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  * Known shortcomings:
  17  * - There is no manual page
  18  * - The syntax of the ACL file is not documented anywhere
  19  * - parse_acl_file() doesn't report fopen() failure properly, fails
  20  *   to check ferror() after fgets() failure, arbitrarily truncates
  21  *   long lines, handles whitespace inconsistently, error messages
  22  *   don't point to the offending file and line, errors in included
  23  *   files are reported, but otherwise ignored, ...
  24  */
  25
  26 #include "qemu/osdep.h"
  27
  28
  29 #include <sys/ioctl.h>
  30 #include <sys/socket.h>
  31 #include <sys/un.h>
  32 #include <sys/prctl.h>
  33
  34 #include <net/if.h>
  35
  36 #include <linux/sockios.h>
  37
  38 #ifndef SIOCBRADDIF
  39 #include <linux/if_bridge.h>
  40 #endif
  41
  42 #include "qemu/queue.h"
  43
  44 #include "net/tap-linux.h"
  45
  46 #ifdef CONFIG_LIBCAP
  47 #include <cap-ng.h>
  48 #endif
  49
  50 #define DEFAULT_ACL_FILE CONFIG_QEMU_CONFDIR "/bridge.conf"
  51
  52 enum {
  53     ACL_ALLOW = 0,
  54     ACL_ALLOW_ALL,
  55     ACL_DENY,
  56     ACL_DENY_ALL,
  57 };
  58
  59 typedef struct ACLRule {
  60     int type;
  61     char iface[IFNAMSIZ];
  62     QSIMPLEQ_ENTRY(ACLRule) entry;
  63 } ACLRule;
  64
  65 typedef QSIMPLEQ_HEAD(ACLList, ACLRule) ACLList;
  66
  67 static void usage(void)
  68 {
  69     fprintf(stderr,
  70             "Usage: qemu-bridge-helper [--use-vnet] --br=bridge --fd=unixfd\n");
  71 }
  72
  73 static int parse_acl_file(const char *filename, ACLList *acl_list)
  74 {
  75     FILE *f;
  76     char line[4096];
  77     ACLRule *acl_rule;
  78
  79     f = fopen(filename, "r");
  80     if (f == NULL) {
  81         return -1;
  82     }
  83
  84     while (fgets(line, sizeof(line), f) != NULL) {
  85         char *ptr = line;
  86         char *cmd, *arg, *argend;
  87
  88         while (g_ascii_isspace(*ptr)) {
  89             ptr++;
  90         }
  91
  92         /* skip comments and empty lines */
  93         if (*ptr == '#' || *ptr == 0) {
  94             continue;
  95         }
  96
  97         cmd = ptr;
  98         arg = strchr(cmd, ' ');
  99         if (arg == NULL) {
 100             arg = strchr(cmd, '\t');
 101         }
 102
 103         if (arg == NULL) {
 104             fprintf(stderr, "Invalid config line:\n  %s\n", line);
 105             fclose(f);
 106             errno = EINVAL;
 107             return -1;
 108         }
 109
 110         *arg = 0;
 111         arg++;
 112         while (g_ascii_isspace(*arg)) {
 113             arg++;
 114         }
 115
 116         argend = arg + strlen(arg);
 117         while (arg != argend && g_ascii_isspace(*(argend - 1))) {
 118             argend--;
 119         }
 120         *argend = 0;
 121
 122         if (!g_str_equal(cmd, "include") && strlen(arg) >= IFNAMSIZ) {
 123             fprintf(stderr, "name `%s' too long: %zu\n", arg, strlen(arg));
 124             fclose(f);
 125             errno = EINVAL;
 126             return -1;
 127         }
 128
 129         if (strcmp(cmd, "deny") == 0) {
 130             acl_rule = g_malloc(sizeof(*acl_rule));
 131             if (strcmp(arg, "all") == 0) {
 132                 acl_rule->type = ACL_DENY_ALL;
 133             } else {
 134                 acl_rule->type = ACL_DENY;
 135                 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg);
 136             }
 137             QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry);
 138         } else if (strcmp(cmd, "allow") == 0) {
 139             acl_rule = g_malloc(sizeof(*acl_rule));
 140             if (strcmp(arg, "all") == 0) {
 141                 acl_rule->type = ACL_ALLOW_ALL;
 142             } else {
 143                 acl_rule->type = ACL_ALLOW;
 144                 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg);
 145             }
 146             QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry);
 147         } else if (strcmp(cmd, "include") == 0) {
 148             /* ignore errors */
 149             parse_acl_file(arg, acl_list);
 150         } else {
 151             fprintf(stderr, "Unknown command `%s'\n", cmd);
 152             fclose(f);
 153             errno = EINVAL;
 154             return -1;
 155         }
 156     }
 157
 158     fclose(f);
 159
 160     return 0;
 161 }
 162
 163 static bool has_vnet_hdr(int fd)
 164 {
 165     unsigned int features = 0;
 166
 167     if (ioctl(fd, TUNGETFEATURES, &features) == -1) {
 168         return false;
 169     }
 170
 171     if (!(features & IFF_VNET_HDR)) {
 172         return false;
 173     }
 174
 175     return true;
 176 }
 177
 178 static void prep_ifreq(struct ifreq *ifr, const char *ifname)
 179 {
 180     memset(ifr, 0, sizeof(*ifr));
 181     snprintf(ifr->ifr_name, IFNAMSIZ, "%s", ifname);
 182 }
 183
 184 static int send_fd(int c, int fd)
 185 {
 186     char msgbuf[CMSG_SPACE(sizeof(fd))];
 187     struct msghdr msg = {
 188         .msg_control = msgbuf,
 189         .msg_controllen = sizeof(msgbuf),
 190     };
 191     struct cmsghdr *cmsg;
 192     struct iovec iov;
 193     char req[1] = { 0x00 };
 194
 195     cmsg = CMSG_FIRSTHDR(&msg);
 196     cmsg->cmsg_level = SOL_SOCKET;
 197     cmsg->cmsg_type = SCM_RIGHTS;
 198     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 199     msg.msg_controllen = cmsg->cmsg_len;
 200
 201     iov.iov_base = req;
 202     iov.iov_len = sizeof(req);
 203
 204     msg.msg_iov = &iov;
 205     msg.msg_iovlen = 1;
 206     memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
 207
 208     return sendmsg(c, &msg, 0);
 209 }
 210
 211 #ifdef CONFIG_LIBCAP
 212 static int drop_privileges(void)
 213 {
 214     /* clear all capabilities */
 215     capng_clear(CAPNG_SELECT_BOTH);
 216
 217     if (capng_update(CAPNG_ADD, CAPNG_EFFECTIVE | CAPNG_PERMITTED,
 218                      CAP_NET_ADMIN) < 0) {
 219         return -1;
 220     }
 221
 222     /* change to calling user's real uid and gid, retaining supplemental
 223      * groups and CAP_NET_ADMIN */
 224     if (capng_change_id(getuid(), getgid(), CAPNG_CLEAR_BOUNDING)) {
 225         return -1;
 226     }
 227
 228     return 0;
 229 }
 230 #endif
 231
 232 int main(int argc, char **argv)
 233 {
 234     struct ifreq ifr;
 235 #ifndef SIOCBRADDIF
 236     unsigned long ifargs[4];
 237 #endif
 238     int ifindex;
 239     int fd = -1, ctlfd = -1, unixfd = -1;
 240     int use_vnet = 0;
 241     int mtu;
 242     const char *bridge = NULL;
 243     char iface[IFNAMSIZ];
 244     int index;
 245     ACLRule *acl_rule;
 246     ACLList acl_list;
 247     int access_allowed, access_denied;
 248     int ret = EXIT_SUCCESS;
 249
 250 #ifdef CONFIG_LIBCAP
 251     /* if we're run from an suid binary, immediately drop privileges preserving
 252      * cap_net_admin */
 253     if (geteuid() == 0 && getuid() != geteuid()) {
 254         if (drop_privileges() == -1) {
 255             fprintf(stderr, "failed to drop privileges\n");
 256             return 1;
 257         }
 258     }
 259 #endif
 260
 261     /* parse arguments */
 262     for (index = 1; index < argc; index++) {
 263         if (strcmp(argv[index], "--use-vnet") == 0) {
 264             use_vnet = 1;
 265         } else if (strncmp(argv[index], "--br=", 5) == 0) {
 266             bridge = &argv[index][5];
 267         } else if (strncmp(argv[index], "--fd=", 5) == 0) {
 268             unixfd = atoi(&argv[index][5]);
 269         } else {
 270             usage();
 271             return EXIT_FAILURE;
 272         }
 273     }
 274
 275     if (bridge == NULL || unixfd == -1) {
 276         usage();
 277         return EXIT_FAILURE;
 278     }
 279     if (strlen(bridge) >= IFNAMSIZ) {
 280         fprintf(stderr, "name `%s' too long: %zu\n", bridge, strlen(bridge));
 281         return EXIT_FAILURE;
 282     }
 283
 284     /* parse default acl file */
 285     QSIMPLEQ_INIT(&acl_list);
 286     if (parse_acl_file(DEFAULT_ACL_FILE, &acl_list) == -1) {
 287         fprintf(stderr, "failed to parse default acl file `%s'\n",
 288                 DEFAULT_ACL_FILE);
 289         ret = EXIT_FAILURE;
 290         goto cleanup;
 291     }
 292
 293     /* validate bridge against acl -- default policy is to deny
 294      * according acl policy if we have a deny and allow both
 295      * then deny should always win over allow
 296      */
 297     access_allowed = 0;
 298     access_denied = 0;
 299     QSIMPLEQ_FOREACH(acl_rule, &acl_list, entry) {
 300         switch (acl_rule->type) {
 301         case ACL_ALLOW_ALL:
 302             access_allowed = 1;
 303             break;
 304         case ACL_ALLOW:
 305             if (strcmp(bridge, acl_rule->iface) == 0) {
 306                 access_allowed = 1;
 307             }
 308             break;
 309         case ACL_DENY_ALL:
 310             access_denied = 1;
 311             break;
 312         case ACL_DENY:
 313             if (strcmp(bridge, acl_rule->iface) == 0) {
 314                 access_denied = 1;
 315             }
 316             break;
 317         }
 318     }
 319
 320     if ((access_allowed == 0) || (access_denied == 1)) {
 321         fprintf(stderr, "access denied by acl file\n");
 322         ret = EXIT_FAILURE;
 323         goto cleanup;
 324     }
 325
 326     /* open a socket to use to control the network interfaces */
 327     ctlfd = socket(AF_INET, SOCK_STREAM, 0);
 328     if (ctlfd == -1) {
 329         fprintf(stderr, "failed to open control socket: %s\n", strerror(errno));
 330         ret = EXIT_FAILURE;
 331         goto cleanup;
 332     }
 333
 334     /* open the tap device */
 335     fd = open("/dev/net/tun", O_RDWR);
 336     if (fd == -1) {
 337         fprintf(stderr, "failed to open /dev/net/tun: %s\n", strerror(errno));
 338         ret = EXIT_FAILURE;
 339         goto cleanup;
 340     }
 341
 342     /* request a tap device, disable PI, and add vnet header support if
 343      * requested and it's available. */
 344     prep_ifreq(&ifr, "tap%d");
 345     ifr.ifr_flags = IFF_TAP|IFF_NO_PI;
 346     if (use_vnet && has_vnet_hdr(fd)) {
 347         ifr.ifr_flags |= IFF_VNET_HDR;
 348     }
 349
 350     if (ioctl(fd, TUNSETIFF, &ifr) == -1) {
 351         fprintf(stderr, "failed to create tun device: %s\n", strerror(errno));
 352         ret = EXIT_FAILURE;
 353         goto cleanup;
 354     }
 355
 356     /* save tap device name */
 357     snprintf(iface, sizeof(iface), "%s", ifr.ifr_name);
 358
 359     /* get the mtu of the bridge */
 360     prep_ifreq(&ifr, bridge);
 361     if (ioctl(ctlfd, SIOCGIFMTU, &ifr) == -1) {
 362         fprintf(stderr, "failed to get mtu of bridge `%s': %s\n",
 363                 bridge, strerror(errno));
 364         ret = EXIT_FAILURE;
 365         goto cleanup;
 366     }
 367
 368     /* save mtu */
 369     mtu = ifr.ifr_mtu;
 370
 371     /* set the mtu of the interface based on the bridge */
 372     prep_ifreq(&ifr, iface);
 373     ifr.ifr_mtu = mtu;
 374     if (ioctl(ctlfd, SIOCSIFMTU, &ifr) == -1) {
 375         fprintf(stderr, "failed to set mtu of device `%s' to %d: %s\n",
 376                 iface, mtu, strerror(errno));
 377         ret = EXIT_FAILURE;
 378         goto cleanup;
 379     }
 380
 381     /* Linux uses the lowest enslaved MAC address as the MAC address of
 382      * the bridge.  Set MAC address to a high value so that it doesn't
 383      * affect the MAC address of the bridge.
 384      */
 385     if (ioctl(ctlfd, SIOCGIFHWADDR, &ifr) < 0) {
 386         fprintf(stderr, "failed to get MAC address of device `%s': %s\n",
 387                 iface, strerror(errno));
 388         ret = EXIT_FAILURE;
 389         goto cleanup;
 390     }
 391     ifr.ifr_hwaddr.sa_data[0] = 0xFE;
 392     if (ioctl(ctlfd, SIOCSIFHWADDR, &ifr) < 0) {
 393         fprintf(stderr, "failed to set MAC address of device `%s': %s\n",
 394                 iface, strerror(errno));
 395         ret = EXIT_FAILURE;
 396         goto cleanup;
 397     }
 398
 399     /* add the interface to the bridge */
 400     prep_ifreq(&ifr, bridge);
 401     ifindex = if_nametoindex(iface);
 402 #ifndef SIOCBRADDIF
 403     ifargs[0] = BRCTL_ADD_IF;
 404     ifargs[1] = ifindex;
 405     ifargs[2] = 0;
 406     ifargs[3] = 0;
 407     ifr.ifr_data = (void *)ifargs;
 408     ret = ioctl(ctlfd, SIOCDEVPRIVATE, &ifr);
 409 #else
 410     ifr.ifr_ifindex = ifindex;
 411     ret = ioctl(ctlfd, SIOCBRADDIF, &ifr);
 412 #endif
 413     if (ret == -1) {
 414         fprintf(stderr, "failed to add interface `%s' to bridge `%s': %s\n",
 415                 iface, bridge, strerror(errno));
 416         ret = EXIT_FAILURE;
 417         goto cleanup;
 418     }
 419
 420     /* bring the interface up */
 421     prep_ifreq(&ifr, iface);
 422     if (ioctl(ctlfd, SIOCGIFFLAGS, &ifr) == -1) {
 423         fprintf(stderr, "failed to get interface flags for `%s': %s\n",
 424                 iface, strerror(errno));
 425         ret = EXIT_FAILURE;
 426         goto cleanup;
 427     }
 428
 429     ifr.ifr_flags |= IFF_UP;
 430     if (ioctl(ctlfd, SIOCSIFFLAGS, &ifr) == -1) {
 431         fprintf(stderr, "failed to bring up interface `%s': %s\n",
 432                 iface, strerror(errno));
 433         ret = EXIT_FAILURE;
 434         goto cleanup;
 435     }
 436
 437     /* write fd to the domain socket */
 438     if (send_fd(unixfd, fd) == -1) {
 439         fprintf(stderr, "failed to write fd to unix socket: %s\n",
 440                 strerror(errno));
 441         ret = EXIT_FAILURE;
 442         goto cleanup;
 443     }
 444
 445     /* ... */
 446
 447     /* profit! */
 448
 449 cleanup:
 450     if (fd >= 0) {
 451         close(fd);
 452     }
 453     if (ctlfd >= 0) {
 454         close(ctlfd);
 455     }
 456     while ((acl_rule = QSIMPLEQ_FIRST(&acl_list)) != NULL) {
 457         QSIMPLEQ_REMOVE_HEAD(&acl_list, entry);
 458         g_free(acl_rule);
 459     }
 460
 461     return ret;
 462 }