root / net / tap-linux.c @ 396d2cfc
History | View | Annotate | Download (6 kB)
1 | c28b1c10 | Mark McLoughlin | /*
|
---|---|---|---|
2 | c28b1c10 | Mark McLoughlin | * QEMU System Emulator
|
3 | c28b1c10 | Mark McLoughlin | *
|
4 | c28b1c10 | Mark McLoughlin | * Copyright (c) 2003-2008 Fabrice Bellard
|
5 | c28b1c10 | Mark McLoughlin | * Copyright (c) 2009 Red Hat, Inc.
|
6 | c28b1c10 | Mark McLoughlin | *
|
7 | c28b1c10 | Mark McLoughlin | * Permission is hereby granted, free of charge, to any person obtaining a copy
|
8 | c28b1c10 | Mark McLoughlin | * of this software and associated documentation files (the "Software"), to deal
|
9 | c28b1c10 | Mark McLoughlin | * in the Software without restriction, including without limitation the rights
|
10 | c28b1c10 | Mark McLoughlin | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
11 | c28b1c10 | Mark McLoughlin | * copies of the Software, and to permit persons to whom the Software is
|
12 | c28b1c10 | Mark McLoughlin | * furnished to do so, subject to the following conditions:
|
13 | c28b1c10 | Mark McLoughlin | *
|
14 | c28b1c10 | Mark McLoughlin | * The above copyright notice and this permission notice shall be included in
|
15 | c28b1c10 | Mark McLoughlin | * all copies or substantial portions of the Software.
|
16 | c28b1c10 | Mark McLoughlin | *
|
17 | c28b1c10 | Mark McLoughlin | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
18 | c28b1c10 | Mark McLoughlin | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
19 | c28b1c10 | Mark McLoughlin | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
20 | c28b1c10 | Mark McLoughlin | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
21 | c28b1c10 | Mark McLoughlin | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
22 | c28b1c10 | Mark McLoughlin | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
23 | c28b1c10 | Mark McLoughlin | * THE SOFTWARE.
|
24 | c28b1c10 | Mark McLoughlin | */
|
25 | c28b1c10 | Mark McLoughlin | |
26 | 1422e32d | Paolo Bonzini | #include "tap_int.h" |
27 | 1422e32d | Paolo Bonzini | #include "tap-linux.h" |
28 | c28b1c10 | Mark McLoughlin | #include "net/tap.h" |
29 | c28b1c10 | Mark McLoughlin | |
30 | c28b1c10 | Mark McLoughlin | #include <net/if.h> |
31 | c28b1c10 | Mark McLoughlin | #include <sys/ioctl.h> |
32 | c28b1c10 | Mark McLoughlin | |
33 | 9c17d615 | Paolo Bonzini | #include "sysemu/sysemu.h" |
34 | c28b1c10 | Mark McLoughlin | #include "qemu-common.h" |
35 | 1de7afc9 | Paolo Bonzini | #include "qemu/error-report.h" |
36 | c28b1c10 | Mark McLoughlin | |
37 | 91ca60e0 | Michael Tokarev | #define PATH_NET_TUN "/dev/net/tun" |
38 | 91ca60e0 | Michael Tokarev | |
39 | c28b1c10 | Mark McLoughlin | int tap_open(char *ifname, int ifname_size, int *vnet_hdr, int vnet_hdr_required) |
40 | c28b1c10 | Mark McLoughlin | { |
41 | c28b1c10 | Mark McLoughlin | struct ifreq ifr;
|
42 | c28b1c10 | Mark McLoughlin | int fd, ret;
|
43 | 89e6d68e | Michael S. Tsirkin | int len = sizeof(struct virtio_net_hdr); |
44 | c28b1c10 | Mark McLoughlin | |
45 | 91ca60e0 | Michael Tokarev | TFR(fd = open(PATH_NET_TUN, O_RDWR)); |
46 | c28b1c10 | Mark McLoughlin | if (fd < 0) { |
47 | 91ca60e0 | Michael Tokarev | error_report("could not open %s: %m", PATH_NET_TUN);
|
48 | c28b1c10 | Mark McLoughlin | return -1; |
49 | c28b1c10 | Mark McLoughlin | } |
50 | c28b1c10 | Mark McLoughlin | memset(&ifr, 0, sizeof(ifr)); |
51 | c28b1c10 | Mark McLoughlin | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; |
52 | c28b1c10 | Mark McLoughlin | |
53 | c28b1c10 | Mark McLoughlin | if (*vnet_hdr) {
|
54 | c28b1c10 | Mark McLoughlin | unsigned int features; |
55 | c28b1c10 | Mark McLoughlin | |
56 | c28b1c10 | Mark McLoughlin | if (ioctl(fd, TUNGETFEATURES, &features) == 0 && |
57 | c28b1c10 | Mark McLoughlin | features & IFF_VNET_HDR) { |
58 | c28b1c10 | Mark McLoughlin | *vnet_hdr = 1;
|
59 | c28b1c10 | Mark McLoughlin | ifr.ifr_flags |= IFF_VNET_HDR; |
60 | 6720b35b | Pierre Riteau | } else {
|
61 | 6720b35b | Pierre Riteau | *vnet_hdr = 0;
|
62 | c28b1c10 | Mark McLoughlin | } |
63 | c28b1c10 | Mark McLoughlin | |
64 | c28b1c10 | Mark McLoughlin | if (vnet_hdr_required && !*vnet_hdr) {
|
65 | 1ecda02b | Markus Armbruster | error_report("vnet_hdr=1 requested, but no kernel "
|
66 | 1ecda02b | Markus Armbruster | "support for IFF_VNET_HDR available");
|
67 | c28b1c10 | Mark McLoughlin | close(fd); |
68 | c28b1c10 | Mark McLoughlin | return -1; |
69 | c28b1c10 | Mark McLoughlin | } |
70 | 89e6d68e | Michael S. Tsirkin | /*
|
71 | 89e6d68e | Michael S. Tsirkin | * Make sure vnet header size has the default value: for a persistent
|
72 | 89e6d68e | Michael S. Tsirkin | * tap it might have been modified e.g. by another instance of qemu.
|
73 | 89e6d68e | Michael S. Tsirkin | * Ignore errors since old kernels do not support this ioctl: in this
|
74 | 89e6d68e | Michael S. Tsirkin | * case the header size implicitly has the correct value.
|
75 | 89e6d68e | Michael S. Tsirkin | */
|
76 | 89e6d68e | Michael S. Tsirkin | ioctl(fd, TUNSETVNETHDRSZ, &len); |
77 | c28b1c10 | Mark McLoughlin | } |
78 | c28b1c10 | Mark McLoughlin | |
79 | c28b1c10 | Mark McLoughlin | if (ifname[0] != '\0') |
80 | c28b1c10 | Mark McLoughlin | pstrcpy(ifr.ifr_name, IFNAMSIZ, ifname); |
81 | c28b1c10 | Mark McLoughlin | else
|
82 | c28b1c10 | Mark McLoughlin | pstrcpy(ifr.ifr_name, IFNAMSIZ, "tap%d");
|
83 | c28b1c10 | Mark McLoughlin | ret = ioctl(fd, TUNSETIFF, (void *) &ifr);
|
84 | c28b1c10 | Mark McLoughlin | if (ret != 0) { |
85 | 93a7320e | Luiz Capitulino | if (ifname[0] != '\0') { |
86 | 93a7320e | Luiz Capitulino | error_report("could not configure %s (%s): %m", PATH_NET_TUN, ifr.ifr_name);
|
87 | 93a7320e | Luiz Capitulino | } else {
|
88 | 93a7320e | Luiz Capitulino | error_report("could not configure %s: %m", PATH_NET_TUN);
|
89 | 93a7320e | Luiz Capitulino | } |
90 | c28b1c10 | Mark McLoughlin | close(fd); |
91 | c28b1c10 | Mark McLoughlin | return -1; |
92 | c28b1c10 | Mark McLoughlin | } |
93 | c28b1c10 | Mark McLoughlin | pstrcpy(ifname, ifname_size, ifr.ifr_name); |
94 | c28b1c10 | Mark McLoughlin | fcntl(fd, F_SETFL, O_NONBLOCK); |
95 | c28b1c10 | Mark McLoughlin | return fd;
|
96 | c28b1c10 | Mark McLoughlin | } |
97 | 15ac913b | Mark McLoughlin | |
98 | f157ed20 | Michael S. Tsirkin | /* sndbuf implements a kind of flow control for tap.
|
99 | f157ed20 | Michael S. Tsirkin | * Unfortunately when it's enabled, and packets are sent
|
100 | f157ed20 | Michael S. Tsirkin | * to other guests on the same host, the receiver
|
101 | f157ed20 | Michael S. Tsirkin | * can lock up the transmitter indefinitely.
|
102 | f157ed20 | Michael S. Tsirkin | *
|
103 | f157ed20 | Michael S. Tsirkin | * To avoid packet loss, sndbuf should be set to a value lower than the tx
|
104 | f157ed20 | Michael S. Tsirkin | * queue capacity of any destination network interface.
|
105 | 15ac913b | Mark McLoughlin | * Ethernet NICs generally have txqueuelen=1000, so 1Mb is
|
106 | f157ed20 | Michael S. Tsirkin | * a good value, given a 1500 byte MTU.
|
107 | 15ac913b | Mark McLoughlin | */
|
108 | f157ed20 | Michael S. Tsirkin | #define TAP_DEFAULT_SNDBUF 0 |
109 | 15ac913b | Mark McLoughlin | |
110 | 08c573a8 | Laszlo Ersek | int tap_set_sndbuf(int fd, const NetdevTapOptions *tap) |
111 | 15ac913b | Mark McLoughlin | { |
112 | 15ac913b | Mark McLoughlin | int sndbuf;
|
113 | 15ac913b | Mark McLoughlin | |
114 | 08c573a8 | Laszlo Ersek | sndbuf = !tap->has_sndbuf ? TAP_DEFAULT_SNDBUF : |
115 | 08c573a8 | Laszlo Ersek | tap->sndbuf > INT_MAX ? INT_MAX : |
116 | 08c573a8 | Laszlo Ersek | tap->sndbuf; |
117 | 08c573a8 | Laszlo Ersek | |
118 | 15ac913b | Mark McLoughlin | if (!sndbuf) {
|
119 | 15ac913b | Mark McLoughlin | sndbuf = INT_MAX; |
120 | 15ac913b | Mark McLoughlin | } |
121 | 15ac913b | Mark McLoughlin | |
122 | 08c573a8 | Laszlo Ersek | if (ioctl(fd, TUNSETSNDBUF, &sndbuf) == -1 && tap->has_sndbuf) { |
123 | 1ecda02b | Markus Armbruster | error_report("TUNSETSNDBUF ioctl failed: %s", strerror(errno));
|
124 | 15ac913b | Mark McLoughlin | return -1; |
125 | 15ac913b | Mark McLoughlin | } |
126 | 15ac913b | Mark McLoughlin | return 0; |
127 | 15ac913b | Mark McLoughlin | } |
128 | dc69004c | Mark McLoughlin | |
129 | dc69004c | Mark McLoughlin | int tap_probe_vnet_hdr(int fd) |
130 | dc69004c | Mark McLoughlin | { |
131 | dc69004c | Mark McLoughlin | struct ifreq ifr;
|
132 | dc69004c | Mark McLoughlin | |
133 | dc69004c | Mark McLoughlin | if (ioctl(fd, TUNGETIFF, &ifr) != 0) { |
134 | 1ecda02b | Markus Armbruster | error_report("TUNGETIFF ioctl() failed: %s", strerror(errno));
|
135 | dc69004c | Mark McLoughlin | return 0; |
136 | dc69004c | Mark McLoughlin | } |
137 | dc69004c | Mark McLoughlin | |
138 | dc69004c | Mark McLoughlin | return ifr.ifr_flags & IFF_VNET_HDR;
|
139 | dc69004c | Mark McLoughlin | } |
140 | 1faac1f7 | Mark McLoughlin | |
141 | 9c282718 | Mark McLoughlin | int tap_probe_has_ufo(int fd) |
142 | 9c282718 | Mark McLoughlin | { |
143 | 9c282718 | Mark McLoughlin | unsigned offload;
|
144 | 9c282718 | Mark McLoughlin | |
145 | 9c282718 | Mark McLoughlin | offload = TUN_F_CSUM | TUN_F_UFO; |
146 | 9c282718 | Mark McLoughlin | |
147 | 9c282718 | Mark McLoughlin | if (ioctl(fd, TUNSETOFFLOAD, offload) < 0) |
148 | 9c282718 | Mark McLoughlin | return 0; |
149 | 9c282718 | Mark McLoughlin | |
150 | 9c282718 | Mark McLoughlin | return 1; |
151 | 9c282718 | Mark McLoughlin | } |
152 | 9c282718 | Mark McLoughlin | |
153 | 445d892f | Michael S. Tsirkin | /* Verify that we can assign given length */
|
154 | 445d892f | Michael S. Tsirkin | int tap_probe_vnet_hdr_len(int fd, int len) |
155 | 445d892f | Michael S. Tsirkin | { |
156 | 445d892f | Michael S. Tsirkin | int orig;
|
157 | 445d892f | Michael S. Tsirkin | if (ioctl(fd, TUNGETVNETHDRSZ, &orig) == -1) { |
158 | 445d892f | Michael S. Tsirkin | return 0; |
159 | 445d892f | Michael S. Tsirkin | } |
160 | 445d892f | Michael S. Tsirkin | if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) { |
161 | 445d892f | Michael S. Tsirkin | return 0; |
162 | 445d892f | Michael S. Tsirkin | } |
163 | 445d892f | Michael S. Tsirkin | /* Restore original length: we can't handle failure. */
|
164 | 445d892f | Michael S. Tsirkin | if (ioctl(fd, TUNSETVNETHDRSZ, &orig) == -1) { |
165 | 445d892f | Michael S. Tsirkin | fprintf(stderr, "TUNGETVNETHDRSZ ioctl() failed: %s. Exiting.\n",
|
166 | 445d892f | Michael S. Tsirkin | strerror(errno)); |
167 | 445d892f | Michael S. Tsirkin | assert(0);
|
168 | 445d892f | Michael S. Tsirkin | return -errno;
|
169 | 445d892f | Michael S. Tsirkin | } |
170 | 445d892f | Michael S. Tsirkin | return 1; |
171 | 445d892f | Michael S. Tsirkin | } |
172 | 445d892f | Michael S. Tsirkin | |
173 | 445d892f | Michael S. Tsirkin | void tap_fd_set_vnet_hdr_len(int fd, int len) |
174 | 445d892f | Michael S. Tsirkin | { |
175 | 445d892f | Michael S. Tsirkin | if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) { |
176 | 445d892f | Michael S. Tsirkin | fprintf(stderr, "TUNSETVNETHDRSZ ioctl() failed: %s. Exiting.\n",
|
177 | 445d892f | Michael S. Tsirkin | strerror(errno)); |
178 | 445d892f | Michael S. Tsirkin | assert(0);
|
179 | 445d892f | Michael S. Tsirkin | } |
180 | 445d892f | Michael S. Tsirkin | } |
181 | 445d892f | Michael S. Tsirkin | |
182 | 1faac1f7 | Mark McLoughlin | void tap_fd_set_offload(int fd, int csum, int tso4, |
183 | 1faac1f7 | Mark McLoughlin | int tso6, int ecn, int ufo) |
184 | 1faac1f7 | Mark McLoughlin | { |
185 | 1faac1f7 | Mark McLoughlin | unsigned int offload = 0; |
186 | 1faac1f7 | Mark McLoughlin | |
187 | 2e50326c | Pierre Riteau | /* Check if our kernel supports TUNSETOFFLOAD */
|
188 | 2e50326c | Pierre Riteau | if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) { |
189 | 2e50326c | Pierre Riteau | return;
|
190 | 2e50326c | Pierre Riteau | } |
191 | 2e50326c | Pierre Riteau | |
192 | 1faac1f7 | Mark McLoughlin | if (csum) {
|
193 | 1faac1f7 | Mark McLoughlin | offload |= TUN_F_CSUM; |
194 | 1faac1f7 | Mark McLoughlin | if (tso4)
|
195 | 1faac1f7 | Mark McLoughlin | offload |= TUN_F_TSO4; |
196 | 1faac1f7 | Mark McLoughlin | if (tso6)
|
197 | 1faac1f7 | Mark McLoughlin | offload |= TUN_F_TSO6; |
198 | 1faac1f7 | Mark McLoughlin | if ((tso4 || tso6) && ecn)
|
199 | 1faac1f7 | Mark McLoughlin | offload |= TUN_F_TSO_ECN; |
200 | 1faac1f7 | Mark McLoughlin | if (ufo)
|
201 | 1faac1f7 | Mark McLoughlin | offload |= TUN_F_UFO; |
202 | 1faac1f7 | Mark McLoughlin | } |
203 | 1faac1f7 | Mark McLoughlin | |
204 | 1faac1f7 | Mark McLoughlin | if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) { |
205 | 1faac1f7 | Mark McLoughlin | offload &= ~TUN_F_UFO; |
206 | 1faac1f7 | Mark McLoughlin | if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) { |
207 | 1faac1f7 | Mark McLoughlin | fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
|
208 | 1faac1f7 | Mark McLoughlin | strerror(errno)); |
209 | 1faac1f7 | Mark McLoughlin | } |
210 | 1faac1f7 | Mark McLoughlin | } |
211 | 1faac1f7 | Mark McLoughlin | } |