xref: /freebsd/sys/dev/virtio/network/virtio_net.h (revision 97a4045aaf36079f54ee05a8cb314b8a90f20af8)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * This header is BSD licensed so anyone can use the definitions to implement
5  * compatible drivers/servers.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of IBM nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL IBM OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #ifndef _VIRTIO_NET_H
32 #define _VIRTIO_NET_H
33 
34 /* The feature bitmap for virtio net */
35 #define VIRTIO_NET_F_CSUM		 0x000001 /* Host handles pkts w/ partial csum */
36 #define VIRTIO_NET_F_GUEST_CSUM		 0x000002 /* Guest handles pkts w/ partial csum*/
37 #define VIRTIO_NET_F_CTRL_GUEST_OFFLOADS 0x000004 /* Dynamic offload configuration. */
38 #define VIRTIO_NET_F_MTU		 0x000008 /* Initial MTU advice */
39 #define VIRTIO_NET_F_MAC		 0x000020 /* Host has given MAC address. */
40 #define VIRTIO_NET_F_GSO		 0x000040 /* Host handles pkts w/ any GSO type */
41 #define VIRTIO_NET_F_GUEST_TSO4		 0x000080 /* Guest can handle TSOv4 in. */
42 #define VIRTIO_NET_F_GUEST_TSO6		 0x000100 /* Guest can handle TSOv6 in. */
43 #define VIRTIO_NET_F_GUEST_ECN		 0x000200 /* Guest can handle TSO[6] w/ ECN in. */
44 #define VIRTIO_NET_F_GUEST_UFO		 0x000400 /* Guest can handle UFO in. */
45 #define VIRTIO_NET_F_HOST_TSO4		 0x000800 /* Host can handle TSOv4 in. */
46 #define VIRTIO_NET_F_HOST_TSO6		 0x001000 /* Host can handle TSOv6 in. */
47 #define VIRTIO_NET_F_HOST_ECN		 0x002000 /* Host can handle TSO[6] w/ ECN in. */
48 #define VIRTIO_NET_F_HOST_UFO		 0x004000 /* Host can handle UFO in. */
49 #define VIRTIO_NET_F_MRG_RXBUF		 0x008000 /* Host can merge receive buffers. */
50 #define VIRTIO_NET_F_STATUS		 0x010000 /* virtio_net_config.status available*/
51 #define VIRTIO_NET_F_CTRL_VQ		 0x020000 /* Control channel available */
52 #define VIRTIO_NET_F_CTRL_RX		 0x040000 /* Control channel RX mode support */
53 #define VIRTIO_NET_F_CTRL_VLAN		 0x080000 /* Control channel VLAN filtering */
54 #define VIRTIO_NET_F_CTRL_RX_EXTRA	 0x100000 /* Extra RX mode control support */
55 #define VIRTIO_NET_F_GUEST_ANNOUNCE	 0x200000 /* Announce device on network */
56 #define VIRTIO_NET_F_MQ			 0x400000 /* Device supports Receive Flow Steering */
57 #define VIRTIO_NET_F_CTRL_MAC_ADDR	 0x800000 /* Set MAC address */
58 #define VIRTIO_NET_F_SPEED_DUPLEX	 (1ULL << 63) /* Device set linkspeed and duplex */
59 
60 #define VIRTIO_NET_S_LINK_UP	1	/* Link is up */
61 #define VIRTIO_NET_S_ANNOUNCE	2	/* Announcement is needed */
62 
63 struct virtio_net_config {
64 	/* The config defining mac address (if VIRTIO_NET_F_MAC) */
65 	uint8_t		mac[ETHER_ADDR_LEN];
66 	/* See VIRTIO_NET_F_STATUS and VIRTIO_NET_S_* above */
67 	uint16_t	status;
68 	/* Maximum number of each of transmit and receive queues;
69 	 * see VIRTIO_NET_F_MQ and VIRTIO_NET_CTRL_MQ.
70 	 * Legal values are between 1 and 0x8000.
71 	 */
72 	uint16_t	max_virtqueue_pairs;
73 	/* Default maximum transmit unit advice */
74 	uint16_t	mtu;
75 	/*
76 	 * speed, in units of 1Mb. All values 0 to INT_MAX are legal.
77 	 * Any other value stands for unknown.
78 	 */
79 	uint32_t	speed;
80 	/*
81 	 * 0x00 - half duplex
82 	 * 0x01 - full duplex
83 	 * Any other value stands for unknown.
84 	 */
85 	uint8_t		duplex;
86 } __packed;
87 
88 /*
89  * This header comes first in the scatter-gather list.  If you don't
90  * specify GSO or CSUM features, you can simply ignore the header.
91  *
92  * This is bitwise-equivalent to the legacy struct virtio_net_hdr_mrg_rxbuf,
93  * only flattened.
94  */
95 struct virtio_net_hdr_v1 {
96 #define VIRTIO_NET_HDR_F_NEEDS_CSUM	1	/* Use csum_start, csum_offset */
97 #define VIRTIO_NET_HDR_F_DATA_VALID	2	/* Csum is valid */
98 	uint8_t flags;
99 #define VIRTIO_NET_HDR_GSO_NONE		0	/* Not a GSO frame */
100 #define VIRTIO_NET_HDR_GSO_TCPV4	1	/* GSO frame, IPv4 TCP (TSO) */
101 #define VIRTIO_NET_HDR_GSO_UDP		3	/* GSO frame, IPv4 UDP (UFO) */
102 #define VIRTIO_NET_HDR_GSO_TCPV6	4	/* GSO frame, IPv6 TCP */
103 #define VIRTIO_NET_HDR_GSO_ECN		0x80	/* TCP has ECN set */
104 	uint8_t gso_type;
105 	uint16_t hdr_len;	/* Ethernet + IP + tcp/udp hdrs */
106 	uint16_t gso_size;	/* Bytes to append to hdr_len per frame */
107 	uint16_t csum_start;	/* Position to start checksumming from */
108 	uint16_t csum_offset;	/* Offset after that to place checksum */
109 	uint16_t num_buffers;	/* Number of merged rx buffers */
110 };
111 
112 /*
113  * This header comes first in the scatter-gather list.
114  * For legacy virtio, if VIRTIO_F_ANY_LAYOUT is not negotiated, it must
115  * be the first element of the scatter-gather list.  If you don't
116  * specify GSO or CSUM features, you can simply ignore the header.
117  */
118 struct virtio_net_hdr {
119 	/* See VIRTIO_NET_HDR_F_* */
120 	uint8_t	flags;
121 	/* See VIRTIO_NET_HDR_GSO_* */
122 	uint8_t gso_type;
123 	uint16_t hdr_len;	/* Ethernet + IP + tcp/udp hdrs */
124 	uint16_t gso_size;	/* Bytes to append to hdr_len per frame */
125 	uint16_t csum_start;	/* Position to start checksumming from */
126 	uint16_t csum_offset;	/* Offset after that to place checksum */
127 };
128 
129 /*
130  * This is the version of the header to use when the MRG_RXBUF
131  * feature has been negotiated.
132  */
133 struct virtio_net_hdr_mrg_rxbuf {
134 	struct virtio_net_hdr hdr;
135 	uint16_t num_buffers;	/* Number of merged rx buffers */
136 };
137 
138 /*
139  * Control virtqueue data structures
140  *
141  * The control virtqueue expects a header in the first sg entry
142  * and an ack/status response in the last entry.  Data for the
143  * command goes in between.
144  */
145 struct virtio_net_ctrl_hdr {
146 	uint8_t class;
147 	uint8_t cmd;
148 } __packed;
149 
150 #define VIRTIO_NET_OK	0
151 #define VIRTIO_NET_ERR	1
152 
153 /*
154  * Control the RX mode, ie. promiscuous, allmulti, etc...
155  * All commands require an "out" sg entry containing a 1 byte
156  * state value, zero = disable, non-zero = enable.  Commands
157  * 0 and 1 are supported with the VIRTIO_NET_F_CTRL_RX feature.
158  * Commands 2-5 are added with VIRTIO_NET_F_CTRL_RX_EXTRA.
159  */
160 #define VIRTIO_NET_CTRL_RX	0
161 #define VIRTIO_NET_CTRL_RX_PROMISC	0
162 #define VIRTIO_NET_CTRL_RX_ALLMULTI	1
163 #define VIRTIO_NET_CTRL_RX_ALLUNI	2
164 #define VIRTIO_NET_CTRL_RX_NOMULTI	3
165 #define VIRTIO_NET_CTRL_RX_NOUNI	4
166 #define VIRTIO_NET_CTRL_RX_NOBCAST	5
167 
168 /*
169  * Control the MAC filter table.
170  *
171  * The MAC filter table is managed by the hypervisor, the guest should
172  * assume the size is infinite.  Filtering should be considered
173  * non-perfect, ie. based on hypervisor resources, the guest may
174  * received packets from sources not specified in the filter list.
175  *
176  * In addition to the class/cmd header, the TABLE_SET command requires
177  * two out scatterlists.  Each contains a 4 byte count of entries followed
178  * by a concatenated byte stream of the ETH_ALEN MAC addresses.  The
179  * first sg list contains unicast addresses, the second is for multicast.
180  * This functionality is present if the VIRTIO_NET_F_CTRL_RX feature
181  * is available.
182  *
183  * The ADDR_SET command requests one out scatterlist, it contains a
184  * 6 bytes MAC address. This functionality is present if the
185  * VIRTIO_NET_F_CTRL_MAC_ADDR feature is available.
186  */
187 struct virtio_net_ctrl_mac {
188 	uint32_t	entries;
189 	uint8_t		macs[][ETHER_ADDR_LEN];
190 } __packed;
191 
192 #define VIRTIO_NET_CTRL_MAC	1
193 #define VIRTIO_NET_CTRL_MAC_TABLE_SET	0
194 #define VIRTIO_NET_CTRL_MAC_ADDR_SET	1
195 
196 /*
197  * Control VLAN filtering
198  *
199  * The VLAN filter table is controlled via a simple ADD/DEL interface.
200  * VLAN IDs not added may be filtered by the hypervisor.  Del is the
201  * opposite of add.  Both commands expect an out entry containing a 2
202  * byte VLAN ID.  VLAN filtering is available with the
203  * VIRTIO_NET_F_CTRL_VLAN feature bit.
204  */
205 #define VIRTIO_NET_CTRL_VLAN	2
206 #define VIRTIO_NET_CTRL_VLAN_ADD	0
207 #define VIRTIO_NET_CTRL_VLAN_DEL	1
208 
209 /*
210  * Control link announce acknowledgement
211  *
212  * The command VIRTIO_NET_CTRL_ANNOUNCE_ACK is used to indicate that
213  * driver has received the notification; device would clear the
214  * VIRTIO_NET_S_ANNOUNCE bit in the status field after it receives
215  * this command.
216  */
217 #define VIRTIO_NET_CTRL_ANNOUNCE	3
218 #define VIRTIO_NET_CTRL_ANNOUNCE_ACK	0
219 
220 /*
221  * Control Receive Flow Steering
222  *
223  * The command VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET enables Receive Flow
224  * Steering, specifying the number of the transmit and receive queues
225  * that will be used. After the command is consumed and acked by the
226  * device, the device will not steer new packets on receive virtqueues
227  * other than specified nor read from transmit virtqueues other than
228  * specified. Accordingly, driver should not transmit new packets on
229  * virtqueues other than specified.
230  */
231 struct virtio_net_ctrl_mq {
232 	uint16_t	virtqueue_pairs;
233 } __packed;
234 
235 #define VIRTIO_NET_CTRL_MQ	4
236 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET		0
237 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN		1
238 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX		0x8000
239 
240 /*
241  * Control network offloads
242  *
243  * Reconfigures the network offloads that Guest can handle.
244  *
245  * Available with the VIRTIO_NET_F_CTRL_GUEST_OFFLOADS feature bit.
246  *
247  * Command data format matches the feature bit mask exactly.
248  *
249  * See VIRTIO_NET_F_GUEST_* for the list of offloads
250  * that can be enabled/disabled.
251  */
252 #define VIRTIO_NET_CTRL_GUEST_OFFLOADS		5
253 #define VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET	0
254 
255 /*
256  * Use the checksum offset in the VirtIO header to set the
257  * correct CSUM_* flags.
258  */
259 static inline int
virtio_net_rx_csum_by_offset(struct mbuf * m,uint16_t eth_type,int ip_start,struct virtio_net_hdr * hdr)260 virtio_net_rx_csum_by_offset(struct mbuf *m, uint16_t eth_type, int ip_start,
261 			struct virtio_net_hdr *hdr)
262 {
263 #if defined(INET) || defined(INET6)
264 	int offset = hdr->csum_start + hdr->csum_offset;
265 #endif
266 
267 	/* Only do a basic sanity check on the offset. */
268 	switch (eth_type) {
269 #if defined(INET)
270 	case ETHERTYPE_IP:
271 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
272 			return (1);
273 		break;
274 #endif
275 #if defined(INET6)
276 	case ETHERTYPE_IPV6:
277 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
278 			return (1);
279 		break;
280 #endif
281 	default:
282 		/* Here we should increment the rx_csum_bad_ethtype counter. */
283 		return (1);
284 	}
285 
286 	/*
287 	 * Use the offset to determine the appropriate CSUM_* flags. This is
288 	 * a bit dirty, but we can get by with it since the checksum offsets
289 	 * happen to be different. We assume the host host does not do IPv4
290 	 * header checksum offloading.
291 	 */
292 	switch (hdr->csum_offset) {
293 	case offsetof(struct udphdr, uh_sum):
294 	case offsetof(struct tcphdr, th_sum):
295 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
296 		m->m_pkthdr.csum_data = 0xFFFF;
297 		break;
298 	default:
299 		/* Here we should increment the rx_csum_bad_offset counter. */
300 		return (1);
301 	}
302 
303 	return (0);
304 }
305 
306 static inline int
virtio_net_rx_csum_by_parse(struct mbuf * m,uint16_t eth_type,int ip_start,struct virtio_net_hdr * hdr)307 virtio_net_rx_csum_by_parse(struct mbuf *m, uint16_t eth_type, int ip_start,
308 		       struct virtio_net_hdr *hdr)
309 {
310 	int offset, proto;
311 
312 	switch (eth_type) {
313 #if defined(INET)
314 	case ETHERTYPE_IP: {
315 		struct ip *ip;
316 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
317 			return (1);
318 		ip = (struct ip *)(m->m_data + ip_start);
319 		proto = ip->ip_p;
320 		offset = ip_start + (ip->ip_hl << 2);
321 		break;
322 	}
323 #endif
324 #if defined(INET6)
325 	case ETHERTYPE_IPV6:
326 		if (__predict_false(m->m_len < ip_start +
327 		    sizeof(struct ip6_hdr)))
328 			return (1);
329 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
330 		if (__predict_false(offset < 0))
331 			return (1);
332 		break;
333 #endif
334 	default:
335 		/* Here we should increment the rx_csum_bad_ethtype counter. */
336 		return (1);
337 	}
338 
339 	switch (proto) {
340 	case IPPROTO_TCP:
341 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
342 			return (1);
343 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
344 		m->m_pkthdr.csum_data = 0xFFFF;
345 		break;
346 	case IPPROTO_UDP:
347 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
348 			return (1);
349 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
350 		m->m_pkthdr.csum_data = 0xFFFF;
351 		break;
352 	default:
353 		/*
354 		 * For the remaining protocols, FreeBSD does not support
355 		 * checksum offloading, so the checksum will be recomputed.
356 		 */
357 #if 0
358 		if_printf(ifp, "cksum offload of unsupported "
359 		    "protocol eth_type=%#x proto=%d csum_start=%d "
360 		    "csum_offset=%d\n", __func__, eth_type, proto,
361 		    hdr->csum_start, hdr->csum_offset);
362 #endif
363 		break;
364 	}
365 
366 	return (0);
367 }
368 
369 /*
370  * Set the appropriate CSUM_* flags. Unfortunately, the information
371  * provided is not directly useful to us. The VirtIO header gives the
372  * offset of the checksum, which is all Linux needs, but this is not
373  * how FreeBSD does things. We are forced to peek inside the packet
374  * a bit.
375  *
376  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
377  * could accept the offsets and let the stack figure it out.
378  */
379 static inline int
virtio_net_rx_csum(struct mbuf * m,struct virtio_net_hdr * hdr)380 virtio_net_rx_csum(struct mbuf *m, struct virtio_net_hdr *hdr)
381 {
382 	struct ether_header *eh;
383 	struct ether_vlan_header *evh;
384 	uint16_t eth_type;
385 	int offset, error;
386 
387 	if ((hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
388 	    VIRTIO_NET_HDR_F_DATA_VALID)) == 0) {
389 		return (0);
390 	}
391 
392 	eh = mtod(m, struct ether_header *);
393 	eth_type = ntohs(eh->ether_type);
394 	if (eth_type == ETHERTYPE_VLAN) {
395 		/* BMV: We should handle nested VLAN tags too. */
396 		evh = mtod(m, struct ether_vlan_header *);
397 		eth_type = ntohs(evh->evl_proto);
398 		offset = sizeof(struct ether_vlan_header);
399 	} else
400 		offset = sizeof(struct ether_header);
401 
402 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
403 		error = virtio_net_rx_csum_by_offset(m, eth_type, offset, hdr);
404 	else
405 		error = virtio_net_rx_csum_by_parse(m, eth_type, offset, hdr);
406 
407 	return (error);
408 }
409 
410 static inline int
virtio_net_tx_offload_ctx(struct mbuf * m,int * etype,int * proto,int * start)411 virtio_net_tx_offload_ctx(struct mbuf *m, int *etype, int *proto, int *start)
412 {
413 	struct ether_vlan_header *evh;
414 #if defined(INET) || defined(INET6)
415 	int offset;
416 #endif
417 
418 	evh = mtod(m, struct ether_vlan_header *);
419 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
420 		/* BMV: We should handle nested VLAN tags too. */
421 		*etype = ntohs(evh->evl_proto);
422 #if defined(INET) || defined(INET6)
423 		offset = sizeof(struct ether_vlan_header);
424 #endif
425 	} else {
426 		*etype = ntohs(evh->evl_encap_proto);
427 #if defined(INET) || defined(INET6)
428 		offset = sizeof(struct ether_header);
429 #endif
430 	}
431 
432 	switch (*etype) {
433 #if defined(INET)
434 	case ETHERTYPE_IP: {
435 		struct ip *ip, iphdr;
436 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
437 			m_copydata(m, offset, sizeof(struct ip),
438 			    (caddr_t) &iphdr);
439 			ip = &iphdr;
440 		} else
441 			ip = (struct ip *)(m->m_data + offset);
442 		*proto = ip->ip_p;
443 		*start = offset + (ip->ip_hl << 2);
444 		break;
445 	}
446 #endif
447 #if defined(INET6)
448 	case ETHERTYPE_IPV6:
449 		*proto = -1;
450 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
451 		/* Assert the network stack sent us a valid packet. */
452 		KASSERT(*start > offset,
453 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
454 		    *start, offset, *proto));
455 		break;
456 #endif
457 	default:
458 		/* Here we should increment the tx_csum_bad_ethtype counter. */
459 		return (EINVAL);
460 	}
461 
462 	return (0);
463 }
464 
465 static inline int
virtio_net_tx_offload_tso(if_t ifp,struct mbuf * m,int eth_type,int offset,bool allow_ecn,struct virtio_net_hdr * hdr)466 virtio_net_tx_offload_tso(if_t ifp, struct mbuf *m, int eth_type,
467 		     int offset, bool allow_ecn, struct virtio_net_hdr *hdr)
468 {
469 	static struct timeval lastecn;
470 	static int curecn;
471 	struct tcphdr *tcp, tcphdr;
472 
473 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
474 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
475 		tcp = &tcphdr;
476 	} else
477 		tcp = (struct tcphdr *)(m->m_data + offset);
478 
479 	hdr->hdr_len = offset + (tcp->th_off << 2);
480 	hdr->gso_size = m->m_pkthdr.tso_segsz;
481 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
482 	    VIRTIO_NET_HDR_GSO_TCPV6;
483 
484 	if (tcp->th_flags & TH_CWR) {
485 		/*
486 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
487 		 * ECN support is not on a per-interface basis, but globally via
488 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
489 		 */
490 		if (!allow_ecn) {
491 			if (ppsratecheck(&lastecn, &curecn, 1))
492 				if_printf(ifp,
493 				    "TSO with ECN not negotiated with host\n");
494 			return (ENOTSUP);
495 		}
496 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
497 	}
498 
499 	/* Here we should increment tx_tso counter. */
500 
501 	return (0);
502 }
503 
504 static inline struct mbuf *
virtio_net_tx_offload(if_t ifp,struct mbuf * m,bool allow_ecn,struct virtio_net_hdr * hdr)505 virtio_net_tx_offload(if_t ifp, struct mbuf *m, bool allow_ecn,
506 		 struct virtio_net_hdr *hdr)
507 {
508 	int flags, etype, csum_start, proto, error;
509 
510 	flags = m->m_pkthdr.csum_flags;
511 
512 	error = virtio_net_tx_offload_ctx(m, &etype, &proto, &csum_start);
513 	if (error)
514 		goto drop;
515 
516 	if ((etype == ETHERTYPE_IP && (flags & (CSUM_TCP | CSUM_UDP))) ||
517 	    (etype == ETHERTYPE_IPV6 &&
518 	        (flags & (CSUM_TCP_IPV6 | CSUM_UDP_IPV6)))) {
519 		/*
520 		 * We could compare the IP protocol vs the CSUM_ flag too,
521 		 * but that really should not be necessary.
522 		 */
523 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
524 		hdr->csum_start = csum_start;
525 		hdr->csum_offset = m->m_pkthdr.csum_data;
526 		/* Here we should increment the tx_csum counter. */
527 	}
528 
529 	if (flags & CSUM_TSO) {
530 		if (__predict_false(proto != IPPROTO_TCP)) {
531 			/* Likely failed to correctly parse the mbuf.
532 			 * Here we should increment the tx_tso_not_tcp
533 			 * counter. */
534 			goto drop;
535 		}
536 
537 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
538 		    ("%s: mbuf %p TSO without checksum offload %#x",
539 		    __func__, m, flags));
540 
541 		error = virtio_net_tx_offload_tso(ifp, m, etype, csum_start,
542 					     allow_ecn, hdr);
543 		if (error)
544 			goto drop;
545 	}
546 
547 	return (m);
548 
549 drop:
550 	m_freem(m);
551 	return (NULL);
552 }
553 
554 #endif /* _VIRTIO_NET_H */
555