]> git.proxmox.com Git - mirror_ovs.git/blob - lib/packets.c
Always treat datapath ports as 32 bits.
[mirror_ovs.git] / lib / packets.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17 #include <config.h>
18 #include "packets.h"
19 #include <assert.h>
20 #include <arpa/inet.h>
21 #include <sys/socket.h>
22 #include <netinet/in.h>
23 #include <stdlib.h>
24 #include "byte-order.h"
25 #include "csum.h"
26 #include "flow.h"
27 #include "hmap.h"
28 #include "dynamic-string.h"
29 #include "ofpbuf.h"
30
31 const struct in6_addr in6addr_exact = IN6ADDR_EXACT_INIT;
32
33 /* Parses 's' as a 16-digit hexadecimal number representing a datapath ID. On
34 * success stores the dpid into '*dpidp' and returns true, on failure stores 0
35 * into '*dpidp' and returns false.
36 *
37 * Rejects an all-zeros dpid as invalid. */
38 bool
39 dpid_from_string(const char *s, uint64_t *dpidp)
40 {
41 *dpidp = (strlen(s) == 16 && strspn(s, "0123456789abcdefABCDEF") == 16
42 ? strtoull(s, NULL, 16)
43 : 0);
44 return *dpidp != 0;
45 }
46
47 /* Returns true if 'ea' is a reserved address, that a bridge must never
48 * forward, false otherwise.
49 *
50 * If you change this function's behavior, please update corresponding
51 * documentation in vswitch.xml at the same time. */
52 bool
53 eth_addr_is_reserved(const uint8_t ea[ETH_ADDR_LEN])
54 {
55 struct eth_addr_node {
56 struct hmap_node hmap_node;
57 uint64_t ea64;
58 };
59
60 static struct eth_addr_node nodes[] = {
61 /* STP, IEEE pause frames, and other reserved protocols. */
62 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000000ULL },
63 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000001ULL },
64 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000002ULL },
65 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000003ULL },
66 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000004ULL },
67 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000005ULL },
68 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000006ULL },
69 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000007ULL },
70 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000008ULL },
71 { HMAP_NODE_NULL_INITIALIZER, 0x0108c2000009ULL },
72 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000aULL },
73 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000bULL },
74 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000cULL },
75 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000dULL },
76 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000eULL },
77 { HMAP_NODE_NULL_INITIALIZER, 0x0108c200000fULL },
78
79 /* Extreme protocols. */
80 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000000ULL }, /* EDP. */
81 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000004ULL }, /* EAPS. */
82 { HMAP_NODE_NULL_INITIALIZER, 0x00e02b000006ULL }, /* EAPS. */
83
84 /* Cisco protocols. */
85 { HMAP_NODE_NULL_INITIALIZER, 0x01000c000000ULL }, /* ISL. */
86 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccccULL }, /* PAgP, UDLD, CDP,
87 * DTP, VTP. */
88 { HMAP_NODE_NULL_INITIALIZER, 0x01000ccccccdULL }, /* PVST+. */
89 { HMAP_NODE_NULL_INITIALIZER, 0x01000ccdcdcdULL }, /* STP Uplink Fast,
90 * FlexLink. */
91
92 /* Cisco CFM. */
93 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc0ULL },
94 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc1ULL },
95 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc2ULL },
96 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc3ULL },
97 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc4ULL },
98 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc5ULL },
99 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc6ULL },
100 { HMAP_NODE_NULL_INITIALIZER, 0x01000cccccc7ULL },
101 };
102
103 static struct hmap addrs = HMAP_INITIALIZER(&addrs);
104 struct eth_addr_node *node;
105 uint64_t ea64;
106
107 if (hmap_is_empty(&addrs)) {
108 for (node = nodes; node < &nodes[ARRAY_SIZE(nodes)]; node++) {
109 hmap_insert(&addrs, &node->hmap_node,
110 hash_2words(node->ea64, node->ea64 >> 32));
111 }
112 }
113
114 ea64 = eth_addr_to_uint64(ea);
115 HMAP_FOR_EACH_IN_BUCKET (node, hmap_node, hash_2words(ea64, ea64 >> 32),
116 &addrs) {
117 if (node->ea64 == ea64) {
118 return true;
119 }
120 }
121 return false;
122 }
123
124 bool
125 eth_addr_from_string(const char *s, uint8_t ea[ETH_ADDR_LEN])
126 {
127 if (sscanf(s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(ea))
128 == ETH_ADDR_SCAN_COUNT) {
129 return true;
130 } else {
131 memset(ea, 0, ETH_ADDR_LEN);
132 return false;
133 }
134 }
135
136 /* Fills 'b' with a Reverse ARP packet with Ethernet source address 'eth_src'.
137 * This function is used by Open vSwitch to compose packets in cases where
138 * context is important but content doesn't (or shouldn't) matter.
139 *
140 * The returned packet has enough headroom to insert an 802.1Q VLAN header if
141 * desired. */
142 void
143 compose_rarp(struct ofpbuf *b, const uint8_t eth_src[ETH_ADDR_LEN])
144 {
145 struct eth_header *eth;
146 struct arp_eth_header *arp;
147
148 ofpbuf_clear(b);
149 ofpbuf_prealloc_tailroom(b, ETH_HEADER_LEN + VLAN_HEADER_LEN
150 + ARP_ETH_HEADER_LEN);
151 ofpbuf_reserve(b, VLAN_HEADER_LEN);
152 eth = ofpbuf_put_uninit(b, sizeof *eth);
153 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
154 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
155 eth->eth_type = htons(ETH_TYPE_RARP);
156
157 arp = ofpbuf_put_uninit(b, sizeof *arp);
158 arp->ar_hrd = htons(ARP_HRD_ETHERNET);
159 arp->ar_pro = htons(ARP_PRO_IP);
160 arp->ar_hln = sizeof arp->ar_sha;
161 arp->ar_pln = sizeof arp->ar_spa;
162 arp->ar_op = htons(ARP_OP_RARP);
163 memcpy(arp->ar_sha, eth_src, ETH_ADDR_LEN);
164 arp->ar_spa = htonl(0);
165 memcpy(arp->ar_tha, eth_src, ETH_ADDR_LEN);
166 arp->ar_tpa = htonl(0);
167 }
168
169 /* Insert VLAN header according to given TCI. Packet passed must be Ethernet
170 * packet. Ignores the CFI bit of 'tci' using 0 instead.
171 *
172 * Also sets 'packet->l2' to point to the new Ethernet header. */
173 void
174 eth_push_vlan(struct ofpbuf *packet, ovs_be16 tci)
175 {
176 struct eth_header *eh = packet->data;
177 struct vlan_eth_header *veh;
178
179 /* Insert new 802.1Q header. */
180 struct vlan_eth_header tmp;
181 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
182 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
183 tmp.veth_type = htons(ETH_TYPE_VLAN);
184 tmp.veth_tci = tci & htons(~VLAN_CFI);
185 tmp.veth_next_type = eh->eth_type;
186
187 veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN);
188 memcpy(veh, &tmp, sizeof tmp);
189
190 packet->l2 = packet->data;
191 }
192
193 /* Removes outermost VLAN header (if any is present) from 'packet'.
194 *
195 * 'packet->l2' must initially point to 'packet''s Ethernet header. */
196 void
197 eth_pop_vlan(struct ofpbuf *packet)
198 {
199 struct vlan_eth_header *veh = packet->l2;
200 if (packet->size >= sizeof *veh
201 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
202 struct eth_header tmp;
203
204 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
205 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
206 tmp.eth_type = veh->veth_next_type;
207
208 ofpbuf_pull(packet, VLAN_HEADER_LEN);
209 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
210 memcpy(packet->data, &tmp, sizeof tmp);
211 }
212 }
213
214 /* Converts hex digits in 'hex' to an Ethernet packet in '*packetp'. The
215 * caller must free '*packetp'. On success, returns NULL. On failure, returns
216 * an error message and stores NULL in '*packetp'. */
217 const char *
218 eth_from_hex(const char *hex, struct ofpbuf **packetp)
219 {
220 struct ofpbuf *packet;
221
222 packet = *packetp = ofpbuf_new(strlen(hex) / 2);
223
224 if (ofpbuf_put_hex(packet, hex, NULL)[0] != '\0') {
225 ofpbuf_delete(packet);
226 *packetp = NULL;
227 return "Trailing garbage in packet data";
228 }
229
230 if (packet->size < ETH_HEADER_LEN) {
231 ofpbuf_delete(packet);
232 *packetp = NULL;
233 return "Packet data too short for Ethernet";
234 }
235
236 return NULL;
237 }
238
239 void
240 eth_format_masked(const uint8_t eth[ETH_ADDR_LEN],
241 const uint8_t mask[ETH_ADDR_LEN], struct ds *s)
242 {
243 ds_put_format(s, ETH_ADDR_FMT, ETH_ADDR_ARGS(eth));
244 if (mask && !eth_mask_is_exact(mask)) {
245 ds_put_format(s, "/"ETH_ADDR_FMT, ETH_ADDR_ARGS(mask));
246 }
247 }
248
249 void
250 eth_addr_bitand(const uint8_t src[ETH_ADDR_LEN],
251 const uint8_t mask[ETH_ADDR_LEN],
252 uint8_t dst[ETH_ADDR_LEN])
253 {
254 int i;
255
256 for (i = 0; i < ETH_ADDR_LEN; i++) {
257 dst[i] = src[i] & mask[i];
258 }
259 }
260
261 /* Given the IP netmask 'netmask', returns the number of bits of the IP address
262 * that it specifies, that is, the number of 1-bits in 'netmask'.
263 *
264 * If 'netmask' is not a CIDR netmask (see ip_is_cidr()), the return value will
265 * still be in the valid range but isn't otherwise meaningful. */
266 int
267 ip_count_cidr_bits(ovs_be32 netmask)
268 {
269 return 32 - ctz(ntohl(netmask));
270 }
271
272 void
273 ip_format_masked(ovs_be32 ip, ovs_be32 mask, struct ds *s)
274 {
275 ds_put_format(s, IP_FMT, IP_ARGS(&ip));
276 if (mask != htonl(UINT32_MAX)) {
277 if (ip_is_cidr(mask)) {
278 ds_put_format(s, "/%d", ip_count_cidr_bits(mask));
279 } else {
280 ds_put_format(s, "/"IP_FMT, IP_ARGS(&mask));
281 }
282 }
283 }
284
285
286 /* Stores the string representation of the IPv6 address 'addr' into the
287 * character array 'addr_str', which must be at least INET6_ADDRSTRLEN
288 * bytes long. */
289 void
290 format_ipv6_addr(char *addr_str, const struct in6_addr *addr)
291 {
292 inet_ntop(AF_INET6, addr, addr_str, INET6_ADDRSTRLEN);
293 }
294
295 void
296 print_ipv6_addr(struct ds *string, const struct in6_addr *addr)
297 {
298 char *dst;
299
300 ds_reserve(string, string->length + INET6_ADDRSTRLEN);
301
302 dst = string->string + string->length;
303 format_ipv6_addr(dst, addr);
304 string->length += strlen(dst);
305 }
306
307 void
308 print_ipv6_masked(struct ds *s, const struct in6_addr *addr,
309 const struct in6_addr *mask)
310 {
311 print_ipv6_addr(s, addr);
312 if (mask && !ipv6_mask_is_exact(mask)) {
313 if (ipv6_is_cidr(mask)) {
314 int cidr_bits = ipv6_count_cidr_bits(mask);
315 ds_put_format(s, "/%d", cidr_bits);
316 } else {
317 ds_put_char(s, '/');
318 print_ipv6_addr(s, mask);
319 }
320 }
321 }
322
323 struct in6_addr ipv6_addr_bitand(const struct in6_addr *a,
324 const struct in6_addr *b)
325 {
326 int i;
327 struct in6_addr dst;
328
329 #ifdef s6_addr32
330 for (i=0; i<4; i++) {
331 dst.s6_addr32[i] = a->s6_addr32[i] & b->s6_addr32[i];
332 }
333 #else
334 for (i=0; i<16; i++) {
335 dst.s6_addr[i] = a->s6_addr[i] & b->s6_addr[i];
336 }
337 #endif
338
339 return dst;
340 }
341
342 /* Returns an in6_addr consisting of 'mask' high-order 1-bits and 128-N
343 * low-order 0-bits. */
344 struct in6_addr
345 ipv6_create_mask(int mask)
346 {
347 struct in6_addr netmask;
348 uint8_t *netmaskp = &netmask.s6_addr[0];
349
350 memset(&netmask, 0, sizeof netmask);
351 while (mask > 8) {
352 *netmaskp = 0xff;
353 netmaskp++;
354 mask -= 8;
355 }
356
357 if (mask) {
358 *netmaskp = 0xff << (8 - mask);
359 }
360
361 return netmask;
362 }
363
364 /* Given the IPv6 netmask 'netmask', returns the number of bits of the IPv6
365 * address that it specifies, that is, the number of 1-bits in 'netmask'.
366 * 'netmask' must be a CIDR netmask (see ipv6_is_cidr()).
367 *
368 * If 'netmask' is not a CIDR netmask (see ipv6_is_cidr()), the return value
369 * will still be in the valid range but isn't otherwise meaningful. */
370 int
371 ipv6_count_cidr_bits(const struct in6_addr *netmask)
372 {
373 int i;
374 int count = 0;
375 const uint8_t *netmaskp = &netmask->s6_addr[0];
376
377 for (i=0; i<16; i++) {
378 if (netmaskp[i] == 0xff) {
379 count += 8;
380 } else {
381 uint8_t nm;
382
383 for(nm = netmaskp[i]; nm; nm <<= 1) {
384 count++;
385 }
386 break;
387 }
388
389 }
390
391 return count;
392 }
393
394 /* Returns true if 'netmask' is a CIDR netmask, that is, if it consists of N
395 * high-order 1-bits and 128-N low-order 0-bits. */
396 bool
397 ipv6_is_cidr(const struct in6_addr *netmask)
398 {
399 const uint8_t *netmaskp = &netmask->s6_addr[0];
400 int i;
401
402 for (i=0; i<16; i++) {
403 if (netmaskp[i] != 0xff) {
404 uint8_t x = ~netmaskp[i];
405 if (x & (x + 1)) {
406 return false;
407 }
408 while (++i < 16) {
409 if (netmaskp[i]) {
410 return false;
411 }
412 }
413 }
414 }
415
416 return true;
417 }
418
419 /* Populates 'b' with an Ethernet II packet headed with the given 'eth_dst',
420 * 'eth_src' and 'eth_type' parameters. A payload of 'size' bytes is allocated
421 * in 'b' and returned. This payload may be populated with appropriate
422 * information by the caller. Sets 'b''s 'l2' and 'l3' pointers to the
423 * Ethernet header and payload respectively.
424 *
425 * The returned packet has enough headroom to insert an 802.1Q VLAN header if
426 * desired. */
427 void *
428 eth_compose(struct ofpbuf *b, const uint8_t eth_dst[ETH_ADDR_LEN],
429 const uint8_t eth_src[ETH_ADDR_LEN], uint16_t eth_type,
430 size_t size)
431 {
432 void *data;
433 struct eth_header *eth;
434
435 ofpbuf_clear(b);
436
437 ofpbuf_prealloc_tailroom(b, ETH_HEADER_LEN + VLAN_HEADER_LEN + size);
438 ofpbuf_reserve(b, VLAN_HEADER_LEN);
439 eth = ofpbuf_put_uninit(b, ETH_HEADER_LEN);
440 data = ofpbuf_put_uninit(b, size);
441
442 memcpy(eth->eth_dst, eth_dst, ETH_ADDR_LEN);
443 memcpy(eth->eth_src, eth_src, ETH_ADDR_LEN);
444 eth->eth_type = htons(eth_type);
445
446 b->l2 = eth;
447 b->l3 = data;
448
449 return data;
450 }
451
452 static void
453 packet_set_ipv4_addr(struct ofpbuf *packet, ovs_be32 *addr, ovs_be32 new_addr)
454 {
455 struct ip_header *nh = packet->l3;
456
457 if (nh->ip_proto == IPPROTO_TCP && packet->l7) {
458 struct tcp_header *th = packet->l4;
459
460 th->tcp_csum = recalc_csum32(th->tcp_csum, *addr, new_addr);
461 } else if (nh->ip_proto == IPPROTO_UDP && packet->l7) {
462 struct udp_header *uh = packet->l4;
463
464 if (uh->udp_csum) {
465 uh->udp_csum = recalc_csum32(uh->udp_csum, *addr, new_addr);
466 if (!uh->udp_csum) {
467 uh->udp_csum = htons(0xffff);
468 }
469 }
470 }
471 nh->ip_csum = recalc_csum32(nh->ip_csum, *addr, new_addr);
472 *addr = new_addr;
473 }
474
475 /* Modifies the IPv4 header fields of 'packet' to be consistent with 'src',
476 * 'dst', 'tos', and 'ttl'. Updates 'packet''s L4 checksums as appropriate.
477 * 'packet' must contain a valid IPv4 packet with correctly populated l[347]
478 * markers. */
479 void
480 packet_set_ipv4(struct ofpbuf *packet, ovs_be32 src, ovs_be32 dst,
481 uint8_t tos, uint8_t ttl)
482 {
483 struct ip_header *nh = packet->l3;
484
485 if (nh->ip_src != src) {
486 packet_set_ipv4_addr(packet, &nh->ip_src, src);
487 }
488
489 if (nh->ip_dst != dst) {
490 packet_set_ipv4_addr(packet, &nh->ip_dst, dst);
491 }
492
493 if (nh->ip_tos != tos) {
494 uint8_t *field = &nh->ip_tos;
495
496 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t) *field),
497 htons((uint16_t) tos));
498 *field = tos;
499 }
500
501 if (nh->ip_ttl != ttl) {
502 uint8_t *field = &nh->ip_ttl;
503
504 nh->ip_csum = recalc_csum16(nh->ip_csum, htons(*field << 8),
505 htons(ttl << 8));
506 *field = ttl;
507 }
508 }
509
510 static void
511 packet_set_port(ovs_be16 *port, ovs_be16 new_port, ovs_be16 *csum)
512 {
513 if (*port != new_port) {
514 *csum = recalc_csum16(*csum, *port, new_port);
515 *port = new_port;
516 }
517 }
518
519 /* Sets the TCP source and destination port ('src' and 'dst' respectively) of
520 * the TCP header contained in 'packet'. 'packet' must be a valid TCP packet
521 * with its l4 marker properly populated. */
522 void
523 packet_set_tcp_port(struct ofpbuf *packet, ovs_be16 src, ovs_be16 dst)
524 {
525 struct tcp_header *th = packet->l4;
526
527 packet_set_port(&th->tcp_src, src, &th->tcp_csum);
528 packet_set_port(&th->tcp_dst, dst, &th->tcp_csum);
529 }
530
531 /* Sets the UDP source and destination port ('src' and 'dst' respectively) of
532 * the UDP header contained in 'packet'. 'packet' must be a valid UDP packet
533 * with its l4 marker properly populated. */
534 void
535 packet_set_udp_port(struct ofpbuf *packet, ovs_be16 src, ovs_be16 dst)
536 {
537 struct udp_header *uh = packet->l4;
538
539 if (uh->udp_csum) {
540 packet_set_port(&uh->udp_src, src, &uh->udp_csum);
541 packet_set_port(&uh->udp_dst, dst, &uh->udp_csum);
542
543 if (!uh->udp_csum) {
544 uh->udp_csum = htons(0xffff);
545 }
546 } else {
547 uh->udp_src = src;
548 uh->udp_dst = dst;
549 }
550 }
551
552 /* If 'packet' is a TCP packet, returns the TCP flags. Otherwise, returns 0.
553 *
554 * 'flow' must be the flow corresponding to 'packet' and 'packet''s header
555 * pointers must be properly initialized (e.g. with flow_extract()). */
556 uint8_t
557 packet_get_tcp_flags(const struct ofpbuf *packet, const struct flow *flow)
558 {
559 if ((flow->dl_type == htons(ETH_TYPE_IP) ||
560 flow->dl_type == htons(ETH_TYPE_IPV6)) &&
561 flow->nw_proto == IPPROTO_TCP && packet->l7) {
562 const struct tcp_header *tcp = packet->l4;
563 return TCP_FLAGS(tcp->tcp_ctl);
564 } else {
565 return 0;
566 }
567 }
568
569 /* Appends a string representation of the TCP flags value 'tcp_flags'
570 * (e.g. obtained via packet_get_tcp_flags() or TCP_FLAGS) to 's', in the
571 * format used by tcpdump. */
572 void
573 packet_format_tcp_flags(struct ds *s, uint8_t tcp_flags)
574 {
575 if (!tcp_flags) {
576 ds_put_cstr(s, "none");
577 return;
578 }
579
580 if (tcp_flags & TCP_SYN) {
581 ds_put_char(s, 'S');
582 }
583 if (tcp_flags & TCP_FIN) {
584 ds_put_char(s, 'F');
585 }
586 if (tcp_flags & TCP_PSH) {
587 ds_put_char(s, 'P');
588 }
589 if (tcp_flags & TCP_RST) {
590 ds_put_char(s, 'R');
591 }
592 if (tcp_flags & TCP_URG) {
593 ds_put_char(s, 'U');
594 }
595 if (tcp_flags & TCP_ACK) {
596 ds_put_char(s, '.');
597 }
598 if (tcp_flags & 0x40) {
599 ds_put_cstr(s, "[40]");
600 }
601 if (tcp_flags & 0x80) {
602 ds_put_cstr(s, "[80]");
603 }
604 }