2 * QEMU RX packets abstractions
4 * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
6 * Developed by Daynix Computing LTD (http://www.daynix.com)
9 * Dmitry Fleytman <dmitry@daynix.com>
10 * Tamir Shomer <tamirs@daynix.com>
11 * Yan Vugenfirer <yan@daynix.com>
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.
18 #include "qemu/osdep.h"
20 #include "net_rx_pkt.h"
21 #include "net/checksum.h"
25 struct virtio_net_hdr virt_hdr
;
26 uint8_t ehdr_buf
[sizeof(struct eth_header
) + sizeof(struct vlan_header
)];
28 uint16_t vec_len_total
;
34 eth_pkt_types_e packet_type
;
36 /* Analysis results */
46 eth_ip6_hdr_info ip6hdr_info
;
47 eth_ip4_hdr_info ip4hdr_info
;
48 eth_l4_hdr_info l4hdr_info
;
51 void net_rx_pkt_init(struct NetRxPkt
**pkt
, bool has_virt_hdr
)
53 struct NetRxPkt
*p
= g_malloc0(sizeof *p
);
54 p
->has_virt_hdr
= has_virt_hdr
;
60 void net_rx_pkt_uninit(struct NetRxPkt
*pkt
)
62 if (pkt
->vec_len_total
!= 0) {
69 struct virtio_net_hdr
*net_rx_pkt_get_vhdr(struct NetRxPkt
*pkt
)
72 return &pkt
->virt_hdr
;
76 net_rx_pkt_iovec_realloc(struct NetRxPkt
*pkt
,
79 if (pkt
->vec_len_total
< new_iov_len
) {
81 pkt
->vec
= g_malloc(sizeof(*pkt
->vec
) * new_iov_len
);
82 pkt
->vec_len_total
= new_iov_len
;
87 net_rx_pkt_pull_data(struct NetRxPkt
*pkt
,
88 const struct iovec
*iov
, int iovcnt
,
91 uint32_t pllen
= iov_size(iov
, iovcnt
) - ploff
;
93 if (pkt
->ehdr_buf_len
) {
94 net_rx_pkt_iovec_realloc(pkt
, iovcnt
+ 1);
96 pkt
->vec
[0].iov_base
= pkt
->ehdr_buf
;
97 pkt
->vec
[0].iov_len
= pkt
->ehdr_buf_len
;
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;
103 net_rx_pkt_iovec_realloc(pkt
, iovcnt
);
105 pkt
->tot_len
= pllen
;
106 pkt
->vec_len
= iov_copy(pkt
->vec
, pkt
->vec_len_total
,
107 iov
, iovcnt
, ploff
, pkt
->tot_len
);
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
);
115 trace_net_rx_pkt_parsed(pkt
->isip4
, pkt
->isip6
, pkt
->isudp
, pkt
->istcp
,
116 pkt
->l3hdr_off
, pkt
->l4hdr_off
, pkt
->l5hdr_off
);
119 void net_rx_pkt_attach_iovec(struct NetRxPkt
*pkt
,
120 const struct iovec
*iov
, int iovcnt
,
121 size_t iovoff
, bool strip_vlan
)
124 uint16_t ploff
= iovoff
;
128 pkt
->ehdr_buf_len
= eth_strip_vlan(iov
, iovcnt
, iovoff
, pkt
->ehdr_buf
,
131 pkt
->ehdr_buf_len
= 0;
136 net_rx_pkt_pull_data(pkt
, iov
, iovcnt
, ploff
);
139 void net_rx_pkt_attach_iovec_ex(struct NetRxPkt
*pkt
,
140 const struct iovec
*iov
, int iovcnt
,
141 size_t iovoff
, bool strip_vlan
,
145 uint16_t ploff
= iovoff
;
149 pkt
->ehdr_buf_len
= eth_strip_vlan_ex(iov
, iovcnt
, iovoff
, vet
,
153 pkt
->ehdr_buf_len
= 0;
158 net_rx_pkt_pull_data(pkt
, iov
, iovcnt
, ploff
);
161 void net_rx_pkt_dump(struct NetRxPkt
*pkt
)
163 #ifdef NET_RX_PKT_DEBUG
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
);
171 void net_rx_pkt_set_packet_type(struct NetRxPkt
*pkt
,
172 eth_pkt_types_e packet_type
)
176 pkt
->packet_type
= packet_type
;
180 eth_pkt_types_e
net_rx_pkt_get_packet_type(struct NetRxPkt
*pkt
)
184 return pkt
->packet_type
;
187 size_t net_rx_pkt_get_total_len(struct NetRxPkt
*pkt
)
194 void net_rx_pkt_set_protocols(struct NetRxPkt
*pkt
, const void *data
,
197 const struct iovec iov
= {
198 .iov_base
= (void *)data
,
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
);
210 void net_rx_pkt_get_protocols(struct NetRxPkt
*pkt
,
211 bool *isip4
, bool *isip6
,
212 bool *isudp
, bool *istcp
)
222 size_t net_rx_pkt_get_l3_hdr_offset(struct NetRxPkt
*pkt
)
225 return pkt
->l3hdr_off
;
228 size_t net_rx_pkt_get_l4_hdr_offset(struct NetRxPkt
*pkt
)
231 return pkt
->l4hdr_off
;
234 size_t net_rx_pkt_get_l5_hdr_offset(struct NetRxPkt
*pkt
)
237 return pkt
->l5hdr_off
;
240 eth_ip6_hdr_info
*net_rx_pkt_get_ip6_info(struct NetRxPkt
*pkt
)
242 return &pkt
->ip6hdr_info
;
245 eth_ip4_hdr_info
*net_rx_pkt_get_ip4_info(struct NetRxPkt
*pkt
)
247 return &pkt
->ip4hdr_info
;
250 eth_l4_hdr_info
*net_rx_pkt_get_l4_info(struct NetRxPkt
*pkt
)
252 return &pkt
->l4hdr_info
;
256 _net_rx_rss_add_chunk(uint8_t *rss_input
, size_t *bytes_written
,
257 void *ptr
, size_t size
)
259 memcpy(&rss_input
[*bytes_written
], ptr
, size
);
260 trace_net_rx_pkt_rss_add_chunk(ptr
, size
, *bytes_written
);
261 *bytes_written
+= size
;
265 _net_rx_rss_prepare_ip4(uint8_t *rss_input
,
266 struct NetRxPkt
*pkt
,
267 size_t *bytes_written
)
269 struct ip_header
*ip4_hdr
= &pkt
->ip4hdr_info
.ip4_hdr
;
271 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
272 &ip4_hdr
->ip_src
, sizeof(uint32_t));
274 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
275 &ip4_hdr
->ip_dst
, sizeof(uint32_t));
279 _net_rx_rss_prepare_ip6(uint8_t *rss_input
,
280 struct NetRxPkt
*pkt
,
281 bool ipv6ex
, size_t *bytes_written
)
283 eth_ip6_hdr_info
*ip6info
= &pkt
->ip6hdr_info
;
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
));
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
));
297 _net_rx_rss_prepare_tcp(uint8_t *rss_input
,
298 struct NetRxPkt
*pkt
,
299 size_t *bytes_written
)
301 struct tcp_header
*tcphdr
= &pkt
->l4hdr_info
.hdr
.tcp
;
303 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
304 &tcphdr
->th_sport
, sizeof(uint16_t));
306 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
307 &tcphdr
->th_dport
, sizeof(uint16_t));
311 _net_rx_rss_prepare_udp(uint8_t *rss_input
,
312 struct NetRxPkt
*pkt
,
313 size_t *bytes_written
)
315 struct udp_header
*udphdr
= &pkt
->l4hdr_info
.hdr
.udp
;
317 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
318 &udphdr
->uh_sport
, sizeof(uint16_t));
320 _net_rx_rss_add_chunk(rss_input
, bytes_written
,
321 &udphdr
->uh_dport
, sizeof(uint16_t));
325 net_rx_pkt_calc_rss_hash(struct NetRxPkt
*pkt
,
326 NetRxPktRssType type
,
329 uint8_t rss_input
[36];
330 size_t rss_length
= 0;
331 uint32_t rss_hash
= 0;
332 net_toeplitz_key key_data
;
337 trace_net_rx_pkt_rss_ip4();
338 _net_rx_rss_prepare_ip4(&rss_input
[0], pkt
, &rss_length
);
340 case NetPktRssIpV4Tcp
:
343 trace_net_rx_pkt_rss_ip4_tcp();
344 _net_rx_rss_prepare_ip4(&rss_input
[0], pkt
, &rss_length
);
345 _net_rx_rss_prepare_tcp(&rss_input
[0], pkt
, &rss_length
);
347 case NetPktRssIpV6Tcp
:
350 trace_net_rx_pkt_rss_ip6_tcp();
351 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, false, &rss_length
);
352 _net_rx_rss_prepare_tcp(&rss_input
[0], pkt
, &rss_length
);
356 trace_net_rx_pkt_rss_ip6();
357 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, false, &rss_length
);
359 case NetPktRssIpV6Ex
:
361 trace_net_rx_pkt_rss_ip6_ex();
362 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, true, &rss_length
);
364 case NetPktRssIpV6TcpEx
:
367 trace_net_rx_pkt_rss_ip6_ex_tcp();
368 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, true, &rss_length
);
369 _net_rx_rss_prepare_tcp(&rss_input
[0], pkt
, &rss_length
);
371 case NetPktRssIpV4Udp
:
374 trace_net_rx_pkt_rss_ip4_udp();
375 _net_rx_rss_prepare_ip4(&rss_input
[0], pkt
, &rss_length
);
376 _net_rx_rss_prepare_udp(&rss_input
[0], pkt
, &rss_length
);
378 case NetPktRssIpV6Udp
:
381 trace_net_rx_pkt_rss_ip6_udp();
382 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, false, &rss_length
);
383 _net_rx_rss_prepare_udp(&rss_input
[0], pkt
, &rss_length
);
385 case NetPktRssIpV6UdpEx
:
388 trace_net_rx_pkt_rss_ip6_ex_udp();
389 _net_rx_rss_prepare_ip6(&rss_input
[0], pkt
, true, &rss_length
);
390 _net_rx_rss_prepare_udp(&rss_input
[0], pkt
, &rss_length
);
397 net_toeplitz_key_init(&key_data
, key
);
398 net_toeplitz_add(&rss_hash
, rss_input
, rss_length
, &key_data
);
400 trace_net_rx_pkt_rss_hash(rss_length
, rss_hash
);
405 uint16_t net_rx_pkt_get_ip_id(struct NetRxPkt
*pkt
)
410 return be16_to_cpu(pkt
->ip4hdr_info
.ip4_hdr
.ip_id
);
416 bool net_rx_pkt_is_tcp_ack(struct NetRxPkt
*pkt
)
421 return TCP_HEADER_FLAGS(&pkt
->l4hdr_info
.hdr
.tcp
) & TCP_FLAG_ACK
;
427 bool net_rx_pkt_has_tcp_data(struct NetRxPkt
*pkt
)
432 return pkt
->l4hdr_info
.has_tcp_data
;
438 struct iovec
*net_rx_pkt_get_iovec(struct NetRxPkt
*pkt
)
445 uint16_t net_rx_pkt_get_iovec_len(struct NetRxPkt
*pkt
)
452 void net_rx_pkt_set_vhdr(struct NetRxPkt
*pkt
,
453 struct virtio_net_hdr
*vhdr
)
457 memcpy(&pkt
->virt_hdr
, vhdr
, sizeof pkt
->virt_hdr
);
460 void net_rx_pkt_set_vhdr_iovec(struct NetRxPkt
*pkt
,
461 const struct iovec
*iov
, int iovcnt
)
465 iov_to_buf(iov
, iovcnt
, 0, &pkt
->virt_hdr
, sizeof pkt
->virt_hdr
);
468 bool net_rx_pkt_is_vlan_stripped(struct NetRxPkt
*pkt
)
472 return pkt
->ehdr_buf_len
? true : false;
475 bool net_rx_pkt_has_virt_hdr(struct NetRxPkt
*pkt
)
479 return pkt
->has_virt_hdr
;
482 uint16_t net_rx_pkt_get_vlan_tag(struct NetRxPkt
*pkt
)
489 bool net_rx_pkt_validate_l3_csum(struct NetRxPkt
*pkt
, bool *csum_valid
)
495 trace_net_rx_pkt_l3_csum_validate_entry();
498 trace_net_rx_pkt_l3_csum_validate_not_ip4();
502 csl
= pkt
->l4hdr_off
- pkt
->l3hdr_off
;
504 cntr
= net_checksum_add_iov(pkt
->vec
, pkt
->vec_len
,
508 csum
= net_checksum_finish(cntr
);
510 *csum_valid
= (csum
== 0);
512 trace_net_rx_pkt_l3_csum_validate_csum(pkt
->l3hdr_off
, csl
,
513 cntr
, csum
, *csum_valid
);
519 _net_rx_pkt_calc_l4_csum(struct NetRxPkt
*pkt
)
526 trace_net_rx_pkt_l4_csum_calc_entry();
530 csl
= be16_to_cpu(pkt
->l4hdr_info
.hdr
.udp
.uh_ulen
);
531 trace_net_rx_pkt_l4_csum_calc_ip4_udp();
533 csl
= be16_to_cpu(pkt
->ip4hdr_info
.ip4_hdr
.ip_len
) -
534 IP_HDR_GET_LEN(&pkt
->ip4hdr_info
.ip4_hdr
);
535 trace_net_rx_pkt_l4_csum_calc_ip4_tcp();
538 cntr
= eth_calc_ip4_pseudo_hdr_csum(&pkt
->ip4hdr_info
.ip4_hdr
,
540 trace_net_rx_pkt_l4_csum_calc_ph_csum(cntr
, csl
);
543 csl
= be16_to_cpu(pkt
->l4hdr_info
.hdr
.udp
.uh_ulen
);
544 trace_net_rx_pkt_l4_csum_calc_ip6_udp();
546 struct ip6_header
*ip6hdr
= &pkt
->ip6hdr_info
.ip6_hdr
;
547 size_t full_ip6hdr_len
= pkt
->l4hdr_off
- pkt
->l3hdr_off
;
548 size_t ip6opts_len
= full_ip6hdr_len
- sizeof(struct ip6_header
);
550 csl
= be16_to_cpu(ip6hdr
->ip6_ctlun
.ip6_un1
.ip6_un1_plen
) -
552 trace_net_rx_pkt_l4_csum_calc_ip6_tcp();
555 cntr
= eth_calc_ip6_pseudo_hdr_csum(&pkt
->ip6hdr_info
.ip6_hdr
, csl
,
556 pkt
->ip6hdr_info
.l4proto
, &cso
);
557 trace_net_rx_pkt_l4_csum_calc_ph_csum(cntr
, csl
);
560 cntr
+= net_checksum_add_iov(pkt
->vec
, pkt
->vec_len
,
561 pkt
->l4hdr_off
, csl
, cso
);
563 csum
= net_checksum_finish_nozero(cntr
);
565 trace_net_rx_pkt_l4_csum_calc_csum(pkt
->l4hdr_off
, csl
, cntr
, csum
);
570 bool net_rx_pkt_validate_l4_csum(struct NetRxPkt
*pkt
, bool *csum_valid
)
574 trace_net_rx_pkt_l4_csum_validate_entry();
576 if (!pkt
->istcp
&& !pkt
->isudp
) {
577 trace_net_rx_pkt_l4_csum_validate_not_xxp();
581 if (pkt
->isudp
&& (pkt
->l4hdr_info
.hdr
.udp
.uh_sum
== 0)) {
582 trace_net_rx_pkt_l4_csum_validate_udp_with_no_checksum();
586 if (pkt
->isip4
&& pkt
->ip4hdr_info
.fragment
) {
587 trace_net_rx_pkt_l4_csum_validate_ip4_fragment();
591 csum
= _net_rx_pkt_calc_l4_csum(pkt
);
593 *csum_valid
= ((csum
== 0) || (csum
== 0xFFFF));
595 trace_net_rx_pkt_l4_csum_validate_csum(*csum_valid
);
600 bool net_rx_pkt_fix_l4_csum(struct NetRxPkt
*pkt
)
605 trace_net_rx_pkt_l4_csum_fix_entry();
608 l4_cso
= offsetof(struct tcp_header
, th_sum
);
609 trace_net_rx_pkt_l4_csum_fix_tcp(l4_cso
);
610 } else if (pkt
->isudp
) {
611 if (pkt
->l4hdr_info
.hdr
.udp
.uh_sum
== 0) {
612 trace_net_rx_pkt_l4_csum_fix_udp_with_no_checksum();
615 l4_cso
= offsetof(struct udp_header
, uh_sum
);
616 trace_net_rx_pkt_l4_csum_fix_udp(l4_cso
);
618 trace_net_rx_pkt_l4_csum_fix_not_xxp();
622 if (pkt
->isip4
&& pkt
->ip4hdr_info
.fragment
) {
623 trace_net_rx_pkt_l4_csum_fix_ip4_fragment();
627 /* Set zero to checksum word */
628 iov_from_buf(pkt
->vec
, pkt
->vec_len
,
629 pkt
->l4hdr_off
+ l4_cso
,
630 &csum
, sizeof(csum
));
632 /* Calculate L4 checksum */
633 csum
= cpu_to_be16(_net_rx_pkt_calc_l4_csum(pkt
));
635 /* Set calculated checksum to checksum word */
636 iov_from_buf(pkt
->vec
, pkt
->vec_len
,
637 pkt
->l4hdr_off
+ l4_cso
,
638 &csum
, sizeof(csum
));
640 trace_net_rx_pkt_l4_csum_fix_csum(pkt
->l4hdr_off
+ l4_cso
, csum
);