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 #include "qemu/cutils.h" 44 45 #include "net/tap-linux.h" 46 47 #ifdef CONFIG_LIBCAP_NG 48 #include <cap-ng.h> 49 #endif 50 51 #define DEFAULT_ACL_FILE CONFIG_QEMU_CONFDIR "/bridge.conf" 52 53 enum { 54 ACL_ALLOW = 0, 55 ACL_ALLOW_ALL, 56 ACL_DENY, 57 ACL_DENY_ALL, 58 }; 59 60 typedef struct ACLRule { 61 int type; 62 char iface[IFNAMSIZ]; 63 QSIMPLEQ_ENTRY(ACLRule) entry; 64 } ACLRule; 65 66 typedef QSIMPLEQ_HEAD(ACLList, ACLRule) ACLList; 67 68 static void usage(void) 69 { 70 fprintf(stderr, 71 "Usage: qemu-bridge-helper [--use-vnet] --br=bridge --fd=unixfd\n"); 72 } 73 74 static int parse_acl_file(const char *filename, ACLList *acl_list) 75 { 76 FILE *f; 77 char line[4096]; 78 ACLRule *acl_rule; 79 80 f = fopen(filename, "r"); 81 if (f == NULL) { 82 return -1; 83 } 84 85 while (fgets(line, sizeof(line), f) != NULL) { 86 char *ptr = line; 87 char *cmd, *arg, *argend; 88 89 while (g_ascii_isspace(*ptr)) { 90 ptr++; 91 } 92 93 /* skip comments and empty lines */ 94 if (*ptr == '#' || *ptr == 0) { 95 continue; 96 } 97 98 cmd = ptr; 99 arg = strchr(cmd, ' '); 100 if (arg == NULL) { 101 arg = strchr(cmd, '\t'); 102 } 103 104 if (arg == NULL) { 105 fprintf(stderr, "Invalid config line:\n %s\n", line); 106 goto err; 107 } 108 109 *arg = 0; 110 arg++; 111 while (g_ascii_isspace(*arg)) { 112 arg++; 113 } 114 115 argend = arg + strlen(arg); 116 while (arg != argend && g_ascii_isspace(*(argend - 1))) { 117 argend--; 118 } 119 *argend = 0; 120 121 if (!g_str_equal(cmd, "include") && strlen(arg) >= IFNAMSIZ) { 122 fprintf(stderr, "name `%s' too long: %zu\n", arg, strlen(arg)); 123 goto err; 124 } 125 126 if (strcmp(cmd, "deny") == 0) { 127 acl_rule = g_malloc(sizeof(*acl_rule)); 128 if (strcmp(arg, "all") == 0) { 129 acl_rule->type = ACL_DENY_ALL; 130 } else { 131 acl_rule->type = ACL_DENY; 132 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg); 133 } 134 QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry); 135 } else if (strcmp(cmd, "allow") == 0) { 136 acl_rule = g_malloc(sizeof(*acl_rule)); 137 if (strcmp(arg, "all") == 0) { 138 acl_rule->type = ACL_ALLOW_ALL; 139 } else { 140 acl_rule->type = ACL_ALLOW; 141 snprintf(acl_rule->iface, IFNAMSIZ, "%s", arg); 142 } 143 QSIMPLEQ_INSERT_TAIL(acl_list, acl_rule, entry); 144 } else if (strcmp(cmd, "include") == 0) { 145 /* ignore errors */ 146 parse_acl_file(arg, acl_list); 147 } else { 148 fprintf(stderr, "Unknown command `%s'\n", cmd); 149 goto err; 150 } 151 } 152 153 fclose(f); 154 return 0; 155 156 err: 157 fclose(f); 158 errno = EINVAL; 159 return -1; 160 161 } 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_NG 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 g_autofree char *acl_file = NULL; 250 251 #ifdef CONFIG_LIBCAP_NG 252 /* if we're run from an suid binary, immediately drop privileges preserving 253 * cap_net_admin */ 254 if (geteuid() == 0 && getuid() != geteuid()) { 255 if (drop_privileges() == -1) { 256 fprintf(stderr, "failed to drop privileges\n"); 257 return 1; 258 } 259 } 260 #endif 261 262 qemu_init_exec_dir(argv[0]); 263 264 /* parse arguments */ 265 for (index = 1; index < argc; index++) { 266 if (strcmp(argv[index], "--use-vnet") == 0) { 267 use_vnet = 1; 268 } else if (strncmp(argv[index], "--br=", 5) == 0) { 269 bridge = &argv[index][5]; 270 } else if (strncmp(argv[index], "--fd=", 5) == 0) { 271 unixfd = atoi(&argv[index][5]); 272 } else { 273 usage(); 274 return EXIT_FAILURE; 275 } 276 } 277 278 if (bridge == NULL || unixfd == -1) { 279 usage(); 280 return EXIT_FAILURE; 281 } 282 if (strlen(bridge) >= IFNAMSIZ) { 283 fprintf(stderr, "name `%s' too long: %zu\n", bridge, strlen(bridge)); 284 return EXIT_FAILURE; 285 } 286 287 /* parse default acl file */ 288 QSIMPLEQ_INIT(&acl_list); 289 acl_file = get_relocated_path(DEFAULT_ACL_FILE); 290 if (parse_acl_file(acl_file, &acl_list) == -1) { 291 fprintf(stderr, "failed to parse default acl file `%s'\n", 292 acl_file); 293 ret = EXIT_FAILURE; 294 goto cleanup; 295 } 296 297 /* validate bridge against acl -- default policy is to deny 298 * according acl policy if we have a deny and allow both 299 * then deny should always win over allow 300 */ 301 access_allowed = 0; 302 access_denied = 0; 303 QSIMPLEQ_FOREACH(acl_rule, &acl_list, entry) { 304 switch (acl_rule->type) { 305 case ACL_ALLOW_ALL: 306 access_allowed = 1; 307 break; 308 case ACL_ALLOW: 309 if (strcmp(bridge, acl_rule->iface) == 0) { 310 access_allowed = 1; 311 } 312 break; 313 case ACL_DENY_ALL: 314 access_denied = 1; 315 break; 316 case ACL_DENY: 317 if (strcmp(bridge, acl_rule->iface) == 0) { 318 access_denied = 1; 319 } 320 break; 321 } 322 } 323 324 if ((access_allowed == 0) || (access_denied == 1)) { 325 fprintf(stderr, "access denied by acl file\n"); 326 ret = EXIT_FAILURE; 327 goto cleanup; 328 } 329 330 /* open a socket to use to control the network interfaces */ 331 ctlfd = socket(AF_INET, SOCK_STREAM, 0); 332 if (ctlfd == -1) { 333 fprintf(stderr, "failed to open control socket: %s\n", strerror(errno)); 334 ret = EXIT_FAILURE; 335 goto cleanup; 336 } 337 338 /* open the tap device */ 339 fd = open("/dev/net/tun", O_RDWR); 340 if (fd == -1) { 341 fprintf(stderr, "failed to open /dev/net/tun: %s\n", strerror(errno)); 342 ret = EXIT_FAILURE; 343 goto cleanup; 344 } 345 346 /* request a tap device, disable PI, and add vnet header support if 347 * requested and it's available. */ 348 prep_ifreq(&ifr, "tap%d"); 349 ifr.ifr_flags = IFF_TAP|IFF_NO_PI; 350 if (use_vnet && has_vnet_hdr(fd)) { 351 ifr.ifr_flags |= IFF_VNET_HDR; 352 } 353 354 if (ioctl(fd, TUNSETIFF, &ifr) == -1) { 355 fprintf(stderr, "failed to create tun device: %s\n", strerror(errno)); 356 ret = EXIT_FAILURE; 357 goto cleanup; 358 } 359 360 /* save tap device name */ 361 snprintf(iface, sizeof(iface), "%s", ifr.ifr_name); 362 363 /* get the mtu of the bridge */ 364 prep_ifreq(&ifr, bridge); 365 if (ioctl(ctlfd, SIOCGIFMTU, &ifr) == -1) { 366 fprintf(stderr, "failed to get mtu of bridge `%s': %s\n", 367 bridge, strerror(errno)); 368 ret = EXIT_FAILURE; 369 goto cleanup; 370 } 371 372 /* save mtu */ 373 mtu = ifr.ifr_mtu; 374 375 /* set the mtu of the interface based on the bridge */ 376 prep_ifreq(&ifr, iface); 377 ifr.ifr_mtu = mtu; 378 if (ioctl(ctlfd, SIOCSIFMTU, &ifr) == -1) { 379 fprintf(stderr, "failed to set mtu of device `%s' to %d: %s\n", 380 iface, mtu, strerror(errno)); 381 ret = EXIT_FAILURE; 382 goto cleanup; 383 } 384 385 /* Linux uses the lowest enslaved MAC address as the MAC address of 386 * the bridge. Set MAC address to a high value so that it doesn't 387 * affect the MAC address of the bridge. 388 */ 389 if (ioctl(ctlfd, SIOCGIFHWADDR, &ifr) < 0) { 390 fprintf(stderr, "failed to get MAC address of device `%s': %s\n", 391 iface, strerror(errno)); 392 ret = EXIT_FAILURE; 393 goto cleanup; 394 } 395 ifr.ifr_hwaddr.sa_data[0] = 0xFE; 396 if (ioctl(ctlfd, SIOCSIFHWADDR, &ifr) < 0) { 397 fprintf(stderr, "failed to set MAC address of device `%s': %s\n", 398 iface, strerror(errno)); 399 ret = EXIT_FAILURE; 400 goto cleanup; 401 } 402 403 /* add the interface to the bridge */ 404 prep_ifreq(&ifr, bridge); 405 ifindex = if_nametoindex(iface); 406 #ifndef SIOCBRADDIF 407 ifargs[0] = BRCTL_ADD_IF; 408 ifargs[1] = ifindex; 409 ifargs[2] = 0; 410 ifargs[3] = 0; 411 ifr.ifr_data = (void *)ifargs; 412 ret = ioctl(ctlfd, SIOCDEVPRIVATE, &ifr); 413 #else 414 ifr.ifr_ifindex = ifindex; 415 ret = ioctl(ctlfd, SIOCBRADDIF, &ifr); 416 #endif 417 if (ret == -1) { 418 fprintf(stderr, "failed to add interface `%s' to bridge `%s': %s\n", 419 iface, bridge, strerror(errno)); 420 ret = EXIT_FAILURE; 421 goto cleanup; 422 } 423 424 /* bring the interface up */ 425 prep_ifreq(&ifr, iface); 426 if (ioctl(ctlfd, SIOCGIFFLAGS, &ifr) == -1) { 427 fprintf(stderr, "failed to get interface flags for `%s': %s\n", 428 iface, strerror(errno)); 429 ret = EXIT_FAILURE; 430 goto cleanup; 431 } 432 433 ifr.ifr_flags |= IFF_UP; 434 if (ioctl(ctlfd, SIOCSIFFLAGS, &ifr) == -1) { 435 fprintf(stderr, "failed to bring up interface `%s': %s\n", 436 iface, strerror(errno)); 437 ret = EXIT_FAILURE; 438 goto cleanup; 439 } 440 441 /* write fd to the domain socket */ 442 if (send_fd(unixfd, fd) == -1) { 443 fprintf(stderr, "failed to write fd to unix socket: %s\n", 444 strerror(errno)); 445 ret = EXIT_FAILURE; 446 goto cleanup; 447 } 448 449 /* ... */ 450 451 /* profit! */ 452 453 cleanup: 454 if (fd >= 0) { 455 close(fd); 456 } 457 if (ctlfd >= 0) { 458 close(ctlfd); 459 } 460 while ((acl_rule = QSIMPLEQ_FIRST(&acl_list)) != NULL) { 461 QSIMPLEQ_REMOVE_HEAD(&acl_list, entry); 462 g_free(acl_rule); 463 } 464 465 return ret; 466 } 467