qemu-bridge-helper.c

   1 /*
   2  * QEMU Bridge Helper
   3  *
   4  * Copyright IBM, Corp. 2011
   5  *
   6  * Authors:
   7  * Anthony Liguori   <aliguori@us.ibm.com>
   8  * Richa Marwaha     <rmarwah@linux.vnet.ibm.com>
   9  * Corey Bryant      <coreyb@linux.vnet.ibm.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include "qemu/osdep.h"
  17
  18
  19 #include <sys/ioctl.h>
  20 #include <sys/socket.h>
  21 #include <sys/un.h>
  22 #include <sys/prctl.h>
  23
  24 #include <net/if.h>
  25
  26 #include <linux/sockios.h>
  27
  28 #ifndef SIOCBRADDIF
  29 #include <linux/if_bridge.h>
  30 #endif
  31
  32 #include "qemu/queue.h"
  33
  34 #include "net/tap-linux.h"
  35
  36 #ifdef CONFIG_LIBCAP
  37 #include <cap-ng.h>
  38 #endif
  39
  40 #define DEFAULT_ACL_FILE CONFIG_QEMU_CONFDIR "/bridge.conf"
  41
  42 enum {
  43     ACL_ALLOW = 0,
  44     ACL_ALLOW_ALL,
  45     ACL_DENY,
  46     ACL_DENY_ALL,
  47 };
  48
  49 typedef struct ACLRule {
  50     int type;
  51     char iface[IFNAMSIZ];
  52     QSIMPLEQ_ENTRY(ACLRule) entry;
  53 } ACLRule;
  54
  55 typedef QSIMPLEQ_HEAD(ACLList, ACLRule) ACLList;
  56
  57 static void usage(void)
  58 {
  59     fprintf(stderr,
  60             "Usage: qemu-bridge-helper [--use-vnet] --br=bridge --fd=unixfd\n");
  61 }
  62
  63 static int parse_acl_file(const char *filename, ACLList *acl_list)
  64 {
  65     FILE *f;
  66     char line[4096];
  67     ACLRule *acl_rule;
  68
  69     f = fopen(filename, "r");
  70     if (f == NULL) {
  71         return -1;
  72     }
  73
  74     while (fgets(line, sizeof(line), f) != NULL) {
  75         char *ptr = line;
  76         char *cmd, *arg, *argend;
  77
  78         while (isspace(*ptr)) {
  79             ptr++;
  80         }
  81
  82         /* skip comments and empty lines */
  83         if (*ptr == '#' || *ptr == 0) {
  84             continue;
  85         }
  86
  87         cmd = ptr;
  88         arg = strchr(cmd, ' ');
  89         if (arg == NULL) {
  90             arg = strchr(cmd, '\t');
  91         }
  92
  93         if (arg == NULL) {
  94             fprintf(stderr, "Invalid config line:\n  %s\n", line);
  95             fclose(f);
  96             errno = EINVAL;
  97             return -1;
  98         }
  99
 100         *arg = 0;
 101         arg++;
 102         while (isspace(*arg)) {
 103             arg++;
 104         }
 105
 106         argend = arg + strlen(arg);
 107         while (arg != argend && isspace(*(argend - 1))) {
 108             argend--;
 109         }
 110         *argend = 0;
 111
 112         if (!g_str_equal(cmd, "include") && strlen(arg) >= IFNAMSIZ) {
 113             fprintf(stderr, "name `%s' too long: %zu\n", arg, strlen(arg));
 114             fclose(f);
 115             errno = EINVAL;
 116             return -1;
 117         }
 118
 119         if (strcmp(cmd, "deny") == 0) {
 120             acl_rule = g_malloc(sizeof(*acl_rule));
 121             if (strcmp(arg, "all") == 0) {
 122                 acl_rule->type = ACL_DENY_ALL;
 123             } else {
 124                 acl_rule->type = ACL_DENY;
 125                 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg);
 126             }
 127             QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry);
 128         } else if (strcmp(cmd, "allow") == 0) {
 129             acl_rule = g_malloc(sizeof(*acl_rule));
 130             if (strcmp(arg, "all") == 0) {
 131                 acl_rule->type = ACL_ALLOW_ALL;
 132             } else {
 133                 acl_rule->type = ACL_ALLOW;
 134                 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg);
 135             }
 136             QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry);
 137         } else if (strcmp(cmd, "include") == 0) {
 138             /* ignore errors */
 139             parse_acl_file(arg, acl_list);
 140         } else {
 141             fprintf(stderr, "Unknown command `%s'\n", cmd);
 142             fclose(f);
 143             errno = EINVAL;
 144             return -1;
 145         }
 146     }
 147
 148     fclose(f);
 149
 150     return 0;
 151 }
 152
 153 static bool has_vnet_hdr(int fd)
 154 {
 155     unsigned int features = 0;
 156
 157     if (ioctl(fd, TUNGETFEATURES, &features) == -1) {
 158         return false;
 159     }
 160
 161     if (!(features & IFF_VNET_HDR)) {
 162         return false;
 163     }
 164
 165     return true;
 166 }
 167
 168 static void prep_ifreq(struct ifreq *ifr, const char *ifname)
 169 {
 170     memset(ifr, 0, sizeof(*ifr));
 171     snprintf(ifr->ifr_name, IFNAMSIZ, "%s", ifname);
 172 }
 173
 174 static int send_fd(int c, int fd)
 175 {
 176     char msgbuf[CMSG_SPACE(sizeof(fd))];
 177     struct msghdr msg = {
 178         .msg_control = msgbuf,
 179         .msg_controllen = sizeof(msgbuf),
 180     };
 181     struct cmsghdr *cmsg;
 182     struct iovec iov;
 183     char req[1] = { 0x00 };
 184
 185     cmsg = CMSG_FIRSTHDR(&msg);
 186     cmsg->cmsg_level = SOL_SOCKET;
 187     cmsg->cmsg_type = SCM_RIGHTS;
 188     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 189     msg.msg_controllen = cmsg->cmsg_len;
 190
 191     iov.iov_base = req;
 192     iov.iov_len = sizeof(req);
 193
 194     msg.msg_iov = &iov;
 195     msg.msg_iovlen = 1;
 196     memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
 197
 198     return sendmsg(c, &msg, 0);
 199 }
 200
 201 #ifdef CONFIG_LIBCAP
 202 static int drop_privileges(void)
 203 {
 204     /* clear all capabilities */
 205     capng_clear(CAPNG_SELECT_BOTH);
 206
 207     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 208                      CAP_NET_ADMIN) < 0) {
 209         return -1;
 210     }
 211
 212     /* change to calling user's real uid and gid, retaining supplemental
 213      * groups and CAP_NET_ADMIN */
 214     if (capng_change_id(getuid(), getgid(), CAPNG_CLEAR_BOUNDING)) {
 215         return -1;
 216     }
 217
 218     return 0;
 219 }
 220 #endif
 221
 222 int main(int argc, char **argv)
 223 {
 224     struct ifreq ifr;
 225 #ifndef SIOCBRADDIF
 226     unsigned long ifargs[4];
 227 #endif
 228     int ifindex;
 229     int fd = -1, ctlfd = -1, unixfd = -1;
 230     int use_vnet = 0;
 231     int mtu;
 232     const char *bridge = NULL;
 233     char iface[IFNAMSIZ];
 234     int index;
 235     ACLRule *acl_rule;
 236     ACLList acl_list;
 237     int access_allowed, access_denied;
 238     int ret = EXIT_SUCCESS;
 239
 240 #ifdef CONFIG_LIBCAP
 241     /* if we're run from an suid binary, immediately drop privileges preserving
 242      * cap_net_admin */
 243     if (geteuid() == 0 && getuid() != geteuid()) {
 244         if (drop_privileges() == -1) {
 245             fprintf(stderr, "failed to drop privileges\n");
 246             return 1;
 247         }
 248     }
 249 #endif
 250
 251     /* parse arguments */
 252     for (index = 1; index < argc; index++) {
 253         if (strcmp(argv[index], "--use-vnet") == 0) {
 254             use_vnet = 1;
 255         } else if (strncmp(argv[index], "--br=", 5) == 0) {
 256             bridge = &argv[index][5];
 257         } else if (strncmp(argv[index], "--fd=", 5) == 0) {
 258             unixfd = atoi(&argv[index][5]);
 259         } else {
 260             usage();
 261             return EXIT_FAILURE;
 262         }
 263     }
 264
 265     if (bridge == NULL || unixfd == -1) {
 266         usage();
 267         return EXIT_FAILURE;
 268     }
 269     if (strlen(bridge) >= IFNAMSIZ) {
 270         fprintf(stderr, "name `%s' too long: %zu\n", bridge, strlen(bridge));
 271         return EXIT_FAILURE;
 272     }
 273
 274     /* parse default acl file */
 275     QSIMPLEQ_INIT(&acl_list);
 276     if (parse_acl_file(DEFAULT_ACL_FILE, &acl_list) == -1) {
 277         fprintf(stderr, "failed to parse default acl file `%s'\n",
 278                 DEFAULT_ACL_FILE);
 279         ret = EXIT_FAILURE;
 280         goto cleanup;
 281     }
 282
 283     /* validate bridge against acl -- default policy is to deny
 284      * according acl policy if we have a deny and allow both
 285      * then deny should always win over allow
 286      */
 287     access_allowed = 0;
 288     access_denied = 0;
 289     QSIMPLEQ_FOREACH(acl_rule, &acl_list, entry) {
 290         switch (acl_rule->type) {
 291         case ACL_ALLOW_ALL:
 292             access_allowed = 1;
 293             break;
 294         case ACL_ALLOW:
 295             if (strcmp(bridge, acl_rule->iface) == 0) {
 296                 access_allowed = 1;
 297             }
 298             break;
 299         case ACL_DENY_ALL:
 300             access_denied = 1;
 301             break;
 302         case ACL_DENY:
 303             if (strcmp(bridge, acl_rule->iface) == 0) {
 304                 access_denied = 1;
 305             }
 306             break;
 307         }
 308     }
 309
 310     if ((access_allowed == 0) || (access_denied == 1)) {
 311         fprintf(stderr, "access denied by acl file\n");
 312         ret = EXIT_FAILURE;
 313         goto cleanup;
 314     }
 315
 316     /* open a socket to use to control the network interfaces */
 317     ctlfd = socket(AF_INET, SOCK_STREAM, 0);
 318     if (ctlfd == -1) {
 319         fprintf(stderr, "failed to open control socket: %s\n", strerror(errno));
 320         ret = EXIT_FAILURE;
 321         goto cleanup;
 322     }
 323
 324     /* open the tap device */
 325     fd = open("/dev/net/tun", O_RDWR);
 326     if (fd == -1) {
 327         fprintf(stderr, "failed to open /dev/net/tun: %s\n", strerror(errno));
 328         ret = EXIT_FAILURE;
 329         goto cleanup;
 330     }
 331
 332     /* request a tap device, disable PI, and add vnet header support if
 333      * requested and it's available. */
 334     prep_ifreq(&ifr, "tap%d");
 335     ifr.ifr_flags = IFF_TAP|IFF_NO_PI;
 336     if (use_vnet && has_vnet_hdr(fd)) {
 337         ifr.ifr_flags |= IFF_VNET_HDR;
 338     }
 339
 340     if (ioctl(fd, TUNSETIFF, &ifr) == -1) {
 341         fprintf(stderr, "failed to create tun device: %s\n", strerror(errno));
 342         ret = EXIT_FAILURE;
 343         goto cleanup;
 344     }
 345
 346     /* save tap device name */
 347     snprintf(iface, sizeof(iface), "%s", ifr.ifr_name);
 348
 349     /* get the mtu of the bridge */
 350     prep_ifreq(&ifr, bridge);
 351     if (ioctl(ctlfd, SIOCGIFMTU, &ifr) == -1) {
 352         fprintf(stderr, "failed to get mtu of bridge `%s': %s\n",
 353                 bridge, strerror(errno));
 354         ret = EXIT_FAILURE;
 355         goto cleanup;
 356     }
 357
 358     /* save mtu */
 359     mtu = ifr.ifr_mtu;
 360
 361     /* set the mtu of the interface based on the bridge */
 362     prep_ifreq(&ifr, iface);
 363     ifr.ifr_mtu = mtu;
 364     if (ioctl(ctlfd, SIOCSIFMTU, &ifr) == -1) {
 365         fprintf(stderr, "failed to set mtu of device `%s' to %d: %s\n",
 366                 iface, mtu, strerror(errno));
 367         ret = EXIT_FAILURE;
 368         goto cleanup;
 369     }
 370
 371     /* Linux uses the lowest enslaved MAC address as the MAC address of
 372      * the bridge.  Set MAC address to a high value so that it doesn't
 373      * affect the MAC address of the bridge.
 374      */
 375     if (ioctl(ctlfd, SIOCGIFHWADDR, &ifr) < 0) {
 376         fprintf(stderr, "failed to get MAC address of device `%s': %s\n",
 377                 iface, strerror(errno));
 378         ret = EXIT_FAILURE;
 379         goto cleanup;
 380     }
 381     ifr.ifr_hwaddr.sa_data[0] = 0xFE;
 382     if (ioctl(ctlfd, SIOCSIFHWADDR, &ifr) < 0) {
 383         fprintf(stderr, "failed to set MAC address of device `%s': %s\n",
 384                 iface, strerror(errno));
 385         ret = EXIT_FAILURE;
 386         goto cleanup;
 387     }
 388
 389     /* add the interface to the bridge */
 390     prep_ifreq(&ifr, bridge);
 391     ifindex = if_nametoindex(iface);
 392 #ifndef SIOCBRADDIF
 393     ifargs[0] = BRCTL_ADD_IF;
 394     ifargs[1] = ifindex;
 395     ifargs[2] = 0;
 396     ifargs[3] = 0;
 397     ifr.ifr_data = (void *)ifargs;
 398     ret = ioctl(ctlfd, SIOCDEVPRIVATE, &ifr);
 399 #else
 400     ifr.ifr_ifindex = ifindex;
 401     ret = ioctl(ctlfd, SIOCBRADDIF, &ifr);
 402 #endif
 403     if (ret == -1) {
 404         fprintf(stderr, "failed to add interface `%s' to bridge `%s': %s\n",
 405                 iface, bridge, strerror(errno));
 406         ret = EXIT_FAILURE;
 407         goto cleanup;
 408     }
 409
 410     /* bring the interface up */
 411     prep_ifreq(&ifr, iface);
 412     if (ioctl(ctlfd, SIOCGIFFLAGS, &ifr) == -1) {
 413         fprintf(stderr, "failed to get interface flags for `%s': %s\n",
 414                 iface, strerror(errno));
 415         ret = EXIT_FAILURE;
 416         goto cleanup;
 417     }
 418
 419     ifr.ifr_flags |= IFF_UP;
 420     if (ioctl(ctlfd, SIOCSIFFLAGS, &ifr) == -1) {
 421         fprintf(stderr, "failed to bring up interface `%s': %s\n",
 422                 iface, strerror(errno));
 423         ret = EXIT_FAILURE;
 424         goto cleanup;
 425     }
 426
 427     /* write fd to the domain socket */
 428     if (send_fd(unixfd, fd) == -1) {
 429         fprintf(stderr, "failed to write fd to unix socket: %s\n",
 430                 strerror(errno));
 431         ret = EXIT_FAILURE;
 432         goto cleanup;
 433     }
 434
 435     /* ... */
 436
 437     /* profit! */
 438
 439 cleanup:
 440     if (fd >= 0) {
 441         close(fd);
 442     }
 443     if (ctlfd >= 0) {
 444         close(ctlfd);
 445     }
 446     while ((acl_rule = QSIMPLEQ_FIRST(&acl_list)) != NULL) {
 447         QSIMPLEQ_REMOVE_HEAD(&acl_list, entry);
 448         g_free(acl_rule);
 449     }
 450
 451     return ret;
 452 }