xref: /f-stack/dpdk/lib/librte_net/rte_ip.h (revision 031be553)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   Copyright 2014 6WIND S.A.
6  *   All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 1982, 1986, 1990, 1993
37  *      The Regents of the University of California.  All rights reserved.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. All advertising materials mentioning features or use of this software
48  *    must display the following acknowledgement:
49  *      This product includes software developed by the University of
50  *      California, Berkeley and its contributors.
51  * 4. Neither the name of the University nor the names of its contributors
52  *    may be used to endorse or promote products derived from this software
53  *    without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  *      @(#)in.h        8.3 (Berkeley) 1/3/94
68  * $FreeBSD: src/sys/netinet/in.h,v 1.82 2003/10/25 09:37:10 ume Exp $
69  */
70 
71 #ifndef _RTE_IP_H_
72 #define _RTE_IP_H_
73 
74 /**
75  * @file
76  *
77  * IP-related defines
78  */
79 
80 #include <stdint.h>
81 #include <netinet/in.h>
82 
83 #include <rte_byteorder.h>
84 #include <rte_mbuf.h>
85 
86 #ifdef __cplusplus
87 extern "C" {
88 #endif
89 
90 /**
91  * IPv4 Header
92  */
93 struct ipv4_hdr {
94 	uint8_t  version_ihl;		/**< version and header length */
95 	uint8_t  type_of_service;	/**< type of service */
96 	uint16_t total_length;		/**< length of packet */
97 	uint16_t packet_id;		/**< packet ID */
98 	uint16_t fragment_offset;	/**< fragmentation offset */
99 	uint8_t  time_to_live;		/**< time to live */
100 	uint8_t  next_proto_id;		/**< protocol ID */
101 	uint16_t hdr_checksum;		/**< header checksum */
102 	uint32_t src_addr;		/**< source address */
103 	uint32_t dst_addr;		/**< destination address */
104 } __attribute__((__packed__));
105 
106 /** Create IPv4 address */
107 #define IPv4(a,b,c,d) ((uint32_t)(((a) & 0xff) << 24) | \
108 					   (((b) & 0xff) << 16) | \
109 					   (((c) & 0xff) << 8)  | \
110 					   ((d) & 0xff))
111 
112 /** Maximal IPv4 packet length (including a header) */
113 #define IPV4_MAX_PKT_LEN        65535
114 
115 /** Internet header length mask for version_ihl field */
116 #define IPV4_HDR_IHL_MASK	(0x0f)
117 /**
118  * Internet header length field multiplier (IHL field specifies overall header
119  * length in number of 4-byte words)
120  */
121 #define IPV4_IHL_MULTIPLIER	(4)
122 
123 /* Fragment Offset * Flags. */
124 #define	IPV4_HDR_DF_SHIFT	14
125 #define	IPV4_HDR_MF_SHIFT	13
126 #define	IPV4_HDR_FO_SHIFT	3
127 
128 #define	IPV4_HDR_DF_FLAG	(1 << IPV4_HDR_DF_SHIFT)
129 #define	IPV4_HDR_MF_FLAG	(1 << IPV4_HDR_MF_SHIFT)
130 
131 #define	IPV4_HDR_OFFSET_MASK	((1 << IPV4_HDR_MF_SHIFT) - 1)
132 
133 #define	IPV4_HDR_OFFSET_UNITS	8
134 
135 /*
136  * IPv4 address types
137  */
138 #define IPV4_ANY              ((uint32_t)0x00000000) /**< 0.0.0.0 */
139 #define IPV4_LOOPBACK         ((uint32_t)0x7f000001) /**< 127.0.0.1 */
140 #define IPV4_BROADCAST        ((uint32_t)0xe0000000) /**< 224.0.0.0 */
141 #define IPV4_ALLHOSTS_GROUP   ((uint32_t)0xe0000001) /**< 224.0.0.1 */
142 #define IPV4_ALLRTRS_GROUP    ((uint32_t)0xe0000002) /**< 224.0.0.2 */
143 #define IPV4_MAX_LOCAL_GROUP  ((uint32_t)0xe00000ff) /**< 224.0.0.255 */
144 
145 /*
146  * IPv4 Multicast-related macros
147  */
148 #define IPV4_MIN_MCAST  IPv4(224, 0, 0, 0)          /**< Minimal IPv4-multicast address */
149 #define IPV4_MAX_MCAST  IPv4(239, 255, 255, 255)    /**< Maximum IPv4 multicast address */
150 
151 #define IS_IPV4_MCAST(x) \
152 	((x) >= IPV4_MIN_MCAST && (x) <= IPV4_MAX_MCAST) /**< check if IPv4 address is multicast */
153 
154 /**
155  * @internal Calculate a sum of all words in the buffer.
156  * Helper routine for the rte_raw_cksum().
157  *
158  * @param buf
159  *   Pointer to the buffer.
160  * @param len
161  *   Length of the buffer.
162  * @param sum
163  *   Initial value of the sum.
164  * @return
165  *   sum += Sum of all words in the buffer.
166  */
167 static inline uint32_t
168 __rte_raw_cksum(const void *buf, size_t len, uint32_t sum)
169 {
170 	/* workaround gcc strict-aliasing warning */
171 	uintptr_t ptr = (uintptr_t)buf;
172 	typedef uint16_t __attribute__((__may_alias__)) u16_p;
173 	const u16_p *u16 = (const u16_p *)ptr;
174 
175 	while (len >= (sizeof(*u16) * 4)) {
176 		sum += u16[0];
177 		sum += u16[1];
178 		sum += u16[2];
179 		sum += u16[3];
180 		len -= sizeof(*u16) * 4;
181 		u16 += 4;
182 	}
183 	while (len >= sizeof(*u16)) {
184 		sum += *u16;
185 		len -= sizeof(*u16);
186 		u16 += 1;
187 	}
188 
189 	/* if length is in odd bytes */
190 	if (len == 1)
191 		sum += *((const uint8_t *)u16);
192 
193 	return sum;
194 }
195 
196 /**
197  * @internal Reduce a sum to the non-complemented checksum.
198  * Helper routine for the rte_raw_cksum().
199  *
200  * @param sum
201  *   Value of the sum.
202  * @return
203  *   The non-complemented checksum.
204  */
205 static inline uint16_t
206 __rte_raw_cksum_reduce(uint32_t sum)
207 {
208 	sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
209 	sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
210 	return (uint16_t)sum;
211 }
212 
213 /**
214  * Process the non-complemented checksum of a buffer.
215  *
216  * @param buf
217  *   Pointer to the buffer.
218  * @param len
219  *   Length of the buffer.
220  * @return
221  *   The non-complemented checksum.
222  */
223 static inline uint16_t
224 rte_raw_cksum(const void *buf, size_t len)
225 {
226 	uint32_t sum;
227 
228 	sum = __rte_raw_cksum(buf, len, 0);
229 	return __rte_raw_cksum_reduce(sum);
230 }
231 
232 /**
233  * Compute the raw (non complemented) checksum of a packet.
234  *
235  * @param m
236  *   The pointer to the mbuf.
237  * @param off
238  *   The offset in bytes to start the checksum.
239  * @param len
240  *   The length in bytes of the data to checksum.
241  * @param cksum
242  *   A pointer to the checksum, filled on success.
243  * @return
244  *   0 on success, -1 on error (bad length or offset).
245  */
246 static inline int
247 rte_raw_cksum_mbuf(const struct rte_mbuf *m, uint32_t off, uint32_t len,
248 	uint16_t *cksum)
249 {
250 	const struct rte_mbuf *seg;
251 	const char *buf;
252 	uint32_t sum, tmp;
253 	uint32_t seglen, done;
254 
255 	/* easy case: all data in the first segment */
256 	if (off + len <= rte_pktmbuf_data_len(m)) {
257 		*cksum = rte_raw_cksum(rte_pktmbuf_mtod_offset(m,
258 				const char *, off), len);
259 		return 0;
260 	}
261 
262 	if (unlikely(off + len > rte_pktmbuf_pkt_len(m)))
263 		return -1; /* invalid params, return a dummy value */
264 
265 	/* else browse the segment to find offset */
266 	seglen = 0;
267 	for (seg = m; seg != NULL; seg = seg->next) {
268 		seglen = rte_pktmbuf_data_len(seg);
269 		if (off < seglen)
270 			break;
271 		off -= seglen;
272 	}
273 	seglen -= off;
274 	buf = rte_pktmbuf_mtod_offset(seg, const char *, off);
275 	if (seglen >= len) {
276 		/* all in one segment */
277 		*cksum = rte_raw_cksum(buf, len);
278 		return 0;
279 	}
280 
281 	/* hard case: process checksum of several segments */
282 	sum = 0;
283 	done = 0;
284 	for (;;) {
285 		tmp = __rte_raw_cksum(buf, seglen, 0);
286 		if (done & 1)
287 			tmp = rte_bswap16(tmp);
288 		sum += tmp;
289 		done += seglen;
290 		if (done == len)
291 			break;
292 		seg = seg->next;
293 		buf = rte_pktmbuf_mtod(seg, const char *);
294 		seglen = rte_pktmbuf_data_len(seg);
295 		if (seglen > len - done)
296 			seglen = len - done;
297 	}
298 
299 	*cksum = __rte_raw_cksum_reduce(sum);
300 	return 0;
301 }
302 
303 /**
304  * Process the IPv4 checksum of an IPv4 header.
305  *
306  * The checksum field must be set to 0 by the caller.
307  *
308  * @param ipv4_hdr
309  *   The pointer to the contiguous IPv4 header.
310  * @return
311  *   The complemented checksum to set in the IP packet.
312  */
313 static inline uint16_t
314 rte_ipv4_cksum(const struct ipv4_hdr *ipv4_hdr)
315 {
316 	uint16_t cksum;
317 	cksum = rte_raw_cksum(ipv4_hdr, sizeof(struct ipv4_hdr));
318 	return (cksum == 0xffff) ? cksum : ~cksum;
319 }
320 
321 /**
322  * Process the pseudo-header checksum of an IPv4 header.
323  *
324  * The checksum field must be set to 0 by the caller.
325  *
326  * Depending on the ol_flags, the pseudo-header checksum expected by the
327  * drivers is not the same. For instance, when TSO is enabled, the IP
328  * payload length must not be included in the packet.
329  *
330  * When ol_flags is 0, it computes the standard pseudo-header checksum.
331  *
332  * @param ipv4_hdr
333  *   The pointer to the contiguous IPv4 header.
334  * @param ol_flags
335  *   The ol_flags of the associated mbuf.
336  * @return
337  *   The non-complemented checksum to set in the L4 header.
338  */
339 static inline uint16_t
340 rte_ipv4_phdr_cksum(const struct ipv4_hdr *ipv4_hdr, uint64_t ol_flags)
341 {
342 	struct ipv4_psd_header {
343 		uint32_t src_addr; /* IP address of source host. */
344 		uint32_t dst_addr; /* IP address of destination host. */
345 		uint8_t  zero;     /* zero. */
346 		uint8_t  proto;    /* L4 protocol type. */
347 		uint16_t len;      /* L4 length. */
348 	} psd_hdr;
349 
350 	psd_hdr.src_addr = ipv4_hdr->src_addr;
351 	psd_hdr.dst_addr = ipv4_hdr->dst_addr;
352 	psd_hdr.zero = 0;
353 	psd_hdr.proto = ipv4_hdr->next_proto_id;
354 	if (ol_flags & PKT_TX_TCP_SEG) {
355 		psd_hdr.len = 0;
356 	} else {
357 		psd_hdr.len = rte_cpu_to_be_16(
358 			(uint16_t)(rte_be_to_cpu_16(ipv4_hdr->total_length)
359 				- sizeof(struct ipv4_hdr)));
360 	}
361 	return rte_raw_cksum(&psd_hdr, sizeof(psd_hdr));
362 }
363 
364 /**
365  * Process the IPv4 UDP or TCP checksum.
366  *
367  * The IPv4 header should not contains options. The IP and layer 4
368  * checksum must be set to 0 in the packet by the caller.
369  *
370  * @param ipv4_hdr
371  *   The pointer to the contiguous IPv4 header.
372  * @param l4_hdr
373  *   The pointer to the beginning of the L4 header.
374  * @return
375  *   The complemented checksum to set in the IP packet.
376  */
377 static inline uint16_t
378 rte_ipv4_udptcp_cksum(const struct ipv4_hdr *ipv4_hdr, const void *l4_hdr)
379 {
380 	uint32_t cksum;
381 	uint32_t l4_len;
382 
383 	l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) -
384 		sizeof(struct ipv4_hdr);
385 
386 	cksum = rte_raw_cksum(l4_hdr, l4_len);
387 	cksum += rte_ipv4_phdr_cksum(ipv4_hdr, 0);
388 
389 	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
390 	cksum = (~cksum) & 0xffff;
391 	if (cksum == 0)
392 		cksum = 0xffff;
393 
394 	return cksum;
395 }
396 
397 /**
398  * IPv6 Header
399  */
400 struct ipv6_hdr {
401 	uint32_t vtc_flow;     /**< IP version, traffic class & flow label. */
402 	uint16_t payload_len;  /**< IP packet length - includes sizeof(ip_header). */
403 	uint8_t  proto;        /**< Protocol, next header. */
404 	uint8_t  hop_limits;   /**< Hop limits. */
405 	uint8_t  src_addr[16]; /**< IP address of source host. */
406 	uint8_t  dst_addr[16]; /**< IP address of destination host(s). */
407 } __attribute__((__packed__));
408 
409 /* IPv6 vtc_flow: IPv / TC / flow_label */
410 #define IPV6_HDR_FL_SHIFT 0
411 #define IPV6_HDR_TC_SHIFT 20
412 #define IPV6_HDR_FL_MASK ((1u << IPV6_HDR_TC_SHIFT) - 1)
413 #define IPV6_HDR_TC_MASK (0xf << IPV6_HDR_TC_SHIFT)
414 
415 /**
416  * Process the pseudo-header checksum of an IPv6 header.
417  *
418  * Depending on the ol_flags, the pseudo-header checksum expected by the
419  * drivers is not the same. For instance, when TSO is enabled, the IPv6
420  * payload length must not be included in the packet.
421  *
422  * When ol_flags is 0, it computes the standard pseudo-header checksum.
423  *
424  * @param ipv6_hdr
425  *   The pointer to the contiguous IPv6 header.
426  * @param ol_flags
427  *   The ol_flags of the associated mbuf.
428  * @return
429  *   The non-complemented checksum to set in the L4 header.
430  */
431 static inline uint16_t
432 rte_ipv6_phdr_cksum(const struct ipv6_hdr *ipv6_hdr, uint64_t ol_flags)
433 {
434 	uint32_t sum;
435 	struct {
436 		uint32_t len;   /* L4 length. */
437 		uint32_t proto; /* L4 protocol - top 3 bytes must be zero */
438 	} psd_hdr;
439 
440 	psd_hdr.proto = (ipv6_hdr->proto << 24);
441 	if (ol_flags & PKT_TX_TCP_SEG) {
442 		psd_hdr.len = 0;
443 	} else {
444 		psd_hdr.len = ipv6_hdr->payload_len;
445 	}
446 
447 	sum = __rte_raw_cksum(ipv6_hdr->src_addr,
448 		sizeof(ipv6_hdr->src_addr) + sizeof(ipv6_hdr->dst_addr),
449 		0);
450 	sum = __rte_raw_cksum(&psd_hdr, sizeof(psd_hdr), sum);
451 	return __rte_raw_cksum_reduce(sum);
452 }
453 
454 /**
455  * Process the IPv6 UDP or TCP checksum.
456  *
457  * The IPv4 header should not contains options. The layer 4 checksum
458  * must be set to 0 in the packet by the caller.
459  *
460  * @param ipv6_hdr
461  *   The pointer to the contiguous IPv6 header.
462  * @param l4_hdr
463  *   The pointer to the beginning of the L4 header.
464  * @return
465  *   The complemented checksum to set in the IP packet.
466  */
467 static inline uint16_t
468 rte_ipv6_udptcp_cksum(const struct ipv6_hdr *ipv6_hdr, const void *l4_hdr)
469 {
470 	uint32_t cksum;
471 	uint32_t l4_len;
472 
473 	l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
474 
475 	cksum = rte_raw_cksum(l4_hdr, l4_len);
476 	cksum += rte_ipv6_phdr_cksum(ipv6_hdr, 0);
477 
478 	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
479 	cksum = (~cksum) & 0xffff;
480 	if (cksum == 0)
481 		cksum = 0xffff;
482 
483 	return cksum;
484 }
485 
486 #ifdef __cplusplus
487 }
488 #endif
489 
490 #endif /* _RTE_IP_H_ */
491