xref: /dpdk/drivers/net/virtio/virtio_user/vhost_kernel_tap.c (revision 200bc52e5aa0d72e70464c9cd22b55cf536ed13c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2016 Intel Corporation
3  */
4 
5 #include <unistd.h>
6 #include <sys/types.h>
7 #include <sys/stat.h>
8 #include <fcntl.h>
9 #include <net/if.h>
10 #include <net/if_arp.h>
11 #include <errno.h>
12 #include <string.h>
13 #include <limits.h>
14 
15 #include <rte_ether.h>
16 
17 #include "vhost_kernel_tap.h"
18 #include "../virtio_logs.h"
19 #include "../virtio_pci.h"
20 
21 static int
22 vhost_kernel_tap_set_offload(int fd, uint64_t features)
23 {
24 	unsigned int offload = 0;
25 
26 	if (features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
27 		offload |= TUN_F_CSUM;
28 		if (features & (1ULL << VIRTIO_NET_F_GUEST_TSO4))
29 			offload |= TUN_F_TSO4;
30 		if (features & (1ULL << VIRTIO_NET_F_GUEST_TSO6))
31 			offload |= TUN_F_TSO6;
32 		if (features & ((1ULL << VIRTIO_NET_F_GUEST_TSO4) |
33 			(1ULL << VIRTIO_NET_F_GUEST_TSO6)) &&
34 			(features & (1ULL << VIRTIO_NET_F_GUEST_ECN)))
35 			offload |= TUN_F_TSO_ECN;
36 		if (features & (1ULL << VIRTIO_NET_F_GUEST_UFO))
37 			offload |= TUN_F_UFO;
38 	}
39 
40 	if (offload != 0) {
41 		/* Check if our kernel supports TUNSETOFFLOAD */
42 		if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) {
43 			PMD_DRV_LOG(ERR, "Kernel does't support TUNSETOFFLOAD\n");
44 			return -ENOTSUP;
45 		}
46 
47 		if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
48 			offload &= ~TUN_F_UFO;
49 			if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
50 				PMD_DRV_LOG(ERR, "TUNSETOFFLOAD ioctl() failed: %s\n",
51 					strerror(errno));
52 				return -1;
53 			}
54 		}
55 	}
56 
57 	return 0;
58 }
59 
60 int
61 vhost_kernel_open_tap(char **p_ifname, int hdr_size, int req_mq,
62 			 const char *mac, uint64_t features)
63 {
64 	unsigned int tap_features;
65 	char *tap_name = NULL;
66 	int sndbuf = INT_MAX;
67 	struct ifreq ifr;
68 	int tapfd;
69 
70 	/* TODO:
71 	 * 1. verify we can get/set vnet_hdr_len, tap_probe_vnet_hdr_len
72 	 * 2. get number of memory regions from vhost module parameter
73 	 * max_mem_regions, supported in newer version linux kernel
74 	 */
75 	tapfd = open(PATH_NET_TUN, O_RDWR);
76 	if (tapfd < 0) {
77 		PMD_DRV_LOG(ERR, "fail to open %s: %s",
78 			    PATH_NET_TUN, strerror(errno));
79 		return -1;
80 	}
81 
82 	/* Construct ifr */
83 	memset(&ifr, 0, sizeof(ifr));
84 	ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
85 
86 	if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
87 		PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
88 		goto error;
89 	}
90 	if (tap_features & IFF_ONE_QUEUE)
91 		ifr.ifr_flags |= IFF_ONE_QUEUE;
92 
93 	/* Let tap instead of vhost-net handle vnet header, as the latter does
94 	 * not support offloading. And in this case, we should not set feature
95 	 * bit VHOST_NET_F_VIRTIO_NET_HDR.
96 	 */
97 	if (tap_features & IFF_VNET_HDR) {
98 		ifr.ifr_flags |= IFF_VNET_HDR;
99 	} else {
100 		PMD_DRV_LOG(ERR, "TAP does not support IFF_VNET_HDR");
101 		goto error;
102 	}
103 
104 	if (req_mq)
105 		ifr.ifr_flags |= IFF_MULTI_QUEUE;
106 
107 	if (*p_ifname)
108 		strncpy(ifr.ifr_name, *p_ifname, IFNAMSIZ - 1);
109 	else
110 		strncpy(ifr.ifr_name, "tap%d", IFNAMSIZ - 1);
111 	if (ioctl(tapfd, TUNSETIFF, (void *)&ifr) == -1) {
112 		PMD_DRV_LOG(ERR, "TUNSETIFF failed: %s", strerror(errno));
113 		goto error;
114 	}
115 
116 	tap_name = strdup(ifr.ifr_name);
117 	if (!tap_name) {
118 		PMD_DRV_LOG(ERR, "strdup ifname failed: %s", strerror(errno));
119 		goto error;
120 	}
121 
122 	fcntl(tapfd, F_SETFL, O_NONBLOCK);
123 
124 	if (ioctl(tapfd, TUNSETVNETHDRSZ, &hdr_size) < 0) {
125 		PMD_DRV_LOG(ERR, "TUNSETVNETHDRSZ failed: %s", strerror(errno));
126 		goto error;
127 	}
128 
129 	if (ioctl(tapfd, TUNSETSNDBUF, &sndbuf) < 0) {
130 		PMD_DRV_LOG(ERR, "TUNSETSNDBUF failed: %s", strerror(errno));
131 		goto error;
132 	}
133 
134 	vhost_kernel_tap_set_offload(tapfd, features);
135 
136 	memset(&ifr, 0, sizeof(ifr));
137 	ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
138 	memcpy(ifr.ifr_hwaddr.sa_data, mac, RTE_ETHER_ADDR_LEN);
139 	if (ioctl(tapfd, SIOCSIFHWADDR, (void *)&ifr) == -1) {
140 		PMD_DRV_LOG(ERR, "SIOCSIFHWADDR failed: %s", strerror(errno));
141 		goto error;
142 	}
143 
144 	free(*p_ifname);
145 	*p_ifname = tap_name;
146 
147 	return tapfd;
148 error:
149 	free(tap_name);
150 	close(tapfd);
151 	return -1;
152 }
153