2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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>
25 #include "byte-order.h"
26 #include "openflow/nicira-ext.h"
27 #include "openflow/openflow.h"
32 struct dpif_flow_stats;
34 struct flow_wildcards;
40 /* This sequence number should be incremented whenever anything involving flows
41 * or the wildcarding of flows changes. This will cause build assertion
42 * failures in places which likely need to be updated. */
43 #define FLOW_WC_SEQ 35
45 /* Number of Open vSwitch extension 32-bit registers. */
47 BUILD_ASSERT_DECL(FLOW_N_REGS <= NXM_NX_MAX_REGS);
48 BUILD_ASSERT_DECL(FLOW_N_REGS % 2 == 0); /* Even. */
50 /* Number of OpenFlow 1.5+ 64-bit registers.
52 * Each of these overlays a pair of Open vSwitch 32-bit registers, so there
53 * are half as many of them.*/
54 #define FLOW_N_XREGS (FLOW_N_REGS / 2)
56 /* Used for struct flow's dl_type member for frames that have no Ethernet
57 * type, that is, pure 802.2 frames. */
58 #define FLOW_DL_TYPE_NONE 0x5ff
60 /* Fragment bits, used for IPv4 and IPv6, always zero for non-IP flows. */
61 #define FLOW_NW_FRAG_ANY (1 << 0) /* Set for any IP frag. */
62 #define FLOW_NW_FRAG_LATER (1 << 1) /* Set for IP frag with nonzero offset. */
63 #define FLOW_NW_FRAG_MASK (FLOW_NW_FRAG_ANY | FLOW_NW_FRAG_LATER)
65 BUILD_ASSERT_DECL(FLOW_NW_FRAG_ANY == NX_IP_FRAG_ANY);
66 BUILD_ASSERT_DECL(FLOW_NW_FRAG_LATER == NX_IP_FRAG_LATER);
68 BUILD_ASSERT_DECL(FLOW_TNL_F_OAM == NX_TUN_FLAG_OAM);
70 const char *flow_tun_flag_to_string(uint32_t flags);
72 /* Maximum number of supported MPLS labels. */
73 #define FLOW_MAX_MPLS_LABELS 3
76 * A flow in the network.
78 * Must be initialized to all zeros to make any compiler-induced padding
79 * zeroed. Helps also in keeping unused fields (such as mutually exclusive
80 * IPv4 and IPv6 addresses) zeroed out.
82 * The meaning of 'in_port' is context-dependent. In most cases, it is a
83 * 16-bit OpenFlow 1.0 port number. In the software datapath interface (dpif)
84 * layer and its implementations (e.g. dpif-netlink, dpif-netdev), it is
85 * instead a 32-bit datapath port number.
87 * The fields are organized in four segments to facilitate staged lookup, where
88 * lower layer fields are first used to determine if the later fields need to
89 * be looked at. This enables better wildcarding for datapath flows.
91 * NOTE: Order of the fields is significant, any change in the order must be
92 * reflected in miniflow_extract()!
96 struct flow_tnl tunnel; /* Encapsulating tunnel parameters. */
97 ovs_be64 metadata; /* OpenFlow Metadata. */
98 uint32_t regs[FLOW_N_REGS]; /* Registers. */
99 uint32_t skb_priority; /* Packet priority for QoS. */
100 uint32_t pkt_mark; /* Packet mark. */
101 uint32_t dp_hash; /* Datapath computed hash value. The exact
102 * computation is opaque to the user space. */
103 union flow_in_port in_port; /* Input port.*/
104 uint32_t recirc_id; /* Must be exact match. */
105 uint16_t ct_state; /* Connection tracking state. */
106 uint16_t ct_zone; /* Connection tracking zone. */
107 uint32_t ct_mark; /* Connection mark.*/
108 uint8_t pad1[4]; /* Pad to 64 bits. */
109 ovs_u128 ct_label; /* Connection label. */
110 uint32_t conj_id; /* Conjunction ID. */
111 ofp_port_t actset_output; /* Output port in action set. */
112 uint8_t pad2[2]; /* Pad to 64 bits. */
114 /* L2, Order the same as in the Ethernet header! (64-bit aligned) */
115 struct eth_addr dl_dst; /* Ethernet destination address. */
116 struct eth_addr dl_src; /* Ethernet source address. */
117 ovs_be16 dl_type; /* Ethernet frame type. */
118 ovs_be16 vlan_tci; /* If 802.1Q, TCI | VLAN_CFI; otherwise 0. */
119 ovs_be32 mpls_lse[ROUND_UP(FLOW_MAX_MPLS_LABELS, 2)]; /* MPLS label stack
121 /* L3 (64-bit aligned) */
122 ovs_be32 nw_src; /* IPv4 source address. */
123 ovs_be32 nw_dst; /* IPv4 destination address. */
124 struct in6_addr ipv6_src; /* IPv6 source address. */
125 struct in6_addr ipv6_dst; /* IPv6 destination address. */
126 ovs_be32 ipv6_label; /* IPv6 flow label. */
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 struct in6_addr nd_target; /* IPv6 neighbor discovery (ND) target. */
132 struct eth_addr arp_sha; /* ARP/ND source hardware address. */
133 struct eth_addr arp_tha; /* ARP/ND target hardware address. */
134 ovs_be16 tcp_flags; /* TCP flags. With L3 to avoid matching L4. */
135 ovs_be16 pad3; /* Pad to 64 bits. */
137 /* L4 (64-bit aligned) */
138 ovs_be16 tp_src; /* TCP/UDP/SCTP source port/ICMP type. */
139 ovs_be16 tp_dst; /* TCP/UDP/SCTP destination port/ICMP code. */
140 ovs_be32 igmp_group_ip4; /* IGMP group IPv4 address.
141 * Keep last for BUILD_ASSERT_DECL below. */
143 BUILD_ASSERT_DECL(sizeof(struct flow) % sizeof(uint64_t) == 0);
144 BUILD_ASSERT_DECL(sizeof(struct flow_tnl) % sizeof(uint64_t) == 0);
146 #define FLOW_U64S (sizeof(struct flow) / sizeof(uint64_t))
148 /* Some flow fields are mutually exclusive or only appear within the flow
149 * pipeline. IPv6 headers are bigger than IPv4 and MPLS, and IPv6 ND packets
150 * are bigger than TCP,UDP and IGMP packets. */
151 #define FLOW_MAX_PACKET_U64S (FLOW_U64S \
152 /* Unused in datapath */ - FLOW_U64_SIZE(regs) \
153 - FLOW_U64_SIZE(metadata) \
154 /* L2.5/3 */ - FLOW_U64_SIZE(nw_src) /* incl. nw_dst */ \
155 - FLOW_U64_SIZE(mpls_lse) \
156 /* L4 */ - FLOW_U64_SIZE(tp_src) \
159 /* Remember to update FLOW_WC_SEQ when changing 'struct flow'. */
160 BUILD_ASSERT_DECL(offsetof(struct flow, igmp_group_ip4) + sizeof(uint32_t)
161 == sizeof(struct flow_tnl) + 216
162 && FLOW_WC_SEQ == 35);
164 /* Incremental points at which flow classification may be performed in
166 * This is located here since this is dependent on the structure of the
167 * struct flow defined above:
168 * Each offset must be on a distinct, successive U64 boundary strictly
169 * within the struct flow. */
171 FLOW_SEGMENT_1_ENDS_AT = offsetof(struct flow, dl_dst),
172 FLOW_SEGMENT_2_ENDS_AT = offsetof(struct flow, nw_src),
173 FLOW_SEGMENT_3_ENDS_AT = offsetof(struct flow, tp_src),
175 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT % sizeof(uint64_t) == 0);
176 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT % sizeof(uint64_t) == 0);
177 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT % sizeof(uint64_t) == 0);
178 BUILD_ASSERT_DECL( 0 < FLOW_SEGMENT_1_ENDS_AT);
179 BUILD_ASSERT_DECL(FLOW_SEGMENT_1_ENDS_AT < FLOW_SEGMENT_2_ENDS_AT);
180 BUILD_ASSERT_DECL(FLOW_SEGMENT_2_ENDS_AT < FLOW_SEGMENT_3_ENDS_AT);
181 BUILD_ASSERT_DECL(FLOW_SEGMENT_3_ENDS_AT < sizeof(struct flow));
183 extern const uint8_t flow_segment_u64s[];
185 #define FLOW_U64_OFFSET(FIELD) \
186 (offsetof(struct flow, FIELD) / sizeof(uint64_t))
187 #define FLOW_U64_OFFREM(FIELD) \
188 (offsetof(struct flow, FIELD) % sizeof(uint64_t))
190 /* Number of 64-bit units spanned by a 'FIELD'. */
191 #define FLOW_U64_SIZE(FIELD) \
192 DIV_ROUND_UP(FLOW_U64_OFFREM(FIELD) + MEMBER_SIZEOF(struct flow, FIELD), \
195 void flow_extract(struct dp_packet *, struct flow *);
197 void flow_zero_wildcards(struct flow *, const struct flow_wildcards *);
198 void flow_unwildcard_tp_ports(const struct flow *, struct flow_wildcards *);
199 void flow_get_metadata(const struct flow *, struct match *flow_metadata);
201 const char *ct_state_to_string(uint32_t state);
202 char *flow_to_string(const struct flow *);
203 void format_flags(struct ds *ds, const char *(*bit_to_string)(uint32_t),
204 uint32_t flags, char del);
205 void format_flags_masked(struct ds *ds, const char *name,
206 const char *(*bit_to_string)(uint32_t),
207 uint32_t flags, uint32_t mask, uint32_t max_mask);
208 int parse_flags(const char *s, const char *(*bit_to_string)(uint32_t),
209 char end, const char *field_name, char **res_string,
210 uint32_t *res_flags, uint32_t allowed, uint32_t *res_mask);
212 void flow_format(struct ds *, const struct flow *);
213 void flow_print(FILE *, const struct flow *);
214 static inline int flow_compare_3way(const struct flow *, const struct flow *);
215 static inline bool flow_equal(const struct flow *, const struct flow *);
216 static inline size_t flow_hash(const struct flow *, uint32_t basis);
218 void flow_set_dl_vlan(struct flow *, ovs_be16 vid);
219 void flow_set_vlan_vid(struct flow *, ovs_be16 vid);
220 void flow_set_vlan_pcp(struct flow *, uint8_t pcp);
222 int flow_count_mpls_labels(const struct flow *, struct flow_wildcards *);
223 int flow_count_common_mpls_labels(const struct flow *a, int an,
224 const struct flow *b, int bn,
225 struct flow_wildcards *wc);
226 void flow_push_mpls(struct flow *, int n, ovs_be16 mpls_eth_type,
227 struct flow_wildcards *);
228 bool flow_pop_mpls(struct flow *, int n, ovs_be16 eth_type,
229 struct flow_wildcards *);
230 void flow_set_mpls_label(struct flow *, int idx, ovs_be32 label);
231 void flow_set_mpls_ttl(struct flow *, int idx, uint8_t ttl);
232 void flow_set_mpls_tc(struct flow *, int idx, uint8_t tc);
233 void flow_set_mpls_bos(struct flow *, int idx, uint8_t stack);
234 void flow_set_mpls_lse(struct flow *, int idx, ovs_be32 lse);
236 void flow_compose(struct dp_packet *, const struct flow *);
238 static inline uint64_t
239 flow_get_xreg(const struct flow *flow, int idx)
241 return ((uint64_t) flow->regs[idx * 2] << 32) | flow->regs[idx * 2 + 1];
245 flow_set_xreg(struct flow *flow, int idx, uint64_t value)
247 flow->regs[idx * 2] = value >> 32;
248 flow->regs[idx * 2 + 1] = value;
252 flow_compare_3way(const struct flow *a, const struct flow *b)
254 return memcmp(a, b, sizeof *a);
258 flow_equal(const struct flow *a, const struct flow *b)
260 return !flow_compare_3way(a, b);
264 flow_hash(const struct flow *flow, uint32_t basis)
266 return hash_words64((const uint64_t *)flow,
267 sizeof *flow / sizeof(uint64_t), basis);
270 static inline uint16_t
271 ofp_to_u16(ofp_port_t ofp_port)
273 return (OVS_FORCE uint16_t) ofp_port;
276 static inline uint32_t
277 odp_to_u32(odp_port_t odp_port)
279 return (OVS_FORCE uint32_t) odp_port;
282 static inline uint32_t
283 ofp11_to_u32(ofp11_port_t ofp11_port)
285 return (OVS_FORCE uint32_t) ofp11_port;
288 static inline ofp_port_t
289 u16_to_ofp(uint16_t port)
291 return OFP_PORT_C(port);
294 static inline odp_port_t
295 u32_to_odp(uint32_t port)
297 return ODP_PORT_C(port);
300 static inline ofp11_port_t
301 u32_to_ofp11(uint32_t port)
303 return OFP11_PORT_C(port);
306 static inline uint32_t
307 hash_ofp_port(ofp_port_t ofp_port)
309 return hash_int(ofp_to_u16(ofp_port), 0);
312 static inline uint32_t
313 hash_odp_port(odp_port_t odp_port)
315 return hash_int(odp_to_u32(odp_port), 0);
318 /* Wildcards for a flow.
320 * A 1-bit in each bit in 'masks' indicates that the corresponding bit of
321 * the flow is significant (must match). A 0-bit indicates that the
322 * corresponding bit of the flow is wildcarded (need not match). */
323 struct flow_wildcards {
327 #define WC_MASK_FIELD(WC, FIELD) \
328 memset(&(WC)->masks.FIELD, 0xff, sizeof (WC)->masks.FIELD)
329 #define WC_MASK_FIELD_MASK(WC, FIELD, MASK) \
330 ((WC)->masks.FIELD |= (MASK))
331 #define WC_UNMASK_FIELD(WC, FIELD) \
332 memset(&(WC)->masks.FIELD, 0, sizeof (WC)->masks.FIELD)
334 void flow_wildcards_init_catchall(struct flow_wildcards *);
336 void flow_wildcards_init_for_packet(struct flow_wildcards *,
337 const struct flow *);
339 void flow_wildcards_clear_non_packet_fields(struct flow_wildcards *);
341 bool flow_wildcards_is_catchall(const struct flow_wildcards *);
343 void flow_wildcards_set_reg_mask(struct flow_wildcards *,
344 int idx, uint32_t mask);
345 void flow_wildcards_set_xreg_mask(struct flow_wildcards *,
346 int idx, uint64_t mask);
348 void flow_wildcards_and(struct flow_wildcards *dst,
349 const struct flow_wildcards *src1,
350 const struct flow_wildcards *src2);
351 void flow_wildcards_or(struct flow_wildcards *dst,
352 const struct flow_wildcards *src1,
353 const struct flow_wildcards *src2);
354 bool flow_wildcards_has_extra(const struct flow_wildcards *,
355 const struct flow_wildcards *);
356 uint32_t flow_wildcards_hash(const struct flow_wildcards *, uint32_t basis);
357 bool flow_wildcards_equal(const struct flow_wildcards *,
358 const struct flow_wildcards *);
359 uint32_t flow_hash_5tuple(const struct flow *flow, uint32_t basis);
360 uint32_t flow_hash_symmetric_l4(const struct flow *flow, uint32_t basis);
361 uint32_t flow_hash_symmetric_l3l4(const struct flow *flow, uint32_t basis,
362 bool inc_udp_ports );
364 /* Initialize a flow with random fields that matter for nx_hash_fields. */
365 void flow_random_hash_fields(struct flow *);
366 void flow_mask_hash_fields(const struct flow *, struct flow_wildcards *,
367 enum nx_hash_fields);
368 uint32_t flow_hash_fields(const struct flow *, enum nx_hash_fields,
370 const char *flow_hash_fields_to_str(enum nx_hash_fields);
371 bool flow_hash_fields_valid(enum nx_hash_fields);
373 uint32_t flow_hash_in_wildcards(const struct flow *,
374 const struct flow_wildcards *,
377 bool flow_equal_except(const struct flow *a, const struct flow *b,
378 const struct flow_wildcards *);
380 /* Bitmap for flow values. For each 1-bit the corresponding flow value is
381 * explicitly specified, other values are zeroes.
383 * map_t must be wide enough to hold any member of struct flow. */
384 typedef unsigned long long map_t;
385 #define MAP_T_BITS (sizeof(map_t) * CHAR_BIT)
386 #define MAP_1 (map_t)1
387 #define MAP_MAX TYPE_MAXIMUM(map_t)
389 #define MAP_IS_SET(MAP, IDX) ((MAP) & (MAP_1 << (IDX)))
391 /* Iterate through the indices of all 1-bits in 'MAP'. */
392 #define MAP_FOR_EACH_INDEX(IDX, MAP) \
393 ULLONG_FOR_EACH_1(IDX, MAP)
395 #define FLOWMAP_UNITS DIV_ROUND_UP(FLOW_U64S, MAP_T_BITS)
398 map_t bits[FLOWMAP_UNITS];
401 #define FLOWMAP_EMPTY_INITIALIZER { { 0 } }
403 static inline void flowmap_init(struct flowmap *);
404 static inline bool flowmap_equal(struct flowmap, struct flowmap);
405 static inline bool flowmap_is_set(const struct flowmap *, size_t idx);
406 static inline bool flowmap_are_set(const struct flowmap *, size_t idx,
407 unsigned int n_bits);
408 static inline void flowmap_set(struct flowmap *, size_t idx,
409 unsigned int n_bits);
410 static inline void flowmap_clear(struct flowmap *, size_t idx,
411 unsigned int n_bits);
412 static inline struct flowmap flowmap_or(struct flowmap, struct flowmap);
413 static inline struct flowmap flowmap_and(struct flowmap, struct flowmap);
414 static inline bool flowmap_is_empty(struct flowmap);
415 static inline unsigned int flowmap_n_1bits(struct flowmap);
417 #define FLOWMAP_HAS_FIELD(FM, FIELD) \
418 flowmap_are_set(FM, FLOW_U64_OFFSET(FIELD), FLOW_U64_SIZE(FIELD))
420 #define FLOWMAP_SET(FM, FIELD) \
421 flowmap_set(FM, FLOW_U64_OFFSET(FIELD), FLOW_U64_SIZE(FIELD))
423 #define FLOWMAP_SET__(FM, FIELD, SIZE) \
424 flowmap_set(FM, FLOW_U64_OFFSET(FIELD), \
425 DIV_ROUND_UP(SIZE, sizeof(uint64_t)))
427 /* XXX: Only works for full 64-bit units. */
428 #define FLOWMAP_CLEAR(FM, FIELD) \
429 BUILD_ASSERT_DECL(FLOW_U64_OFFREM(FIELD) == 0); \
430 BUILD_ASSERT_DECL(sizeof(((struct flow *)0)->FIELD) % sizeof(uint64_t) == 0); \
431 flowmap_clear(FM, FLOW_U64_OFFSET(FIELD), FLOW_U64_SIZE(FIELD))
433 /* Iterate through all units in 'FMAP'. */
434 #define FLOWMAP_FOR_EACH_UNIT(UNIT) \
435 for ((UNIT) = 0; (UNIT) < FLOWMAP_UNITS; (UNIT)++)
437 /* Iterate through all map units in 'FMAP'. */
438 #define FLOWMAP_FOR_EACH_MAP(MAP, FLOWMAP) \
439 for (size_t unit__ = 0; \
440 unit__ < FLOWMAP_UNITS && ((MAP) = (FLOWMAP).bits[unit__], true); \
444 static inline bool flowmap_next_index(struct flowmap_aux *, size_t *idx);
446 #define FLOWMAP_AUX_INITIALIZER(FLOWMAP) { .unit = 0, .map = (FLOWMAP) }
448 /* Iterate through all struct flow u64 indices specified by 'MAP'. This is a
449 * slower but easier version of the FLOWMAP_FOR_EACH_MAP() &
450 * MAP_FOR_EACH_INDEX() combination. */
451 #define FLOWMAP_FOR_EACH_INDEX(IDX, MAP) \
452 for (struct flowmap_aux aux__ = FLOWMAP_AUX_INITIALIZER(MAP); \
453 flowmap_next_index(&aux__, &(IDX));)
455 /* Flowmap inline implementations. */
457 flowmap_init(struct flowmap *fm)
459 memset(fm, 0, sizeof *fm);
463 flowmap_equal(struct flowmap a, struct flowmap b)
465 return !memcmp(&a, &b, sizeof a);
469 flowmap_is_set(const struct flowmap *fm, size_t idx)
471 return (fm->bits[idx / MAP_T_BITS] & (MAP_1 << (idx % MAP_T_BITS))) != 0;
474 /* Returns 'true' if any of the 'n_bits' bits starting at 'idx' are set in
475 * 'fm'. 'n_bits' can be at most MAP_T_BITS. */
477 flowmap_are_set(const struct flowmap *fm, size_t idx, unsigned int n_bits)
479 map_t n_bits_mask = (MAP_1 << n_bits) - 1;
480 size_t unit = idx / MAP_T_BITS;
484 if (fm->bits[unit] & (n_bits_mask << idx)) {
487 /* The seemingly unnecessary bounds check on 'unit' is a workaround for a
488 * false-positive array out of bounds error by GCC 4.9. */
489 if (unit + 1 < FLOWMAP_UNITS && idx + n_bits > MAP_T_BITS) {
490 /* Check the remaining bits from the next unit. */
491 return fm->bits[unit + 1] & (n_bits_mask >> (MAP_T_BITS - idx));
496 /* Set the 'n_bits' consecutive bits in 'fm', starting at bit 'idx'.
497 * 'n_bits' can be at most MAP_T_BITS. */
499 flowmap_set(struct flowmap *fm, size_t idx, unsigned int n_bits)
501 map_t n_bits_mask = (MAP_1 << n_bits) - 1;
502 size_t unit = idx / MAP_T_BITS;
506 fm->bits[unit] |= n_bits_mask << idx;
507 /* The seemingly unnecessary bounds check on 'unit' is a workaround for a
508 * false-positive array out of bounds error by GCC 4.9. */
509 if (unit + 1 < FLOWMAP_UNITS && idx + n_bits > MAP_T_BITS) {
510 /* 'MAP_T_BITS - idx' bits were set on 'unit', set the remaining
511 * bits from the next unit. */
512 fm->bits[unit + 1] |= n_bits_mask >> (MAP_T_BITS - idx);
516 /* Clears the 'n_bits' consecutive bits in 'fm', starting at bit 'idx'.
517 * 'n_bits' can be at most MAP_T_BITS. */
519 flowmap_clear(struct flowmap *fm, size_t idx, unsigned int n_bits)
521 map_t n_bits_mask = (MAP_1 << n_bits) - 1;
522 size_t unit = idx / MAP_T_BITS;
526 fm->bits[unit] &= ~(n_bits_mask << idx);
527 /* The seemingly unnecessary bounds check on 'unit' is a workaround for a
528 * false-positive array out of bounds error by GCC 4.9. */
529 if (unit + 1 < FLOWMAP_UNITS && idx + n_bits > MAP_T_BITS) {
530 /* 'MAP_T_BITS - idx' bits were cleared on 'unit', clear the
531 * remaining bits from the next unit. */
532 fm->bits[unit + 1] &= ~(n_bits_mask >> (MAP_T_BITS - idx));
536 /* OR the bits in the flowmaps. */
537 static inline struct flowmap
538 flowmap_or(struct flowmap a, struct flowmap b)
543 FLOWMAP_FOR_EACH_UNIT (unit) {
544 map.bits[unit] = a.bits[unit] | b.bits[unit];
549 /* AND the bits in the flowmaps. */
550 static inline struct flowmap
551 flowmap_and(struct flowmap a, struct flowmap b)
556 FLOWMAP_FOR_EACH_UNIT (unit) {
557 map.bits[unit] = a.bits[unit] & b.bits[unit];
563 flowmap_is_empty(struct flowmap fm)
567 FLOWMAP_FOR_EACH_MAP (map, fm) {
575 static inline unsigned int
576 flowmap_n_1bits(struct flowmap fm)
578 unsigned int n_1bits = 0;
581 FLOWMAP_FOR_EACH_UNIT (unit) {
582 n_1bits += count_1bits(fm.bits[unit]);
593 flowmap_next_index(struct flowmap_aux *aux, size_t *idx)
596 map_t *map = &aux->map.bits[aux->unit];
598 *idx = aux->unit * MAP_T_BITS + raw_ctz(*map);
599 *map = zero_rightmost_1bit(*map);
602 if (++aux->unit >= FLOWMAP_UNITS) {
609 /* Compressed flow. */
611 /* A sparse representation of a "struct flow".
613 * A "struct flow" is fairly large and tends to be mostly zeros. Sparse
614 * representation has two advantages. First, it saves memory and, more
615 * importantly, minimizes the number of accessed cache lines. Second, it saves
616 * time when the goal is to iterate over only the nonzero parts of the struct.
618 * The map member hold one bit for each uint64_t in a "struct flow". Each
619 * 0-bit indicates that the corresponding uint64_t is zero, each 1-bit that it
620 * *may* be nonzero (see below how this applies to minimasks).
622 * The values indicated by 'map' always follow the miniflow in memory. The
623 * user of the miniflow is responsible for always having enough storage after
624 * the struct miniflow corresponding to the number of 1-bits in maps.
626 * Elements in values array are allowed to be zero. This is useful for "struct
627 * minimatch", for which ensuring that the miniflow and minimask members have
628 * same maps allows optimization. This allowance applies only to a miniflow
629 * that is not a mask. That is, a minimask may NOT have zero elements in its
632 * A miniflow is always dynamically allocated so that the maps are followed by
633 * at least as many elements as there are 1-bits in maps. */
637 * uint64_t values[n];
638 * where 'n' is miniflow_n_values(miniflow). */
640 BUILD_ASSERT_DECL(sizeof(struct miniflow) % sizeof(uint64_t) == 0);
642 #define MINIFLOW_VALUES_SIZE(COUNT) ((COUNT) * sizeof(uint64_t))
644 static inline uint64_t *miniflow_values(struct miniflow *mf)
646 return (uint64_t *)(mf + 1);
649 static inline const uint64_t *miniflow_get_values(const struct miniflow *mf)
651 return (const uint64_t *)(mf + 1);
656 /* The 'dst' must follow with buffer space for FLOW_U64S 64-bit units.
657 * 'dst->map' is ignored on input and set on output to indicate which fields
659 void miniflow_extract(struct dp_packet *packet, struct miniflow *dst);
660 void miniflow_map_init(struct miniflow *, const struct flow *);
661 void flow_wc_map(const struct flow *, struct flowmap *);
662 size_t miniflow_alloc(struct miniflow *dsts[], size_t n,
663 const struct miniflow *src);
664 void miniflow_init(struct miniflow *, const struct flow *);
665 void miniflow_clone(struct miniflow *, const struct miniflow *,
667 struct miniflow * miniflow_create(const struct flow *);
669 void miniflow_expand(const struct miniflow *, struct flow *);
671 static inline uint64_t flow_u64_value(const struct flow *flow, size_t index)
673 return ((uint64_t *)flow)[index];
676 static inline uint64_t *flow_u64_lvalue(struct flow *flow, size_t index)
678 return &((uint64_t *)flow)[index];
682 miniflow_n_values(const struct miniflow *flow)
684 return flowmap_n_1bits(flow->map);
687 struct flow_for_each_in_maps_aux {
688 const struct flow *flow;
689 struct flowmap_aux map_aux;
693 flow_values_get_next_in_maps(struct flow_for_each_in_maps_aux *aux,
698 if (flowmap_next_index(&aux->map_aux, &idx)) {
699 *value = flow_u64_value(aux->flow, idx);
705 /* Iterate through all flow u64 values specified by 'MAPS'. */
706 #define FLOW_FOR_EACH_IN_MAPS(VALUE, FLOW, MAPS) \
707 for (struct flow_for_each_in_maps_aux aux__ \
708 = { (FLOW), FLOWMAP_AUX_INITIALIZER(MAPS) }; \
709 flow_values_get_next_in_maps(&aux__, &(VALUE));)
711 struct mf_for_each_in_map_aux {
715 const uint64_t *values;
719 mf_get_next_in_map(struct mf_for_each_in_map_aux *aux,
725 while (OVS_UNLIKELY(!*(map = &aux->map.bits[aux->unit]))) {
726 /* Skip remaining data in the previous unit. */
727 aux->values += count_1bits(aux->fmap.bits[aux->unit]);
728 if (++aux->unit == FLOWMAP_UNITS) {
733 rm1bit = rightmost_1bit(*map);
735 fmap = &aux->fmap.bits[aux->unit];
737 if (OVS_LIKELY(*fmap & rm1bit)) {
738 map_t trash = *fmap & (rm1bit - 1);
741 /* count_1bits() is fast for systems where speed matters (e.g.,
742 * DPDK), so we don't try avoid using it.
743 * Advance 'aux->values' to point to the value for 'rm1bit'. */
744 aux->values += count_1bits(trash);
746 *value = *aux->values;
753 /* Iterate through miniflow u64 values specified by 'FLOWMAP'. */
754 #define MINIFLOW_FOR_EACH_IN_FLOWMAP(VALUE, FLOW, FLOWMAP) \
755 for (struct mf_for_each_in_map_aux aux__ = \
756 { 0, (FLOW)->map, (FLOWMAP), miniflow_get_values(FLOW) }; \
757 mf_get_next_in_map(&aux__, &(VALUE));)
759 /* This can be used when it is known that 'idx' is set in 'map'. */
760 static inline const uint64_t *
761 miniflow_values_get__(const uint64_t *values, map_t map, size_t idx)
763 return values + count_1bits(map & ((MAP_1 << idx) - 1));
766 /* This can be used when it is known that 'u64_idx' is set in
767 * the map of 'mf'. */
768 static inline const uint64_t *
769 miniflow_get__(const struct miniflow *mf, size_t idx)
771 const uint64_t *values = miniflow_get_values(mf);
772 const map_t *map = mf->map.bits;
774 while (idx >= MAP_T_BITS) {
776 values += count_1bits(*map++);
778 return miniflow_values_get__(values, *map, idx);
781 #define MINIFLOW_IN_MAP(MF, IDX) flowmap_is_set(&(MF)->map, IDX)
783 /* Get the value of the struct flow 'FIELD' as up to 8 byte wide integer type
784 * 'TYPE' from miniflow 'MF'. */
785 #define MINIFLOW_GET_TYPE(MF, TYPE, FIELD) \
786 (MINIFLOW_IN_MAP(MF, FLOW_U64_OFFSET(FIELD)) \
787 ? ((OVS_FORCE const TYPE *)miniflow_get__(MF, FLOW_U64_OFFSET(FIELD))) \
788 [FLOW_U64_OFFREM(FIELD) / sizeof(TYPE)] \
791 #define MINIFLOW_GET_U128(FLOW, FIELD) \
792 (ovs_u128) { .u64 = { \
793 (MINIFLOW_IN_MAP(FLOW, FLOW_U64_OFFSET(FIELD)) ? \
794 *miniflow_get__(FLOW, FLOW_U64_OFFSET(FIELD)) : 0), \
795 (MINIFLOW_IN_MAP(FLOW, FLOW_U64_OFFSET(FIELD) + 1) ? \
796 *miniflow_get__(FLOW, FLOW_U64_OFFSET(FIELD) + 1) : 0) } }
798 #define MINIFLOW_GET_U8(FLOW, FIELD) \
799 MINIFLOW_GET_TYPE(FLOW, uint8_t, FIELD)
800 #define MINIFLOW_GET_U16(FLOW, FIELD) \
801 MINIFLOW_GET_TYPE(FLOW, uint16_t, FIELD)
802 #define MINIFLOW_GET_BE16(FLOW, FIELD) \
803 MINIFLOW_GET_TYPE(FLOW, ovs_be16, FIELD)
804 #define MINIFLOW_GET_U32(FLOW, FIELD) \
805 MINIFLOW_GET_TYPE(FLOW, uint32_t, FIELD)
806 #define MINIFLOW_GET_BE32(FLOW, FIELD) \
807 MINIFLOW_GET_TYPE(FLOW, ovs_be32, FIELD)
808 #define MINIFLOW_GET_U64(FLOW, FIELD) \
809 MINIFLOW_GET_TYPE(FLOW, uint64_t, FIELD)
810 #define MINIFLOW_GET_BE64(FLOW, FIELD) \
811 MINIFLOW_GET_TYPE(FLOW, ovs_be64, FIELD)
813 static inline uint64_t miniflow_get(const struct miniflow *,
814 unsigned int u64_ofs);
815 static inline uint32_t miniflow_get_u32(const struct miniflow *,
816 unsigned int u32_ofs);
817 static inline ovs_be32 miniflow_get_be32(const struct miniflow *,
818 unsigned int be32_ofs);
819 static inline uint16_t miniflow_get_vid(const struct miniflow *);
820 static inline uint16_t miniflow_get_tcp_flags(const struct miniflow *);
821 static inline ovs_be64 miniflow_get_metadata(const struct miniflow *);
823 bool miniflow_equal(const struct miniflow *a, const struct miniflow *b);
824 bool miniflow_equal_in_minimask(const struct miniflow *a,
825 const struct miniflow *b,
826 const struct minimask *);
827 bool miniflow_equal_flow_in_minimask(const struct miniflow *a,
828 const struct flow *b,
829 const struct minimask *);
830 uint32_t miniflow_hash_5tuple(const struct miniflow *flow, uint32_t basis);
833 /* Compressed flow wildcards. */
835 /* A sparse representation of a "struct flow_wildcards".
837 * See the large comment on struct miniflow for details.
839 * Note: While miniflow can have zero data for a 1-bit in the map,
840 * a minimask may not! We rely on this in the implementation. */
842 struct miniflow masks;
845 void minimask_init(struct minimask *, const struct flow_wildcards *);
846 struct minimask * minimask_create(const struct flow_wildcards *);
847 void minimask_combine(struct minimask *dst,
848 const struct minimask *a, const struct minimask *b,
849 uint64_t storage[FLOW_U64S]);
851 void minimask_expand(const struct minimask *, struct flow_wildcards *);
853 static inline uint32_t minimask_get_u32(const struct minimask *,
854 unsigned int u32_ofs);
855 static inline ovs_be32 minimask_get_be32(const struct minimask *,
856 unsigned int be32_ofs);
857 static inline uint16_t minimask_get_vid_mask(const struct minimask *);
858 static inline ovs_be64 minimask_get_metadata_mask(const struct minimask *);
860 bool minimask_equal(const struct minimask *a, const struct minimask *b);
861 bool minimask_has_extra(const struct minimask *, const struct minimask *);
864 /* Returns true if 'mask' matches every packet, false if 'mask' fixes any bits
867 minimask_is_catchall(const struct minimask *mask)
869 /* For every 1-bit in mask's map, the corresponding value is non-zero,
870 * so the only way the mask can not fix any bits or fields is for the
871 * map the be zero. */
872 return flowmap_is_empty(mask->masks.map);
875 /* Returns the uint64_t that would be at byte offset '8 * u64_ofs' if 'flow'
876 * were expanded into a "struct flow". */
877 static inline uint64_t miniflow_get(const struct miniflow *flow,
878 unsigned int u64_ofs)
880 return MINIFLOW_IN_MAP(flow, u64_ofs) ? *miniflow_get__(flow, u64_ofs) : 0;
883 static inline uint32_t miniflow_get_u32(const struct miniflow *flow,
884 unsigned int u32_ofs)
886 uint64_t value = miniflow_get(flow, u32_ofs / 2);
889 return (u32_ofs & 1) ? value : value >> 32;
891 return (u32_ofs & 1) ? value >> 32 : value;
895 static inline ovs_be32 miniflow_get_be32(const struct miniflow *flow,
896 unsigned int be32_ofs)
898 return (OVS_FORCE ovs_be32)miniflow_get_u32(flow, be32_ofs);
901 /* Returns the VID within the vlan_tci member of the "struct flow" represented
903 static inline uint16_t
904 miniflow_get_vid(const struct miniflow *flow)
906 ovs_be16 tci = MINIFLOW_GET_BE16(flow, vlan_tci);
907 return vlan_tci_to_vid(tci);
910 /* Returns the uint32_t that would be at byte offset '4 * u32_ofs' if 'mask'
911 * were expanded into a "struct flow_wildcards". */
912 static inline uint32_t
913 minimask_get_u32(const struct minimask *mask, unsigned int u32_ofs)
915 return miniflow_get_u32(&mask->masks, u32_ofs);
918 static inline ovs_be32
919 minimask_get_be32(const struct minimask *mask, unsigned int be32_ofs)
921 return (OVS_FORCE ovs_be32)minimask_get_u32(mask, be32_ofs);
924 /* Returns the VID mask within the vlan_tci member of the "struct
925 * flow_wildcards" represented by 'mask'. */
926 static inline uint16_t
927 minimask_get_vid_mask(const struct minimask *mask)
929 return miniflow_get_vid(&mask->masks);
932 /* Returns the value of the "tcp_flags" field in 'flow'. */
933 static inline uint16_t
934 miniflow_get_tcp_flags(const struct miniflow *flow)
936 return ntohs(MINIFLOW_GET_BE16(flow, tcp_flags));
939 /* Returns the value of the OpenFlow 1.1+ "metadata" field in 'flow'. */
940 static inline ovs_be64
941 miniflow_get_metadata(const struct miniflow *flow)
943 return MINIFLOW_GET_BE64(flow, metadata);
946 /* Returns the mask for the OpenFlow 1.1+ "metadata" field in 'mask'.
948 * The return value is all-1-bits if 'mask' matches on the whole value of the
949 * metadata field, all-0-bits if 'mask' entirely wildcards the metadata field,
950 * or some other value if the metadata field is partially matched, partially
952 static inline ovs_be64
953 minimask_get_metadata_mask(const struct minimask *mask)
955 return MINIFLOW_GET_BE64(&mask->masks, metadata);
958 /* Perform a bitwise OR of miniflow 'src' flow data specified in 'subset' with
959 * the equivalent fields in 'dst', storing the result in 'dst'. 'subset' must
960 * be a subset of 'src's map. */
962 flow_union_with_miniflow_subset(struct flow *dst, const struct miniflow *src,
963 struct flowmap subset)
965 uint64_t *dst_u64 = (uint64_t *) dst;
966 const uint64_t *p = miniflow_get_values(src);
969 FLOWMAP_FOR_EACH_MAP (map, subset) {
972 MAP_FOR_EACH_INDEX(idx, map) {
973 dst_u64[idx] |= *p++;
975 dst_u64 += MAP_T_BITS;
979 /* Perform a bitwise OR of miniflow 'src' flow data with the equivalent
980 * fields in 'dst', storing the result in 'dst'. */
982 flow_union_with_miniflow(struct flow *dst, const struct miniflow *src)
984 flow_union_with_miniflow_subset(dst, src, src->map);
988 pkt_metadata_from_flow(struct pkt_metadata *md, const struct flow *flow)
990 md->recirc_id = flow->recirc_id;
991 md->dp_hash = flow->dp_hash;
992 flow_tnl_copy__(&md->tunnel, &flow->tunnel);
993 md->skb_priority = flow->skb_priority;
994 md->pkt_mark = flow->pkt_mark;
995 md->in_port = flow->in_port;
996 md->ct_state = flow->ct_state;
997 md->ct_zone = flow->ct_zone;
998 md->ct_mark = flow->ct_mark;
999 md->ct_label = flow->ct_label;
1002 static inline bool is_ip_any(const struct flow *flow)
1004 return dl_type_is_ip_any(flow->dl_type);
1007 static inline bool is_icmpv4(const struct flow *flow)
1009 return (flow->dl_type == htons(ETH_TYPE_IP)
1010 && flow->nw_proto == IPPROTO_ICMP);
1013 static inline bool is_icmpv6(const struct flow *flow)
1015 return (flow->dl_type == htons(ETH_TYPE_IPV6)
1016 && flow->nw_proto == IPPROTO_ICMPV6);
1019 static inline bool is_igmp(const struct flow *flow)
1021 return (flow->dl_type == htons(ETH_TYPE_IP)
1022 && flow->nw_proto == IPPROTO_IGMP);
1025 static inline bool is_mld(const struct flow *flow)
1027 return is_icmpv6(flow)
1028 && (flow->tp_src == htons(MLD_QUERY)
1029 || flow->tp_src == htons(MLD_REPORT)
1030 || flow->tp_src == htons(MLD_DONE)
1031 || flow->tp_src == htons(MLD2_REPORT));
1034 static inline bool is_mld_query(const struct flow *flow)
1036 return is_icmpv6(flow) && flow->tp_src == htons(MLD_QUERY);
1039 static inline bool is_mld_report(const struct flow *flow)
1041 return is_mld(flow) && !is_mld_query(flow);
1044 static inline bool is_stp(const struct flow *flow)
1046 return (eth_addr_equals(flow->dl_dst, eth_addr_stp)
1047 && flow->dl_type == htons(FLOW_DL_TYPE_NONE));