]>
Commit | Line | Data |
---|---|---|
c28b1c10 MM |
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 | ||
1422e32d PB |
26 | #include "tap_int.h" |
27 | #include "tap-linux.h" | |
c28b1c10 | 28 | #include "net/tap.h" |
c28b1c10 MM |
29 | |
30 | #include <net/if.h> | |
31 | #include <sys/ioctl.h> | |
32 | ||
9c17d615 | 33 | #include "sysemu/sysemu.h" |
c28b1c10 | 34 | #include "qemu-common.h" |
1de7afc9 | 35 | #include "qemu/error-report.h" |
c28b1c10 | 36 | |
91ca60e0 MT |
37 | #define PATH_NET_TUN "/dev/net/tun" |
38 | ||
264986e2 JW |
39 | int tap_open(char *ifname, int ifname_size, int *vnet_hdr, |
40 | int vnet_hdr_required, int mq_required) | |
c28b1c10 MM |
41 | { |
42 | struct ifreq ifr; | |
43 | int fd, ret; | |
89e6d68e | 44 | int len = sizeof(struct virtio_net_hdr); |
d26e445c | 45 | unsigned int features; |
c28b1c10 | 46 | |
91ca60e0 | 47 | TFR(fd = open(PATH_NET_TUN, O_RDWR)); |
c28b1c10 | 48 | if (fd < 0) { |
91ca60e0 | 49 | error_report("could not open %s: %m", PATH_NET_TUN); |
c28b1c10 MM |
50 | return -1; |
51 | } | |
52 | memset(&ifr, 0, sizeof(ifr)); | |
53 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
54 | ||
1f149e72 KK |
55 | if (ioctl(fd, TUNGETFEATURES, &features) == -1) { |
56 | error_report("warning: TUNGETFEATURES failed: %s", strerror(errno)); | |
57 | features = 0; | |
58 | } | |
59 | ||
60 | if (features & IFF_ONE_QUEUE) { | |
d26e445c PL |
61 | ifr.ifr_flags |= IFF_ONE_QUEUE; |
62 | } | |
c28b1c10 | 63 | |
d26e445c | 64 | if (*vnet_hdr) { |
1f149e72 | 65 | if (features & IFF_VNET_HDR) { |
c28b1c10 MM |
66 | *vnet_hdr = 1; |
67 | ifr.ifr_flags |= IFF_VNET_HDR; | |
6720b35b PR |
68 | } else { |
69 | *vnet_hdr = 0; | |
c28b1c10 MM |
70 | } |
71 | ||
72 | if (vnet_hdr_required && !*vnet_hdr) { | |
1ecda02b MA |
73 | error_report("vnet_hdr=1 requested, but no kernel " |
74 | "support for IFF_VNET_HDR available"); | |
c28b1c10 MM |
75 | close(fd); |
76 | return -1; | |
77 | } | |
89e6d68e MT |
78 | /* |
79 | * Make sure vnet header size has the default value: for a persistent | |
80 | * tap it might have been modified e.g. by another instance of qemu. | |
81 | * Ignore errors since old kernels do not support this ioctl: in this | |
82 | * case the header size implicitly has the correct value. | |
83 | */ | |
84 | ioctl(fd, TUNSETVNETHDRSZ, &len); | |
c28b1c10 MM |
85 | } |
86 | ||
94fdc6d0 | 87 | if (mq_required) { |
1f149e72 | 88 | if (!(features & IFF_MULTI_QUEUE)) { |
94fdc6d0 JW |
89 | error_report("multiqueue required, but no kernel " |
90 | "support for IFF_MULTI_QUEUE available"); | |
91 | close(fd); | |
92 | return -1; | |
93 | } else { | |
94 | ifr.ifr_flags |= IFF_MULTI_QUEUE; | |
95 | } | |
96 | } | |
97 | ||
c28b1c10 MM |
98 | if (ifname[0] != '\0') |
99 | pstrcpy(ifr.ifr_name, IFNAMSIZ, ifname); | |
100 | else | |
101 | pstrcpy(ifr.ifr_name, IFNAMSIZ, "tap%d"); | |
102 | ret = ioctl(fd, TUNSETIFF, (void *) &ifr); | |
103 | if (ret != 0) { | |
93a7320e LC |
104 | if (ifname[0] != '\0') { |
105 | error_report("could not configure %s (%s): %m", PATH_NET_TUN, ifr.ifr_name); | |
106 | } else { | |
107 | error_report("could not configure %s: %m", PATH_NET_TUN); | |
108 | } | |
c28b1c10 MM |
109 | close(fd); |
110 | return -1; | |
111 | } | |
112 | pstrcpy(ifname, ifname_size, ifr.ifr_name); | |
113 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
114 | return fd; | |
115 | } | |
15ac913b | 116 | |
f157ed20 MT |
117 | /* sndbuf implements a kind of flow control for tap. |
118 | * Unfortunately when it's enabled, and packets are sent | |
119 | * to other guests on the same host, the receiver | |
120 | * can lock up the transmitter indefinitely. | |
121 | * | |
122 | * To avoid packet loss, sndbuf should be set to a value lower than the tx | |
123 | * queue capacity of any destination network interface. | |
15ac913b | 124 | * Ethernet NICs generally have txqueuelen=1000, so 1Mb is |
f157ed20 | 125 | * a good value, given a 1500 byte MTU. |
15ac913b | 126 | */ |
f157ed20 | 127 | #define TAP_DEFAULT_SNDBUF 0 |
15ac913b | 128 | |
08c573a8 | 129 | int tap_set_sndbuf(int fd, const NetdevTapOptions *tap) |
15ac913b MM |
130 | { |
131 | int sndbuf; | |
132 | ||
08c573a8 LE |
133 | sndbuf = !tap->has_sndbuf ? TAP_DEFAULT_SNDBUF : |
134 | tap->sndbuf > INT_MAX ? INT_MAX : | |
135 | tap->sndbuf; | |
136 | ||
15ac913b MM |
137 | if (!sndbuf) { |
138 | sndbuf = INT_MAX; | |
139 | } | |
140 | ||
08c573a8 | 141 | if (ioctl(fd, TUNSETSNDBUF, &sndbuf) == -1 && tap->has_sndbuf) { |
1ecda02b | 142 | error_report("TUNSETSNDBUF ioctl failed: %s", strerror(errno)); |
15ac913b MM |
143 | return -1; |
144 | } | |
145 | return 0; | |
146 | } | |
dc69004c MM |
147 | |
148 | int tap_probe_vnet_hdr(int fd) | |
149 | { | |
150 | struct ifreq ifr; | |
151 | ||
152 | if (ioctl(fd, TUNGETIFF, &ifr) != 0) { | |
1ecda02b | 153 | error_report("TUNGETIFF ioctl() failed: %s", strerror(errno)); |
dc69004c MM |
154 | return 0; |
155 | } | |
156 | ||
157 | return ifr.ifr_flags & IFF_VNET_HDR; | |
158 | } | |
1faac1f7 | 159 | |
9c282718 MM |
160 | int tap_probe_has_ufo(int fd) |
161 | { | |
162 | unsigned offload; | |
163 | ||
164 | offload = TUN_F_CSUM | TUN_F_UFO; | |
165 | ||
166 | if (ioctl(fd, TUNSETOFFLOAD, offload) < 0) | |
167 | return 0; | |
168 | ||
169 | return 1; | |
170 | } | |
171 | ||
445d892f MT |
172 | /* Verify that we can assign given length */ |
173 | int tap_probe_vnet_hdr_len(int fd, int len) | |
174 | { | |
175 | int orig; | |
176 | if (ioctl(fd, TUNGETVNETHDRSZ, &orig) == -1) { | |
177 | return 0; | |
178 | } | |
179 | if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) { | |
180 | return 0; | |
181 | } | |
182 | /* Restore original length: we can't handle failure. */ | |
183 | if (ioctl(fd, TUNSETVNETHDRSZ, &orig) == -1) { | |
184 | fprintf(stderr, "TUNGETVNETHDRSZ ioctl() failed: %s. Exiting.\n", | |
185 | strerror(errno)); | |
28a65891 | 186 | abort(); |
445d892f MT |
187 | return -errno; |
188 | } | |
189 | return 1; | |
190 | } | |
191 | ||
192 | void tap_fd_set_vnet_hdr_len(int fd, int len) | |
193 | { | |
194 | if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) { | |
195 | fprintf(stderr, "TUNSETVNETHDRSZ ioctl() failed: %s. Exiting.\n", | |
196 | strerror(errno)); | |
28a65891 | 197 | abort(); |
445d892f MT |
198 | } |
199 | } | |
200 | ||
1faac1f7 MM |
201 | void tap_fd_set_offload(int fd, int csum, int tso4, |
202 | int tso6, int ecn, int ufo) | |
203 | { | |
204 | unsigned int offload = 0; | |
205 | ||
2e50326c PR |
206 | /* Check if our kernel supports TUNSETOFFLOAD */ |
207 | if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) { | |
208 | return; | |
209 | } | |
210 | ||
1faac1f7 MM |
211 | if (csum) { |
212 | offload |= TUN_F_CSUM; | |
213 | if (tso4) | |
214 | offload |= TUN_F_TSO4; | |
215 | if (tso6) | |
216 | offload |= TUN_F_TSO6; | |
217 | if ((tso4 || tso6) && ecn) | |
218 | offload |= TUN_F_TSO_ECN; | |
219 | if (ufo) | |
220 | offload |= TUN_F_UFO; | |
221 | } | |
222 | ||
223 | if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) { | |
224 | offload &= ~TUN_F_UFO; | |
225 | if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) { | |
226 | fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n", | |
227 | strerror(errno)); | |
228 | } | |
229 | } | |
230 | } | |
94fdc6d0 JW |
231 | |
232 | /* Enable a specific queue of tap. */ | |
233 | int tap_fd_enable(int fd) | |
234 | { | |
235 | struct ifreq ifr; | |
236 | int ret; | |
237 | ||
238 | memset(&ifr, 0, sizeof(ifr)); | |
239 | ||
240 | ifr.ifr_flags = IFF_ATTACH_QUEUE; | |
241 | ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr); | |
242 | ||
243 | if (ret != 0) { | |
244 | error_report("could not enable queue"); | |
245 | } | |
246 | ||
247 | return ret; | |
248 | } | |
249 | ||
250 | /* Disable a specific queue of tap/ */ | |
251 | int tap_fd_disable(int fd) | |
252 | { | |
253 | struct ifreq ifr; | |
254 | int ret; | |
255 | ||
256 | memset(&ifr, 0, sizeof(ifr)); | |
257 | ||
258 | ifr.ifr_flags = IFF_DETACH_QUEUE; | |
259 | ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr); | |
260 | ||
261 | if (ret != 0) { | |
262 | error_report("could not disable queue"); | |
263 | } | |
264 | ||
265 | return ret; | |
266 | } | |
267 | ||
e5dc0b40 JW |
268 | int tap_fd_get_ifname(int fd, char *ifname) |
269 | { | |
270 | struct ifreq ifr; | |
271 | ||
272 | if (ioctl(fd, TUNGETIFF, &ifr) != 0) { | |
273 | error_report("TUNGETIFF ioctl() failed: %s", | |
274 | strerror(errno)); | |
275 | return -1; | |
276 | } | |
277 | ||
278 | pstrcpy(ifname, sizeof(ifr.ifr_name), ifr.ifr_name); | |
279 | return 0; | |
280 | } |