1/*
2 * QEMU RX packets abstractions
3 *
4 * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
5 *
6 * Developed by Daynix Computing LTD (http://www.daynix.com)
7 *
8 * Authors:
9 * Dmitry Fleytman <dmitry@daynix.com>
10 * Tamir Shomer <tamirs@daynix.com>
11 * Yan Vugenfirer <yan@daynix.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2 or later.
14 * See the COPYING file in the top-level directory.
15 *
16 */
17
18#include "qemu/osdep.h"
19#include "trace.h"
20#include "net_rx_pkt.h"
21#include "net/checksum.h"
22#include "net/tap.h"
23
24struct NetRxPkt {
25 struct virtio_net_hdr virt_hdr;
26 uint8_t ehdr_buf[sizeof(struct eth_header) + sizeof(struct vlan_header)];
27 struct iovec *vec;
28 uint16_t vec_len_total;
29 uint16_t vec_len;
30 uint32_t tot_len;
31 uint16_t tci;
32 size_t ehdr_buf_len;
33 bool has_virt_hdr;
34 eth_pkt_types_e packet_type;
35
36 /* Analysis results */
37 bool isip4;
38 bool isip6;
39 bool isudp;
40 bool istcp;
41
42 size_t l3hdr_off;
43 size_t l4hdr_off;
44 size_t l5hdr_off;
45
46 eth_ip6_hdr_info ip6hdr_info;
47 eth_ip4_hdr_info ip4hdr_info;
48 eth_l4_hdr_info l4hdr_info;
49};
50
51void net_rx_pkt_init(struct NetRxPkt **pkt, bool has_virt_hdr)
52{
53 struct NetRxPkt *p = g_malloc0(sizeof *p);
54 p->has_virt_hdr = has_virt_hdr;
55 p->vec = NULL;
56 p->vec_len_total = 0;
57 *pkt = p;
58}
59
60void net_rx_pkt_uninit(struct NetRxPkt *pkt)
61{
62 if (pkt->vec_len_total != 0) {
63 g_free(pkt->vec);
64 }
65
66 g_free(pkt);
67}
68
69struct virtio_net_hdr *net_rx_pkt_get_vhdr(struct NetRxPkt *pkt)
70{
71 assert(pkt);
72 return &pkt->virt_hdr;
73}
74
75static inline void
76net_rx_pkt_iovec_realloc(struct NetRxPkt *pkt,
77 int new_iov_len)
78{
79 if (pkt->vec_len_total < new_iov_len) {
80 g_free(pkt->vec);
81 pkt->vec = g_malloc(sizeof(*pkt->vec) * new_iov_len);
82 pkt->vec_len_total = new_iov_len;
83 }
84}
85
86static void
87net_rx_pkt_pull_data(struct NetRxPkt *pkt,
88 const struct iovec *iov, int iovcnt,
89 size_t ploff)
90{
91 uint32_t pllen = iov_size(iov, iovcnt) - ploff;
92
93 if (pkt->ehdr_buf_len) {
94 net_rx_pkt_iovec_realloc(pkt, iovcnt + 1);
95
96 pkt->vec[0].iov_base = pkt->ehdr_buf;
97 pkt->vec[0].iov_len = pkt->ehdr_buf_len;
98
99 pkt->tot_len = pllen + pkt->ehdr_buf_len;
100 pkt->vec_len = iov_copy(pkt->vec + 1, pkt->vec_len_total - 1,
101 iov, iovcnt, ploff, pllen) + 1;
102 } else {
103 net_rx_pkt_iovec_realloc(pkt, iovcnt);
104
105 pkt->tot_len = pllen;
106 pkt->vec_len = iov_copy(pkt->vec, pkt->vec_len_total,
107 iov, iovcnt, ploff, pkt->tot_len);
108 }
109
110 eth_get_protocols(pkt->vec, pkt->vec_len, &pkt->isip4, &pkt->isip6,
111 &pkt->isudp, &pkt->istcp,
112 &pkt->l3hdr_off, &pkt->l4hdr_off, &pkt->l5hdr_off,
113 &pkt->ip6hdr_info, &pkt->ip4hdr_info, &pkt->l4hdr_info);
114
115 trace_net_rx_pkt_parsed(pkt->isip4, pkt->isip6, pkt->isudp, pkt->istcp,
116 pkt->l3hdr_off, pkt->l4hdr_off, pkt->l5hdr_off);
117}
118
119void net_rx_pkt_attach_iovec(struct NetRxPkt *pkt,
120 const struct iovec *iov, int iovcnt,
121 size_t iovoff, bool strip_vlan)
122{
123 uint16_t tci = 0;
124 uint16_t ploff = iovoff;
125 assert(pkt);
126
127 if (strip_vlan) {
128 pkt->ehdr_buf_len = eth_strip_vlan(iov, iovcnt, iovoff, pkt->ehdr_buf,
129 &ploff, &tci);
130 } else {
131 pkt->ehdr_buf_len = 0;
132 }
133
134 pkt->tci = tci;
135
136 net_rx_pkt_pull_data(pkt, iov, iovcnt, ploff);
137}
138
139void net_rx_pkt_attach_iovec_ex(struct NetRxPkt *pkt,
140 const struct iovec *iov, int iovcnt,
141 size_t iovoff, bool strip_vlan,
142 uint16_t vet)
143{
144 uint16_t tci = 0;
145 uint16_t ploff = iovoff;
146 assert(pkt);
147
148 if (strip_vlan) {
149 pkt->ehdr_buf_len = eth_strip_vlan_ex(iov, iovcnt, iovoff, vet,
150 pkt->ehdr_buf,
151 &ploff, &tci);
152 } else {
153 pkt->ehdr_buf_len = 0;
154 }
155
156 pkt->tci = tci;
157
158 net_rx_pkt_pull_data(pkt, iov, iovcnt, ploff);
159}
160
161void net_rx_pkt_dump(struct NetRxPkt *pkt)
162{
163#ifdef NET_RX_PKT_DEBUG
164 assert(pkt);
165
166 printf("RX PKT: tot_len: %d, ehdr_buf_len: %lu, vlan_tag: %d\n",
167 pkt->tot_len, pkt->ehdr_buf_len, pkt->tci);
168#endif
169}
170
171void net_rx_pkt_set_packet_type(struct NetRxPkt *pkt,
172 eth_pkt_types_e packet_type)
173{
174 assert(pkt);
175
176 pkt->packet_type = packet_type;
177
178}
179
180eth_pkt_types_e net_rx_pkt_get_packet_type(struct NetRxPkt *pkt)
181{
182 assert(pkt);
183
184 return pkt->packet_type;
185}
186
187size_t net_rx_pkt_get_total_len(struct NetRxPkt *pkt)
188{
189 assert(pkt);
190
191 return pkt->tot_len;
192}
193
194void net_rx_pkt_set_protocols(struct NetRxPkt *pkt, const void *data,
195 size_t len)
196{
197 const struct iovec iov = {
198 .iov_base = (void *)data,
199 .iov_len = len
200 };
201
202 assert(pkt);
203
204 eth_get_protocols(&iov, 1, &pkt->isip4, &pkt->isip6,
205 &pkt->isudp, &pkt->istcp,
206 &pkt->l3hdr_off, &pkt->l4hdr_off, &pkt->l5hdr_off,
207 &pkt->ip6hdr_info, &pkt->ip4hdr_info, &pkt->l4hdr_info);
208}
209
210void net_rx_pkt_get_protocols(struct NetRxPkt *pkt,
211 bool *isip4, bool *isip6,
212 bool *isudp, bool *istcp)
213{
214 assert(pkt);
215
216 *isip4 = pkt->isip4;
217 *isip6 = pkt->isip6;
218 *isudp = pkt->isudp;
219 *istcp = pkt->istcp;
220}
221
222size_t net_rx_pkt_get_l3_hdr_offset(struct NetRxPkt *pkt)
223{
224 assert(pkt);
225 return pkt->l3hdr_off;
226}
227
228size_t net_rx_pkt_get_l4_hdr_offset(struct NetRxPkt *pkt)
229{
230 assert(pkt);
231 return pkt->l4hdr_off;
232}
233
234size_t net_rx_pkt_get_l5_hdr_offset(struct NetRxPkt *pkt)
235{
236 assert(pkt);
237 return pkt->l5hdr_off;
238}
239
240eth_ip6_hdr_info *net_rx_pkt_get_ip6_info(struct NetRxPkt *pkt)
241{
242 return &pkt->ip6hdr_info;
243}
244
245eth_ip4_hdr_info *net_rx_pkt_get_ip4_info(struct NetRxPkt *pkt)
246{
247 return &pkt->ip4hdr_info;
248}
249
250eth_l4_hdr_info *net_rx_pkt_get_l4_info(struct NetRxPkt *pkt)
251{
252 return &pkt->l4hdr_info;
253}
254
255static inline void
256_net_rx_rss_add_chunk(uint8_t *rss_input, size_t *bytes_written,
257 void *ptr, size_t size)
258{
259 memcpy(&rss_input[*bytes_written], ptr, size);
260 trace_net_rx_pkt_rss_add_chunk(ptr, size, *bytes_written);
261 *bytes_written += size;
262}
263
264static inline void
265_net_rx_rss_prepare_ip4(uint8_t *rss_input,
266 struct NetRxPkt *pkt,
267 size_t *bytes_written)
268{
269 struct ip_header *ip4_hdr = &pkt->ip4hdr_info.ip4_hdr;
270
271 _net_rx_rss_add_chunk(rss_input, bytes_written,
272 &ip4_hdr->ip_src, sizeof(uint32_t));
273
274 _net_rx_rss_add_chunk(rss_input, bytes_written,
275 &ip4_hdr->ip_dst, sizeof(uint32_t));
276}
277
278static inline void
279_net_rx_rss_prepare_ip6(uint8_t *rss_input,
280 struct NetRxPkt *pkt,
281 bool ipv6ex, size_t *bytes_written)
282{
283 eth_ip6_hdr_info *ip6info = &pkt->ip6hdr_info;
284
285 _net_rx_rss_add_chunk(rss_input, bytes_written,
286 (ipv6ex && ip6info->rss_ex_src_valid) ? &ip6info->rss_ex_src
287 : &ip6info->ip6_hdr.ip6_src,
288 sizeof(struct in6_address));
289
290 _net_rx_rss_add_chunk(rss_input, bytes_written,
291 (ipv6ex && ip6info->rss_ex_dst_valid) ? &ip6info->rss_ex_dst
292 : &ip6info->ip6_hdr.ip6_dst,
293 sizeof(struct in6_address));
294}
295
296static inline void
297_net_rx_rss_prepare_tcp(uint8_t *rss_input,
298 struct NetRxPkt *pkt,
299 size_t *bytes_written)
300{
301 struct tcp_header *tcphdr = &pkt->l4hdr_info.hdr.tcp;
302
303 _net_rx_rss_add_chunk(rss_input, bytes_written,
304 &tcphdr->th_sport, sizeof(uint16_t));
305
306 _net_rx_rss_add_chunk(rss_input, bytes_written,
307 &tcphdr->th_dport, sizeof(uint16_t));
308}
309
310uint32_t
311net_rx_pkt_calc_rss_hash(struct NetRxPkt *pkt,
312 NetRxPktRssType type,
313 uint8_t *key)
314{
315 uint8_t rss_input[36];
316 size_t rss_length = 0;
317 uint32_t rss_hash = 0;
318 net_toeplitz_key key_data;
319
320 switch (type) {
321 case NetPktRssIpV4:
322 assert(pkt->isip4);
323 trace_net_rx_pkt_rss_ip4();
324 _net_rx_rss_prepare_ip4(&rss_input[0], pkt, &rss_length);
325 break;
326 case NetPktRssIpV4Tcp:
327 assert(pkt->isip4);
328 assert(pkt->istcp);
329 trace_net_rx_pkt_rss_ip4_tcp();
330 _net_rx_rss_prepare_ip4(&rss_input[0], pkt, &rss_length);
331 _net_rx_rss_prepare_tcp(&rss_input[0], pkt, &rss_length);
332 break;
333 case NetPktRssIpV6Tcp:
334 assert(pkt->isip6);
335 assert(pkt->istcp);
336 trace_net_rx_pkt_rss_ip6_tcp();
337 _net_rx_rss_prepare_ip6(&rss_input[0], pkt, true, &rss_length);
338 _net_rx_rss_prepare_tcp(&rss_input[0], pkt, &rss_length);
339 break;
340 case NetPktRssIpV6:
341 assert(pkt->isip6);
342 trace_net_rx_pkt_rss_ip6();
343 _net_rx_rss_prepare_ip6(&rss_input[0], pkt, false, &rss_length);
344 break;
345 case NetPktRssIpV6Ex:
346 assert(pkt->isip6);
347 trace_net_rx_pkt_rss_ip6_ex();
348 _net_rx_rss_prepare_ip6(&rss_input[0], pkt, true, &rss_length);
349 break;
350 default:
351 assert(false);
352 break;
353 }
354
355 net_toeplitz_key_init(&key_data, key);
356 net_toeplitz_add(&rss_hash, rss_input, rss_length, &key_data);
357
358 trace_net_rx_pkt_rss_hash(rss_length, rss_hash);
359
360 return rss_hash;
361}
362
363uint16_t net_rx_pkt_get_ip_id(struct NetRxPkt *pkt)
364{
365 assert(pkt);
366
367 if (pkt->isip4) {
368 return be16_to_cpu(pkt->ip4hdr_info.ip4_hdr.ip_id);
369 }
370
371 return 0;
372}
373
374bool net_rx_pkt_is_tcp_ack(struct NetRxPkt *pkt)
375{
376 assert(pkt);
377
378 if (pkt->istcp) {
379 return TCP_HEADER_FLAGS(&pkt->l4hdr_info.hdr.tcp) & TCP_FLAG_ACK;
380 }
381
382 return false;
383}
384
385bool net_rx_pkt_has_tcp_data(struct NetRxPkt *pkt)
386{
387 assert(pkt);
388
389 if (pkt->istcp) {
390 return pkt->l4hdr_info.has_tcp_data;
391 }
392
393 return false;
394}
395
396struct iovec *net_rx_pkt_get_iovec(struct NetRxPkt *pkt)
397{
398 assert(pkt);
399
400 return pkt->vec;
401}
402
403uint16_t net_rx_pkt_get_iovec_len(struct NetRxPkt *pkt)
404{
405 assert(pkt);
406
407 return pkt->vec_len;
408}
409
410void net_rx_pkt_set_vhdr(struct NetRxPkt *pkt,
411 struct virtio_net_hdr *vhdr)
412{
413 assert(pkt);
414
415 memcpy(&pkt->virt_hdr, vhdr, sizeof pkt->virt_hdr);
416}
417
418void net_rx_pkt_set_vhdr_iovec(struct NetRxPkt *pkt,
419 const struct iovec *iov, int iovcnt)
420{
421 assert(pkt);
422
423 iov_to_buf(iov, iovcnt, 0, &pkt->virt_hdr, sizeof pkt->virt_hdr);
424}
425
426bool net_rx_pkt_is_vlan_stripped(struct NetRxPkt *pkt)
427{
428 assert(pkt);
429
430 return pkt->ehdr_buf_len ? true : false;
431}
432
433bool net_rx_pkt_has_virt_hdr(struct NetRxPkt *pkt)
434{
435 assert(pkt);
436
437 return pkt->has_virt_hdr;
438}
439
440uint16_t net_rx_pkt_get_vlan_tag(struct NetRxPkt *pkt)
441{
442 assert(pkt);
443
444 return pkt->tci;
445}
446
447bool net_rx_pkt_validate_l3_csum(struct NetRxPkt *pkt, bool *csum_valid)
448{
449 uint32_t cntr;
450 uint16_t csum;
451 uint32_t csl;
452
453 trace_net_rx_pkt_l3_csum_validate_entry();
454
455 if (!pkt->isip4) {
456 trace_net_rx_pkt_l3_csum_validate_not_ip4();
457 return false;
458 }
459
460 csl = pkt->l4hdr_off - pkt->l3hdr_off;
461
462 cntr = net_checksum_add_iov(pkt->vec, pkt->vec_len,
463 pkt->l3hdr_off,
464 csl, 0);
465
466 csum = net_checksum_finish(cntr);
467
468 *csum_valid = (csum == 0);
469
470 trace_net_rx_pkt_l3_csum_validate_csum(pkt->l3hdr_off, csl,
471 cntr, csum, *csum_valid);
472
473 return true;
474}
475
476static uint16_t
477_net_rx_pkt_calc_l4_csum(struct NetRxPkt *pkt)
478{
479 uint32_t cntr;
480 uint16_t csum;
481 uint16_t csl;
482 uint32_t cso;
483
484 trace_net_rx_pkt_l4_csum_calc_entry();
485
486 if (pkt->isip4) {
487 if (pkt->isudp) {
488 csl = be16_to_cpu(pkt->l4hdr_info.hdr.udp.uh_ulen);
489 trace_net_rx_pkt_l4_csum_calc_ip4_udp();
490 } else {
491 csl = be16_to_cpu(pkt->ip4hdr_info.ip4_hdr.ip_len) -
492 IP_HDR_GET_LEN(&pkt->ip4hdr_info.ip4_hdr);
493 trace_net_rx_pkt_l4_csum_calc_ip4_tcp();
494 }
495
496 cntr = eth_calc_ip4_pseudo_hdr_csum(&pkt->ip4hdr_info.ip4_hdr,
497 csl, &cso);
498 trace_net_rx_pkt_l4_csum_calc_ph_csum(cntr, csl);
499 } else {
500 if (pkt->isudp) {
501 csl = be16_to_cpu(pkt->l4hdr_info.hdr.udp.uh_ulen);
502 trace_net_rx_pkt_l4_csum_calc_ip6_udp();
503 } else {
504 struct ip6_header *ip6hdr = &pkt->ip6hdr_info.ip6_hdr;
505 size_t full_ip6hdr_len = pkt->l4hdr_off - pkt->l3hdr_off;
506 size_t ip6opts_len = full_ip6hdr_len - sizeof(struct ip6_header);
507
508 csl = be16_to_cpu(ip6hdr->ip6_ctlun.ip6_un1.ip6_un1_plen) -
509 ip6opts_len;
510 trace_net_rx_pkt_l4_csum_calc_ip6_tcp();
511 }
512
513 cntr = eth_calc_ip6_pseudo_hdr_csum(&pkt->ip6hdr_info.ip6_hdr, csl,
514 pkt->ip6hdr_info.l4proto, &cso);
515 trace_net_rx_pkt_l4_csum_calc_ph_csum(cntr, csl);
516 }
517
518 cntr += net_checksum_add_iov(pkt->vec, pkt->vec_len,
519 pkt->l4hdr_off, csl, cso);
520
521 csum = net_checksum_finish_nozero(cntr);
522
523 trace_net_rx_pkt_l4_csum_calc_csum(pkt->l4hdr_off, csl, cntr, csum);
524
525 return csum;
526}
527
528bool net_rx_pkt_validate_l4_csum(struct NetRxPkt *pkt, bool *csum_valid)
529{
530 uint16_t csum;
531
532 trace_net_rx_pkt_l4_csum_validate_entry();
533
534 if (!pkt->istcp && !pkt->isudp) {
535 trace_net_rx_pkt_l4_csum_validate_not_xxp();
536 return false;
537 }
538
539 if (pkt->isudp && (pkt->l4hdr_info.hdr.udp.uh_sum == 0)) {
540 trace_net_rx_pkt_l4_csum_validate_udp_with_no_checksum();
541 return false;
542 }
543
544 if (pkt->isip4 && pkt->ip4hdr_info.fragment) {
545 trace_net_rx_pkt_l4_csum_validate_ip4_fragment();
546 return false;
547 }
548
549 csum = _net_rx_pkt_calc_l4_csum(pkt);
550
551 *csum_valid = ((csum == 0) || (csum == 0xFFFF));
552
553 trace_net_rx_pkt_l4_csum_validate_csum(*csum_valid);
554
555 return true;
556}
557
558bool net_rx_pkt_fix_l4_csum(struct NetRxPkt *pkt)
559{
560 uint16_t csum = 0;
561 uint32_t l4_cso;
562
563 trace_net_rx_pkt_l4_csum_fix_entry();
564
565 if (pkt->istcp) {
566 l4_cso = offsetof(struct tcp_header, th_sum);
567 trace_net_rx_pkt_l4_csum_fix_tcp(l4_cso);
568 } else if (pkt->isudp) {
569 if (pkt->l4hdr_info.hdr.udp.uh_sum == 0) {
570 trace_net_rx_pkt_l4_csum_fix_udp_with_no_checksum();
571 return false;
572 }
573 l4_cso = offsetof(struct udp_header, uh_sum);
574 trace_net_rx_pkt_l4_csum_fix_udp(l4_cso);
575 } else {
576 trace_net_rx_pkt_l4_csum_fix_not_xxp();
577 return false;
578 }
579
580 if (pkt->isip4 && pkt->ip4hdr_info.fragment) {
581 trace_net_rx_pkt_l4_csum_fix_ip4_fragment();
582 return false;
583 }
584
585 /* Set zero to checksum word */
586 iov_from_buf(pkt->vec, pkt->vec_len,
587 pkt->l4hdr_off + l4_cso,
588 &csum, sizeof(csum));
589
590 /* Calculate L4 checksum */
591 csum = cpu_to_be16(_net_rx_pkt_calc_l4_csum(pkt));
592
593 /* Set calculated checksum to checksum word */
594 iov_from_buf(pkt->vec, pkt->vec_len,
595 pkt->l4hdr_off + l4_cso,
596 &csum, sizeof(csum));
597
598 trace_net_rx_pkt_l4_csum_fix_csum(pkt->l4hdr_off + l4_cso, csum);
599
600 return true;
601}
602