master
c 360 lines 9.05 KB
Raw
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009 Red Hat, Inc.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27 #include "tap_int.h"
28 #include "tap-linux.h"
29 #include "net/tap.h"
30
31 #include <net/if.h>
32 #include <sys/ioctl.h>
33
34 #include "qapi/error.h"
35 #include "qemu/error-report.h"
36 #include "qemu/cutils.h"
37
38 #define PATH_NET_TUN "/dev/net/tun"
39
40 int tap_open(char *ifname, int ifname_size, int *vnet_hdr,
41 int vnet_hdr_required, int mq_required, Error **errp)
42 {
43 struct ifreq ifr;
44 int fd, ret;
45 int len = sizeof(struct virtio_net_hdr);
46 unsigned int features;
47
48
49 ret = if_nametoindex(ifname);
50 if (ret) {
51 g_autofree char *file = g_strdup_printf("/dev/tap%d", ret);
52 fd = open(file, O_RDWR);
53 } else {
54 fd = -1;
55 }
56
57 if (fd < 0) {
58 fd = RETRY_ON_EINTR(open(PATH_NET_TUN, O_RDWR));
59 if (fd < 0) {
60 error_setg_file_open(errp, errno, PATH_NET_TUN);
61 return -1;
62 }
63 }
64 memset(&ifr, 0, sizeof(ifr));
65 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
66
67 if (ioctl(fd, TUNGETFEATURES, &features) == -1) {
68 warn_report("TUNGETFEATURES failed: %s", strerror(errno));
69 features = 0;
70 }
71
72 if (features & IFF_ONE_QUEUE) {
73 ifr.ifr_flags |= IFF_ONE_QUEUE;
74 }
75
76 if (*vnet_hdr) {
77 if (features & IFF_VNET_HDR) {
78 *vnet_hdr = 1;
79 ifr.ifr_flags |= IFF_VNET_HDR;
80 } else {
81 *vnet_hdr = 0;
82 }
83
84 if (vnet_hdr_required && !*vnet_hdr) {
85 error_setg(errp, "vnet_hdr=1 requested, but no kernel "
86 "support for IFF_VNET_HDR available");
87 close(fd);
88 return -1;
89 }
90 /*
91 * Make sure vnet header size has the default value: for a persistent
92 * tap it might have been modified e.g. by another instance of qemu.
93 * Ignore errors since old kernels do not support this ioctl: in this
94 * case the header size implicitly has the correct value.
95 */
96 ioctl(fd, TUNSETVNETHDRSZ, &len);
97 }
98
99 if (mq_required) {
100 if (!(features & IFF_MULTI_QUEUE)) {
101 error_setg(errp, "multiqueue required, but no kernel "
102 "support for IFF_MULTI_QUEUE available");
103 close(fd);
104 return -1;
105 } else {
106 ifr.ifr_flags |= IFF_MULTI_QUEUE;
107 }
108 }
109
110 if (ifname[0] != '\0')
111 pstrcpy(ifr.ifr_name, IFNAMSIZ, ifname);
112 else
113 pstrcpy(ifr.ifr_name, IFNAMSIZ, "tap%d");
114 ret = ioctl(fd, TUNSETIFF, (void *) &ifr);
115 if (ret != 0) {
116 if (ifname[0] != '\0') {
117 error_setg_errno(errp, errno, "could not configure %s (%s)",
118 PATH_NET_TUN, ifr.ifr_name);
119 } else {
120 error_setg_errno(errp, errno, "could not configure %s",
121 PATH_NET_TUN);
122 }
123 close(fd);
124 return -1;
125 }
126 pstrcpy(ifname, ifname_size, ifr.ifr_name);
127
128 if (!qemu_set_blocking(fd, false, errp)) {
129 close(fd);
130 return -1;
131 }
132
133 return fd;
134 }
135
136 /* sndbuf implements a kind of flow control for tap.
137 * Unfortunately when it's enabled, and packets are sent
138 * to other guests on the same host, the receiver
139 * can lock up the transmitter indefinitely.
140 *
141 * To avoid packet loss, sndbuf should be set to a value lower than the tx
142 * queue capacity of any destination network interface.
143 * Ethernet NICs generally have txqueuelen=1000, so 1Mb is
144 * a good value, given a 1500 byte MTU.
145 */
146 bool tap_set_sndbuf(int fd, int sndbuf, Error **errp)
147 {
148 if (ioctl(fd, TUNSETSNDBUF, &sndbuf) == -1) {
149 error_setg_errno(errp, errno, "TUNSETSNDBUF ioctl failed");
150 return false;
151 }
152
153 return true;
154 }
155
156 int tap_probe_vnet_hdr(int fd, Error **errp)
157 {
158 struct ifreq ifr;
159 memset(&ifr, 0, sizeof(ifr));
160
161 if (ioctl(fd, TUNGETIFF, &ifr) != 0) {
162 /* TUNGETIFF is available since kernel v2.6.27 */
163 error_setg_errno(errp, errno,
164 "Unable to query TUNGETIFF on FD %d", fd);
165 return -1;
166 }
167
168 return ifr.ifr_flags & IFF_VNET_HDR;
169 }
170
171 int tap_probe_has_ufo(int fd)
172 {
173 unsigned offload;
174
175 offload = TUN_F_CSUM | TUN_F_UFO;
176
177 if (ioctl(fd, TUNSETOFFLOAD, offload) < 0)
178 return 0;
179
180 return 1;
181 }
182
183 int tap_probe_has_uso(int fd)
184 {
185 unsigned offload;
186
187 offload = TUN_F_CSUM | TUN_F_USO4 | TUN_F_USO6;
188
189 if (ioctl(fd, TUNSETOFFLOAD, offload) < 0) {
190 return 0;
191 }
192 return 1;
193 }
194
195 bool tap_probe_has_tunnel(int fd)
196 {
197 unsigned offload;
198
199 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_UDP_TUNNEL_GSO;
200 if (ioctl(fd, TUNSETOFFLOAD, offload) < 0) {
201 return false;
202 }
203 return true;
204 }
205
206 void tap_fd_set_vnet_hdr_len(int fd, int len)
207 {
208 int ret;
209
210 ret = ioctl(fd, TUNSETVNETHDRSZ, &len);
211 if (ret != 0) {
212 error_report("TUNSETVNETHDRSZ ioctl() failed: %s.", strerror(errno));
213 }
214 }
215
216 int tap_fd_set_vnet_le(int fd, int is_le)
217 {
218 int arg = is_le ? 1 : 0;
219
220 if (!ioctl(fd, TUNSETVNETLE, &arg)) {
221 return 0;
222 }
223
224 /* Check if our kernel supports TUNSETVNETLE */
225 if (errno == EINVAL) {
226 return -errno;
227 }
228
229 error_report("TUNSETVNETLE ioctl() failed: %s.", strerror(errno));
230 abort();
231 }
232
233 int tap_fd_set_vnet_be(int fd, int is_be)
234 {
235 int arg = is_be ? 1 : 0;
236
237 if (!ioctl(fd, TUNSETVNETBE, &arg)) {
238 return 0;
239 }
240
241 /* Check if our kernel supports TUNSETVNETBE */
242 if (errno == EINVAL) {
243 return -errno;
244 }
245
246 error_report("TUNSETVNETBE ioctl() failed: %s.", strerror(errno));
247 abort();
248 }
249
250 void tap_fd_set_offload(int fd, const NetOffloads *ol)
251 {
252 unsigned int offload = 0;
253
254 /* Check if our kernel supports TUNSETOFFLOAD */
255 if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) {
256 return;
257 }
258
259 if (ol->csum) {
260 offload |= TUN_F_CSUM;
261 if (ol->tso4) {
262 offload |= TUN_F_TSO4;
263 }
264 if (ol->tso6) {
265 offload |= TUN_F_TSO6;
266 }
267 if ((ol->tso4 || ol->tso6) && ol->ecn) {
268 offload |= TUN_F_TSO_ECN;
269 }
270 if (ol->ufo) {
271 offload |= TUN_F_UFO;
272 }
273 if (ol->uso4) {
274 offload |= TUN_F_USO4;
275 }
276 if (ol->uso6) {
277 offload |= TUN_F_USO6;
278 }
279 if (ol->tnl) {
280 offload |= TUN_F_UDP_TUNNEL_GSO;
281 }
282 if (ol->tnl_csum) {
283 offload |= TUN_F_UDP_TUNNEL_GSO_CSUM;
284 }
285 }
286
287 if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
288 offload &= ~(TUN_F_USO4 | TUN_F_USO6);
289 if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
290 offload &= ~TUN_F_UFO;
291 if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
292 fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
293 strerror(errno));
294 }
295 }
296 }
297 }
298
299 /* Enable a specific queue of tap. */
300 int tap_fd_enable(int fd)
301 {
302 struct ifreq ifr;
303 int ret;
304
305 memset(&ifr, 0, sizeof(ifr));
306
307 ifr.ifr_flags = IFF_ATTACH_QUEUE;
308 ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr);
309
310 if (ret != 0) {
311 error_report("could not enable queue");
312 }
313
314 return ret;
315 }
316
317 /* Disable a specific queue of tap/ */
318 int tap_fd_disable(int fd)
319 {
320 struct ifreq ifr;
321 int ret;
322
323 memset(&ifr, 0, sizeof(ifr));
324
325 ifr.ifr_flags = IFF_DETACH_QUEUE;
326 ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr);
327
328 if (ret != 0) {
329 error_report("could not disable queue");
330 }
331
332 return ret;
333 }
334
335 int tap_fd_get_ifname(int fd, char *ifname)
336 {
337 struct ifreq ifr;
338
339 if (ioctl(fd, TUNGETIFF, &ifr) != 0) {
340 error_report("TUNGETIFF ioctl() failed: %s",
341 strerror(errno));
342 return -1;
343 }
344
345 pstrcpy(ifname, sizeof(ifr.ifr_name), ifr.ifr_name);
346 return 0;
347 }
348
349 int tap_fd_set_steering_ebpf(int fd, int prog_fd)
350 {
351 if (ioctl(fd, TUNSETSTEERINGEBPF, (void *) &prog_fd) != 0) {
352 error_report("Issue while setting TUNSETSTEERINGEBPF:"
353 " %s with fd: %d, prog_fd: %d",
354 strerror(errno), fd, prog_fd);
355
356 return -1;
357 }
358
359 return 0;
360 }