLOW_GET_TYPE(FLOW, ovs_be16, offsetof(struct flow, FIELD))
440 #define MINIFLOW_GET_U32(FLOW, FIELD) \
441 MINIFLOW_GET_TYPE(FLOW, uint32_t, offsetof(struct flow, FIELD))
442 #define MINIFLOW_GET_BE32(FLOW, FIELD) \
443 MINIFLOW_GET_TYPE(FLOW, ovs_be32, offsetof(struct flow, FIELD))
445 static inline uint16_t miniflow_get_vid(const struct miniflow *);
446 static inline uint16_t miniflow_get_tcp_flags(const struct miniflow *);
447 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
449 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
450 bool miniflow_equal_in_minimask(const struct miniflow *a,
451 const struct miniflow *b,
452 const struct minimask *);
453 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
454 const struct flow *b,
455 const struct minimask *);
456 uint32_t miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis);
459 /* Compressed flow wildcards. */
461 /* A sparse representation of a "struct flow_wildcards".
463 * See the large comment on struct miniflow for details.
465 * Note: While miniflow can have zero data for a 1-bit in the map,
466 * a minimask may not! We rely on this in the implementation. */
468 struct miniflow masks;
471 void minimask_init(struct minimask *, const struct flow_wildcards *);
472 void minimask_clone(struct minimask *, const struct minimask *);
473 void minimask_move(struct minimask *dst, struct minimask *src);
474 void minimask_combine(struct minimask *dst,
475 const struct minimask *a, const struct minimask *b,
476 uint32_t storage[FLOW_U32S]);
477 void minimask_destroy(struct minimask *);
479 void minimask_expand(const struct minimask *, struct flow_wildcards *);
481 uint32_t minimask_get(const struct minimask *, unsigned int u32_ofs);
482 static inline uint16_t minimask_get_vid_mask(const struct minimask *);
483 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
485 bool minimask_equal(const struct minimask *a, const struct minimask *b);
486 bool minimask_has_extra(const struct minimask *, const struct minimask *);
487 bool minimask_is_catchall(const struct minimask *);
491 /* Returns the VID within the vlan_tci member of the "struct flow" represented
493 static inline uint16_t
494 miniflow_get_vid(const struct miniflow *flow)
496 ovs_be16 tci = MINIFLOW_GET_BE16(flow, vlan_tci);
497 return vlan_tci_to_vid(tci);
500 /* Returns the VID mask within the vlan_tci member of the "struct
501 * flow_wildcards" represented by 'mask'. */
502 static inline uint16_t
503 minimask_get_vid_mask(const struct minimask *mask)
505 return miniflow_get_vid(&mask->masks);
508 /* Returns the value of the "tcp_flags" field in 'flow'. */
509 static inline uint16_t
510 miniflow_get_tcp_flags(const struct miniflow *flow)
512 return ntohs(MINIFLOW_GET_BE16(flow, tcp_flags));
515 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
516 static inline ovs_be64
517 miniflow_get_metadata(const struct miniflow *flow)
527 enum { MD_OFS = offsetof(struct flow, metadata) };
528 BUILD_ASSERT_DECL(MD_OFS % sizeof(uint32_t) == 0);
529 value.hi = MINIFLOW_GET_TYPE(flow, ovs_be32, MD_OFS);
530 value.lo = MINIFLOW_GET_TYPE(flow, ovs_be32, MD_OFS + 4);
535 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
537 * The return value is all-1-bits if 'mask' matches on the whole value of the
538 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
539 * or some other value if the metadata field is partially matched, partially
541 static inline ovs_be64
542 minimask_get_metadata_mask(const struct minimask *mask)
544 return miniflow_get_metadata(&mask->masks);
547 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
548 * fields in 'dst', storing the result in 'dst'. */
550 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
552 uint32_t *dst_u32 = (uint32_t *) dst;
553 const uint32_t *p = (uint32_t *)src->values;
556 for (map = src->map; map; map = zero_rightmost_1bit(map)) {
557 dst_u32[raw_ctz(map)] |= *p++;
561 static inline struct pkt_metadata
562 pkt_metadata_from_flow(const struct flow *flow)
564 struct pkt_metadata md;
566 md.recirc_id = flow->recirc_id;
567 md.dp_hash = flow->dp_hash;
568 md.tunnel = flow->tunnel;
569 md.skb_priority = flow->skb_priority;
570 md.pkt_mark = flow->pkt_mark;
571 md.in_port = flow->in_port;
576 static inline bool is_ip_any(const struct flow *flow)
578 return dl_type_is_ip_any(flow->dl_type);
581 static inline bool is_icmpv4(const struct flow *flow)
583 return (flow->dl_type == htons(ETH_TYPE_IP)
584 && flow->nw_proto == IPPROTO_ICMP);
587 static inline bool is_icmpv6(const struct flow *flow)
589 return (flow->dl_type == htons(ETH_TYPE_IPV6)
590 && flow->nw_proto == IPPROTO_ICMPV6);
593 static inline bool is_stp(const struct flow *flow)
595 return (eth_addr_equals(flow->dl_dst, eth_addr_stp)
596 && flow->dl_type == htons(FLOW_DL_TYPE_NONE));