]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/lib/librte_net/rte_net.c
import 15.2.0 Octopus source
[ceph.git] / ceph / src / spdk / dpdk / lib / librte_net / rte_net.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 */
4
5 #include <stdint.h>
6
7 #include <rte_mbuf.h>
8 #include <rte_mbuf_ptype.h>
9 #include <rte_byteorder.h>
10 #include <rte_ether.h>
11 #include <rte_ip.h>
12 #include <rte_tcp.h>
13 #include <rte_udp.h>
14 #include <rte_sctp.h>
15 #include <rte_gre.h>
16 #include <rte_mpls.h>
17 #include <rte_net.h>
18
19 /* get l3 packet type from ip6 next protocol */
20 static uint32_t
21 ptype_l3_ip6(uint8_t ip6_proto)
22 {
23 static const uint32_t ip6_ext_proto_map[256] = {
24 [IPPROTO_HOPOPTS] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
25 [IPPROTO_ROUTING] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
26 [IPPROTO_FRAGMENT] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
27 [IPPROTO_ESP] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
28 [IPPROTO_AH] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
29 [IPPROTO_DSTOPTS] = RTE_PTYPE_L3_IPV6_EXT - RTE_PTYPE_L3_IPV6,
30 };
31
32 return RTE_PTYPE_L3_IPV6 + ip6_ext_proto_map[ip6_proto];
33 }
34
35 /* get l3 packet type from ip version and header length */
36 static uint32_t
37 ptype_l3_ip(uint8_t ipv_ihl)
38 {
39 static const uint32_t ptype_l3_ip_proto_map[256] = {
40 [0x45] = RTE_PTYPE_L3_IPV4,
41 [0x46] = RTE_PTYPE_L3_IPV4_EXT,
42 [0x47] = RTE_PTYPE_L3_IPV4_EXT,
43 [0x48] = RTE_PTYPE_L3_IPV4_EXT,
44 [0x49] = RTE_PTYPE_L3_IPV4_EXT,
45 [0x4A] = RTE_PTYPE_L3_IPV4_EXT,
46 [0x4B] = RTE_PTYPE_L3_IPV4_EXT,
47 [0x4C] = RTE_PTYPE_L3_IPV4_EXT,
48 [0x4D] = RTE_PTYPE_L3_IPV4_EXT,
49 [0x4E] = RTE_PTYPE_L3_IPV4_EXT,
50 [0x4F] = RTE_PTYPE_L3_IPV4_EXT,
51 };
52
53 return ptype_l3_ip_proto_map[ipv_ihl];
54 }
55
56 /* get l4 packet type from proto */
57 static uint32_t
58 ptype_l4(uint8_t proto)
59 {
60 static const uint32_t ptype_l4_proto[256] = {
61 [IPPROTO_UDP] = RTE_PTYPE_L4_UDP,
62 [IPPROTO_TCP] = RTE_PTYPE_L4_TCP,
63 [IPPROTO_SCTP] = RTE_PTYPE_L4_SCTP,
64 };
65
66 return ptype_l4_proto[proto];
67 }
68
69 /* get inner l3 packet type from ip6 next protocol */
70 static uint32_t
71 ptype_inner_l3_ip6(uint8_t ip6_proto)
72 {
73 static const uint32_t ptype_inner_ip6_ext_proto_map[256] = {
74 [IPPROTO_HOPOPTS] = RTE_PTYPE_INNER_L3_IPV6_EXT -
75 RTE_PTYPE_INNER_L3_IPV6,
76 [IPPROTO_ROUTING] = RTE_PTYPE_INNER_L3_IPV6_EXT -
77 RTE_PTYPE_INNER_L3_IPV6,
78 [IPPROTO_FRAGMENT] = RTE_PTYPE_INNER_L3_IPV6_EXT -
79 RTE_PTYPE_INNER_L3_IPV6,
80 [IPPROTO_ESP] = RTE_PTYPE_INNER_L3_IPV6_EXT -
81 RTE_PTYPE_INNER_L3_IPV6,
82 [IPPROTO_AH] = RTE_PTYPE_INNER_L3_IPV6_EXT -
83 RTE_PTYPE_INNER_L3_IPV6,
84 [IPPROTO_DSTOPTS] = RTE_PTYPE_INNER_L3_IPV6_EXT -
85 RTE_PTYPE_INNER_L3_IPV6,
86 };
87
88 return RTE_PTYPE_INNER_L3_IPV6 +
89 ptype_inner_ip6_ext_proto_map[ip6_proto];
90 }
91
92 /* get inner l3 packet type from ip version and header length */
93 static uint32_t
94 ptype_inner_l3_ip(uint8_t ipv_ihl)
95 {
96 static const uint32_t ptype_inner_l3_ip_proto_map[256] = {
97 [0x45] = RTE_PTYPE_INNER_L3_IPV4,
98 [0x46] = RTE_PTYPE_INNER_L3_IPV4_EXT,
99 [0x47] = RTE_PTYPE_INNER_L3_IPV4_EXT,
100 [0x48] = RTE_PTYPE_INNER_L3_IPV4_EXT,
101 [0x49] = RTE_PTYPE_INNER_L3_IPV4_EXT,
102 [0x4A] = RTE_PTYPE_INNER_L3_IPV4_EXT,
103 [0x4B] = RTE_PTYPE_INNER_L3_IPV4_EXT,
104 [0x4C] = RTE_PTYPE_INNER_L3_IPV4_EXT,
105 [0x4D] = RTE_PTYPE_INNER_L3_IPV4_EXT,
106 [0x4E] = RTE_PTYPE_INNER_L3_IPV4_EXT,
107 [0x4F] = RTE_PTYPE_INNER_L3_IPV4_EXT,
108 };
109
110 return ptype_inner_l3_ip_proto_map[ipv_ihl];
111 }
112
113 /* get inner l4 packet type from proto */
114 static uint32_t
115 ptype_inner_l4(uint8_t proto)
116 {
117 static const uint32_t ptype_inner_l4_proto[256] = {
118 [IPPROTO_UDP] = RTE_PTYPE_INNER_L4_UDP,
119 [IPPROTO_TCP] = RTE_PTYPE_INNER_L4_TCP,
120 [IPPROTO_SCTP] = RTE_PTYPE_INNER_L4_SCTP,
121 };
122
123 return ptype_inner_l4_proto[proto];
124 }
125
126 /* get the tunnel packet type if any, update proto and off. */
127 static uint32_t
128 ptype_tunnel(uint16_t *proto, const struct rte_mbuf *m,
129 uint32_t *off)
130 {
131 switch (*proto) {
132 case IPPROTO_GRE: {
133 static const uint8_t opt_len[16] = {
134 [0x0] = 4,
135 [0x1] = 8,
136 [0x2] = 8,
137 [0x8] = 8,
138 [0x3] = 12,
139 [0x9] = 12,
140 [0xa] = 12,
141 [0xb] = 16,
142 };
143 const struct gre_hdr *gh;
144 struct gre_hdr gh_copy;
145 uint16_t flags;
146
147 gh = rte_pktmbuf_read(m, *off, sizeof(*gh), &gh_copy);
148 if (unlikely(gh == NULL))
149 return 0;
150
151 flags = rte_be_to_cpu_16(*(const uint16_t *)gh);
152 flags >>= 12;
153 if (opt_len[flags] == 0)
154 return 0;
155
156 *off += opt_len[flags];
157 *proto = gh->proto;
158 if (*proto == rte_cpu_to_be_16(ETHER_TYPE_TEB))
159 return RTE_PTYPE_TUNNEL_NVGRE;
160 else
161 return RTE_PTYPE_TUNNEL_GRE;
162 }
163 case IPPROTO_IPIP:
164 *proto = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
165 return RTE_PTYPE_TUNNEL_IP;
166 case IPPROTO_IPV6:
167 *proto = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
168 return RTE_PTYPE_TUNNEL_IP; /* IP is also valid for IPv6 */
169 default:
170 return 0;
171 }
172 }
173
174 /* get the ipv4 header length */
175 static uint8_t
176 ip4_hlen(const struct ipv4_hdr *hdr)
177 {
178 return (hdr->version_ihl & 0xf) * 4;
179 }
180
181 /* parse ipv6 extended headers, update offset and return next proto */
182 int __rte_experimental
183 rte_net_skip_ip6_ext(uint16_t proto, const struct rte_mbuf *m, uint32_t *off,
184 int *frag)
185 {
186 struct ext_hdr {
187 uint8_t next_hdr;
188 uint8_t len;
189 };
190 const struct ext_hdr *xh;
191 struct ext_hdr xh_copy;
192 unsigned int i;
193
194 *frag = 0;
195
196 #define MAX_EXT_HDRS 5
197 for (i = 0; i < MAX_EXT_HDRS; i++) {
198 switch (proto) {
199 case IPPROTO_HOPOPTS:
200 case IPPROTO_ROUTING:
201 case IPPROTO_DSTOPTS:
202 xh = rte_pktmbuf_read(m, *off, sizeof(*xh),
203 &xh_copy);
204 if (xh == NULL)
205 return -1;
206 *off += (xh->len + 1) * 8;
207 proto = xh->next_hdr;
208 break;
209 case IPPROTO_FRAGMENT:
210 xh = rte_pktmbuf_read(m, *off, sizeof(*xh),
211 &xh_copy);
212 if (xh == NULL)
213 return -1;
214 *off += 8;
215 proto = xh->next_hdr;
216 *frag = 1;
217 return proto; /* this is always the last ext hdr */
218 case IPPROTO_NONE:
219 return 0;
220 default:
221 return proto;
222 }
223 }
224 return -1;
225 }
226
227 /* parse mbuf data to get packet type */
228 uint32_t rte_net_get_ptype(const struct rte_mbuf *m,
229 struct rte_net_hdr_lens *hdr_lens, uint32_t layers)
230 {
231 struct rte_net_hdr_lens local_hdr_lens;
232 const struct ether_hdr *eh;
233 struct ether_hdr eh_copy;
234 uint32_t pkt_type = RTE_PTYPE_L2_ETHER;
235 uint32_t off = 0;
236 uint16_t proto;
237 int ret;
238
239 if (hdr_lens == NULL)
240 hdr_lens = &local_hdr_lens;
241
242 eh = rte_pktmbuf_read(m, off, sizeof(*eh), &eh_copy);
243 if (unlikely(eh == NULL))
244 return 0;
245 proto = eh->ether_type;
246 off = sizeof(*eh);
247 hdr_lens->l2_len = off;
248
249 if ((layers & RTE_PTYPE_L2_MASK) == 0)
250 return 0;
251
252 if (proto == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
253 goto l3; /* fast path if packet is IPv4 */
254
255 if (proto == rte_cpu_to_be_16(ETHER_TYPE_VLAN)) {
256 const struct vlan_hdr *vh;
257 struct vlan_hdr vh_copy;
258
259 pkt_type = RTE_PTYPE_L2_ETHER_VLAN;
260 vh = rte_pktmbuf_read(m, off, sizeof(*vh), &vh_copy);
261 if (unlikely(vh == NULL))
262 return pkt_type;
263 off += sizeof(*vh);
264 hdr_lens->l2_len += sizeof(*vh);
265 proto = vh->eth_proto;
266 } else if (proto == rte_cpu_to_be_16(ETHER_TYPE_QINQ)) {
267 const struct vlan_hdr *vh;
268 struct vlan_hdr vh_copy;
269
270 pkt_type = RTE_PTYPE_L2_ETHER_QINQ;
271 vh = rte_pktmbuf_read(m, off + sizeof(*vh), sizeof(*vh),
272 &vh_copy);
273 if (unlikely(vh == NULL))
274 return pkt_type;
275 off += 2 * sizeof(*vh);
276 hdr_lens->l2_len += 2 * sizeof(*vh);
277 proto = vh->eth_proto;
278 } else if ((proto == rte_cpu_to_be_16(ETHER_TYPE_MPLS)) ||
279 (proto == rte_cpu_to_be_16(ETHER_TYPE_MPLSM))) {
280 unsigned int i;
281 const struct mpls_hdr *mh;
282 struct mpls_hdr mh_copy;
283
284 #define MAX_MPLS_HDR 5
285 for (i = 0; i < MAX_MPLS_HDR; i++) {
286 mh = rte_pktmbuf_read(m, off + (i * sizeof(*mh)),
287 sizeof(*mh), &mh_copy);
288 if (unlikely(mh == NULL))
289 return pkt_type;
290 }
291 if (i == MAX_MPLS_HDR)
292 return pkt_type;
293 pkt_type = RTE_PTYPE_L2_ETHER_MPLS;
294 hdr_lens->l2_len += (sizeof(*mh) * i);
295 return pkt_type;
296 }
297
298 l3:
299 if ((layers & RTE_PTYPE_L3_MASK) == 0)
300 return pkt_type;
301
302 if (proto == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
303 const struct ipv4_hdr *ip4h;
304 struct ipv4_hdr ip4h_copy;
305
306 ip4h = rte_pktmbuf_read(m, off, sizeof(*ip4h), &ip4h_copy);
307 if (unlikely(ip4h == NULL))
308 return pkt_type;
309
310 pkt_type |= ptype_l3_ip(ip4h->version_ihl);
311 hdr_lens->l3_len = ip4_hlen(ip4h);
312 off += hdr_lens->l3_len;
313
314 if ((layers & RTE_PTYPE_L4_MASK) == 0)
315 return pkt_type;
316
317 if (ip4h->fragment_offset & rte_cpu_to_be_16(
318 IPV4_HDR_OFFSET_MASK | IPV4_HDR_MF_FLAG)) {
319 pkt_type |= RTE_PTYPE_L4_FRAG;
320 hdr_lens->l4_len = 0;
321 return pkt_type;
322 }
323 proto = ip4h->next_proto_id;
324 pkt_type |= ptype_l4(proto);
325 } else if (proto == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
326 const struct ipv6_hdr *ip6h;
327 struct ipv6_hdr ip6h_copy;
328 int frag = 0;
329
330 ip6h = rte_pktmbuf_read(m, off, sizeof(*ip6h), &ip6h_copy);
331 if (unlikely(ip6h == NULL))
332 return pkt_type;
333
334 proto = ip6h->proto;
335 hdr_lens->l3_len = sizeof(*ip6h);
336 off += hdr_lens->l3_len;
337 pkt_type |= ptype_l3_ip6(proto);
338 if ((pkt_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV6_EXT) {
339 ret = rte_net_skip_ip6_ext(proto, m, &off, &frag);
340 if (ret < 0)
341 return pkt_type;
342 proto = ret;
343 hdr_lens->l3_len = off - hdr_lens->l2_len;
344 }
345 if (proto == 0)
346 return pkt_type;
347
348 if ((layers & RTE_PTYPE_L4_MASK) == 0)
349 return pkt_type;
350
351 if (frag) {
352 pkt_type |= RTE_PTYPE_L4_FRAG;
353 hdr_lens->l4_len = 0;
354 return pkt_type;
355 }
356 pkt_type |= ptype_l4(proto);
357 }
358
359 if ((pkt_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP) {
360 hdr_lens->l4_len = sizeof(struct udp_hdr);
361 return pkt_type;
362 } else if ((pkt_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP) {
363 const struct tcp_hdr *th;
364 struct tcp_hdr th_copy;
365
366 th = rte_pktmbuf_read(m, off, sizeof(*th), &th_copy);
367 if (unlikely(th == NULL))
368 return pkt_type & (RTE_PTYPE_L2_MASK |
369 RTE_PTYPE_L3_MASK);
370 hdr_lens->l4_len = (th->data_off & 0xf0) >> 2;
371 return pkt_type;
372 } else if ((pkt_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_SCTP) {
373 hdr_lens->l4_len = sizeof(struct sctp_hdr);
374 return pkt_type;
375 } else {
376 uint32_t prev_off = off;
377
378 hdr_lens->l4_len = 0;
379
380 if ((layers & RTE_PTYPE_TUNNEL_MASK) == 0)
381 return pkt_type;
382
383 pkt_type |= ptype_tunnel(&proto, m, &off);
384 hdr_lens->tunnel_len = off - prev_off;
385 }
386
387 /* same job for inner header: we need to duplicate the code
388 * because the packet types do not have the same value.
389 */
390 if ((layers & RTE_PTYPE_INNER_L2_MASK) == 0)
391 return pkt_type;
392
393 hdr_lens->inner_l2_len = 0;
394 if (proto == rte_cpu_to_be_16(ETHER_TYPE_TEB)) {
395 eh = rte_pktmbuf_read(m, off, sizeof(*eh), &eh_copy);
396 if (unlikely(eh == NULL))
397 return pkt_type;
398 pkt_type |= RTE_PTYPE_INNER_L2_ETHER;
399 proto = eh->ether_type;
400 off += sizeof(*eh);
401 hdr_lens->inner_l2_len = sizeof(*eh);
402 }
403
404 if (proto == rte_cpu_to_be_16(ETHER_TYPE_VLAN)) {
405 const struct vlan_hdr *vh;
406 struct vlan_hdr vh_copy;
407
408 pkt_type &= ~RTE_PTYPE_INNER_L2_MASK;
409 pkt_type |= RTE_PTYPE_INNER_L2_ETHER_VLAN;
410 vh = rte_pktmbuf_read(m, off, sizeof(*vh), &vh_copy);
411 if (unlikely(vh == NULL))
412 return pkt_type;
413 off += sizeof(*vh);
414 hdr_lens->inner_l2_len += sizeof(*vh);
415 proto = vh->eth_proto;
416 } else if (proto == rte_cpu_to_be_16(ETHER_TYPE_QINQ)) {
417 const struct vlan_hdr *vh;
418 struct vlan_hdr vh_copy;
419
420 pkt_type &= ~RTE_PTYPE_INNER_L2_MASK;
421 pkt_type |= RTE_PTYPE_INNER_L2_ETHER_QINQ;
422 vh = rte_pktmbuf_read(m, off + sizeof(*vh), sizeof(*vh),
423 &vh_copy);
424 if (unlikely(vh == NULL))
425 return pkt_type;
426 off += 2 * sizeof(*vh);
427 hdr_lens->inner_l2_len += 2 * sizeof(*vh);
428 proto = vh->eth_proto;
429 }
430
431 if ((layers & RTE_PTYPE_INNER_L3_MASK) == 0)
432 return pkt_type;
433
434 if (proto == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
435 const struct ipv4_hdr *ip4h;
436 struct ipv4_hdr ip4h_copy;
437
438 ip4h = rte_pktmbuf_read(m, off, sizeof(*ip4h), &ip4h_copy);
439 if (unlikely(ip4h == NULL))
440 return pkt_type;
441
442 pkt_type |= ptype_inner_l3_ip(ip4h->version_ihl);
443 hdr_lens->inner_l3_len = ip4_hlen(ip4h);
444 off += hdr_lens->inner_l3_len;
445
446 if ((layers & RTE_PTYPE_INNER_L4_MASK) == 0)
447 return pkt_type;
448 if (ip4h->fragment_offset &
449 rte_cpu_to_be_16(IPV4_HDR_OFFSET_MASK |
450 IPV4_HDR_MF_FLAG)) {
451 pkt_type |= RTE_PTYPE_INNER_L4_FRAG;
452 hdr_lens->inner_l4_len = 0;
453 return pkt_type;
454 }
455 proto = ip4h->next_proto_id;
456 pkt_type |= ptype_inner_l4(proto);
457 } else if (proto == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
458 const struct ipv6_hdr *ip6h;
459 struct ipv6_hdr ip6h_copy;
460 int frag = 0;
461
462 ip6h = rte_pktmbuf_read(m, off, sizeof(*ip6h), &ip6h_copy);
463 if (unlikely(ip6h == NULL))
464 return pkt_type;
465
466 proto = ip6h->proto;
467 hdr_lens->inner_l3_len = sizeof(*ip6h);
468 off += hdr_lens->inner_l3_len;
469 pkt_type |= ptype_inner_l3_ip6(proto);
470 if ((pkt_type & RTE_PTYPE_INNER_L3_MASK) ==
471 RTE_PTYPE_INNER_L3_IPV6_EXT) {
472 uint32_t prev_off;
473
474 prev_off = off;
475 ret = rte_net_skip_ip6_ext(proto, m, &off, &frag);
476 if (ret < 0)
477 return pkt_type;
478 proto = ret;
479 hdr_lens->inner_l3_len += off - prev_off;
480 }
481 if (proto == 0)
482 return pkt_type;
483
484 if ((layers & RTE_PTYPE_INNER_L4_MASK) == 0)
485 return pkt_type;
486
487 if (frag) {
488 pkt_type |= RTE_PTYPE_INNER_L4_FRAG;
489 hdr_lens->inner_l4_len = 0;
490 return pkt_type;
491 }
492 pkt_type |= ptype_inner_l4(proto);
493 }
494
495 if ((pkt_type & RTE_PTYPE_INNER_L4_MASK) == RTE_PTYPE_INNER_L4_UDP) {
496 hdr_lens->inner_l4_len = sizeof(struct udp_hdr);
497 } else if ((pkt_type & RTE_PTYPE_INNER_L4_MASK) ==
498 RTE_PTYPE_INNER_L4_TCP) {
499 const struct tcp_hdr *th;
500 struct tcp_hdr th_copy;
501
502 th = rte_pktmbuf_read(m, off, sizeof(*th), &th_copy);
503 if (unlikely(th == NULL))
504 return pkt_type & (RTE_PTYPE_INNER_L2_MASK |
505 RTE_PTYPE_INNER_L3_MASK);
506 hdr_lens->inner_l4_len = (th->data_off & 0xf0) >> 2;
507 } else if ((pkt_type & RTE_PTYPE_INNER_L4_MASK) ==
508 RTE_PTYPE_INNER_L4_SCTP) {
509 hdr_lens->inner_l4_len = sizeof(struct sctp_hdr);
510 } else {
511 hdr_lens->inner_l4_len = 0;
512 }
513
514 return pkt_type;
515 }