monitor/qmp: Update comment for commit 4eaca8de268
[qemu/armbru.git] / hw / net / net_tx_pkt.c
blob162f802dd77e09b89c0cb65583e82ddfcc1d3487
1 /*
2 * QEMU TX packets abstractions
4 * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
6 * Developed by Daynix Computing LTD (http://www.daynix.com)
8 * Authors:
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"
19 #include "net_tx_pkt.h"
20 #include "net/eth.h"
21 #include "net/checksum.h"
22 #include "net/tap.h"
23 #include "net/net.h"
24 #include "hw/pci/pci.h"
26 enum {
27 NET_TX_PKT_VHDR_FRAG = 0,
28 NET_TX_PKT_L2HDR_FRAG,
29 NET_TX_PKT_L3HDR_FRAG,
30 NET_TX_PKT_PL_START_FRAG
33 /* TX packet private context */
34 struct NetTxPkt {
35 PCIDevice *pci_dev;
37 struct virtio_net_hdr virt_hdr;
38 bool has_virt_hdr;
40 struct iovec *raw;
41 uint32_t raw_frags;
42 uint32_t max_raw_frags;
44 struct iovec *vec;
46 uint8_t l2_hdr[ETH_MAX_L2_HDR_LEN];
47 uint8_t l3_hdr[ETH_MAX_IP_DGRAM_LEN];
49 uint32_t payload_len;
51 uint32_t payload_frags;
52 uint32_t max_payload_frags;
54 uint16_t hdr_len;
55 eth_pkt_types_e packet_type;
56 uint8_t l4proto;
58 bool is_loopback;
61 void net_tx_pkt_init(struct NetTxPkt **pkt, PCIDevice *pci_dev,
62 uint32_t max_frags, bool has_virt_hdr)
64 struct NetTxPkt *p = g_malloc0(sizeof *p);
66 p->pci_dev = pci_dev;
68 p->vec = g_new(struct iovec, max_frags + NET_TX_PKT_PL_START_FRAG);
70 p->raw = g_new(struct iovec, max_frags);
72 p->max_payload_frags = max_frags;
73 p->max_raw_frags = max_frags;
74 p->has_virt_hdr = has_virt_hdr;
75 p->vec[NET_TX_PKT_VHDR_FRAG].iov_base = &p->virt_hdr;
76 p->vec[NET_TX_PKT_VHDR_FRAG].iov_len =
77 p->has_virt_hdr ? sizeof p->virt_hdr : 0;
78 p->vec[NET_TX_PKT_L2HDR_FRAG].iov_base = &p->l2_hdr;
79 p->vec[NET_TX_PKT_L3HDR_FRAG].iov_base = &p->l3_hdr;
81 *pkt = p;
84 void net_tx_pkt_uninit(struct NetTxPkt *pkt)
86 if (pkt) {
87 g_free(pkt->vec);
88 g_free(pkt->raw);
89 g_free(pkt);
93 void net_tx_pkt_update_ip_hdr_checksum(struct NetTxPkt *pkt)
95 uint16_t csum;
96 assert(pkt);
97 struct ip_header *ip_hdr;
98 ip_hdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
100 ip_hdr->ip_len = cpu_to_be16(pkt->payload_len +
101 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
103 ip_hdr->ip_sum = 0;
104 csum = net_raw_checksum((uint8_t *)ip_hdr,
105 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len);
106 ip_hdr->ip_sum = cpu_to_be16(csum);
109 void net_tx_pkt_update_ip_checksums(struct NetTxPkt *pkt)
111 uint16_t csum;
112 uint32_t cntr, cso;
113 assert(pkt);
114 uint8_t gso_type = pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN;
115 void *ip_hdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
117 if (pkt->payload_len + pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len >
118 ETH_MAX_IP_DGRAM_LEN) {
119 return;
122 if (gso_type == VIRTIO_NET_HDR_GSO_TCPV4 ||
123 gso_type == VIRTIO_NET_HDR_GSO_UDP) {
124 /* Calculate IP header checksum */
125 net_tx_pkt_update_ip_hdr_checksum(pkt);
127 /* Calculate IP pseudo header checksum */
128 cntr = eth_calc_ip4_pseudo_hdr_csum(ip_hdr, pkt->payload_len, &cso);
129 csum = cpu_to_be16(~net_checksum_finish(cntr));
130 } else if (gso_type == VIRTIO_NET_HDR_GSO_TCPV6) {
131 /* Calculate IP pseudo header checksum */
132 cntr = eth_calc_ip6_pseudo_hdr_csum(ip_hdr, pkt->payload_len,
133 IP_PROTO_TCP, &cso);
134 csum = cpu_to_be16(~net_checksum_finish(cntr));
135 } else {
136 return;
139 iov_from_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
140 pkt->virt_hdr.csum_offset, &csum, sizeof(csum));
143 static void net_tx_pkt_calculate_hdr_len(struct NetTxPkt *pkt)
145 pkt->hdr_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +
146 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
149 static bool net_tx_pkt_parse_headers(struct NetTxPkt *pkt)
151 struct iovec *l2_hdr, *l3_hdr;
152 size_t bytes_read;
153 size_t full_ip6hdr_len;
154 uint16_t l3_proto;
156 assert(pkt);
158 l2_hdr = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
159 l3_hdr = &pkt->vec[NET_TX_PKT_L3HDR_FRAG];
161 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, 0, l2_hdr->iov_base,
162 ETH_MAX_L2_HDR_LEN);
163 if (bytes_read < sizeof(struct eth_header)) {
164 l2_hdr->iov_len = 0;
165 return false;
168 l2_hdr->iov_len = sizeof(struct eth_header);
169 switch (be16_to_cpu(PKT_GET_ETH_HDR(l2_hdr->iov_base)->h_proto)) {
170 case ETH_P_VLAN:
171 l2_hdr->iov_len += sizeof(struct vlan_header);
172 break;
173 case ETH_P_DVLAN:
174 l2_hdr->iov_len += 2 * sizeof(struct vlan_header);
175 break;
178 if (bytes_read < l2_hdr->iov_len) {
179 l2_hdr->iov_len = 0;
180 l3_hdr->iov_len = 0;
181 pkt->packet_type = ETH_PKT_UCAST;
182 return false;
183 } else {
184 l2_hdr->iov_len = ETH_MAX_L2_HDR_LEN;
185 l2_hdr->iov_len = eth_get_l2_hdr_length(l2_hdr->iov_base);
186 pkt->packet_type = get_eth_packet_type(l2_hdr->iov_base);
189 l3_proto = eth_get_l3_proto(l2_hdr, 1, l2_hdr->iov_len);
191 switch (l3_proto) {
192 case ETH_P_IP:
193 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
194 l3_hdr->iov_base, sizeof(struct ip_header));
196 if (bytes_read < sizeof(struct ip_header)) {
197 l3_hdr->iov_len = 0;
198 return false;
201 l3_hdr->iov_len = IP_HDR_GET_LEN(l3_hdr->iov_base);
203 if (l3_hdr->iov_len < sizeof(struct ip_header)) {
204 l3_hdr->iov_len = 0;
205 return false;
208 pkt->l4proto = IP_HDR_GET_P(l3_hdr->iov_base);
210 if (IP_HDR_GET_LEN(l3_hdr->iov_base) != sizeof(struct ip_header)) {
211 /* copy optional IPv4 header data if any*/
212 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags,
213 l2_hdr->iov_len + sizeof(struct ip_header),
214 l3_hdr->iov_base + sizeof(struct ip_header),
215 l3_hdr->iov_len - sizeof(struct ip_header));
216 if (bytes_read < l3_hdr->iov_len - sizeof(struct ip_header)) {
217 l3_hdr->iov_len = 0;
218 return false;
222 break;
224 case ETH_P_IPV6:
226 eth_ip6_hdr_info hdrinfo;
228 if (!eth_parse_ipv6_hdr(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
229 &hdrinfo)) {
230 l3_hdr->iov_len = 0;
231 return false;
234 pkt->l4proto = hdrinfo.l4proto;
235 full_ip6hdr_len = hdrinfo.full_hdr_len;
237 if (full_ip6hdr_len > ETH_MAX_IP_DGRAM_LEN) {
238 l3_hdr->iov_len = 0;
239 return false;
242 bytes_read = iov_to_buf(pkt->raw, pkt->raw_frags, l2_hdr->iov_len,
243 l3_hdr->iov_base, full_ip6hdr_len);
245 if (bytes_read < full_ip6hdr_len) {
246 l3_hdr->iov_len = 0;
247 return false;
248 } else {
249 l3_hdr->iov_len = full_ip6hdr_len;
251 break;
253 default:
254 l3_hdr->iov_len = 0;
255 break;
258 net_tx_pkt_calculate_hdr_len(pkt);
259 return true;
262 static void net_tx_pkt_rebuild_payload(struct NetTxPkt *pkt)
264 pkt->payload_len = iov_size(pkt->raw, pkt->raw_frags) - pkt->hdr_len;
265 pkt->payload_frags = iov_copy(&pkt->vec[NET_TX_PKT_PL_START_FRAG],
266 pkt->max_payload_frags,
267 pkt->raw, pkt->raw_frags,
268 pkt->hdr_len, pkt->payload_len);
271 bool net_tx_pkt_parse(struct NetTxPkt *pkt)
273 if (net_tx_pkt_parse_headers(pkt)) {
274 net_tx_pkt_rebuild_payload(pkt);
275 return true;
276 } else {
277 return false;
281 struct virtio_net_hdr *net_tx_pkt_get_vhdr(struct NetTxPkt *pkt)
283 assert(pkt);
284 return &pkt->virt_hdr;
287 static uint8_t net_tx_pkt_get_gso_type(struct NetTxPkt *pkt,
288 bool tso_enable)
290 uint8_t rc = VIRTIO_NET_HDR_GSO_NONE;
291 uint16_t l3_proto;
293 l3_proto = eth_get_l3_proto(&pkt->vec[NET_TX_PKT_L2HDR_FRAG], 1,
294 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len);
296 if (!tso_enable) {
297 goto func_exit;
300 rc = eth_get_gso_type(l3_proto, pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base,
301 pkt->l4proto);
303 func_exit:
304 return rc;
307 void net_tx_pkt_build_vheader(struct NetTxPkt *pkt, bool tso_enable,
308 bool csum_enable, uint32_t gso_size)
310 struct tcp_hdr l4hdr;
311 assert(pkt);
313 /* csum has to be enabled if tso is. */
314 assert(csum_enable || !tso_enable);
316 pkt->virt_hdr.gso_type = net_tx_pkt_get_gso_type(pkt, tso_enable);
318 switch (pkt->virt_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
319 case VIRTIO_NET_HDR_GSO_NONE:
320 pkt->virt_hdr.hdr_len = 0;
321 pkt->virt_hdr.gso_size = 0;
322 break;
324 case VIRTIO_NET_HDR_GSO_UDP:
325 pkt->virt_hdr.gso_size = gso_size;
326 pkt->virt_hdr.hdr_len = pkt->hdr_len + sizeof(struct udp_header);
327 break;
329 case VIRTIO_NET_HDR_GSO_TCPV4:
330 case VIRTIO_NET_HDR_GSO_TCPV6:
331 iov_to_buf(&pkt->vec[NET_TX_PKT_PL_START_FRAG], pkt->payload_frags,
332 0, &l4hdr, sizeof(l4hdr));
333 pkt->virt_hdr.hdr_len = pkt->hdr_len + l4hdr.th_off * sizeof(uint32_t);
334 pkt->virt_hdr.gso_size = gso_size;
335 break;
337 default:
338 g_assert_not_reached();
341 if (csum_enable) {
342 switch (pkt->l4proto) {
343 case IP_PROTO_TCP:
344 pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
345 pkt->virt_hdr.csum_start = pkt->hdr_len;
346 pkt->virt_hdr.csum_offset = offsetof(struct tcp_hdr, th_sum);
347 break;
348 case IP_PROTO_UDP:
349 pkt->virt_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
350 pkt->virt_hdr.csum_start = pkt->hdr_len;
351 pkt->virt_hdr.csum_offset = offsetof(struct udp_hdr, uh_sum);
352 break;
353 default:
354 break;
359 void net_tx_pkt_setup_vlan_header_ex(struct NetTxPkt *pkt,
360 uint16_t vlan, uint16_t vlan_ethtype)
362 bool is_new;
363 assert(pkt);
365 eth_setup_vlan_headers_ex(pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base,
366 vlan, vlan_ethtype, &is_new);
368 /* update l2hdrlen */
369 if (is_new) {
370 pkt->hdr_len += sizeof(struct vlan_header);
371 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len +=
372 sizeof(struct vlan_header);
376 bool net_tx_pkt_add_raw_fragment(struct NetTxPkt *pkt, hwaddr pa,
377 size_t len)
379 hwaddr mapped_len = 0;
380 struct iovec *ventry;
381 assert(pkt);
382 assert(pkt->max_raw_frags > pkt->raw_frags);
384 if (!len) {
385 return true;
388 ventry = &pkt->raw[pkt->raw_frags];
389 mapped_len = len;
391 ventry->iov_base = pci_dma_map(pkt->pci_dev, pa,
392 &mapped_len, DMA_DIRECTION_TO_DEVICE);
394 if ((ventry->iov_base != NULL) && (len == mapped_len)) {
395 ventry->iov_len = mapped_len;
396 pkt->raw_frags++;
397 return true;
398 } else {
399 return false;
403 bool net_tx_pkt_has_fragments(struct NetTxPkt *pkt)
405 return pkt->raw_frags > 0;
408 eth_pkt_types_e net_tx_pkt_get_packet_type(struct NetTxPkt *pkt)
410 assert(pkt);
412 return pkt->packet_type;
415 size_t net_tx_pkt_get_total_len(struct NetTxPkt *pkt)
417 assert(pkt);
419 return pkt->hdr_len + pkt->payload_len;
422 void net_tx_pkt_dump(struct NetTxPkt *pkt)
424 #ifdef NET_TX_PKT_DEBUG
425 assert(pkt);
427 printf("TX PKT: hdr_len: %d, pkt_type: 0x%X, l2hdr_len: %lu, "
428 "l3hdr_len: %lu, payload_len: %u\n", pkt->hdr_len, pkt->packet_type,
429 pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len,
430 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len, pkt->payload_len);
431 #endif
434 void net_tx_pkt_reset(struct NetTxPkt *pkt)
436 int i;
438 /* no assert, as reset can be called before tx_pkt_init */
439 if (!pkt) {
440 return;
443 memset(&pkt->virt_hdr, 0, sizeof(pkt->virt_hdr));
445 assert(pkt->vec);
447 pkt->payload_len = 0;
448 pkt->payload_frags = 0;
450 assert(pkt->raw);
451 for (i = 0; i < pkt->raw_frags; i++) {
452 assert(pkt->raw[i].iov_base);
453 pci_dma_unmap(pkt->pci_dev, pkt->raw[i].iov_base, pkt->raw[i].iov_len,
454 DMA_DIRECTION_TO_DEVICE, 0);
456 pkt->raw_frags = 0;
458 pkt->hdr_len = 0;
459 pkt->l4proto = 0;
462 static void net_tx_pkt_do_sw_csum(struct NetTxPkt *pkt)
464 struct iovec *iov = &pkt->vec[NET_TX_PKT_L2HDR_FRAG];
465 uint32_t csum_cntr;
466 uint16_t csum = 0;
467 uint32_t cso;
468 /* num of iovec without vhdr */
469 uint32_t iov_len = pkt->payload_frags + NET_TX_PKT_PL_START_FRAG - 1;
470 uint16_t csl;
471 struct ip_header *iphdr;
472 size_t csum_offset = pkt->virt_hdr.csum_start + pkt->virt_hdr.csum_offset;
474 /* Put zero to checksum field */
475 iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
477 /* Calculate L4 TCP/UDP checksum */
478 csl = pkt->payload_len;
480 /* add pseudo header to csum */
481 iphdr = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
482 csum_cntr = eth_calc_ip4_pseudo_hdr_csum(iphdr, csl, &cso);
484 /* data checksum */
485 csum_cntr +=
486 net_checksum_add_iov(iov, iov_len, pkt->virt_hdr.csum_start, csl, cso);
488 /* Put the checksum obtained into the packet */
489 csum = cpu_to_be16(net_checksum_finish_nozero(csum_cntr));
490 iov_from_buf(iov, iov_len, csum_offset, &csum, sizeof csum);
493 enum {
494 NET_TX_PKT_FRAGMENT_L2_HDR_POS = 0,
495 NET_TX_PKT_FRAGMENT_L3_HDR_POS,
496 NET_TX_PKT_FRAGMENT_HEADER_NUM
499 #define NET_MAX_FRAG_SG_LIST (64)
501 static size_t net_tx_pkt_fetch_fragment(struct NetTxPkt *pkt,
502 int *src_idx, size_t *src_offset, struct iovec *dst, int *dst_idx)
504 size_t fetched = 0;
505 struct iovec *src = pkt->vec;
507 *dst_idx = NET_TX_PKT_FRAGMENT_HEADER_NUM;
509 while (fetched < IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size)) {
511 /* no more place in fragment iov */
512 if (*dst_idx == NET_MAX_FRAG_SG_LIST) {
513 break;
516 /* no more data in iovec */
517 if (*src_idx == (pkt->payload_frags + NET_TX_PKT_PL_START_FRAG)) {
518 break;
522 dst[*dst_idx].iov_base = src[*src_idx].iov_base + *src_offset;
523 dst[*dst_idx].iov_len = MIN(src[*src_idx].iov_len - *src_offset,
524 IP_FRAG_ALIGN_SIZE(pkt->virt_hdr.gso_size) - fetched);
526 *src_offset += dst[*dst_idx].iov_len;
527 fetched += dst[*dst_idx].iov_len;
529 if (*src_offset == src[*src_idx].iov_len) {
530 *src_offset = 0;
531 (*src_idx)++;
534 (*dst_idx)++;
537 return fetched;
540 static inline void net_tx_pkt_sendv(struct NetTxPkt *pkt,
541 NetClientState *nc, const struct iovec *iov, int iov_cnt)
543 if (pkt->is_loopback) {
544 nc->info->receive_iov(nc, iov, iov_cnt);
545 } else {
546 qemu_sendv_packet(nc, iov, iov_cnt);
550 static bool net_tx_pkt_do_sw_fragmentation(struct NetTxPkt *pkt,
551 NetClientState *nc)
553 struct iovec fragment[NET_MAX_FRAG_SG_LIST];
554 size_t fragment_len = 0;
555 bool more_frags = false;
557 /* some pointers for shorter code */
558 void *l2_iov_base, *l3_iov_base;
559 size_t l2_iov_len, l3_iov_len;
560 int src_idx = NET_TX_PKT_PL_START_FRAG, dst_idx;
561 size_t src_offset = 0;
562 size_t fragment_offset = 0;
564 l2_iov_base = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_base;
565 l2_iov_len = pkt->vec[NET_TX_PKT_L2HDR_FRAG].iov_len;
566 l3_iov_base = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_base;
567 l3_iov_len = pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len;
569 /* Copy headers */
570 fragment[NET_TX_PKT_FRAGMENT_L2_HDR_POS].iov_base = l2_iov_base;
571 fragment[NET_TX_PKT_FRAGMENT_L2_HDR_POS].iov_len = l2_iov_len;
572 fragment[NET_TX_PKT_FRAGMENT_L3_HDR_POS].iov_base = l3_iov_base;
573 fragment[NET_TX_PKT_FRAGMENT_L3_HDR_POS].iov_len = l3_iov_len;
576 /* Put as much data as possible and send */
577 do {
578 fragment_len = net_tx_pkt_fetch_fragment(pkt, &src_idx, &src_offset,
579 fragment, &dst_idx);
581 more_frags = (fragment_offset + fragment_len < pkt->payload_len);
583 eth_setup_ip4_fragmentation(l2_iov_base, l2_iov_len, l3_iov_base,
584 l3_iov_len, fragment_len, fragment_offset, more_frags);
586 eth_fix_ip4_checksum(l3_iov_base, l3_iov_len);
588 net_tx_pkt_sendv(pkt, nc, fragment, dst_idx);
590 fragment_offset += fragment_len;
592 } while (fragment_len && more_frags);
594 return true;
597 bool net_tx_pkt_send(struct NetTxPkt *pkt, NetClientState *nc)
599 assert(pkt);
601 if (!pkt->has_virt_hdr &&
602 pkt->virt_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
603 net_tx_pkt_do_sw_csum(pkt);
607 * Since underlying infrastructure does not support IP datagrams longer
608 * than 64K we should drop such packets and don't even try to send
610 if (VIRTIO_NET_HDR_GSO_NONE != pkt->virt_hdr.gso_type) {
611 if (pkt->payload_len >
612 ETH_MAX_IP_DGRAM_LEN -
613 pkt->vec[NET_TX_PKT_L3HDR_FRAG].iov_len) {
614 return false;
618 if (pkt->has_virt_hdr ||
619 pkt->virt_hdr.gso_type == VIRTIO_NET_HDR_GSO_NONE) {
620 net_tx_pkt_sendv(pkt, nc, pkt->vec,
621 pkt->payload_frags + NET_TX_PKT_PL_START_FRAG);
622 return true;
625 return net_tx_pkt_do_sw_fragmentation(pkt, nc);
628 bool net_tx_pkt_send_loopback(struct NetTxPkt *pkt, NetClientState *nc)
630 bool res;
632 pkt->is_loopback = true;
633 res = net_tx_pkt_send(pkt, nc);
634 pkt->is_loopback = false;
636 return res;