2 * Copyright (c) 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 <arpa/inet.h>
20 #include <sys/socket.h>
21 #include <netinet/in.h>
22 #include <netinet/ip6.h>
23 #include <netinet/icmp6.h>
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "ovs-thread.h"
33 #include "dp-packet.h"
34 #include "unaligned.h"
36 const struct in6_addr in6addr_exact = IN6ADDR_EXACT_INIT;
37 const struct in6_addr in6addr_all_hosts = IN6ADDR_ALL_HOSTS_INIT;
40 flow_tnl_dst(const struct flow_tnl *tnl)
42 return tnl->ip_dst ? in6_addr_mapped_ipv4(tnl->ip_dst) : tnl->ipv6_dst;
46 flow_tnl_src(const struct flow_tnl *tnl)
48 return tnl->ip_src ? in6_addr_mapped_ipv4(tnl->ip_src) : tnl->ipv6_src;
51 /* Parses 's' as a 16-digit hexadecimal number representing a datapath ID. On
52 * success stores the dpid into '*dpidp' and returns true, on failure stores 0
53 * into '*dpidp' and returns false.
55 * Rejects an all-zeros dpid as invalid. */
57 dpid_from_string(const char *s, uint64_t *dpidp)
59 *dpidp = (strlen(s) == 16 && strspn(s, "0123456789abcdefABCDEF") == 16
60 ? strtoull(s, NULL, 16)
65 /* Returns true if 'ea' is a reserved address, that a bridge must never
66 * forward, false otherwise.
68 * If you change this function's behavior, please update corresponding
69 * documentation in vswitch.xml at the same time. */
71 eth_addr_is_reserved(const struct eth_addr ea)
73 struct eth_addr_node {
74 struct hmap_node hmap_node;
78 static struct eth_addr_node nodes[] = {
79 /* STP, IEEE pause frames, and other reserved protocols. */
80 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000000ULL },
81 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000001ULL },
82 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000002ULL },
83 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000003ULL },
84 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000004ULL },
85 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000005ULL },
86 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000006ULL },
87 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000007ULL },
88 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000008ULL },
89 { HMAP_NODE_NULL_INITIALIZER, 0x0180c2000009ULL },
90 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000aULL },
91 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000bULL },
92 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000cULL },
93 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000dULL },
94 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000eULL },
95 { HMAP_NODE_NULL_INITIALIZER, 0x0180c200000fULL },
97 /* Extreme protocols. */
98 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000000ULL }, /* EDP. */
99 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000004ULL }, /* EAPS. */
100 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000006ULL }, /* EAPS. */
102 /* Cisco protocols. */
103 { HMAP_NODE_NULL_INITIALIZER, 0x01000c000000ULL }, /* ISL. */
104 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccccULL }, /* PAgP, UDLD, CDP,
106 { HMAP_NODE_NULL_INITIALIZER, 0x01000ccccccdULL }, /* PVST+. */
107 { HMAP_NODE_NULL_INITIALIZER, 0x01000ccdcdcdULL }, /* STP Uplink Fast,
111 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc0ULL },
112 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc1ULL },
113 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc2ULL },
114 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc3ULL },
115 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc4ULL },
116 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc5ULL },
117 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc6ULL },
118 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc7ULL },
121 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
122 struct eth_addr_node *node;
123 static struct hmap addrs;
126 if (ovsthread_once_start(&once)) {
128 for (node = nodes; node < &nodes[ARRAY_SIZE(nodes)]; node++) {
129 hmap_insert(&addrs, &node->hmap_node, hash_uint64(node->ea64));
131 ovsthread_once_done(&once);
134 ea64 = eth_addr_to_uint64(ea);
135 HMAP_FOR_EACH_IN_BUCKET (node, hmap_node, hash_uint64(ea64), &addrs) {
136 if (node->ea64 == ea64) {
144 eth_addr_from_string(const char *s, struct eth_addr *ea)
146 if (ovs_scan(s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(*ea))) {
154 /* Fills 'b' with a Reverse ARP packet with Ethernet source address 'eth_src'.
155 * This function is used by Open vSwitch to compose packets in cases where
156 * context is important but content doesn't (or shouldn't) matter.
158 * The returned packet has enough headroom to insert an 802.1Q VLAN header if
161 compose_rarp(struct dp_packet *b, const struct eth_addr eth_src)
163 struct eth_header *eth;
164 struct arp_eth_header *arp;
167 dp_packet_prealloc_tailroom(b, 2 + ETH_HEADER_LEN + VLAN_HEADER_LEN
168 + ARP_ETH_HEADER_LEN);
169 dp_packet_reserve(b, 2 + VLAN_HEADER_LEN);
170 eth = dp_packet_put_uninit(b, sizeof *eth);
171 eth->eth_dst = eth_addr_broadcast;
172 eth->eth_src = eth_src;
173 eth->eth_type = htons(ETH_TYPE_RARP);
175 arp = dp_packet_put_uninit(b, sizeof *arp);
176 arp->ar_hrd = htons(ARP_HRD_ETHERNET);
177 arp->ar_pro = htons(ARP_PRO_IP);
178 arp->ar_hln = sizeof arp->ar_sha;
179 arp->ar_pln = sizeof arp->ar_spa;
180 arp->ar_op = htons(ARP_OP_RARP);
181 arp->ar_sha = eth_src;
182 put_16aligned_be32(&arp->ar_spa, htonl(0));
183 arp->ar_tha = eth_src;
184 put_16aligned_be32(&arp->ar_tpa, htonl(0));
186 dp_packet_reset_offsets(b);
187 dp_packet_set_l3(b, arp);
190 /* Insert VLAN header according to given TCI. Packet passed must be Ethernet
191 * packet. Ignores the CFI bit of 'tci' using 0 instead.
193 * Also adjusts the layer offsets accordingly. */
195 eth_push_vlan(struct dp_packet *packet, ovs_be16 tpid, ovs_be16 tci)
197 struct vlan_eth_header *veh;
199 /* Insert new 802.1Q header. */
200 veh = dp_packet_resize_l2(packet, VLAN_HEADER_LEN);
201 memmove(veh, (char *)veh + VLAN_HEADER_LEN, 2 * ETH_ADDR_LEN);
202 veh->veth_type = tpid;
203 veh->veth_tci = tci & htons(~VLAN_CFI);
206 /* Removes outermost VLAN header (if any is present) from 'packet'.
208 * 'packet->l2_5' should initially point to 'packet''s outer-most VLAN header
209 * or may be NULL if there are no VLAN headers. */
211 eth_pop_vlan(struct dp_packet *packet)
213 struct vlan_eth_header *veh = dp_packet_l2(packet);
215 if (veh && dp_packet_size(packet) >= sizeof *veh
216 && eth_type_vlan(veh->veth_type)) {
218 memmove((char *)veh + VLAN_HEADER_LEN, veh, 2 * ETH_ADDR_LEN);
219 dp_packet_resize_l2(packet, -VLAN_HEADER_LEN);
223 /* Set ethertype of the packet. */
225 set_ethertype(struct dp_packet *packet, ovs_be16 eth_type)
227 struct eth_header *eh = dp_packet_l2(packet);
233 if (eth_type_vlan(eh->eth_type)) {
235 char *l2_5 = dp_packet_l2_5(packet);
237 p = ALIGNED_CAST(ovs_be16 *,
238 (l2_5 ? l2_5 : (char *)dp_packet_l3(packet)) - 2);
241 eh->eth_type = eth_type;
245 static bool is_mpls(struct dp_packet *packet)
247 return packet->l2_5_ofs != UINT16_MAX;
250 /* Set time to live (TTL) of an MPLS label stack entry (LSE). */
252 set_mpls_lse_ttl(ovs_be32 *lse, uint8_t ttl)
254 *lse &= ~htonl(MPLS_TTL_MASK);
255 *lse |= htonl((ttl << MPLS_TTL_SHIFT) & MPLS_TTL_MASK);
258 /* Set traffic class (TC) of an MPLS label stack entry (LSE). */
260 set_mpls_lse_tc(ovs_be32 *lse, uint8_t tc)
262 *lse &= ~htonl(MPLS_TC_MASK);
263 *lse |= htonl((tc << MPLS_TC_SHIFT) & MPLS_TC_MASK);
266 /* Set label of an MPLS label stack entry (LSE). */
268 set_mpls_lse_label(ovs_be32 *lse, ovs_be32 label)
270 *lse &= ~htonl(MPLS_LABEL_MASK);
271 *lse |= htonl((ntohl(label) << MPLS_LABEL_SHIFT) & MPLS_LABEL_MASK);
274 /* Set bottom of stack (BoS) bit of an MPLS label stack entry (LSE). */
276 set_mpls_lse_bos(ovs_be32 *lse, uint8_t bos)
278 *lse &= ~htonl(MPLS_BOS_MASK);
279 *lse |= htonl((bos << MPLS_BOS_SHIFT) & MPLS_BOS_MASK);
282 /* Compose an MPLS label stack entry (LSE) from its components:
283 * label, traffic class (TC), time to live (TTL) and
284 * bottom of stack (BoS) bit. */
286 set_mpls_lse_values(uint8_t ttl, uint8_t tc, uint8_t bos, ovs_be32 label)
288 ovs_be32 lse = htonl(0);
289 set_mpls_lse_ttl(&lse, ttl);
290 set_mpls_lse_tc(&lse, tc);
291 set_mpls_lse_bos(&lse, bos);
292 set_mpls_lse_label(&lse, label);
296 /* Set MPLS label stack entry to outermost MPLS header.*/
298 set_mpls_lse(struct dp_packet *packet, ovs_be32 mpls_lse)
300 /* Packet type should be MPLS to set label stack entry. */
301 if (is_mpls(packet)) {
302 struct mpls_hdr *mh = dp_packet_l2_5(packet);
304 /* Update mpls label stack entry. */
305 put_16aligned_be32(&mh->mpls_lse, mpls_lse);
309 /* Push MPLS label stack entry 'lse' onto 'packet' as the outermost MPLS
310 * header. If 'packet' does not already have any MPLS labels, then its
311 * Ethertype is changed to 'ethtype' (which must be an MPLS Ethertype). */
313 push_mpls(struct dp_packet *packet, ovs_be16 ethtype, ovs_be32 lse)
318 if (!eth_type_mpls(ethtype)) {
322 if (!is_mpls(packet)) {
323 /* Set MPLS label stack offset. */
324 packet->l2_5_ofs = packet->l3_ofs;
327 set_ethertype(packet, ethtype);
329 /* Push new MPLS shim header onto packet. */
330 len = packet->l2_5_ofs;
331 header = dp_packet_resize_l2_5(packet, MPLS_HLEN);
332 memmove(header, header + MPLS_HLEN, len);
333 memcpy(header + len, &lse, sizeof lse);
336 /* If 'packet' is an MPLS packet, removes its outermost MPLS label stack entry.
337 * If the label that was removed was the only MPLS label, changes 'packet''s
338 * Ethertype to 'ethtype' (which ordinarily should not be an MPLS
341 pop_mpls(struct dp_packet *packet, ovs_be16 ethtype)
343 if (is_mpls(packet)) {
344 struct mpls_hdr *mh = dp_packet_l2_5(packet);
345 size_t len = packet->l2_5_ofs;
347 set_ethertype(packet, ethtype);
348 if (get_16aligned_be32(&mh->mpls_lse) & htonl(MPLS_BOS_MASK)) {
349 dp_packet_set_l2_5(packet, NULL);
351 /* Shift the l2 header forward. */
352 memmove((char*)dp_packet_data(packet) + MPLS_HLEN, dp_packet_data(packet), len);
353 dp_packet_resize_l2_5(packet, -MPLS_HLEN);
357 /* Converts hex digits in 'hex' to an Ethernet packet in '*packetp'. The
358 * caller must free '*packetp'. On success, returns NULL. On failure, returns
359 * an error message and stores NULL in '*packetp'.
361 * Aligns the L3 header of '*packetp' on a 32-bit boundary. */
363 eth_from_hex(const char *hex, struct dp_packet **packetp)
365 struct dp_packet *packet;
367 /* Use 2 bytes of headroom to 32-bit align the L3 header. */
368 packet = *packetp = dp_packet_new_with_headroom(strlen(hex) / 2, 2);
370 if (dp_packet_put_hex(packet, hex, NULL)[0] != '\0') {
371 dp_packet_delete(packet);
373 return "Trailing garbage in packet data";
376 if (dp_packet_size(packet) < ETH_HEADER_LEN) {
377 dp_packet_delete(packet);
379 return "Packet data too short for Ethernet";
386 eth_format_masked(const struct eth_addr eth,
387 const struct eth_addr *mask, struct ds *s)
389 ds_put_format(s, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth));
390 if (mask && !eth_mask_is_exact(*mask)) {
391 ds_put_format(s, "/"ETH_ADDR_FMT, ETH_ADDR_ARGS(*mask));
395 /* Given the IP netmask 'netmask', returns the number of bits of the IP address
396 * that it specifies, that is, the number of 1-bits in 'netmask'.
398 * If 'netmask' is not a CIDR netmask (see ip_is_cidr()), the return value will
399 * still be in the valid range but isn't otherwise meaningful. */
401 ip_count_cidr_bits(ovs_be32 netmask)
403 return 32 - ctz32(ntohl(netmask));
407 ip_format_masked(ovs_be32 ip, ovs_be32 mask, struct ds *s)
409 ds_put_format(s, IP_FMT, IP_ARGS(ip));
410 if (mask != OVS_BE32_MAX) {
411 if (ip_is_cidr(mask)) {
412 ds_put_format(s, "/%d", ip_count_cidr_bits(mask));
414 ds_put_format(s, "/"IP_FMT, IP_ARGS(mask));
419 /* Parses string 's', which must be an IP address. Stores the IP address into
420 * '*ip'. Returns true if successful, otherwise false. */
422 ip_parse(const char *s, ovs_be32 *ip)
424 return inet_pton(AF_INET, s, ip) == 1;
427 /* Parses string 's', which must be an IP address with an optional netmask or
428 * CIDR prefix length. Stores the IP address into '*ip' and the netmask into
429 * '*mask'. (If 's' does not contain a netmask, 255.255.255.255 is
432 * Returns NULL if successful, otherwise an error message that the caller must
434 char * OVS_WARN_UNUSED_RESULT
435 ip_parse_masked(const char *s, ovs_be32 *ip, ovs_be32 *mask)
440 if (ovs_scan(s, IP_SCAN_FMT"/"IP_SCAN_FMT"%n",
441 IP_SCAN_ARGS(ip), IP_SCAN_ARGS(mask), &n) && !s[n]) {
443 } else if (ovs_scan(s, IP_SCAN_FMT"/%d%n", IP_SCAN_ARGS(ip), &prefix, &n)
445 if (prefix <= 0 || prefix > 32) {
446 return xasprintf("%s: network prefix bits not between 0 and "
449 *mask = be32_prefix_mask(prefix);
450 } else if (ip_parse(s, ip)) {
451 *mask = OVS_BE32_MAX;
453 return xasprintf("%s: invalid IP address", s);
458 /* Similar to ip_parse_masked(), but the mask, if present, must be a CIDR mask
459 * and is returned as a prefix length in '*plen'. */
460 char * OVS_WARN_UNUSED_RESULT
461 ip_parse_cidr(const char *s, ovs_be32 *ip, unsigned int *plen)
466 error = ip_parse_masked(s, ip, &mask);
471 if (!ip_is_cidr(mask)) {
472 return xasprintf("%s: CIDR network required", s);
474 *plen = ip_count_cidr_bits(mask);
478 /* Parses string 's', which must be an IPv6 address. Stores the IPv6 address
479 * into '*ip'. Returns true if successful, otherwise false. */
481 ipv6_parse(const char *s, struct in6_addr *ip)
483 return inet_pton(AF_INET6, s, ip) == 1;
486 /* Parses string 's', which must be an IPv6 address with an optional netmask or
487 * CIDR prefix length. Stores the IPv6 address into '*ip' and the netmask into
488 * '*mask'. (If 's' does not contain a netmask, all-one-bits is assumed.)
490 * Returns NULL if successful, otherwise an error message that the caller must
492 char * OVS_WARN_UNUSED_RESULT
493 ipv6_parse_masked(const char *s, struct in6_addr *ip, struct in6_addr *mask)
495 char ipv6_s[IPV6_SCAN_LEN + 1];
499 if (ovs_scan(s, IPV6_SCAN_FMT"%n", ipv6_s, &n) && ipv6_parse(ipv6_s, ip)) {
502 *mask = in6addr_exact;
503 } else if (ovs_scan(s, "/%d%n", &prefix, &n) && !s[n]) {
504 if (prefix <= 0 || prefix > 128) {
505 return xasprintf("%s: IPv6 network prefix bits not between 0 "
508 *mask = ipv6_create_mask(prefix);
509 } else if (ovs_scan(s, "/"IPV6_SCAN_FMT"%n", ipv6_s, &n)
511 && ipv6_parse(ipv6_s, mask)) {
514 return xasprintf("%s: syntax error expecting IPv6 prefix length "
519 return xasprintf("%s: invalid IPv6 address", s);
522 /* Similar to ipv6_parse_masked(), but the mask, if present, must be a CIDR
523 * mask and is returned as a prefix length in '*plen'. */
524 char * OVS_WARN_UNUSED_RESULT
525 ipv6_parse_cidr(const char *s, struct in6_addr *ip, unsigned int *plen)
527 struct in6_addr mask;
530 error = ipv6_parse_masked(s, ip, &mask);
535 if (!ipv6_is_cidr(&mask)) {
536 return xasprintf("%s: IPv6 CIDR network required", s);
538 *plen = ipv6_count_cidr_bits(&mask);
542 /* Stores the string representation of the IPv6 address 'addr' into the
543 * character array 'addr_str', which must be at least INET6_ADDRSTRLEN
546 ipv6_format_addr(const struct in6_addr *addr, struct ds *s)
550 ds_reserve(s, s->length + INET6_ADDRSTRLEN);
552 dst = s->string + s->length;
553 inet_ntop(AF_INET6, addr, dst, INET6_ADDRSTRLEN);
554 s->length += strlen(dst);
557 /* Same as print_ipv6_addr, but optionally encloses the address in square
560 ipv6_format_addr_bracket(const struct in6_addr *addr, struct ds *s,
566 ipv6_format_addr(addr, s);
573 ipv6_format_mapped(const struct in6_addr *addr, struct ds *s)
575 if (IN6_IS_ADDR_V4MAPPED(addr)) {
576 ds_put_format(s, IP_FMT, addr->s6_addr[12], addr->s6_addr[13],
577 addr->s6_addr[14], addr->s6_addr[15]);
579 ipv6_format_addr(addr, s);
584 ipv6_format_masked(const struct in6_addr *addr, const struct in6_addr *mask,
587 ipv6_format_addr(addr, s);
588 if (mask && !ipv6_mask_is_exact(mask)) {
589 if (ipv6_is_cidr(mask)) {
590 int cidr_bits = ipv6_count_cidr_bits(mask);
591 ds_put_format(s, "/%d", cidr_bits);
594 ipv6_format_addr(mask, s);
599 /* Stores the string representation of the IPv6 address 'addr' into the
600 * character array 'addr_str', which must be at least INET6_ADDRSTRLEN
601 * bytes long. If addr is IPv4-mapped, store an IPv4 dotted-decimal string. */
603 ipv6_string_mapped(char *addr_str, const struct in6_addr *addr)
606 ip = in6_addr_get_mapped_ipv4(addr);
608 return inet_ntop(AF_INET, &ip, addr_str, INET6_ADDRSTRLEN);
610 return inet_ntop(AF_INET6, addr, addr_str, INET6_ADDRSTRLEN);
614 struct in6_addr ipv6_addr_bitand(const struct in6_addr *a,
615 const struct in6_addr *b)
621 for (i=0; i<4; i++) {
622 dst.s6_addr32[i] = a->s6_addr32[i] & b->s6_addr32[i];
625 for (i=0; i<16; i++) {
626 dst.s6_addr[i] = a->s6_addr[i] & b->s6_addr[i];
633 /* Returns an in6_addr consisting of 'mask' high-order 1-bits and 128-N
634 * low-order 0-bits. */
636 ipv6_create_mask(int mask)
638 struct in6_addr netmask;
639 uint8_t *netmaskp = &netmask.s6_addr[0];
641 memset(&netmask, 0, sizeof netmask);
649 *netmaskp = 0xff << (8 - mask);
655 /* Given the IPv6 netmask 'netmask', returns the number of bits of the IPv6
656 * address that it specifies, that is, the number of 1-bits in 'netmask'.
657 * 'netmask' must be a CIDR netmask (see ipv6_is_cidr()).
659 * If 'netmask' is not a CIDR netmask (see ipv6_is_cidr()), the return value
660 * will still be in the valid range but isn't otherwise meaningful. */
662 ipv6_count_cidr_bits(const struct in6_addr *netmask)
666 const uint8_t *netmaskp = &netmask->s6_addr[0];
668 for (i=0; i<16; i++) {
669 if (netmaskp[i] == 0xff) {
674 for(nm = netmaskp[i]; nm; nm <<= 1) {
685 /* Returns true if 'netmask' is a CIDR netmask, that is, if it consists of N
686 * high-order 1-bits and 128-N low-order 0-bits. */
688 ipv6_is_cidr(const struct in6_addr *netmask)
690 const uint8_t *netmaskp = &netmask->s6_addr[0];
693 for (i=0; i<16; i++) {
694 if (netmaskp[i] != 0xff) {
695 uint8_t x = ~netmaskp[i];
710 /* Populates 'b' with an Ethernet II packet headed with the given 'eth_dst',
711 * 'eth_src' and 'eth_type' parameters. A payload of 'size' bytes is allocated
712 * in 'b' and returned. This payload may be populated with appropriate
713 * information by the caller. Sets 'b''s 'frame' pointer and 'l3' offset to
714 * the Ethernet header and payload respectively. Aligns b->l3 on a 32-bit
717 * The returned packet has enough headroom to insert an 802.1Q VLAN header if
720 eth_compose(struct dp_packet *b, const struct eth_addr eth_dst,
721 const struct eth_addr eth_src, uint16_t eth_type,
725 struct eth_header *eth;
729 /* The magic 2 here ensures that the L3 header (when it is added later)
730 * will be 32-bit aligned. */
731 dp_packet_prealloc_tailroom(b, 2 + ETH_HEADER_LEN + VLAN_HEADER_LEN + size);
732 dp_packet_reserve(b, 2 + VLAN_HEADER_LEN);
733 eth = dp_packet_put_uninit(b, ETH_HEADER_LEN);
734 data = dp_packet_put_uninit(b, size);
736 eth->eth_dst = eth_dst;
737 eth->eth_src = eth_src;
738 eth->eth_type = htons(eth_type);
740 dp_packet_reset_offsets(b);
741 dp_packet_set_l3(b, data);
747 packet_set_ipv4_addr(struct dp_packet *packet,
748 ovs_16aligned_be32 *addr, ovs_be32 new_addr)
750 struct ip_header *nh = dp_packet_l3(packet);
751 ovs_be32 old_addr = get_16aligned_be32(addr);
752 size_t l4_size = dp_packet_l4_size(packet);
754 if (nh->ip_proto == IPPROTO_TCP && l4_size >= TCP_HEADER_LEN) {
755 struct tcp_header *th = dp_packet_l4(packet);
757 th->tcp_csum = recalc_csum32(th->tcp_csum, old_addr, new_addr);
758 } else if (nh->ip_proto == IPPROTO_UDP && l4_size >= UDP_HEADER_LEN ) {
759 struct udp_header *uh = dp_packet_l4(packet);
762 uh->udp_csum = recalc_csum32(uh->udp_csum, old_addr, new_addr);
764 uh->udp_csum = htons(0xffff);
768 nh->ip_csum = recalc_csum32(nh->ip_csum, old_addr, new_addr);
769 put_16aligned_be32(addr, new_addr);
772 /* Returns true, if packet contains at least one routing header where
773 * segements_left > 0.
775 * This function assumes that L3 and L4 offsets are set in the packet. */
777 packet_rh_present(struct dp_packet *packet)
779 const struct ovs_16aligned_ip6_hdr *nh;
783 uint8_t *data = dp_packet_l3(packet);
785 remaining = packet->l4_ofs - packet->l3_ofs;
787 if (remaining < sizeof *nh) {
790 nh = ALIGNED_CAST(struct ovs_16aligned_ip6_hdr *, data);
792 remaining -= sizeof *nh;
793 nexthdr = nh->ip6_nxt;
796 if ((nexthdr != IPPROTO_HOPOPTS)
797 && (nexthdr != IPPROTO_ROUTING)
798 && (nexthdr != IPPROTO_DSTOPTS)
799 && (nexthdr != IPPROTO_AH)
800 && (nexthdr != IPPROTO_FRAGMENT)) {
801 /* It's either a terminal header (e.g., TCP, UDP) or one we
802 * don't understand. In either case, we're done with the
803 * packet, so use it to fill in 'nw_proto'. */
807 /* We only verify that at least 8 bytes of the next header are
808 * available, but many of these headers are longer. Ensure that
809 * accesses within the extension header are within those first 8
810 * bytes. All extension headers are required to be at least 8
816 if (nexthdr == IPPROTO_AH) {
817 /* A standard AH definition isn't available, but the fields
818 * we care about are in the same location as the generic
819 * option header--only the header length is calculated
821 const struct ip6_ext *ext_hdr = (struct ip6_ext *)data;
823 nexthdr = ext_hdr->ip6e_nxt;
824 len = (ext_hdr->ip6e_len + 2) * 4;
825 } else if (nexthdr == IPPROTO_FRAGMENT) {
826 const struct ovs_16aligned_ip6_frag *frag_hdr
827 = ALIGNED_CAST(struct ovs_16aligned_ip6_frag *, data);
829 nexthdr = frag_hdr->ip6f_nxt;
830 len = sizeof *frag_hdr;
831 } else if (nexthdr == IPPROTO_ROUTING) {
832 const struct ip6_rthdr *rh = (struct ip6_rthdr *)data;
834 if (rh->ip6r_segleft > 0) {
838 nexthdr = rh->ip6r_nxt;
839 len = (rh->ip6r_len + 1) * 8;
841 const struct ip6_ext *ext_hdr = (struct ip6_ext *)data;
843 nexthdr = ext_hdr->ip6e_nxt;
844 len = (ext_hdr->ip6e_len + 1) * 8;
847 if (remaining < len) {
858 packet_update_csum128(struct dp_packet *packet, uint8_t proto,
859 ovs_16aligned_be32 addr[4], const ovs_be32 new_addr[4])
861 size_t l4_size = dp_packet_l4_size(packet);
863 if (proto == IPPROTO_TCP && l4_size >= TCP_HEADER_LEN) {
864 struct tcp_header *th = dp_packet_l4(packet);
866 th->tcp_csum = recalc_csum128(th->tcp_csum, addr, new_addr);
867 } else if (proto == IPPROTO_UDP && l4_size >= UDP_HEADER_LEN) {
868 struct udp_header *uh = dp_packet_l4(packet);
871 uh->udp_csum = recalc_csum128(uh->udp_csum, addr, new_addr);
873 uh->udp_csum = htons(0xffff);
876 } else if (proto == IPPROTO_ICMPV6 &&
877 l4_size >= sizeof(struct icmp6_header)) {
878 struct icmp6_header *icmp = dp_packet_l4(packet);
880 icmp->icmp6_cksum = recalc_csum128(icmp->icmp6_cksum, addr, new_addr);
885 packet_set_ipv6_addr(struct dp_packet *packet, uint8_t proto,
886 ovs_16aligned_be32 addr[4], const ovs_be32 new_addr[4],
887 bool recalculate_csum)
889 if (recalculate_csum) {
890 packet_update_csum128(packet, proto, addr, new_addr);
892 memcpy(addr, new_addr, sizeof(ovs_be32[4]));
896 packet_set_ipv6_flow_label(ovs_16aligned_be32 *flow_label, ovs_be32 flow_key)
898 ovs_be32 old_label = get_16aligned_be32(flow_label);
899 ovs_be32 new_label = (old_label & htonl(~IPV6_LABEL_MASK)) | flow_key;
900 put_16aligned_be32(flow_label, new_label);
904 packet_set_ipv6_tc(ovs_16aligned_be32 *flow_label, uint8_t tc)
906 ovs_be32 old_label = get_16aligned_be32(flow_label);
907 ovs_be32 new_label = (old_label & htonl(0xF00FFFFF)) | htonl(tc << 20);
908 put_16aligned_be32(flow_label, new_label);
911 /* Modifies the IPv4 header fields of 'packet' to be consistent with 'src',
912 * 'dst', 'tos', and 'ttl'. Updates 'packet''s L4 checksums as appropriate.
913 * 'packet' must contain a valid IPv4 packet with correctly populated l[347]
916 packet_set_ipv4(struct dp_packet *packet, ovs_be32 src, ovs_be32 dst,
917 uint8_t tos, uint8_t ttl)
919 struct ip_header *nh = dp_packet_l3(packet);
921 if (get_16aligned_be32(&nh->ip_src) != src) {
922 packet_set_ipv4_addr(packet, &nh->ip_src, src);
925 if (get_16aligned_be32(&nh->ip_dst) != dst) {
926 packet_set_ipv4_addr(packet, &nh->ip_dst, dst);
929 if (nh->ip_tos != tos) {
930 uint8_t *field = &nh->ip_tos;
932 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t) *field),
933 htons((uint16_t) tos));
937 if (nh->ip_ttl != ttl) {
938 uint8_t *field = &nh->ip_ttl;
940 nh->ip_csum = recalc_csum16(nh->ip_csum, htons(*field << 8),
946 /* Modifies the IPv6 header fields of 'packet' to be consistent with 'src',
947 * 'dst', 'traffic class', and 'next hop'. Updates 'packet''s L4 checksums as
948 * appropriate. 'packet' must contain a valid IPv6 packet with correctly
949 * populated l[34] offsets. */
951 packet_set_ipv6(struct dp_packet *packet, uint8_t proto, const ovs_be32 src[4],
952 const ovs_be32 dst[4], uint8_t key_tc, ovs_be32 key_fl,
955 struct ovs_16aligned_ip6_hdr *nh = dp_packet_l3(packet);
957 if (memcmp(&nh->ip6_src, src, sizeof(ovs_be32[4]))) {
958 packet_set_ipv6_addr(packet, proto, nh->ip6_src.be32, src, true);
961 if (memcmp(&nh->ip6_dst, dst, sizeof(ovs_be32[4]))) {
962 packet_set_ipv6_addr(packet, proto, nh->ip6_dst.be32, dst,
963 !packet_rh_present(packet));
966 packet_set_ipv6_tc(&nh->ip6_flow, key_tc);
968 packet_set_ipv6_flow_label(&nh->ip6_flow, key_fl);
970 nh->ip6_hlim = key_hl;
974 packet_set_port(ovs_be16 *port, ovs_be16 new_port, ovs_be16 *csum)
976 if (*port != new_port) {
977 *csum = recalc_csum16(*csum, *port, new_port);
982 /* Sets the TCP source and destination port ('src' and 'dst' respectively) of
983 * the TCP header contained in 'packet'. 'packet' must be a valid TCP packet
984 * with its l4 offset properly populated. */
986 packet_set_tcp_port(struct dp_packet *packet, ovs_be16 src, ovs_be16 dst)
988 struct tcp_header *th = dp_packet_l4(packet);
990 packet_set_port(&th->tcp_src, src, &th->tcp_csum);
991 packet_set_port(&th->tcp_dst, dst, &th->tcp_csum);
994 /* Sets the UDP source and destination port ('src' and 'dst' respectively) of
995 * the UDP header contained in 'packet'. 'packet' must be a valid UDP packet
996 * with its l4 offset properly populated. */
998 packet_set_udp_port(struct dp_packet *packet, ovs_be16 src, ovs_be16 dst)
1000 struct udp_header *uh = dp_packet_l4(packet);
1003 packet_set_port(&uh->udp_src, src, &uh->udp_csum);
1004 packet_set_port(&uh->udp_dst, dst, &uh->udp_csum);
1006 if (!uh->udp_csum) {
1007 uh->udp_csum = htons(0xffff);
1015 /* Sets the SCTP source and destination port ('src' and 'dst' respectively) of
1016 * the SCTP header contained in 'packet'. 'packet' must be a valid SCTP packet
1017 * with its l4 offset properly populated. */
1019 packet_set_sctp_port(struct dp_packet *packet, ovs_be16 src, ovs_be16 dst)
1021 struct sctp_header *sh = dp_packet_l4(packet);
1022 ovs_be32 old_csum, old_correct_csum, new_csum;
1023 uint16_t tp_len = dp_packet_l4_size(packet);
1025 old_csum = get_16aligned_be32(&sh->sctp_csum);
1026 put_16aligned_be32(&sh->sctp_csum, 0);
1027 old_correct_csum = crc32c((void *)sh, tp_len);
1032 new_csum = crc32c((void *)sh, tp_len);
1033 put_16aligned_be32(&sh->sctp_csum, old_csum ^ old_correct_csum ^ new_csum);
1036 /* Sets the ICMP type and code of the ICMP header contained in 'packet'.
1037 * 'packet' must be a valid ICMP packet with its l4 offset properly
1040 packet_set_icmp(struct dp_packet *packet, uint8_t type, uint8_t code)
1042 struct icmp_header *ih = dp_packet_l4(packet);
1043 ovs_be16 orig_tc = htons(ih->icmp_type << 8 | ih->icmp_code);
1044 ovs_be16 new_tc = htons(type << 8 | code);
1046 if (orig_tc != new_tc) {
1047 ih->icmp_type = type;
1048 ih->icmp_code = code;
1050 ih->icmp_csum = recalc_csum16(ih->icmp_csum, orig_tc, new_tc);
1055 packet_set_nd(struct dp_packet *packet, const ovs_be32 target[4],
1056 const struct eth_addr sll, const struct eth_addr tll) {
1057 struct ovs_nd_msg *ns;
1058 struct ovs_nd_opt *nd_opt;
1059 int bytes_remain = dp_packet_l4_size(packet);
1061 if (OVS_UNLIKELY(bytes_remain < sizeof(*ns))) {
1065 ns = dp_packet_l4(packet);
1066 nd_opt = &ns->options[0];
1067 bytes_remain -= sizeof(*ns);
1069 if (memcmp(&ns->target, target, sizeof(ovs_be32[4]))) {
1070 packet_set_ipv6_addr(packet, IPPROTO_ICMPV6,
1075 while (bytes_remain >= ND_OPT_LEN && nd_opt->nd_opt_len != 0) {
1076 if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LINKADDR
1077 && nd_opt->nd_opt_len == 1) {
1078 if (!eth_addr_equals(nd_opt->nd_opt_mac, sll)) {
1079 ovs_be16 *csum = &(ns->icmph.icmp6_cksum);
1081 *csum = recalc_csum48(*csum, nd_opt->nd_opt_mac, sll);
1082 nd_opt->nd_opt_mac = sll;
1085 /* A packet can only contain one SLL or TLL option */
1087 } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LINKADDR
1088 && nd_opt->nd_opt_len == 1) {
1089 if (!eth_addr_equals(nd_opt->nd_opt_mac, tll)) {
1090 ovs_be16 *csum = &(ns->icmph.icmp6_cksum);
1092 *csum = recalc_csum48(*csum, nd_opt->nd_opt_mac, tll);
1093 nd_opt->nd_opt_mac = tll;
1096 /* A packet can only contain one SLL or TLL option */
1100 nd_opt += nd_opt->nd_opt_len;
1101 bytes_remain -= nd_opt->nd_opt_len * ND_OPT_LEN;
1106 packet_tcp_flag_to_string(uint32_t flag)
1138 /* Appends a string representation of the TCP flags value 'tcp_flags'
1139 * (e.g. from struct flow.tcp_flags or obtained via TCP_FLAGS) to 's', in the
1140 * format used by tcpdump. */
1142 packet_format_tcp_flags(struct ds *s, uint16_t tcp_flags)
1145 ds_put_cstr(s, "none");
1149 if (tcp_flags & TCP_SYN) {
1150 ds_put_char(s, 'S');
1152 if (tcp_flags & TCP_FIN) {
1153 ds_put_char(s, 'F');
1155 if (tcp_flags & TCP_PSH) {
1156 ds_put_char(s, 'P');
1158 if (tcp_flags & TCP_RST) {
1159 ds_put_char(s, 'R');
1161 if (tcp_flags & TCP_URG) {
1162 ds_put_char(s, 'U');
1164 if (tcp_flags & TCP_ACK) {
1165 ds_put_char(s, '.');
1167 if (tcp_flags & TCP_ECE) {
1168 ds_put_cstr(s, "E");
1170 if (tcp_flags & TCP_CWR) {
1171 ds_put_cstr(s, "C");
1173 if (tcp_flags & TCP_NS) {
1174 ds_put_cstr(s, "N");
1176 if (tcp_flags & 0x200) {
1177 ds_put_cstr(s, "[200]");
1179 if (tcp_flags & 0x400) {
1180 ds_put_cstr(s, "[400]");
1182 if (tcp_flags & 0x800) {
1183 ds_put_cstr(s, "[800]");
1187 #define ARP_PACKET_SIZE (2 + ETH_HEADER_LEN + VLAN_HEADER_LEN + \
1190 /* Clears 'b' and replaces its contents by an ARP frame with the specified
1191 * 'arp_op', 'arp_sha', 'arp_tha', 'arp_spa', and 'arp_tpa'. The outer
1192 * Ethernet frame is initialized with Ethernet source 'arp_sha' and destination
1193 * 'arp_tha', except that destination ff:ff:ff:ff:ff:ff is used instead if
1194 * 'broadcast' is true. */
1196 compose_arp(struct dp_packet *b, uint16_t arp_op,
1197 const struct eth_addr arp_sha, const struct eth_addr arp_tha,
1198 bool broadcast, ovs_be32 arp_spa, ovs_be32 arp_tpa)
1200 struct eth_header *eth;
1201 struct arp_eth_header *arp;
1204 dp_packet_prealloc_tailroom(b, ARP_PACKET_SIZE);
1205 dp_packet_reserve(b, 2 + VLAN_HEADER_LEN);
1207 eth = dp_packet_put_uninit(b, sizeof *eth);
1208 eth->eth_dst = broadcast ? eth_addr_broadcast : arp_tha;
1209 eth->eth_src = arp_sha;
1210 eth->eth_type = htons(ETH_TYPE_ARP);
1212 arp = dp_packet_put_uninit(b, sizeof *arp);
1213 arp->ar_hrd = htons(ARP_HRD_ETHERNET);
1214 arp->ar_pro = htons(ARP_PRO_IP);
1215 arp->ar_hln = sizeof arp->ar_sha;
1216 arp->ar_pln = sizeof arp->ar_spa;
1217 arp->ar_op = htons(arp_op);
1218 arp->ar_sha = arp_sha;
1219 arp->ar_tha = arp_tha;
1221 put_16aligned_be32(&arp->ar_spa, arp_spa);
1222 put_16aligned_be32(&arp->ar_tpa, arp_tpa);
1224 dp_packet_reset_offsets(b);
1225 dp_packet_set_l3(b, arp);
1229 compose_nd(struct dp_packet *b, const struct eth_addr eth_src,
1230 struct in6_addr * ipv6_src, struct in6_addr * ipv6_dst)
1232 struct in6_addr sn_addr;
1233 struct eth_addr eth_dst;
1234 struct ovs_nd_msg *ns;
1235 struct ovs_nd_opt *nd_opt;
1237 in6_addr_solicited_node(&sn_addr, ipv6_dst);
1238 ipv6_multicast_to_ethernet(ð_dst, &sn_addr);
1240 eth_compose(b, eth_dst, eth_src, ETH_TYPE_IPV6,
1241 IPV6_HEADER_LEN + ICMP6_HEADER_LEN + ND_OPT_LEN);
1242 packet_set_ipv6(b, IPPROTO_ICMPV6,
1243 ALIGNED_CAST(ovs_be32 *, ipv6_src->s6_addr),
1244 ALIGNED_CAST(ovs_be32 *, sn_addr.s6_addr),
1247 ns = dp_packet_l4(b);
1248 nd_opt = &ns->options[0];
1250 ns->icmph.icmp6_type = ND_NEIGHBOR_SOLICIT;
1251 ns->icmph.icmp6_code = 0;
1253 nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR;
1254 packet_set_nd(b, ALIGNED_CAST(ovs_be32 *, ipv6_dst->s6_addr),
1255 eth_src, eth_addr_zero);
1259 packet_csum_pseudoheader(const struct ip_header *ip)
1261 uint32_t partial = 0;
1263 partial = csum_add32(partial, get_16aligned_be32(&ip->ip_src));
1264 partial = csum_add32(partial, get_16aligned_be32(&ip->ip_dst));
1265 partial = csum_add16(partial, htons(ip->ip_proto));
1266 partial = csum_add16(partial, htons(ntohs(ip->ip_tot_len) -
1267 IP_IHL(ip->ip_ihl_ver) * 4));
1274 packet_csum_pseudoheader6(const struct ovs_16aligned_ip6_hdr *ip6)
1276 uint32_t partial = 0;
1278 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_src.be32[0])));
1279 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_src.be32[1])));
1280 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_src.be32[2])));
1281 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_src.be32[3])));
1282 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_dst.be32[0])));
1283 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_dst.be32[1])));
1284 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_dst.be32[2])));
1285 partial = csum_add32(partial, get_16aligned_be32(&(ip6->ip6_dst.be32[3])));
1287 partial = csum_add16(partial, 0);
1288 partial = csum_add16(partial, ip6->ip6_plen);
1289 partial = csum_add16(partial, 0);
1290 partial = csum_add16(partial, ip6->ip6_nxt);