xref: /freebsd/sys/dev/virtio/network/virtio_net.h (revision 634d9c0d111b630c3d63a1cf25d15c32a37afab8)
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		 (1ULL <<  0) /* Host handles pkts w/ partial csum */
36 #define VIRTIO_NET_F_GUEST_CSUM		 (1ULL <<  1) /* Guest handles pkts w/ partial csum*/
37 #define VIRTIO_NET_F_CTRL_GUEST_OFFLOADS (1ULL <<  2) /* Dynamic offload configuration. */
38 #define VIRTIO_NET_F_MTU		 (1ULL <<  3) /* Initial MTU advice */
39 #define VIRTIO_NET_F_MAC		 (1ULL <<  5) /* Host has given MAC address. */
40 #define VIRTIO_NET_F_GSO		 (1ULL <<  6) /* Host handles pkts w/ any GSO type */
41 #define VIRTIO_NET_F_GUEST_TSO4		 (1ULL <<  7) /* Guest can handle TSOv4 in. */
42 #define VIRTIO_NET_F_GUEST_TSO6		 (1ULL <<  8) /* Guest can handle TSOv6 in. */
43 #define VIRTIO_NET_F_GUEST_ECN		 (1ULL <<  9) /* Guest can handle TSO[6] w/ ECN in. */
44 #define VIRTIO_NET_F_GUEST_UFO		 (1ULL << 10) /* Guest can handle UFO in. */
45 #define VIRTIO_NET_F_HOST_TSO4		 (1ULL << 11) /* Host can handle TSOv4 in. */
46 #define VIRTIO_NET_F_HOST_TSO6		 (1ULL << 12) /* Host can handle TSOv6 in. */
47 #define VIRTIO_NET_F_HOST_ECN		 (1ULL << 13) /* Host can handle TSO[6] w/ ECN in. */
48 #define VIRTIO_NET_F_HOST_UFO		 (1ULL << 14) /* Host can handle UFO in. */
49 #define VIRTIO_NET_F_MRG_RXBUF		 (1ULL << 15) /* Host can merge receive buffers. */
50 #define VIRTIO_NET_F_STATUS		 (1ULL << 16) /* virtio_net_config.status available*/
51 #define VIRTIO_NET_F_CTRL_VQ		 (1ULL << 17) /* Control channel available */
52 #define VIRTIO_NET_F_CTRL_RX		 (1ULL << 18) /* Control channel RX mode support */
53 #define VIRTIO_NET_F_CTRL_VLAN		 (1ULL << 19) /* Control channel VLAN filtering */
54 #define VIRTIO_NET_F_CTRL_RX_EXTRA	 (1ULL << 20) /* Extra RX mode control support */
55 #define VIRTIO_NET_F_GUEST_ANNOUNCE	 (1ULL << 21) /* Announce device on network */
56 #define VIRTIO_NET_F_MQ			 (1ULL << 22) /* Device supports Receive Flow Steering */
57 #define VIRTIO_NET_F_CTRL_MAC_ADDR	 (1ULL << 23) /* Set MAC address */
58 #define VIRTIO_NET_F_SPEED_DUPLEX	 (1ULL << 63) /* Device set linkspeed and duplex */
59 
60 /* virtio net feature flag descriptions for use with printf(9) %b identifier. */
61 #define VIRTIO_NET_FEATURE_BITS \
62     "\20\200CSUM\201GUEST_CSUM\202CTRL_GUEST_OFFLOADS\203MTU\205MAC\206GSO" \
63     "\207GUEST_TSO4\210GUEST_TSO6\211GUEST_ECN\212GUEST_UFO\213HOST_TSO4" \
64     "\214HOST_TSO6\215HOST_ECN\216HOST_UFO\217MRG_RXBUF\220STATUS\221CTRL_VQ" \
65     "\222CTRL_RX\223CTRL_VLAN\224CTRL_RX_EXTRA\225GUEST_ANNOUNCE\226MQ" \
66     "\227CTRL_MAC_ADDR\277SPEED_DUPLEX"
67 
68 #define VIRTIO_NET_S_LINK_UP	1	/* Link is up */
69 #define VIRTIO_NET_S_ANNOUNCE	2	/* Announcement is needed */
70 
71 struct virtio_net_config {
72 	/* The config defining mac address (if VIRTIO_NET_F_MAC) */
73 	uint8_t		mac[ETHER_ADDR_LEN];
74 	/* See VIRTIO_NET_F_STATUS and VIRTIO_NET_S_* above */
75 	uint16_t	status;
76 	/* Maximum number of each of transmit and receive queues;
77 	 * see VIRTIO_NET_F_MQ and VIRTIO_NET_CTRL_MQ.
78 	 * Legal values are between 1 and 0x8000.
79 	 */
80 	uint16_t	max_virtqueue_pairs;
81 	/* Default maximum transmit unit advice */
82 	uint16_t	mtu;
83 	/*
84 	 * speed, in units of 1Mb. All values 0 to INT_MAX are legal.
85 	 * Any other value stands for unknown.
86 	 */
87 	uint32_t	speed;
88 	/*
89 	 * 0x00 - half duplex
90 	 * 0x01 - full duplex
91 	 * Any other value stands for unknown.
92 	 */
93 	uint8_t		duplex;
94 } __packed;
95 
96 /*
97  * This header comes first in the scatter-gather list.  If you don't
98  * specify GSO or CSUM features, you can simply ignore the header.
99  *
100  * This is bitwise-equivalent to the legacy struct virtio_net_hdr_mrg_rxbuf,
101  * only flattened.
102  */
103 struct virtio_net_hdr_v1 {
104 #define VIRTIO_NET_HDR_F_NEEDS_CSUM	1	/* Use csum_start, csum_offset */
105 #define VIRTIO_NET_HDR_F_DATA_VALID	2	/* Csum is valid */
106 	uint8_t flags;
107 #define VIRTIO_NET_HDR_GSO_NONE		0	/* Not a GSO frame */
108 #define VIRTIO_NET_HDR_GSO_TCPV4	1	/* GSO frame, IPv4 TCP (TSO) */
109 #define VIRTIO_NET_HDR_GSO_UDP		3	/* GSO frame, IPv4 UDP (UFO) */
110 #define VIRTIO_NET_HDR_GSO_TCPV6	4	/* GSO frame, IPv6 TCP */
111 #define VIRTIO_NET_HDR_GSO_ECN		0x80	/* TCP has ECN set */
112 	uint8_t gso_type;
113 	uint16_t hdr_len;	/* Ethernet + IP + tcp/udp hdrs */
114 	uint16_t gso_size;	/* Bytes to append to hdr_len per frame */
115 	uint16_t csum_start;	/* Position to start checksumming from */
116 	uint16_t csum_offset;	/* Offset after that to place checksum */
117 	uint16_t num_buffers;	/* Number of merged rx buffers */
118 };
119 
120 /*
121  * This header comes first in the scatter-gather list.
122  * For legacy virtio, if VIRTIO_F_ANY_LAYOUT is not negotiated, it must
123  * be the first element of the scatter-gather list.  If you don't
124  * specify GSO or CSUM features, you can simply ignore the header.
125  */
126 struct virtio_net_hdr {
127 	/* See VIRTIO_NET_HDR_F_* */
128 	uint8_t	flags;
129 	/* See VIRTIO_NET_HDR_GSO_* */
130 	uint8_t gso_type;
131 	uint16_t hdr_len;	/* Ethernet + IP + tcp/udp hdrs */
132 	uint16_t gso_size;	/* Bytes to append to hdr_len per frame */
133 	uint16_t csum_start;	/* Position to start checksumming from */
134 	uint16_t csum_offset;	/* Offset after that to place checksum */
135 };
136 
137 /*
138  * This is the version of the header to use when the MRG_RXBUF
139  * feature has been negotiated.
140  */
141 struct virtio_net_hdr_mrg_rxbuf {
142 	struct virtio_net_hdr hdr;
143 	uint16_t num_buffers;	/* Number of merged rx buffers */
144 };
145 
146 /*
147  * Control virtqueue data structures
148  *
149  * The control virtqueue expects a header in the first sg entry
150  * and an ack/status response in the last entry.  Data for the
151  * command goes in between.
152  */
153 struct virtio_net_ctrl_hdr {
154 	uint8_t class;
155 	uint8_t cmd;
156 } __packed;
157 
158 #define VIRTIO_NET_OK	0
159 #define VIRTIO_NET_ERR	1
160 
161 /*
162  * Control the RX mode, ie. promiscuous, allmulti, etc...
163  * All commands require an "out" sg entry containing a 1 byte
164  * state value, zero = disable, non-zero = enable.  Commands
165  * 0 and 1 are supported with the VIRTIO_NET_F_CTRL_RX feature.
166  * Commands 2-5 are added with VIRTIO_NET_F_CTRL_RX_EXTRA.
167  */
168 #define VIRTIO_NET_CTRL_RX	0
169 #define VIRTIO_NET_CTRL_RX_PROMISC	0
170 #define VIRTIO_NET_CTRL_RX_ALLMULTI	1
171 #define VIRTIO_NET_CTRL_RX_ALLUNI	2
172 #define VIRTIO_NET_CTRL_RX_NOMULTI	3
173 #define VIRTIO_NET_CTRL_RX_NOUNI	4
174 #define VIRTIO_NET_CTRL_RX_NOBCAST	5
175 
176 /*
177  * Control the MAC filter table.
178  *
179  * The MAC filter table is managed by the hypervisor, the guest should
180  * assume the size is infinite.  Filtering should be considered
181  * non-perfect, ie. based on hypervisor resources, the guest may
182  * received packets from sources not specified in the filter list.
183  *
184  * In addition to the class/cmd header, the TABLE_SET command requires
185  * two out scatterlists.  Each contains a 4 byte count of entries followed
186  * by a concatenated byte stream of the ETH_ALEN MAC addresses.  The
187  * first sg list contains unicast addresses, the second is for multicast.
188  * This functionality is present if the VIRTIO_NET_F_CTRL_RX feature
189  * is available.
190  *
191  * The ADDR_SET command requests one out scatterlist, it contains a
192  * 6 bytes MAC address. This functionality is present if the
193  * VIRTIO_NET_F_CTRL_MAC_ADDR feature is available.
194  */
195 struct virtio_net_ctrl_mac {
196 	uint32_t	entries;
197 	uint8_t		macs[][ETHER_ADDR_LEN];
198 } __packed;
199 
200 #define VIRTIO_NET_CTRL_MAC	1
201 #define VIRTIO_NET_CTRL_MAC_TABLE_SET	0
202 #define VIRTIO_NET_CTRL_MAC_ADDR_SET	1
203 
204 /*
205  * Control VLAN filtering
206  *
207  * The VLAN filter table is controlled via a simple ADD/DEL interface.
208  * VLAN IDs not added may be filtered by the hypervisor.  Del is the
209  * opposite of add.  Both commands expect an out entry containing a 2
210  * byte VLAN ID.  VLAN filtering is available with the
211  * VIRTIO_NET_F_CTRL_VLAN feature bit.
212  */
213 #define VIRTIO_NET_CTRL_VLAN	2
214 #define VIRTIO_NET_CTRL_VLAN_ADD	0
215 #define VIRTIO_NET_CTRL_VLAN_DEL	1
216 
217 /*
218  * Control link announce acknowledgement
219  *
220  * The command VIRTIO_NET_CTRL_ANNOUNCE_ACK is used to indicate that
221  * driver has received the notification; device would clear the
222  * VIRTIO_NET_S_ANNOUNCE bit in the status field after it receives
223  * this command.
224  */
225 #define VIRTIO_NET_CTRL_ANNOUNCE	3
226 #define VIRTIO_NET_CTRL_ANNOUNCE_ACK	0
227 
228 /*
229  * Control Receive Flow Steering
230  *
231  * The command VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET enables Receive Flow
232  * Steering, specifying the number of the transmit and receive queues
233  * that will be used. After the command is consumed and acked by the
234  * device, the device will not steer new packets on receive virtqueues
235  * other than specified nor read from transmit virtqueues other than
236  * specified. Accordingly, driver should not transmit new packets on
237  * virtqueues other than specified.
238  */
239 struct virtio_net_ctrl_mq {
240 	uint16_t	virtqueue_pairs;
241 } __packed;
242 
243 #define VIRTIO_NET_CTRL_MQ	4
244 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET		0
245 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN		1
246 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX		0x8000
247 
248 /*
249  * Control network offloads
250  *
251  * Reconfigures the network offloads that Guest can handle.
252  *
253  * Available with the VIRTIO_NET_F_CTRL_GUEST_OFFLOADS feature bit.
254  *
255  * Command data format matches the feature bit mask exactly.
256  *
257  * See VIRTIO_NET_F_GUEST_* for the list of offloads
258  * that can be enabled/disabled.
259  */
260 #define VIRTIO_NET_CTRL_GUEST_OFFLOADS		5
261 #define VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET	0
262 
263 /*
264  * Use the checksum offset in the VirtIO header to set the
265  * correct CSUM_* flags.
266  */
267 static inline int
virtio_net_rx_csum_by_offset(struct mbuf * m,uint16_t eth_type,int ip_start,struct virtio_net_hdr * hdr)268 virtio_net_rx_csum_by_offset(struct mbuf *m, uint16_t eth_type, int ip_start,
269 			struct virtio_net_hdr *hdr)
270 {
271 #if defined(INET) || defined(INET6)
272 	int offset = hdr->csum_start + hdr->csum_offset;
273 #endif
274 
275 	/* Only do a basic sanity check on the offset. */
276 	switch (eth_type) {
277 #if defined(INET)
278 	case ETHERTYPE_IP:
279 		if (__predict_false(offset < ip_start + sizeof(struct ip)))
280 			return (1);
281 		break;
282 #endif
283 #if defined(INET6)
284 	case ETHERTYPE_IPV6:
285 		if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr)))
286 			return (1);
287 		break;
288 #endif
289 	default:
290 		/* Here we should increment the rx_csum_bad_ethtype counter. */
291 		return (1);
292 	}
293 
294 	/*
295 	 * Use the offset to determine the appropriate CSUM_* flags. This is
296 	 * a bit dirty, but we can get by with it since the checksum offsets
297 	 * happen to be different. We assume the host host does not do IPv4
298 	 * header checksum offloading.
299 	 */
300 	switch (hdr->csum_offset) {
301 	case offsetof(struct udphdr, uh_sum):
302 	case offsetof(struct tcphdr, th_sum):
303 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
304 		m->m_pkthdr.csum_data = 0xFFFF;
305 		break;
306 	default:
307 		/* Here we should increment the rx_csum_bad_offset counter. */
308 		return (1);
309 	}
310 
311 	return (0);
312 }
313 
314 static inline int
virtio_net_rx_csum_by_parse(struct mbuf * m,uint16_t eth_type,int ip_start,struct virtio_net_hdr * hdr)315 virtio_net_rx_csum_by_parse(struct mbuf *m, uint16_t eth_type, int ip_start,
316 		       struct virtio_net_hdr *hdr)
317 {
318 	int offset, proto;
319 
320 	switch (eth_type) {
321 #if defined(INET)
322 	case ETHERTYPE_IP: {
323 		struct ip *ip;
324 		if (__predict_false(m->m_len < ip_start + sizeof(struct ip)))
325 			return (1);
326 		ip = (struct ip *)(m->m_data + ip_start);
327 		proto = ip->ip_p;
328 		offset = ip_start + (ip->ip_hl << 2);
329 		break;
330 	}
331 #endif
332 #if defined(INET6)
333 	case ETHERTYPE_IPV6:
334 		if (__predict_false(m->m_len < ip_start +
335 		    sizeof(struct ip6_hdr)))
336 			return (1);
337 		offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto);
338 		if (__predict_false(offset < 0))
339 			return (1);
340 		break;
341 #endif
342 	default:
343 		/* Here we should increment the rx_csum_bad_ethtype counter. */
344 		return (1);
345 	}
346 
347 	switch (proto) {
348 	case IPPROTO_TCP:
349 		if (__predict_false(m->m_len < offset + sizeof(struct tcphdr)))
350 			return (1);
351 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
352 		m->m_pkthdr.csum_data = 0xFFFF;
353 		break;
354 	case IPPROTO_UDP:
355 		if (__predict_false(m->m_len < offset + sizeof(struct udphdr)))
356 			return (1);
357 		m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
358 		m->m_pkthdr.csum_data = 0xFFFF;
359 		break;
360 	default:
361 		/*
362 		 * For the remaining protocols, FreeBSD does not support
363 		 * checksum offloading, so the checksum will be recomputed.
364 		 */
365 #if 0
366 		if_printf(ifp, "cksum offload of unsupported "
367 		    "protocol eth_type=%#x proto=%d csum_start=%d "
368 		    "csum_offset=%d\n", __func__, eth_type, proto,
369 		    hdr->csum_start, hdr->csum_offset);
370 #endif
371 		break;
372 	}
373 
374 	return (0);
375 }
376 
377 /*
378  * Set the appropriate CSUM_* flags. Unfortunately, the information
379  * provided is not directly useful to us. The VirtIO header gives the
380  * offset of the checksum, which is all Linux needs, but this is not
381  * how FreeBSD does things. We are forced to peek inside the packet
382  * a bit.
383  *
384  * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD
385  * could accept the offsets and let the stack figure it out.
386  */
387 static inline int
virtio_net_rx_csum(struct mbuf * m,struct virtio_net_hdr * hdr)388 virtio_net_rx_csum(struct mbuf *m, struct virtio_net_hdr *hdr)
389 {
390 	struct ether_header *eh;
391 	struct ether_vlan_header *evh;
392 	uint16_t eth_type;
393 	int offset, error;
394 
395 	if ((hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM |
396 	    VIRTIO_NET_HDR_F_DATA_VALID)) == 0) {
397 		return (0);
398 	}
399 
400 	eh = mtod(m, struct ether_header *);
401 	eth_type = ntohs(eh->ether_type);
402 	if (eth_type == ETHERTYPE_VLAN) {
403 		/* BMV: We should handle nested VLAN tags too. */
404 		evh = mtod(m, struct ether_vlan_header *);
405 		eth_type = ntohs(evh->evl_proto);
406 		offset = sizeof(struct ether_vlan_header);
407 	} else
408 		offset = sizeof(struct ether_header);
409 
410 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
411 		error = virtio_net_rx_csum_by_offset(m, eth_type, offset, hdr);
412 	else
413 		error = virtio_net_rx_csum_by_parse(m, eth_type, offset, hdr);
414 
415 	return (error);
416 }
417 
418 static inline int
virtio_net_tx_offload_ctx(struct mbuf * m,int * etype,int * proto,int * start)419 virtio_net_tx_offload_ctx(struct mbuf *m, int *etype, int *proto, int *start)
420 {
421 	struct ether_vlan_header *evh;
422 #if defined(INET) || defined(INET6)
423 	int offset;
424 #endif
425 
426 	evh = mtod(m, struct ether_vlan_header *);
427 	if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
428 		/* BMV: We should handle nested VLAN tags too. */
429 		*etype = ntohs(evh->evl_proto);
430 #if defined(INET) || defined(INET6)
431 		offset = sizeof(struct ether_vlan_header);
432 #endif
433 	} else {
434 		*etype = ntohs(evh->evl_encap_proto);
435 #if defined(INET) || defined(INET6)
436 		offset = sizeof(struct ether_header);
437 #endif
438 	}
439 
440 	switch (*etype) {
441 #if defined(INET)
442 	case ETHERTYPE_IP: {
443 		struct ip *ip, iphdr;
444 		if (__predict_false(m->m_len < offset + sizeof(struct ip))) {
445 			m_copydata(m, offset, sizeof(struct ip),
446 			    (caddr_t) &iphdr);
447 			ip = &iphdr;
448 		} else
449 			ip = (struct ip *)(m->m_data + offset);
450 		*proto = ip->ip_p;
451 		*start = offset + (ip->ip_hl << 2);
452 		break;
453 	}
454 #endif
455 #if defined(INET6)
456 	case ETHERTYPE_IPV6:
457 		*proto = -1;
458 		*start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto);
459 		/* Assert the network stack sent us a valid packet. */
460 		KASSERT(*start > offset,
461 		    ("%s: mbuf %p start %d offset %d proto %d", __func__, m,
462 		    *start, offset, *proto));
463 		break;
464 #endif
465 	default:
466 		/* Here we should increment the tx_csum_bad_ethtype counter. */
467 		return (EINVAL);
468 	}
469 
470 	return (0);
471 }
472 
473 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)474 virtio_net_tx_offload_tso(if_t ifp, struct mbuf *m, int eth_type,
475 		     int offset, bool allow_ecn, struct virtio_net_hdr *hdr)
476 {
477 	static struct timeval lastecn;
478 	static int curecn;
479 	struct tcphdr *tcp, tcphdr;
480 
481 	if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) {
482 		m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr);
483 		tcp = &tcphdr;
484 	} else
485 		tcp = (struct tcphdr *)(m->m_data + offset);
486 
487 	hdr->hdr_len = offset + (tcp->th_off << 2);
488 	hdr->gso_size = m->m_pkthdr.tso_segsz;
489 	hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 :
490 	    VIRTIO_NET_HDR_GSO_TCPV6;
491 
492 	if (tcp_get_flags(tcp) & TH_CWR) {
493 		/*
494 		 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD,
495 		 * ECN support is not on a per-interface basis, but globally via
496 		 * the net.inet.tcp.ecn.enable sysctl knob. The default is off.
497 		 */
498 		if (!allow_ecn) {
499 			if (ppsratecheck(&lastecn, &curecn, 1))
500 				if_printf(ifp,
501 				    "TSO with ECN not negotiated with host\n");
502 			return (ENOTSUP);
503 		}
504 		hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
505 	}
506 
507 	/* Here we should increment tx_tso counter. */
508 
509 	return (0);
510 }
511 
512 static inline struct mbuf *
virtio_net_tx_offload(if_t ifp,struct mbuf * m,bool allow_ecn,struct virtio_net_hdr * hdr)513 virtio_net_tx_offload(if_t ifp, struct mbuf *m, bool allow_ecn,
514 		 struct virtio_net_hdr *hdr)
515 {
516 	int flags, etype, csum_start, proto, error;
517 
518 	flags = m->m_pkthdr.csum_flags;
519 
520 	error = virtio_net_tx_offload_ctx(m, &etype, &proto, &csum_start);
521 	if (error)
522 		goto drop;
523 
524 	if ((etype == ETHERTYPE_IP && (flags & (CSUM_TCP | CSUM_UDP))) ||
525 	    (etype == ETHERTYPE_IPV6 &&
526 	        (flags & (CSUM_TCP_IPV6 | CSUM_UDP_IPV6)))) {
527 		/*
528 		 * We could compare the IP protocol vs the CSUM_ flag too,
529 		 * but that really should not be necessary.
530 		 */
531 		hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
532 		hdr->csum_start = csum_start;
533 		hdr->csum_offset = m->m_pkthdr.csum_data;
534 		/* Here we should increment the tx_csum counter. */
535 	}
536 
537 	if (flags & CSUM_TSO) {
538 		if (__predict_false(proto != IPPROTO_TCP)) {
539 			/* Likely failed to correctly parse the mbuf.
540 			 * Here we should increment the tx_tso_not_tcp
541 			 * counter. */
542 			goto drop;
543 		}
544 
545 		KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM,
546 		    ("%s: mbuf %p TSO without checksum offload %#x",
547 		    __func__, m, flags));
548 
549 		error = virtio_net_tx_offload_tso(ifp, m, etype, csum_start,
550 					     allow_ecn, hdr);
551 		if (error)
552 			goto drop;
553 	}
554 
555 	return (m);
556 
557 drop:
558 	m_freem(m);
559 	return (NULL);
560 }
561 
562 #endif /* _VIRTIO_NET_H */
563