net/tap-linux.c

   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 "qemu-common.h"
  28 #include "tap_int.h"
  29 #include "tap-linux.h"
  30 #include "net/tap.h"
  31
  32 #include <net/if.h>
  33 #include <sys/ioctl.h>
  34
  35 #include "qapi/error.h"
  36 #include "qemu/error-report.h"
  37 #include "qemu/cutils.h"
  38
  39 #define PATH_NET_TUN "/dev/net/tun"
  40
  41 int tap_open(char *ifname, int ifname_size, int *vnet_hdr,
  42              int vnet_hdr_required, int mq_required, Error **errp)
  43 {
  44     struct ifreq ifr;
  45     int fd, ret;
  46     int len = sizeof(struct virtio_net_hdr);
  47     unsigned int features;
  48
  49     TFR(fd = open(PATH_NET_TUN, O_RDWR));
  50     if (fd < 0) {
  51         error_setg_errno(errp, errno, "could not open %s", PATH_NET_TUN);
  52         return -1;
  53     }
  54     memset(&ifr, 0, sizeof(ifr));
  55     ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
  56
  57     if (ioctl(fd, TUNGETFEATURES, &features) == -1) {
  58         warn_report("TUNGETFEATURES failed: %s", strerror(errno));
  59         features = 0;
  60     }
  61
  62     if (features & IFF_ONE_QUEUE) {
  63         ifr.ifr_flags |= IFF_ONE_QUEUE;
  64     }
  65
  66     if (*vnet_hdr) {
  67         if (features & IFF_VNET_HDR) {
  68             *vnet_hdr = 1;
  69             ifr.ifr_flags |= IFF_VNET_HDR;
  70         } else {
  71             *vnet_hdr = 0;
  72         }
  73
  74         if (vnet_hdr_required && !*vnet_hdr) {
  75             error_setg(errp, "vnet_hdr=1 requested, but no kernel "
  76                        "support for IFF_VNET_HDR available");
  77             close(fd);
  78             return -1;
  79         }
  80         /*
  81          * Make sure vnet header size has the default value: for a persistent
  82          * tap it might have been modified e.g. by another instance of qemu.
  83          * Ignore errors since old kernels do not support this ioctl: in this
  84          * case the header size implicitly has the correct value.
  85          */
  86         ioctl(fd, TUNSETVNETHDRSZ, &len);
  87     }
  88
  89     if (mq_required) {
  90         if (!(features & IFF_MULTI_QUEUE)) {
  91             error_setg(errp, "multiqueue required, but no kernel "
  92                        "support for IFF_MULTI_QUEUE available");
  93             close(fd);
  94             return -1;
  95         } else {
  96             ifr.ifr_flags |= IFF_MULTI_QUEUE;
  97         }
  98     }
  99
 100     if (ifname[0] != '\0')
 101         pstrcpy(ifr.ifr_name, IFNAMSIZ, ifname);
 102     else
 103         pstrcpy(ifr.ifr_name, IFNAMSIZ, "tap%d");
 104     ret = ioctl(fd, TUNSETIFF, (void *) &ifr);
 105     if (ret != 0) {
 106         if (ifname[0] != '\0') {
 107             error_setg_errno(errp, errno, "could not configure %s (%s)",
 108                              PATH_NET_TUN, ifr.ifr_name);
 109         } else {
 110             error_setg_errno(errp, errno, "could not configure %s",
 111                              PATH_NET_TUN);
 112         }
 113         close(fd);
 114         return -1;
 115     }
 116     pstrcpy(ifname, ifname_size, ifr.ifr_name);
 117     fcntl(fd, F_SETFL, O_NONBLOCK);
 118     return fd;
 119 }
 120
 121 /* sndbuf implements a kind of flow control for tap.
 122  * Unfortunately when it's enabled, and packets are sent
 123  * to other guests on the same host, the receiver
 124  * can lock up the transmitter indefinitely.
 125  *
 126  * To avoid packet loss, sndbuf should be set to a value lower than the tx
 127  * queue capacity of any destination network interface.
 128  * Ethernet NICs generally have txqueuelen=1000, so 1Mb is
 129  * a good value, given a 1500 byte MTU.
 130  */
 131 #define TAP_DEFAULT_SNDBUF 0
 132
 133 void tap_set_sndbuf(int fd, const NetdevTapOptions *tap, Error **errp)
 134 {
 135     int sndbuf;
 136
 137     sndbuf = !tap->has_sndbuf       ? TAP_DEFAULT_SNDBUF :
 138              tap->sndbuf > INT_MAX  ? INT_MAX :
 139              tap->sndbuf;
 140
 141     if (!sndbuf) {
 142         sndbuf = INT_MAX;
 143     }
 144
 145     if (ioctl(fd, TUNSETSNDBUF, &sndbuf) == -1 && tap->has_sndbuf) {
 146         error_setg_errno(errp, errno, "TUNSETSNDBUF ioctl failed");
 147     }
 148 }
 149
 150 int tap_probe_vnet_hdr(int fd, Error **errp)
 151 {
 152     struct ifreq ifr;
 153     memset(&ifr, 0, sizeof(ifr));
 154
 155     if (ioctl(fd, TUNGETIFF, &ifr) != 0) {
 156         /* TUNGETIFF is available since kernel v2.6.27 */
 157         error_setg_errno(errp, errno,
 158                          "Unable to query TUNGETIFF on FD %d", fd);
 159         return -1;
 160     }
 161
 162     return ifr.ifr_flags & IFF_VNET_HDR;
 163 }
 164
 165 int tap_probe_has_ufo(int fd)
 166 {
 167     unsigned offload;
 168
 169     offload = TUN_F_CSUM | TUN_F_UFO;
 170
 171     if (ioctl(fd, TUNSETOFFLOAD, offload) < 0)
 172         return 0;
 173
 174     return 1;
 175 }
 176
 177 /* Verify that we can assign given length */
 178 int tap_probe_vnet_hdr_len(int fd, int len)
 179 {
 180     int orig;
 181     if (ioctl(fd, TUNGETVNETHDRSZ, &orig) == -1) {
 182         return 0;
 183     }
 184     if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) {
 185         return 0;
 186     }
 187     /* Restore original length: we can't handle failure. */
 188     if (ioctl(fd, TUNSETVNETHDRSZ, &orig) == -1) {
 189         fprintf(stderr, "TUNGETVNETHDRSZ ioctl() failed: %s. Exiting.\n",
 190                 strerror(errno));
 191         abort();
 192         return -errno;
 193     }
 194     return 1;
 195 }
 196
 197 void tap_fd_set_vnet_hdr_len(int fd, int len)
 198 {
 199     if (ioctl(fd, TUNSETVNETHDRSZ, &len) == -1) {
 200         fprintf(stderr, "TUNSETVNETHDRSZ ioctl() failed: %s. Exiting.\n",
 201                 strerror(errno));
 202         abort();
 203     }
 204 }
 205
 206 int tap_fd_set_vnet_le(int fd, int is_le)
 207 {
 208     int arg = is_le ? 1 : 0;
 209
 210     if (!ioctl(fd, TUNSETVNETLE, &arg)) {
 211         return 0;
 212     }
 213
 214     /* Check if our kernel supports TUNSETVNETLE */
 215     if (errno == EINVAL) {
 216         return -errno;
 217     }
 218
 219     error_report("TUNSETVNETLE ioctl() failed: %s.", strerror(errno));
 220     abort();
 221 }
 222
 223 int tap_fd_set_vnet_be(int fd, int is_be)
 224 {
 225     int arg = is_be ? 1 : 0;
 226
 227     if (!ioctl(fd, TUNSETVNETBE, &arg)) {
 228         return 0;
 229     }
 230
 231     /* Check if our kernel supports TUNSETVNETBE */
 232     if (errno == EINVAL) {
 233         return -errno;
 234     }
 235
 236     error_report("TUNSETVNETBE ioctl() failed: %s.", strerror(errno));
 237     abort();
 238 }
 239
 240 void tap_fd_set_offload(int fd, int csum, int tso4,
 241                         int tso6, int ecn, int ufo)
 242 {
 243     unsigned int offload = 0;
 244
 245     /* Check if our kernel supports TUNSETOFFLOAD */
 246     if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) {
 247         return;
 248     }
 249
 250     if (csum) {
 251         offload |= TUN_F_CSUM;
 252         if (tso4)
 253             offload |= TUN_F_TSO4;
 254         if (tso6)
 255             offload |= TUN_F_TSO6;
 256         if ((tso4 || tso6) && ecn)
 257             offload |= TUN_F_TSO_ECN;
 258         if (ufo)
 259             offload |= TUN_F_UFO;
 260     }
 261
 262     if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
 263         offload &= ~TUN_F_UFO;
 264         if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
 265             fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
 266                     strerror(errno));
 267         }
 268     }
 269 }
 270
 271 /* Enable a specific queue of tap. */
 272 int tap_fd_enable(int fd)
 273 {
 274     struct ifreq ifr;
 275     int ret;
 276
 277     memset(&ifr, 0, sizeof(ifr));
 278
 279     ifr.ifr_flags = IFF_ATTACH_QUEUE;
 280     ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr);
 281
 282     if (ret != 0) {
 283         error_report("could not enable queue");
 284     }
 285
 286     return ret;
 287 }
 288
 289 /* Disable a specific queue of tap/ */
 290 int tap_fd_disable(int fd)
 291 {
 292     struct ifreq ifr;
 293     int ret;
 294
 295     memset(&ifr, 0, sizeof(ifr));
 296
 297     ifr.ifr_flags = IFF_DETACH_QUEUE;
 298     ret = ioctl(fd, TUNSETQUEUE, (void *) &ifr);
 299
 300     if (ret != 0) {
 301         error_report("could not disable queue");
 302     }
 303
 304     return ret;
 305 }
 306
 307 int tap_fd_get_ifname(int fd, char *ifname)
 308 {
 309     struct ifreq ifr;
 310
 311     if (ioctl(fd, TUNGETIFF, &ifr) != 0) {
 312         error_report("TUNGETIFF ioctl() failed: %s",
 313                      strerror(errno));
 314         return -1;
 315     }
 316
 317     pstrcpy(ifname, sizeof(ifr.ifr_name), ifr.ifr_name);
 318     return 0;
 319 }
 320
 321 int tap_fd_set_steering_ebpf(int fd, int prog_fd)
 322 {
 323     if (ioctl(fd, TUNSETSTEERINGEBPF, (void *) &prog_fd) != 0) {
 324         error_report("Issue while setting TUNSETSTEERINGEBPF:"
 325                     " %s with fd: %d, prog_fd: %d",
 326                     strerror(errno), fd, prog_fd);
 327
 328        return -1;
 329     }
 330
 331     return 0;
 332 }