net_pkt: Name vmxnet3 packet abstractions more generic
[qemu.git] / hw / net / net_tx_pkt.c
1 /*
2 * QEMU TX 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 "hw/hw.h"
20 #include "net_tx_pkt.h"
21 #include "net/eth.h"
22 #include "qemu-common.h"
23 #include "qemu/iov.h"
24 #include "net/checksum.h"
25 #include "net/tap.h"
26 #include "net/net.h"
27
28 enum {
29 NET_TX_PKT_VHDR_FRAG = 0,
30 NET_TX_PKT_L2HDR_FRAG,
31 NET_TX_PKT_L3HDR_FRAG,
32 NET_TX_PKT_PL_START_FRAG
33 };
34
35 /* TX packet private context */
36 struct NetTxPkt {
37 struct virtio_net_hdr virt_hdr;
38 bool has_virt_hdr;
39
40 struct iovec *raw;
41 uint32_t raw_frags;
42 uint32_t max_raw_frags;
43
44 struct iovec *vec;
45
46 uint8_t l2_hdr[ETH_MAX_L2_HDR_LEN];
47
48 uint32_t payload_len;
49
50 uint32_t payload_frags;
51 uint32_t max_payload_frags;
52
53 uint16_t hdr_len;
54 eth_pkt_types_e packet_type;
55 uint8_t l4proto;
56 };
57
58 void net_tx_pkt_init(struct NetTxPkt **pkt, uint32_t max_frags,
59 bool has_virt_hdr)
60 {
61 struct NetTxPkt *p = g_malloc0(sizeof *p);
62
63 p->vec = g_malloc((sizeof *p->vec) *
64 (max_frags + NET_TX_PKT_PL_START_FRAG));
65
66 p->raw = g_malloc((sizeof *p->raw) * max_frags);
67
68 p->max_payload_frags = max_frags;
69 p->max_raw_frags = max_frags;
70 p->has_virt_hdr = has_virt_hdr;
71 p->vec[NET_TX_PKT_VHDR_FRAG].iov_base = &p->virt_hdr;
72 p->vec[NET_TX_PKT_VHDR_FRAG].iov_len =
73 p->has_virt_hdr ? sizeof p->virt_hdr : 0;
74 p->vec[NET_TX_PKT_L2HDR_FRAG].iov_base = &p->l2_hdr;
75 p->vec[NET_TX_PKT_L3HDR_FRAG].iov_base = NULL;
76 p->vec[NET_TX_PKT_L3HDR_FRAG].iov_len = 0;
77
78 *pkt = p;
79 }
80
81 void net_tx_pkt_uninit(struct NetTxPkt *pkt)
82 {
83 if (pkt) {
84 g_free(pkt->vec);
85 g_free(pkt->raw);
86 g_free(pkt);
87 }
88 }
89
90 void net_tx_pkt_update_ip_checksums(struct NetTxPkt *pkt)
91 {
92 uint16_t csum;
93 uint32_t ph_raw_csum;
94 assert(pkt);
95 uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
96 struct ip_header *ip_hdr;
97
98 if (VIRTIO_NET_HDR_GSO_TCPV4 != gso_type &&
99 VIRTIO_NET_HDR_GSO_UDP != gso_type) {
100 return;
101 }
102
103 ip_hdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
104
105 if (pkt->payload_len + pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len >
106 ETH_MAX_IP_DGRAM_LEN) {
107 return;
108 }
109
110 ip_hdr->ip_len = cpu_to_be16(pkt->payload_len +
111 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
112
113 /* Calculate IP header checksum */
114 ip_hdr->ip_sum = 0;
115 csum = net_raw_checksum((uint8_t *)ip_hdr,
116 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
117 ip_hdr->ip_sum = cpu_to_be16(csum);
118
119 /* Calculate IP pseudo header checksum */
120 ph_raw_csum = eth_calc_pseudo_hdr_csum(ip_hdr, pkt->payload_len);
121 csum = cpu_to_be16(~net_checksum_finish(ph_raw_csum));
122 iov_from_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
123 pkt->virt_hdr.csum_offset, &csum, sizeof(csum));
124 }
125
126 static void net_tx_pkt_calculate_hdr_len(struct NetTxPkt *pkt)
127 {
128 pkt->hdr_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +
129 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
130 }
131
132 static bool net_tx_pkt_parse_headers(struct NetTxPkt *pkt)
133 {
134 struct iovec *l2_hdr, *l3_hdr;
135 size_t bytes_read;
136 size_t full_ip6hdr_len;
137 uint16_t l3_proto;
138
139 assert(pkt);
140
141 l2_hdr = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
142 l3_hdr = &pkt->vec[NET_TX_PKT_L3HDR_FRAG];
143
144 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, 0, l2_hdr->iov_base,
145 ETH_MAX_L2_HDR_LEN);
146 if (bytes_read < sizeof(struct eth_header)) {
147 l2_hdr->iov_len = 0;
148 return false;
149 }
150
151 l2_hdr->iov_len = sizeof(struct eth_header);
152 switch (be16_to_cpu(PKT_GET_ETH_HDR(l2_hdr->iov_base)->h_proto)) {
153 case ETH_P_VLAN:
154 l2_hdr->iov_len += sizeof(struct vlan_header);
155 break;
156 case ETH_P_DVLAN:
157 l2_hdr->iov_len += 2 * sizeof(struct vlan_header);
158 break;
159 }
160
161 if (bytes_read < l2_hdr->iov_len) {
162 l2_hdr->iov_len = 0;
163 return false;
164 }
165
166 l3_proto = eth_get_l3_proto(l2_hdr->iov_base, l2_hdr->iov_len);
167
168 switch (l3_proto) {
169 case ETH_P_IP:
170 l3_hdr->iov_base = g_malloc(ETH_MAX_IP4_HDR_LEN);
171
172 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
173 l3_hdr->iov_base, sizeof(struct ip_header));
174
175 if (bytes_read < sizeof(struct ip_header)) {
176 l3_hdr->iov_len = 0;
177 return false;
178 }
179
180 l3_hdr->iov_len = IP_HDR_GET_LEN(l3_hdr->iov_base);
181 pkt->l4proto = ((struct ip_header *) l3_hdr->iov_base)->ip_p;
182
183 /* copy optional IPv4 header data */
184 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags,
185 l2_hdr->iov_len + sizeof(struct ip_header),
186 l3_hdr->iov_base + sizeof(struct ip_header),
187 l3_hdr->iov_len - sizeof(struct ip_header));
188 if (bytes_read < l3_hdr->iov_len - sizeof(struct ip_header)) {
189 l3_hdr->iov_len = 0;
190 return false;
191 }
192 break;
193
194 case ETH_P_IPV6:
195 if (!eth_parse_ipv6_hdr(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
196 &pkt->l4proto, &full_ip6hdr_len)) {
197 l3_hdr->iov_len = 0;
198 return false;
199 }
200
201 l3_hdr->iov_base = g_malloc(full_ip6hdr_len);
202
203 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
204 l3_hdr->iov_base, full_ip6hdr_len);
205
206 if (bytes_read < full_ip6hdr_len) {
207 l3_hdr->iov_len = 0;
208 return false;
209 } else {
210 l3_hdr->iov_len = full_ip6hdr_len;
211 }
212 break;
213
214 default:
215 l3_hdr->iov_len = 0;
216 break;
217 }
218
219 net_tx_pkt_calculate_hdr_len(pkt);
220 pkt->packet_type = get_eth_packet_type(l2_hdr->iov_base);
221 return true;
222 }
223
224 static bool net_tx_pkt_rebuild_payload(struct NetTxPkt *pkt)
225 {
226 size_t payload_len = iov_size(pkt->raw, pkt->raw_frags) - pkt->hdr_len;
227
228 pkt->payload_frags = iov_copy(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
229 pkt->max_payload_frags,
230 pkt->raw, pkt->raw_frags,
231 pkt->hdr_len, payload_len);
232
233 if (pkt->payload_frags != (uint32_t) -1) {
234 pkt->payload_len = payload_len;
235 return true;
236 } else {
237 return false;
238 }
239 }
240
241 bool net_tx_pkt_parse(struct NetTxPkt *pkt)
242 {
243 return net_tx_pkt_parse_headers(pkt) &&
244 net_tx_pkt_rebuild_payload(pkt);
245 }
246
247 struct virtio_net_hdr *net_tx_pkt_get_vhdr(struct NetTxPkt *pkt)
248 {
249 assert(pkt);
250 return &pkt->virt_hdr;
251 }
252
253 static uint8_t net_tx_pkt_get_gso_type(struct NetTxPkt *pkt,
254 bool tso_enable)
255 {
256 uint8_t rc = VIRTIO_NET_HDR_GSO_NONE;
257 uint16_t l3_proto;
258
259 l3_proto = eth_get_l3_proto(pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base,
260 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len);
261
262 if (!tso_enable) {
263 goto func_exit;
264 }
265
266 rc = eth_get_gso_type(l3_proto, pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
267 pkt->l4proto);
268
269 func_exit:
270 return rc;
271 }
272
273 void net_tx_pkt_build_vheader(struct NetTxPkt *pkt, bool tso_enable,
274 bool csum_enable, uint32_t gso_size)
275 {
276 struct tcp_hdr l4hdr;
277 assert(pkt);
278
279 /* csum has to be enabled if tso is. */
280 assert(csum_enable || !tso_enable);
281
282 pkt->virt_hdr.gso_type = net_tx_pkt_get_gso_type(pkt, tso_enable);
283
284 switch (pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
285 case VIRTIO_NET_HDR_GSO_NONE:
286 pkt->virt_hdr.hdr_len = 0;
287 pkt->virt_hdr.gso_size = 0;
288 break;
289
290 case VIRTIO_NET_HDR_GSO_UDP:
291 pkt->virt_hdr.gso_size = IP_FRAG_ALIGN_SIZE(gso_size);
292 pkt->virt_hdr.hdr_len = pkt->hdr_len + sizeof(struct udp_header);
293 break;
294
295 case VIRTIO_NET_HDR_GSO_TCPV4:
296 case VIRTIO_NET_HDR_GSO_TCPV6:
297 iov_to_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
298 0, &l4hdr, sizeof(l4hdr));
299 pkt->virt_hdr.hdr_len = pkt->hdr_len + l4hdr.th_off * sizeof(uint32_t);
300 pkt->virt_hdr.gso_size = IP_FRAG_ALIGN_SIZE(gso_size);
301 break;
302
303 default:
304 g_assert_not_reached();
305 }
306
307 if (csum_enable) {
308 switch (pkt->l4proto) {
309 case IP_PROTO_TCP:
310 pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
311 pkt->virt_hdr.csum_start = pkt->hdr_len;
312 pkt->virt_hdr.csum_offset = offsetof(struct tcp_hdr, th_sum);
313 break;
314 case IP_PROTO_UDP:
315 pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
316 pkt->virt_hdr.csum_start = pkt->hdr_len;
317 pkt->virt_hdr.csum_offset = offsetof(struct udp_hdr, uh_sum);
318 break;
319 default:
320 break;
321 }
322 }
323 }
324
325 void net_tx_pkt_setup_vlan_header(struct NetTxPkt *pkt, uint16_t vlan)
326 {
327 bool is_new;
328 assert(pkt);
329
330 eth_setup_vlan_headers(pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base,
331 vlan, &is_new);
332
333 /* update l2hdrlen */
334 if (is_new) {
335 pkt->hdr_len += sizeof(struct vlan_header);
336 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +=
337 sizeof(struct vlan_header);
338 }
339 }
340
341 bool net_tx_pkt_add_raw_fragment(struct NetTxPkt *pkt, hwaddr pa,
342 size_t len)
343 {
344 hwaddr mapped_len = 0;
345 struct iovec *ventry;
346 assert(pkt);
347 assert(pkt->max_raw_frags > pkt->raw_frags);
348
349 if (!len) {
350 return true;
351 }
352
353 ventry = &pkt->raw[pkt->raw_frags];
354 mapped_len = len;
355
356 ventry->iov_base = cpu_physical_memory_map(pa, &mapped_len, false);
357 ventry->iov_len = mapped_len;
358 pkt->raw_frags += !!ventry->iov_base;
359
360 if ((ventry->iov_base == NULL) || (len != mapped_len)) {
361 return false;
362 }
363
364 return true;
365 }
366
367 eth_pkt_types_e net_tx_pkt_get_packet_type(struct NetTxPkt *pkt)
368 {
369 assert(pkt);
370
371 return pkt->packet_type;
372 }
373
374 size_t net_tx_pkt_get_total_len(struct NetTxPkt *pkt)
375 {
376 assert(pkt);
377
378 return pkt->hdr_len + pkt->payload_len;
379 }
380
381 void net_tx_pkt_dump(struct NetTxPkt *pkt)
382 {
383 #ifdef NET_TX_PKT_DEBUG
384 assert(pkt);
385
386 printf("TX PKT: hdr_len: %d, pkt_type: 0x%X, l2hdr_len: %lu, "
387 "l3hdr_len: %lu, payload_len: %u\n", pkt->hdr_len, pkt->packet_type,
388 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
389 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len, pkt->payload_len);
390 #endif
391 }
392
393 void net_tx_pkt_reset(struct NetTxPkt *pkt)
394 {
395 int i;
396
397 /* no assert, as reset can be called before tx_pkt_init */
398 if (!pkt) {
399 return;
400 }
401
402 memset(&pkt->virt_hdr, 0, sizeof(pkt->virt_hdr));
403
404 g_free(pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base);
405 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base = NULL;
406
407 assert(pkt->vec);
408 for (i = NET_TX_PKT_L2HDR_FRAG;
409 i < pkt->payload_frags + NET_TX_PKT_PL_START_FRAG; i++) {
410 pkt->vec[i].iov_len = 0;
411 }
412 pkt->payload_len = 0;
413 pkt->payload_frags = 0;
414
415 assert(pkt->raw);
416 for (i = 0; i < pkt->raw_frags; i++) {
417 assert(pkt->raw[i].iov_base);
418 cpu_physical_memory_unmap(pkt->raw[i].iov_base, pkt->raw[i].iov_len,
419 false, pkt->raw[i].iov_len);
420 pkt->raw[i].iov_len = 0;
421 }
422 pkt->raw_frags = 0;
423
424 pkt->hdr_len = 0;
425 pkt->packet_type = 0;
426 pkt->l4proto = 0;
427 }
428
429 static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt)
430 {
431 struct iovec *iov = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
432 uint32_t csum_cntr;
433 uint16_t csum = 0;
434 /* num of iovec without vhdr */
435 uint32_t iov_len = pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1;
436 uint16_t csl;
437 struct ip_header *iphdr;
438 size_t csum_offset = pkt->virt_hdr.csum_start + pkt->virt_hdr.csum_offset;
439
440 /* Put zero to checksum field */
441 iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
442
443 /* Calculate L4 TCP/UDP checksum */
444 csl = pkt->payload_len;
445
446 /* data checksum */
447 csum_cntr =
448 net_checksum_add_iov(iov, iov_len, pkt->virt_hdr.csum_start, csl);
449 /* add pseudo header to csum */
450 iphdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
451 csum_cntr += eth_calc_pseudo_hdr_csum(iphdr, csl);
452
453 /* Put the checksum obtained into the packet */
454 csum = cpu_to_be16(net_checksum_finish(csum_cntr));
455 iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
456 }
457
458 enum {
459 NET_TX_PKT_FRAGMENT_L2_HDR_POS = 0,
460 NET_TX_PKT_FRAGMENT_L3_HDR_POS,
461 NET_TX_PKT_FRAGMENT_HEADER_NUM
462 };
463
464 #define NET_MAX_FRAG_SG_LIST (64)
465
466 static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt,
467 int *src_idx, size_t *src_offset, struct iovec *dst, int *dst_idx)
468 {
469 size_t fetched = 0;
470 struct iovec *src = pkt->vec;
471
472 *dst_idx = NET_TX_PKT_FRAGMENT_HEADER_NUM;
473
474 while (fetched < pkt->virt_hdr.gso_size) {
475
476 /* no more place in fragment iov */
477 if (*dst_idx == NET_MAX_FRAG_SG_LIST) {
478 break;
479 }
480
481 /* no more data in iovec */
482 if (*src_idx == (pkt->payload_frags + NET_TX_PKT_PL_START_FRAG)) {
483 break;
484 }
485
486
487 dst[*dst_idx].iov_base = src[*src_idx].iov_base + *src_offset;
488 dst[*dst_idx].iov_len = MIN(src[*src_idx].iov_len - *src_offset,
489 pkt->virt_hdr.gso_size - fetched);
490
491 *src_offset += dst[*dst_idx].iov_len;
492 fetched += dst[*dst_idx].iov_len;
493
494 if (*src_offset == src[*src_idx].iov_len) {
495 *src_offset = 0;
496 (*src_idx)++;
497 }
498
499 (*dst_idx)++;
500 }
501
502 return fetched;
503 }
504
505 static bool net_tx_pkt_do_sw_fragmentation(struct NetTxPkt *pkt,
506 NetClientState *nc)
507 {
508 struct iovec fragment[NET_MAX_FRAG_SG_LIST];
509 size_t fragment_len = 0;
510 bool more_frags = false;
511
512 /* some pointers for shorter code */
513 void *l2_iov_base, *l3_iov_base;
514 size_t l2_iov_len, l3_iov_len;
515 int src_idx = NET_TX_PKT_PL_START_FRAG, dst_idx;
516 size_t src_offset = 0;
517 size_t fragment_offset = 0;
518
519 l2_iov_base = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base;
520 l2_iov_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len;
521 l3_iov_base = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
522 l3_iov_len = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
523
524 /* Copy headers */
525 fragment[NET_TX_PKT_FRAGMENT_L2_HDR_POS].iov_base = l2_iov_base;
526 fragment[NET_TX_PKT_FRAGMENT_L2_HDR_POS].iov_len = l2_iov_len;
527 fragment[NET_TX_PKT_FRAGMENT_L3_HDR_POS].iov_base = l3_iov_base;
528 fragment[NET_TX_PKT_FRAGMENT_L3_HDR_POS].iov_len = l3_iov_len;
529
530
531 /* Put as much data as possible and send */
532 do {
533 fragment_len = net_tx_pkt_fetch_fragment(pkt, &src_idx, &src_offset,
534 fragment, &dst_idx);
535
536 more_frags = (fragment_offset + fragment_len < pkt->payload_len);
537
538 eth_setup_ip4_fragmentation(l2_iov_base, l2_iov_len, l3_iov_base,
539 l3_iov_len, fragment_len, fragment_offset, more_frags);
540
541 eth_fix_ip4_checksum(l3_iov_base, l3_iov_len);
542
543 qemu_sendv_packet(nc, fragment, dst_idx);
544
545 fragment_offset += fragment_len;
546
547 } while (more_frags);
548
549 return true;
550 }
551
552 bool net_tx_pkt_send(struct NetTxPkt *pkt, NetClientState *nc)
553 {
554 assert(pkt);
555
556 if (!pkt->has_virt_hdr &&
557 pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
558 net_tx_pkt_do_sw_csum(pkt);
559 }
560
561 /*
562 * Since underlying infrastructure does not support IP datagrams longer
563 * than 64K we should drop such packets and don't even try to send
564 */
565 if (VIRTIO_NET_HDR_GSO_NONE != pkt->virt_hdr.gso_type) {
566 if (pkt->payload_len >
567 ETH_MAX_IP_DGRAM_LEN -
568 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len) {
569 return false;
570 }
571 }
572
573 if (pkt->has_virt_hdr ||
574 pkt->virt_hdr.gso_type == VIRTIO_NET_HDR_GSO_NONE) {
575 qemu_sendv_packet(nc, pkt->vec,
576 pkt->payload_frags + NET_TX_PKT_PL_START_FRAG);
577 return true;
578 }
579
580 return net_tx_pkt_do_sw_fragmentation(pkt, nc);
581 }