2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
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.
17 #include <sys/types.h>
20 #include <netinet/in.h>
23 #include "byte-order.h"
25 #include "dynamic-string.h"
28 #include "openflow/openflow.h"
29 #include "openvswitch/datapath-protocol.h"
31 #include "unaligned.h"
34 VLOG_DEFINE_THIS_MODULE(flow);
36 static struct arp_eth_header *
37 pull_arp(struct ofpbuf *packet)
39 return ofpbuf_try_pull(packet, ARP_ETH_HEADER_LEN);
42 static struct ip_header *
43 pull_ip(struct ofpbuf *packet)
45 if (packet->size >= IP_HEADER_LEN) {
46 struct ip_header *ip = packet->data;
47 int ip_len = IP_IHL(ip->ip_ihl_ver) * 4;
48 if (ip_len >= IP_HEADER_LEN && packet->size >= ip_len) {
49 return ofpbuf_pull(packet, ip_len);
55 static struct tcp_header *
56 pull_tcp(struct ofpbuf *packet)
58 if (packet->size >= TCP_HEADER_LEN) {
59 struct tcp_header *tcp = packet->data;
60 int tcp_len = TCP_OFFSET(tcp->tcp_ctl) * 4;
61 if (tcp_len >= TCP_HEADER_LEN && packet->size >= tcp_len) {
62 return ofpbuf_pull(packet, tcp_len);
68 static struct udp_header *
69 pull_udp(struct ofpbuf *packet)
71 return ofpbuf_try_pull(packet, UDP_HEADER_LEN);
74 static struct icmp_header *
75 pull_icmp(struct ofpbuf *packet)
77 return ofpbuf_try_pull(packet, ICMP_HEADER_LEN);
81 parse_vlan(struct ofpbuf *b, struct flow *flow)
84 ovs_be16 eth_type; /* ETH_TYPE_VLAN */
88 if (b->size >= sizeof(struct qtag_prefix) + sizeof(ovs_be16)) {
89 struct qtag_prefix *qp = ofpbuf_pull(b, sizeof *qp);
90 flow->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
91 flow->dl_vlan_pcp = vlan_tci_to_pcp(qp->tci);
96 parse_ethertype(struct ofpbuf *b)
98 struct llc_snap_header *llc;
101 proto = *(ovs_be16 *) ofpbuf_pull(b, sizeof proto);
102 if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF) {
106 if (b->size < sizeof *llc) {
107 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
111 if (llc->llc.llc_dsap != LLC_DSAP_SNAP
112 || llc->llc.llc_ssap != LLC_SSAP_SNAP
113 || llc->llc.llc_cntl != LLC_CNTL_SNAP
114 || memcmp(llc->snap.snap_org, SNAP_ORG_ETHERNET,
115 sizeof llc->snap.snap_org)) {
116 return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
119 ofpbuf_pull(b, sizeof *llc);
120 return llc->snap.snap_type;
123 /* Initializes 'flow' members from 'packet', 'tun_id', and 'in_port.
124 * Initializes 'packet' header pointers as follows:
126 * - packet->l2 to the start of the Ethernet header.
128 * - packet->l3 to just past the Ethernet header, or just past the
129 * vlan_header if one is present, to the first byte of the payload of the
132 * - packet->l4 to just past the IPv4 header, if one is present and has a
133 * correct length, and otherwise NULL.
135 * - packet->l7 to just past the TCP or UDP or ICMP header, if one is
136 * present and has a correct length, and otherwise NULL.
139 flow_extract(struct ofpbuf *packet, ovs_be32 tun_id, uint16_t in_port,
142 struct ofpbuf b = *packet;
143 struct eth_header *eth;
146 COVERAGE_INC(flow_extract);
148 memset(flow, 0, sizeof *flow);
149 flow->tun_id = tun_id;
150 flow->in_port = in_port;
151 flow->dl_vlan = htons(OFP_VLAN_NONE);
158 if (b.size < sizeof *eth) {
164 memcpy(flow->dl_src, eth->eth_src, ETH_ADDR_LEN);
165 memcpy(flow->dl_dst, eth->eth_dst, ETH_ADDR_LEN);
167 /* dl_type, dl_vlan, dl_vlan_pcp. */
168 ofpbuf_pull(&b, ETH_ADDR_LEN * 2);
169 if (eth->eth_type == htons(ETH_TYPE_VLAN)) {
170 parse_vlan(&b, flow);
172 flow->dl_type = parse_ethertype(&b);
176 if (flow->dl_type == htons(ETH_TYPE_IP)) {
177 const struct ip_header *nh = pull_ip(&b);
179 flow->nw_src = get_unaligned_u32(&nh->ip_src);
180 flow->nw_dst = get_unaligned_u32(&nh->ip_dst);
181 flow->nw_tos = nh->ip_tos & IP_DSCP_MASK;
182 flow->nw_proto = nh->ip_proto;
184 if (!IP_IS_FRAGMENT(nh->ip_frag_off)) {
185 if (flow->nw_proto == IP_TYPE_TCP) {
186 const struct tcp_header *tcp = pull_tcp(&b);
188 flow->tp_src = tcp->tcp_src;
189 flow->tp_dst = tcp->tcp_dst;
192 } else if (flow->nw_proto == IP_TYPE_UDP) {
193 const struct udp_header *udp = pull_udp(&b);
195 flow->tp_src = udp->udp_src;
196 flow->tp_dst = udp->udp_dst;
199 } else if (flow->nw_proto == IP_TYPE_ICMP) {
200 const struct icmp_header *icmp = pull_icmp(&b);
202 flow->icmp_type = htons(icmp->icmp_type);
203 flow->icmp_code = htons(icmp->icmp_code);
211 } else if (flow->dl_type == htons(ETH_TYPE_ARP)) {
212 const struct arp_eth_header *arp = pull_arp(&b);
213 if (arp && arp->ar_hrd == htons(1)
214 && arp->ar_pro == htons(ETH_TYPE_IP)
215 && arp->ar_hln == ETH_ADDR_LEN
216 && arp->ar_pln == 4) {
217 /* We only match on the lower 8 bits of the opcode. */
218 if (ntohs(arp->ar_op) <= 0xff) {
219 flow->nw_proto = ntohs(arp->ar_op);
222 if ((flow->nw_proto == ARP_OP_REQUEST)
223 || (flow->nw_proto == ARP_OP_REPLY)) {
224 flow->nw_src = arp->ar_spa;
225 flow->nw_dst = arp->ar_tpa;
232 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
233 * arguments must have been initialized through a call to flow_extract().
236 flow_extract_stats(const struct flow *flow, struct ofpbuf *packet,
237 struct odp_flow_stats *stats)
239 memset(stats, '\0', sizeof(*stats));
241 if ((flow->dl_type == htons(ETH_TYPE_IP)) && packet->l4) {
242 if ((flow->nw_proto == IP_TYPE_TCP) && packet->l7) {
243 struct tcp_header *tcp = packet->l4;
244 stats->tcp_flags = TCP_FLAGS(tcp->tcp_ctl);
248 stats->n_bytes = packet->size;
249 stats->n_packets = 1;
252 /* Extract 'flow' with 'wildcards' into the OpenFlow match structure
253 * 'match'. 'flow_format' should be one of NXFF_*. */
255 flow_to_match(const struct flow *flow, uint32_t wildcards,
256 int flow_format, struct ofp_match *match)
258 wildcards &= (flow_format == NXFF_TUN_ID_FROM_COOKIE ? OVSFW_ALL
260 match->wildcards = htonl(wildcards);
262 match->in_port = htons(flow->in_port == ODPP_LOCAL ? OFPP_LOCAL
264 match->dl_vlan = flow->dl_vlan;
265 match->dl_vlan_pcp = flow->dl_vlan_pcp;
266 memcpy(match->dl_src, flow->dl_src, ETH_ADDR_LEN);
267 memcpy(match->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
268 match->dl_type = flow->dl_type;
269 match->nw_src = flow->nw_src;
270 match->nw_dst = flow->nw_dst;
271 match->nw_tos = flow->nw_tos;
272 match->nw_proto = flow->nw_proto;
273 match->tp_src = flow->tp_src;
274 match->tp_dst = flow->tp_dst;
275 memset(match->pad1, '\0', sizeof match->pad1);
276 memset(match->pad2, '\0', sizeof match->pad2);
280 flow_from_match(const struct ofp_match *match, int flow_format,
281 ovs_be64 cookie, struct flow *flow,
282 struct flow_wildcards *wc)
284 uint32_t wildcards = ntohl(match->wildcards) & OVSFW_ALL;
287 if (flow_format != NXFF_TUN_ID_FROM_COOKIE) {
288 wildcards |= NXFW_TUN_ID;
290 if (!(wildcards & NXFW_TUN_ID)) {
291 flow->tun_id = htonl(ntohll(cookie) >> 32);
294 if (wildcards & OFPFW_DL_DST) {
295 /* OpenFlow 1.0 OFPFW_DL_DST covers the whole Ethernet destination, but
296 * internally to OVS it excludes the multicast bit, which has to be set
297 * separately with FWW_ETH_MCAST. */
298 wildcards |= FWW_ETH_MCAST;
300 flow_wildcards_init(wc, wildcards);
302 flow->nw_src = match->nw_src;
303 flow->nw_dst = match->nw_dst;
304 flow->in_port = (match->in_port == htons(OFPP_LOCAL) ? ODPP_LOCAL
305 : ntohs(match->in_port));
306 flow->dl_vlan = match->dl_vlan;
307 flow->dl_vlan_pcp = match->dl_vlan_pcp;
308 flow->dl_type = match->dl_type;
309 flow->tp_src = match->tp_src;
310 flow->tp_dst = match->tp_dst;
311 memcpy(flow->dl_src, match->dl_src, ETH_ADDR_LEN);
312 memcpy(flow->dl_dst, match->dl_dst, ETH_ADDR_LEN);
313 flow->nw_tos = match->nw_tos;
314 flow->nw_proto = match->nw_proto;
318 flow_to_string(const struct flow *flow)
320 struct ds ds = DS_EMPTY_INITIALIZER;
321 flow_format(&ds, flow);
326 flow_format(struct ds *ds, const struct flow *flow)
328 ds_put_format(ds, "tunnel%08"PRIx32":in_port%04"PRIx16
329 ":vlan%"PRIu16":pcp%"PRIu8
330 " mac"ETH_ADDR_FMT"->"ETH_ADDR_FMT
334 " ip"IP_FMT"->"IP_FMT
335 " port%"PRIu16"->%"PRIu16,
338 ntohs(flow->dl_vlan),
340 ETH_ADDR_ARGS(flow->dl_src),
341 ETH_ADDR_ARGS(flow->dl_dst),
342 ntohs(flow->dl_type),
345 IP_ARGS(&flow->nw_src),
346 IP_ARGS(&flow->nw_dst),
348 ntohs(flow->tp_dst));
352 flow_print(FILE *stream, const struct flow *flow)
354 char *s = flow_to_string(flow);
359 /* flow_wildcards functions. */
361 /* Given the wildcard bit count in bits 'shift' through 'shift + 5' (inclusive)
362 * of 'wildcards', returns a 32-bit bit mask with a 1 in each bit that must
363 * match and a 0 in each bit that is wildcarded.
365 * The bits in 'wildcards' are in the format used in enum ofp_flow_wildcards: 0
366 * is exact match, 1 ignores the LSB, 2 ignores the 2 least-significant bits,
367 * ..., 32 and higher wildcard the entire field. This is the *opposite* of the
368 * usual convention where e.g. /24 indicates that 8 bits (not 24 bits) are
371 flow_nw_bits_to_mask(uint32_t wildcards, int shift)
373 wildcards = (wildcards >> shift) & 0x3f;
374 return wildcards < 32 ? htonl(~((1u << wildcards) - 1)) : 0;
377 /* Return 'wildcards' in "normal form":
379 * - Forces unknown bits to 0.
381 * - Forces nw_src and nw_dst masks greater than 32 to exactly 32.
383 static inline uint32_t
384 flow_wildcards_normalize(uint32_t wildcards)
386 wildcards &= wildcards & (OVSFW_ALL | FWW_ALL);
387 if (wildcards & (0x20 << OFPFW_NW_SRC_SHIFT)) {
388 wildcards &= ~(0x1f << OFPFW_NW_SRC_SHIFT);
390 if (wildcards & (0x20 << OFPFW_NW_DST_SHIFT)) {
391 wildcards &= ~(0x1f << OFPFW_NW_DST_SHIFT);
396 /* Initializes 'wc' from 'wildcards', which may be any combination of the
397 * OFPFW_* and OVSFW_* wildcard bits.
399 * All registers (NXM_NX_REG*) are always completely wildcarded, because
400 * 'wildcards' doesn't have enough bits to give the details on which
401 * particular bits should be wildcarded (if any). The caller may use
402 * flow_wildcards_set_reg_mask() to update the register wildcard masks. */
404 flow_wildcards_init(struct flow_wildcards *wc, uint32_t wildcards)
406 wc->wildcards = flow_wildcards_normalize(wildcards) | FWW_REGS;
407 wc->nw_src_mask = flow_nw_bits_to_mask(wc->wildcards, OFPFW_NW_SRC_SHIFT);
408 wc->nw_dst_mask = flow_nw_bits_to_mask(wc->wildcards, OFPFW_NW_DST_SHIFT);
409 memset(wc->reg_masks, 0, sizeof wc->reg_masks);
412 /* Initializes 'wc' as an exact-match set of wildcards; that is, 'wc' does not
413 * wildcard any bits or fields. */
415 flow_wildcards_init_exact(struct flow_wildcards *wc)
418 wc->nw_src_mask = htonl(UINT32_MAX);
419 wc->nw_dst_mask = htonl(UINT32_MAX);
420 memset(wc->reg_masks, 0xff, sizeof wc->reg_masks);
423 /* Returns true if 'wc' is exact-match, false if 'wc' wildcards any bits or
426 flow_wildcards_is_exact(const struct flow_wildcards *wc)
428 return !wc->wildcards;
431 static inline uint32_t
432 combine_nw_bits(uint32_t wb1, uint32_t wb2, int shift)
434 uint32_t sb1 = (wb1 >> shift) & 0x3f;
435 uint32_t sb2 = (wb2 >> shift) & 0x3f;
436 return MAX(sb1, sb2) << shift;
439 /* Initializes 'dst' as the combination of wildcards in 'src1' and 'src2'.
440 * That is, a bit or a field is wildcarded in 'dst' if it is wildcarded in
441 * 'src1' or 'src2' or both. */
443 flow_wildcards_combine(struct flow_wildcards *dst,
444 const struct flow_wildcards *src1,
445 const struct flow_wildcards *src2)
447 uint32_t wb1 = src1->wildcards;
448 uint32_t wb2 = src2->wildcards;
451 dst->wildcards = (wb1 | wb2) & ~(OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK);
452 dst->wildcards |= combine_nw_bits(wb1, wb2, OFPFW_NW_SRC_SHIFT);
453 dst->wildcards |= combine_nw_bits(wb1, wb2, OFPFW_NW_DST_SHIFT);
454 dst->nw_src_mask = src1->nw_src_mask & src2->nw_src_mask;
455 dst->nw_dst_mask = src1->nw_dst_mask & src2->nw_dst_mask;
456 for (i = 0; i < FLOW_N_REGS; i++) {
457 dst->reg_masks[i] = src1->reg_masks[i] & src2->reg_masks[i];
461 /* Returns a hash of the wildcards in 'wc'. */
463 flow_wildcards_hash(const struct flow_wildcards *wc)
465 /* There is no need to include nw_src_mask or nw_dst_mask because they do
466 * not add any information (they can be computed from wc->wildcards). */
467 BUILD_ASSERT_DECL(sizeof wc->wildcards == 4);
468 BUILD_ASSERT_DECL(sizeof wc->reg_masks == 4 * FLOW_N_REGS);
469 BUILD_ASSERT_DECL(offsetof(struct flow_wildcards, wildcards) == 0);
470 BUILD_ASSERT_DECL(offsetof(struct flow_wildcards, reg_masks) == 4);
471 return hash_words((const uint32_t *) wc, 1 + FLOW_N_REGS, 0);
474 /* Returns true if 'a' and 'b' represent the same wildcards, false if they are
477 flow_wildcards_equal(const struct flow_wildcards *a,
478 const struct flow_wildcards *b)
482 if (a->wildcards != b->wildcards) {
486 for (i = 0; i < FLOW_N_REGS; i++) {
487 if (a->reg_masks[i] != b->reg_masks[i]) {
495 /* Returns true if at least one bit or field is wildcarded in 'a' but not in
496 * 'b', false otherwise. */
498 flow_wildcards_has_extra(const struct flow_wildcards *a,
499 const struct flow_wildcards *b)
503 for (i = 0; i < FLOW_N_REGS; i++) {
504 if ((a->reg_masks[i] & b->reg_masks[i]) != b->reg_masks[i]) {
509 #define OFPFW_NW_MASK (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK)
510 return ((a->wildcards & ~(b->wildcards | OFPFW_NW_MASK))
511 || (a->nw_src_mask & b->nw_src_mask) != b->nw_src_mask
512 || (a->nw_dst_mask & b->nw_dst_mask) != b->nw_dst_mask);
516 count_ones(ovs_be32 mask)
519 return __builtin_popcount(mask);
523 for (bits = 0; mask; bits++) {
532 set_nw_mask(struct flow_wildcards *wc, ovs_be32 mask,
533 ovs_be32 *maskp, int shift)
535 int wcbits = 32 - count_ones(mask);
536 if (flow_nw_bits_to_mask(wcbits, 0) == mask) {
537 wc->wildcards &= ~(0x3f << shift);
538 wc->wildcards |= wcbits << shift;
546 /* Sets the IP (or ARP) source wildcard mask to CIDR 'mask' (consisting of N
547 * high-order 1-bit and 32-N low-order 0-bits). Returns true if successful,
548 * false if 'mask' is not a CIDR mask. */
550 flow_wildcards_set_nw_src_mask(struct flow_wildcards *wc, ovs_be32 mask)
552 return set_nw_mask(wc, mask, &wc->nw_src_mask, OFPFW_NW_SRC_SHIFT);
555 /* Sets the IP (or ARP) destination wildcard mask to CIDR 'mask' (consisting of
556 * N high-order 1-bit and 32-N low-order 0-bits). Returns true if successful,
557 * false if 'mask' is not a CIDR mask. */
559 flow_wildcards_set_nw_dst_mask(struct flow_wildcards *wc, ovs_be32 mask)
561 return set_nw_mask(wc, mask, &wc->nw_dst_mask, OFPFW_NW_DST_SHIFT);
564 /* Sets the wildcard mask for register 'idx' in 'wc' to 'mask'.
565 * (A 0-bit indicates a wildcard bit.) */
567 flow_wildcards_set_reg_mask(struct flow_wildcards *wc, int idx, uint32_t mask)
569 if (mask != wc->reg_masks[idx]) {
570 wc->reg_masks[idx] = mask;
571 if (mask != UINT32_MAX) {
572 wc->wildcards |= FWW_REGS;
576 for (i = 0; i < FLOW_N_REGS; i++) {
577 if (wc->reg_masks[i] != UINT32_MAX) {
578 wc->wildcards |= FWW_REGS;
582 wc->wildcards &= ~FWW_REGS;