1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 1982, 1986, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright(c) 2010-2014 Intel Corporation.
5 * Copyright(c) 2014 6WIND S.A.
6 * All rights reserved.
7 */
8
9 #ifndef _RTE_IP_H_
10 #define _RTE_IP_H_
11
12 /**
13 * @file
14 *
15 * IP-related defines
16 */
17
18 #include <stdint.h>
19 #include <sys/types.h>
20 #include <netinet/in.h>
21 #include <netinet/ip.h>
22
23 #include <rte_byteorder.h>
24 #include <rte_mbuf.h>
25
26 #ifdef __cplusplus
27 extern "C" {
28 #endif
29
30 /**
31 * IPv4 Header
32 */
33 struct rte_ipv4_hdr {
34 uint8_t version_ihl; /**< version and header length */
35 uint8_t type_of_service; /**< type of service */
36 rte_be16_t total_length; /**< length of packet */
37 rte_be16_t packet_id; /**< packet ID */
38 rte_be16_t fragment_offset; /**< fragmentation offset */
39 uint8_t time_to_live; /**< time to live */
40 uint8_t next_proto_id; /**< protocol ID */
41 rte_be16_t hdr_checksum; /**< header checksum */
42 rte_be32_t src_addr; /**< source address */
43 rte_be32_t dst_addr; /**< destination address */
44 } __rte_packed;
45
46 /** Create IPv4 address */
47 #define RTE_IPV4(a, b, c, d) ((uint32_t)(((a) & 0xff) << 24) | \
48 (((b) & 0xff) << 16) | \
49 (((c) & 0xff) << 8) | \
50 ((d) & 0xff))
51
52 /** Maximal IPv4 packet length (including a header) */
53 #define RTE_IPV4_MAX_PKT_LEN 65535
54
55 /** Internet header length mask for version_ihl field */
56 #define RTE_IPV4_HDR_IHL_MASK (0x0f)
57 /**
58 * Internet header length field multiplier (IHL field specifies overall header
59 * length in number of 4-byte words)
60 */
61 #define RTE_IPV4_IHL_MULTIPLIER (4)
62
63 /* Type of Service fields */
64 #define RTE_IPV4_HDR_DSCP_MASK (0xfc)
65 #define RTE_IPV4_HDR_ECN_MASK (0x03)
66 #define RTE_IPV4_HDR_ECN_CE RTE_IPV4_HDR_ECN_MASK
67
68 /* Fragment Offset * Flags. */
69 #define RTE_IPV4_HDR_DF_SHIFT 14
70 #define RTE_IPV4_HDR_MF_SHIFT 13
71 #define RTE_IPV4_HDR_FO_SHIFT 3
72
73 #define RTE_IPV4_HDR_DF_FLAG (1 << RTE_IPV4_HDR_DF_SHIFT)
74 #define RTE_IPV4_HDR_MF_FLAG (1 << RTE_IPV4_HDR_MF_SHIFT)
75
76 #define RTE_IPV4_HDR_OFFSET_MASK ((1 << RTE_IPV4_HDR_MF_SHIFT) - 1)
77
78 #define RTE_IPV4_HDR_OFFSET_UNITS 8
79
80 /*
81 * IPv4 address types
82 */
83 #define RTE_IPV4_ANY ((uint32_t)0x00000000) /**< 0.0.0.0 */
84 #define RTE_IPV4_LOOPBACK ((uint32_t)0x7f000001) /**< 127.0.0.1 */
85 #define RTE_IPV4_BROADCAST ((uint32_t)0xe0000000) /**< 224.0.0.0 */
86 #define RTE_IPV4_ALLHOSTS_GROUP ((uint32_t)0xe0000001) /**< 224.0.0.1 */
87 #define RTE_IPV4_ALLRTRS_GROUP ((uint32_t)0xe0000002) /**< 224.0.0.2 */
88 #define RTE_IPV4_MAX_LOCAL_GROUP ((uint32_t)0xe00000ff) /**< 224.0.0.255 */
89
90 /*
91 * IPv4 Multicast-related macros
92 */
93 #define RTE_IPV4_MIN_MCAST \
94 RTE_IPV4(224, 0, 0, 0) /**< Minimal IPv4-multicast address */
95 #define RTE_IPV4_MAX_MCAST \
96 RTE_IPV4(239, 255, 255, 255) /**< Maximum IPv4 multicast address */
97
98 #define RTE_IS_IPV4_MCAST(x) \
99 ((x) >= RTE_IPV4_MIN_MCAST && (x) <= RTE_IPV4_MAX_MCAST)
100 /**< check if IPv4 address is multicast */
101
102 /* IPv4 default fields values */
103 #define RTE_IPV4_MIN_IHL (0x5)
104 #define RTE_IPV4_VHL_DEF ((IPVERSION << 4) | RTE_IPV4_MIN_IHL)
105
106 /**
107 * Get the length of an IPv4 header.
108 *
109 * @param ipv4_hdr
110 * Pointer to the IPv4 header.
111 * @return
112 * The length of the IPv4 header (with options if present) in bytes.
113 */
114 static inline uint8_t
rte_ipv4_hdr_len(const struct rte_ipv4_hdr * ipv4_hdr)115 rte_ipv4_hdr_len(const struct rte_ipv4_hdr *ipv4_hdr)
116 {
117 return (uint8_t)((ipv4_hdr->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
118 RTE_IPV4_IHL_MULTIPLIER);
119 }
120
121 /**
122 * @internal Calculate a sum of all words in the buffer.
123 * Helper routine for the rte_raw_cksum().
124 *
125 * @param buf
126 * Pointer to the buffer.
127 * @param len
128 * Length of the buffer.
129 * @param sum
130 * Initial value of the sum.
131 * @return
132 * sum += Sum of all words in the buffer.
133 */
134 static inline uint32_t
__rte_raw_cksum(const void * buf,size_t len,uint32_t sum)135 __rte_raw_cksum(const void *buf, size_t len, uint32_t sum)
136 {
137 /* workaround gcc strict-aliasing warning */
138 uintptr_t ptr = (uintptr_t)buf;
139 typedef uint16_t __attribute__((__may_alias__)) u16_p;
140 const u16_p *u16_buf = (const u16_p *)ptr;
141
142 while (len >= (sizeof(*u16_buf) * 4)) {
143 sum += u16_buf[0];
144 sum += u16_buf[1];
145 sum += u16_buf[2];
146 sum += u16_buf[3];
147 len -= sizeof(*u16_buf) * 4;
148 u16_buf += 4;
149 }
150 while (len >= sizeof(*u16_buf)) {
151 sum += *u16_buf;
152 len -= sizeof(*u16_buf);
153 u16_buf += 1;
154 }
155
156 /* if length is in odd bytes */
157 if (len == 1) {
158 uint16_t left = 0;
159 *(uint8_t *)&left = *(const uint8_t *)u16_buf;
160 sum += left;
161 }
162
163 return sum;
164 }
165
166 /**
167 * @internal Reduce a sum to the non-complemented checksum.
168 * Helper routine for the rte_raw_cksum().
169 *
170 * @param sum
171 * Value of the sum.
172 * @return
173 * The non-complemented checksum.
174 */
175 static inline uint16_t
__rte_raw_cksum_reduce(uint32_t sum)176 __rte_raw_cksum_reduce(uint32_t sum)
177 {
178 sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
179 sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
180 return (uint16_t)sum;
181 }
182
183 /**
184 * Process the non-complemented checksum of a buffer.
185 *
186 * @param buf
187 * Pointer to the buffer.
188 * @param len
189 * Length of the buffer.
190 * @return
191 * The non-complemented checksum.
192 */
193 static inline uint16_t
rte_raw_cksum(const void * buf,size_t len)194 rte_raw_cksum(const void *buf, size_t len)
195 {
196 uint32_t sum;
197
198 sum = __rte_raw_cksum(buf, len, 0);
199 return __rte_raw_cksum_reduce(sum);
200 }
201
202 /**
203 * Compute the raw (non complemented) checksum of a packet.
204 *
205 * @param m
206 * The pointer to the mbuf.
207 * @param off
208 * The offset in bytes to start the checksum.
209 * @param len
210 * The length in bytes of the data to checksum.
211 * @param cksum
212 * A pointer to the checksum, filled on success.
213 * @return
214 * 0 on success, -1 on error (bad length or offset).
215 */
216 static inline int
rte_raw_cksum_mbuf(const struct rte_mbuf * m,uint32_t off,uint32_t len,uint16_t * cksum)217 rte_raw_cksum_mbuf(const struct rte_mbuf *m, uint32_t off, uint32_t len,
218 uint16_t *cksum)
219 {
220 const struct rte_mbuf *seg;
221 const char *buf;
222 uint32_t sum, tmp;
223 uint32_t seglen, done;
224
225 /* easy case: all data in the first segment */
226 if (off + len <= rte_pktmbuf_data_len(m)) {
227 *cksum = rte_raw_cksum(rte_pktmbuf_mtod_offset(m,
228 const char *, off), len);
229 return 0;
230 }
231
232 if (unlikely(off + len > rte_pktmbuf_pkt_len(m)))
233 return -1; /* invalid params, return a dummy value */
234
235 /* else browse the segment to find offset */
236 seglen = 0;
237 for (seg = m; seg != NULL; seg = seg->next) {
238 seglen = rte_pktmbuf_data_len(seg);
239 if (off < seglen)
240 break;
241 off -= seglen;
242 }
243 RTE_ASSERT(seg != NULL);
244 if (seg == NULL)
245 return -1;
246 seglen -= off;
247 buf = rte_pktmbuf_mtod_offset(seg, const char *, off);
248 if (seglen >= len) {
249 /* all in one segment */
250 *cksum = rte_raw_cksum(buf, len);
251 return 0;
252 }
253
254 /* hard case: process checksum of several segments */
255 sum = 0;
256 done = 0;
257 for (;;) {
258 tmp = __rte_raw_cksum(buf, seglen, 0);
259 if (done & 1)
260 tmp = rte_bswap16((uint16_t)tmp);
261 sum += tmp;
262 done += seglen;
263 if (done == len)
264 break;
265 seg = seg->next;
266 buf = rte_pktmbuf_mtod(seg, const char *);
267 seglen = rte_pktmbuf_data_len(seg);
268 if (seglen > len - done)
269 seglen = len - done;
270 }
271
272 *cksum = __rte_raw_cksum_reduce(sum);
273 return 0;
274 }
275
276 /**
277 * Process the IPv4 checksum of an IPv4 header.
278 *
279 * The checksum field must be set to 0 by the caller.
280 *
281 * @param ipv4_hdr
282 * The pointer to the contiguous IPv4 header.
283 * @return
284 * The complemented checksum to set in the IP packet.
285 */
286 static inline uint16_t
rte_ipv4_cksum(const struct rte_ipv4_hdr * ipv4_hdr)287 rte_ipv4_cksum(const struct rte_ipv4_hdr *ipv4_hdr)
288 {
289 uint16_t cksum;
290 cksum = rte_raw_cksum(ipv4_hdr, rte_ipv4_hdr_len(ipv4_hdr));
291 return (uint16_t)~cksum;
292 }
293
294 /**
295 * Process the pseudo-header checksum of an IPv4 header.
296 *
297 * The checksum field must be set to 0 by the caller.
298 *
299 * Depending on the ol_flags, the pseudo-header checksum expected by the
300 * drivers is not the same. For instance, when TSO is enabled, the IP
301 * payload length must not be included in the packet.
302 *
303 * When ol_flags is 0, it computes the standard pseudo-header checksum.
304 *
305 * @param ipv4_hdr
306 * The pointer to the contiguous IPv4 header.
307 * @param ol_flags
308 * The ol_flags of the associated mbuf.
309 * @return
310 * The non-complemented checksum to set in the L4 header.
311 */
312 static inline uint16_t
rte_ipv4_phdr_cksum(const struct rte_ipv4_hdr * ipv4_hdr,uint64_t ol_flags)313 rte_ipv4_phdr_cksum(const struct rte_ipv4_hdr *ipv4_hdr, uint64_t ol_flags)
314 {
315 struct ipv4_psd_header {
316 uint32_t src_addr; /* IP address of source host. */
317 uint32_t dst_addr; /* IP address of destination host. */
318 uint8_t zero; /* zero. */
319 uint8_t proto; /* L4 protocol type. */
320 uint16_t len; /* L4 length. */
321 } psd_hdr;
322
323 uint32_t l3_len;
324
325 psd_hdr.src_addr = ipv4_hdr->src_addr;
326 psd_hdr.dst_addr = ipv4_hdr->dst_addr;
327 psd_hdr.zero = 0;
328 psd_hdr.proto = ipv4_hdr->next_proto_id;
329 if (ol_flags & PKT_TX_TCP_SEG) {
330 psd_hdr.len = 0;
331 } else {
332 l3_len = rte_be_to_cpu_16(ipv4_hdr->total_length);
333 psd_hdr.len = rte_cpu_to_be_16((uint16_t)(l3_len -
334 rte_ipv4_hdr_len(ipv4_hdr)));
335 }
336 return rte_raw_cksum(&psd_hdr, sizeof(psd_hdr));
337 }
338
339 /**
340 * Process the IPv4 UDP or TCP checksum.
341 *
342 * The IP and layer 4 checksum must be set to 0 in the packet by
343 * the caller.
344 *
345 * @param ipv4_hdr
346 * The pointer to the contiguous IPv4 header.
347 * @param l4_hdr
348 * The pointer to the beginning of the L4 header.
349 * @return
350 * The complemented checksum to set in the IP packet.
351 */
352 static inline uint16_t
rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr * ipv4_hdr,const void * l4_hdr)353 rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr *ipv4_hdr, const void *l4_hdr)
354 {
355 uint32_t cksum;
356 uint32_t l3_len, l4_len;
357 uint8_t ip_hdr_len;
358
359 ip_hdr_len = rte_ipv4_hdr_len(ipv4_hdr);
360 l3_len = rte_be_to_cpu_16(ipv4_hdr->total_length);
361 if (l3_len < ip_hdr_len)
362 return 0;
363
364 l4_len = l3_len - ip_hdr_len;
365
366 cksum = rte_raw_cksum(l4_hdr, l4_len);
367 cksum += rte_ipv4_phdr_cksum(ipv4_hdr, 0);
368
369 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
370 cksum = (~cksum) & 0xffff;
371 /*
372 * Per RFC 768:If the computed checksum is zero for UDP,
373 * it is transmitted as all ones
374 * (the equivalent in one's complement arithmetic).
375 */
376 if (cksum == 0 && ipv4_hdr->next_proto_id == IPPROTO_UDP)
377 cksum = 0xffff;
378
379 return (uint16_t)cksum;
380 }
381
382 /**
383 * IPv6 Header
384 */
385 struct rte_ipv6_hdr {
386 rte_be32_t vtc_flow; /**< IP version, traffic class & flow label. */
387 rte_be16_t payload_len; /**< IP packet length - includes header size */
388 uint8_t proto; /**< Protocol, next header. */
389 uint8_t hop_limits; /**< Hop limits. */
390 uint8_t src_addr[16]; /**< IP address of source host. */
391 uint8_t dst_addr[16]; /**< IP address of destination host(s). */
392 } __rte_packed;
393
394 /* IPv6 vtc_flow: IPv / TC / flow_label */
395 #define RTE_IPV6_HDR_FL_SHIFT 0
396 #define RTE_IPV6_HDR_TC_SHIFT 20
397 #define RTE_IPV6_HDR_FL_MASK ((1u << RTE_IPV6_HDR_TC_SHIFT) - 1)
398 #define RTE_IPV6_HDR_TC_MASK (0xff << RTE_IPV6_HDR_TC_SHIFT)
399 #define RTE_IPV6_HDR_DSCP_MASK (0xfc << RTE_IPV6_HDR_TC_SHIFT)
400 #define RTE_IPV6_HDR_ECN_MASK (0x03 << RTE_IPV6_HDR_TC_SHIFT)
401 #define RTE_IPV6_HDR_ECN_CE RTE_IPV6_HDR_ECN_MASK
402
403 #define RTE_IPV6_MIN_MTU 1280 /**< Minimum MTU for IPv6, see RFC 8200. */
404
405 /**
406 * Process the pseudo-header checksum of an IPv6 header.
407 *
408 * Depending on the ol_flags, the pseudo-header checksum expected by the
409 * drivers is not the same. For instance, when TSO is enabled, the IPv6
410 * payload length must not be included in the packet.
411 *
412 * When ol_flags is 0, it computes the standard pseudo-header checksum.
413 *
414 * @param ipv6_hdr
415 * The pointer to the contiguous IPv6 header.
416 * @param ol_flags
417 * The ol_flags of the associated mbuf.
418 * @return
419 * The non-complemented checksum to set in the L4 header.
420 */
421 static inline uint16_t
rte_ipv6_phdr_cksum(const struct rte_ipv6_hdr * ipv6_hdr,uint64_t ol_flags)422 rte_ipv6_phdr_cksum(const struct rte_ipv6_hdr *ipv6_hdr, uint64_t ol_flags)
423 {
424 uint32_t sum;
425 struct {
426 rte_be32_t len; /* L4 length. */
427 rte_be32_t proto; /* L4 protocol - top 3 bytes must be zero */
428 } psd_hdr;
429
430 psd_hdr.proto = (uint32_t)(ipv6_hdr->proto << 24);
431 if (ol_flags & PKT_TX_TCP_SEG) {
432 psd_hdr.len = 0;
433 } else {
434 psd_hdr.len = ipv6_hdr->payload_len;
435 }
436
437 sum = __rte_raw_cksum(ipv6_hdr->src_addr,
438 sizeof(ipv6_hdr->src_addr) + sizeof(ipv6_hdr->dst_addr),
439 0);
440 sum = __rte_raw_cksum(&psd_hdr, sizeof(psd_hdr), sum);
441 return __rte_raw_cksum_reduce(sum);
442 }
443
444 /**
445 * Process the IPv6 UDP or TCP checksum.
446 *
447 * The IPv4 header should not contains options. The layer 4 checksum
448 * must be set to 0 in the packet by the caller.
449 *
450 * @param ipv6_hdr
451 * The pointer to the contiguous IPv6 header.
452 * @param l4_hdr
453 * The pointer to the beginning of the L4 header.
454 * @return
455 * The complemented checksum to set in the IP packet.
456 */
457 static inline uint16_t
rte_ipv6_udptcp_cksum(const struct rte_ipv6_hdr * ipv6_hdr,const void * l4_hdr)458 rte_ipv6_udptcp_cksum(const struct rte_ipv6_hdr *ipv6_hdr, const void *l4_hdr)
459 {
460 uint32_t cksum;
461 uint32_t l4_len;
462
463 l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
464
465 cksum = rte_raw_cksum(l4_hdr, l4_len);
466 cksum += rte_ipv6_phdr_cksum(ipv6_hdr, 0);
467
468 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
469 cksum = (~cksum) & 0xffff;
470 /*
471 * Per RFC 768: If the computed checksum is zero for UDP,
472 * it is transmitted as all ones
473 * (the equivalent in one's complement arithmetic).
474 */
475 if (cksum == 0 && ipv6_hdr->proto == IPPROTO_UDP)
476 cksum = 0xffff;
477
478 return (uint16_t)cksum;
479 }
480
481 /** IPv6 fragment extension header. */
482 #define RTE_IPV6_EHDR_MF_SHIFT 0
483 #define RTE_IPV6_EHDR_MF_MASK 1
484 #define RTE_IPV6_EHDR_FO_SHIFT 3
485 #define RTE_IPV6_EHDR_FO_MASK (~((1 << RTE_IPV6_EHDR_FO_SHIFT) - 1))
486 #define RTE_IPV6_EHDR_FO_ALIGN (1 << RTE_IPV6_EHDR_FO_SHIFT)
487
488 #define RTE_IPV6_FRAG_USED_MASK (RTE_IPV6_EHDR_MF_MASK | RTE_IPV6_EHDR_FO_MASK)
489
490 #define RTE_IPV6_GET_MF(x) ((x) & RTE_IPV6_EHDR_MF_MASK)
491 #define RTE_IPV6_GET_FO(x) ((x) >> RTE_IPV6_EHDR_FO_SHIFT)
492
493 #define RTE_IPV6_SET_FRAG_DATA(fo, mf) \
494 (((fo) & RTE_IPV6_EHDR_FO_MASK) | ((mf) & RTE_IPV6_EHDR_MF_MASK))
495
496 struct rte_ipv6_fragment_ext {
497 uint8_t next_header; /**< Next header type */
498 uint8_t reserved; /**< Reserved */
499 rte_be16_t frag_data; /**< All fragmentation data */
500 rte_be32_t id; /**< Packet ID */
501 } __rte_packed;
502
503 /* IPv6 fragment extension header size */
504 #define RTE_IPV6_FRAG_HDR_SIZE sizeof(struct rte_ipv6_fragment_ext)
505
506 /**
507 * Parse next IPv6 header extension
508 *
509 * This function checks if proto number is an IPv6 extensions and parses its
510 * data if so, providing information on next header and extension length.
511 *
512 * @param p
513 * Pointer to an extension raw data.
514 * @param proto
515 * Protocol number extracted from the "next header" field from
516 * the IPv6 header or the previous extension.
517 * @param ext_len
518 * Extension data length.
519 * @return
520 * next protocol number if proto is an IPv6 extension, -EINVAL otherwise
521 */
522 __rte_experimental
523 static inline int
rte_ipv6_get_next_ext(const uint8_t * p,int proto,size_t * ext_len)524 rte_ipv6_get_next_ext(const uint8_t *p, int proto, size_t *ext_len)
525 {
526 int next_proto;
527
528 switch (proto) {
529 case IPPROTO_AH:
530 next_proto = *p++;
531 *ext_len = (*p + 2) * sizeof(uint32_t);
532 break;
533
534 case IPPROTO_HOPOPTS:
535 case IPPROTO_ROUTING:
536 case IPPROTO_DSTOPTS:
537 next_proto = *p++;
538 *ext_len = (*p + 1) * sizeof(uint64_t);
539 break;
540
541 case IPPROTO_FRAGMENT:
542 next_proto = *p;
543 *ext_len = RTE_IPV6_FRAG_HDR_SIZE;
544 break;
545
546 default:
547 return -EINVAL;
548 }
549
550 return next_proto;
551 }
552
553 #ifdef __cplusplus
554 }
555 #endif
556
557 #endif /* _RTE_IP_H_ */
558