Harald Welte | f736559 | 2020-04-15 21:48:45 +0200 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
| 2 | #include <unistd.h> |
| 3 | #include <stdint.h> |
| 4 | #include <stdbool.h> |
| 5 | #include <stdlib.h> |
| 6 | #include <string.h> |
| 7 | #include <stdio.h> |
| 8 | #include <errno.h> |
| 9 | #include <sys/types.h> |
| 10 | #include <sys/socket.h> |
| 11 | #include <sys/stat.h> |
| 12 | #include <fcntl.h> |
| 13 | #include <signal.h> |
| 14 | #include <netdb.h> |
| 15 | |
| 16 | #include <netinet/ip.h> |
| 17 | #include <netinet/ip6.h> |
| 18 | |
| 19 | #include <pthread.h> |
| 20 | |
| 21 | #include <linux/if.h> |
| 22 | #include <linux/if_tun.h> |
| 23 | #include <sys/ioctl.h> |
| 24 | |
Harald Welte | f2be099 | 2020-04-20 11:12:42 +0000 | [diff] [blame] | 25 | #include <linux/netlink.h> |
Harald Welte | f736559 | 2020-04-15 21:48:45 +0200 | [diff] [blame] | 26 | #include <netlink/socket.h> |
| 27 | #include <netlink/route/link.h> |
| 28 | |
| 29 | #include <osmocom/core/linuxlist.h> |
| 30 | #include <osmocom/core/talloc.h> |
| 31 | #include <osmocom/core/logging.h> |
| 32 | #include <osmocom/core/utils.h> |
| 33 | |
| 34 | #include "gtp.h" |
| 35 | #include "internal.h" |
| 36 | #include "netns.h" |
| 37 | |
| 38 | /*********************************************************************** |
| 39 | * TUN Device |
| 40 | ***********************************************************************/ |
| 41 | |
| 42 | #define LOGTUN(tun, lvl, fmt, args ...) \ |
| 43 | LOGP(DTUN, lvl, "%s: " fmt, (tun)->devname, ## args) |
| 44 | |
| 45 | /* extracted information from a packet */ |
| 46 | struct pkt_info { |
| 47 | struct sockaddr_storage saddr; |
| 48 | struct sockaddr_storage daddr; |
| 49 | uint8_t proto; |
| 50 | }; |
| 51 | |
| 52 | static int parse_pkt(struct pkt_info *out, const uint8_t *in, unsigned int in_len) |
| 53 | { |
| 54 | const struct iphdr *ip4 = (struct iphdr *) in; |
| 55 | const uint16_t *l4h = NULL; |
| 56 | |
| 57 | memset(out, 0, sizeof(*out)); |
| 58 | |
| 59 | if (ip4->version == 4) { |
| 60 | struct sockaddr_in *saddr4 = (struct sockaddr_in *) &out->saddr; |
| 61 | struct sockaddr_in *daddr4 = (struct sockaddr_in *) &out->daddr; |
| 62 | |
| 63 | if (in_len < sizeof(*ip4) || in_len < 4*ip4->ihl) |
| 64 | return -1; |
| 65 | |
| 66 | saddr4->sin_family = AF_INET; |
| 67 | saddr4->sin_addr.s_addr = ip4->saddr; |
| 68 | |
| 69 | daddr4->sin_family = AF_INET; |
| 70 | daddr4->sin_addr.s_addr = ip4->daddr; |
| 71 | |
| 72 | out->proto = ip4->protocol; |
| 73 | l4h = (const uint16_t *) (in + sizeof(*ip4)); |
| 74 | |
| 75 | switch (out->proto) { |
| 76 | case IPPROTO_TCP: |
| 77 | case IPPROTO_UDP: |
| 78 | case IPPROTO_DCCP: |
| 79 | case IPPROTO_SCTP: |
| 80 | case IPPROTO_UDPLITE: |
| 81 | saddr4->sin_port = ntohs(l4h[0]); |
| 82 | daddr4->sin_port = ntohs(l4h[1]); |
| 83 | break; |
| 84 | default: |
| 85 | break; |
| 86 | } |
| 87 | } else if (ip4->version == 6) { |
| 88 | const struct ip6_hdr *ip6 = (struct ip6_hdr *) in; |
| 89 | struct sockaddr_in6 *saddr6 = (struct sockaddr_in6 *) &out->saddr; |
| 90 | struct sockaddr_in6 *daddr6 = (struct sockaddr_in6 *) &out->daddr; |
| 91 | |
| 92 | if (in_len < sizeof(*ip6)) |
| 93 | return -1; |
| 94 | |
| 95 | saddr6->sin6_family = AF_INET6; |
| 96 | saddr6->sin6_addr = ip6->ip6_src; |
| 97 | |
| 98 | daddr6->sin6_family = AF_INET6; |
| 99 | daddr6->sin6_addr = ip6->ip6_dst; |
| 100 | |
| 101 | /* FIXME: ext hdr */ |
| 102 | out->proto = ip6->ip6_nxt; |
| 103 | l4h = (const uint16_t *) (in + sizeof(*ip6)); |
| 104 | |
| 105 | switch (out->proto) { |
| 106 | case IPPROTO_TCP: |
| 107 | case IPPROTO_UDP: |
| 108 | case IPPROTO_DCCP: |
| 109 | case IPPROTO_SCTP: |
| 110 | case IPPROTO_UDPLITE: |
| 111 | saddr6->sin6_port = ntohs(l4h[0]); |
| 112 | daddr6->sin6_port = ntohs(l4h[1]); |
| 113 | break; |
| 114 | default: |
| 115 | break; |
| 116 | } |
| 117 | } else |
| 118 | return -1; |
| 119 | |
| 120 | return 0; |
| 121 | } |
| 122 | |
| 123 | /* one thread for reading from each TUN device (TUN -> GTP encapsulation) */ |
| 124 | static void *tun_device_thread(void *arg) |
| 125 | { |
| 126 | struct tun_device *tun = (struct tun_device *)arg; |
| 127 | struct gtp_daemon *d = tun->d; |
| 128 | |
| 129 | uint8_t base_buffer[MAX_UDP_PACKET+sizeof(struct gtp1_header)]; |
| 130 | struct gtp1_header *gtph = (struct gtp1_header *)base_buffer; |
| 131 | uint8_t *buffer = base_buffer + sizeof(struct gtp1_header); |
| 132 | |
| 133 | struct sockaddr_storage daddr; |
| 134 | |
| 135 | /* initialize the fixed part of the GTP header */ |
| 136 | gtph->flags = 0x30; |
| 137 | gtph->type = GTP_TPDU; |
| 138 | |
| 139 | while (1) { |
| 140 | struct gtp_tunnel *t; |
| 141 | struct pkt_info pinfo; |
| 142 | int rc, nread, outfd; |
| 143 | |
| 144 | /* 1) read from tun */ |
| 145 | rc = read(tun->fd, buffer, MAX_UDP_PACKET); |
| 146 | if (rc < 0) { |
| 147 | LOGTUN(tun, LOGL_FATAL, "Error readingfrom tun device: %s\n", strerror(errno)); |
| 148 | exit(1); |
| 149 | } |
| 150 | nread = rc; |
| 151 | gtph->length = htons(nread); |
| 152 | |
| 153 | rc = parse_pkt(&pinfo, buffer, nread); |
| 154 | if (rc < 0) { |
| 155 | LOGTUN(tun, LOGL_NOTICE, "Error parsing IP packet: %s\n", |
| 156 | osmo_hexdump(buffer, nread)); |
| 157 | continue; |
| 158 | } |
| 159 | |
| 160 | if (pinfo.saddr.ss_family == AF_INET6 && pinfo.proto == IPPROTO_ICMPV6) { |
| 161 | /* 2) TODO: magic voodoo for IPv6 neighbor discovery */ |
| 162 | } |
| 163 | |
| 164 | /* 3) look-up tunnel based on source IP address (+ filter) */ |
| 165 | pthread_rwlock_rdlock(&d->rwlock); |
| 166 | t = _gtp_tunnel_find_eua(tun, (struct sockaddr *) &pinfo.saddr, pinfo.proto); |
| 167 | if (!t) { |
| 168 | char host[128]; |
| 169 | char port[8]; |
| 170 | pthread_rwlock_unlock(&d->rwlock); |
| 171 | getnameinfo((const struct sockaddr *)&pinfo.saddr, |
| 172 | sizeof(pinfo.saddr), host, sizeof(host), port, sizeof(port), |
| 173 | NI_NUMERICHOST | NI_NUMERICSERV); |
| 174 | LOGTUN(tun, LOGL_NOTICE, "No tunnel found for source address %s:%s\n", host, port); |
| 175 | continue; |
| 176 | } |
| 177 | outfd = t->gtp_ep->fd; |
| 178 | memcpy(&daddr, &t->remote_udp, sizeof(daddr)); |
| 179 | gtph->tid = htonl(t->tx_teid); |
| 180 | pthread_rwlock_unlock(&d->rwlock); |
| 181 | |
| 182 | /* 4) write to GTP/UDP socket */ |
| 183 | rc = sendto(outfd, base_buffer, nread+sizeof(*gtph), 0, |
| 184 | (struct sockaddr *)&daddr, sizeof(daddr)); |
| 185 | if (rc < 0) { |
| 186 | LOGTUN(tun, LOGL_FATAL, "Error Writing to UDP socket: %s\n", strerror(errno)); |
| 187 | exit(1); |
| 188 | } |
| 189 | } |
| 190 | } |
| 191 | |
| 192 | static int tun_open(int flags, const char *name) |
| 193 | { |
| 194 | struct ifreq ifr; |
| 195 | int fd, rc; |
| 196 | |
| 197 | fd = open("/dev/net/tun", O_RDWR); |
| 198 | if (fd < 0) { |
| 199 | LOGP(DTUN, LOGL_ERROR, "Cannot open /dev/net/tun: %s\n", strerror(errno)); |
| 200 | return fd; |
| 201 | } |
| 202 | |
| 203 | memset(&ifr, 0, sizeof(ifr)); |
| 204 | ifr.ifr_flags = IFF_TUN | IFF_NO_PI | flags; |
| 205 | if (name) { |
| 206 | /* if a TUN interface name was specified, put it in the structure; otherwise, |
| 207 | the kernel will try to allocate the "next" device of the specified type */ |
Harald Welte | f23abd7 | 2020-04-20 12:09:32 +0200 | [diff] [blame] | 208 | osmo_strlcpy(ifr.ifr_name, name, IFNAMSIZ); |
Harald Welte | f736559 | 2020-04-15 21:48:45 +0200 | [diff] [blame] | 209 | } |
| 210 | |
| 211 | /* try to create the device */ |
| 212 | rc = ioctl(fd, TUNSETIFF, (void *) &ifr); |
| 213 | if (rc < 0) { |
| 214 | close(fd); |
| 215 | return rc; |
| 216 | } |
| 217 | |
| 218 | /* FIXME: read name back from device? */ |
| 219 | /* FIXME: SIOCSIFTXQLEN / SIOCSIFFLAGS */ |
| 220 | |
| 221 | return fd; |
| 222 | } |
| 223 | |
| 224 | static struct tun_device * |
| 225 | _tun_device_create(struct gtp_daemon *d, const char *devname, const char *netns_name) |
| 226 | { |
| 227 | struct rtnl_link *link; |
| 228 | struct tun_device *tun; |
| 229 | sigset_t oldmask; |
| 230 | int rc; |
| 231 | |
| 232 | tun = talloc_zero(d, struct tun_device); |
| 233 | if (!tun) |
| 234 | return NULL; |
| 235 | |
| 236 | tun->d = d; |
| 237 | tun->use_count = 1; |
| 238 | tun->devname = talloc_strdup(tun, devname); |
| 239 | |
| 240 | if (netns_name) { |
| 241 | tun->netns_name = talloc_strdup(tun, netns_name); |
| 242 | tun->netns_fd = get_nsfd(tun->netns_name); |
| 243 | if (tun->netns_fd < 0) { |
| 244 | LOGTUN(tun, LOGL_ERROR, "Cannot obtain netns file descriptor: %s\n", |
| 245 | strerror(errno)); |
| 246 | goto err_free; |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | /* temporarily switch to specified namespace to create tun device */ |
| 251 | if (tun->netns_name) { |
| 252 | rc = switch_ns(tun->netns_fd, &oldmask); |
| 253 | if (rc < 0) { |
| 254 | LOGTUN(tun, LOGL_ERROR, "Cannot switch to netns '%s': %s\n", |
| 255 | tun->netns_name, strerror(errno)); |
| 256 | goto err_close_ns; |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | tun->fd = tun_open(0, tun->devname); |
| 261 | if (tun->fd < 0) { |
| 262 | LOGTUN(tun, LOGL_ERROR, "Cannot open TUN device: %s\n", strerror(errno)); |
| 263 | goto err_restore_ns; |
| 264 | } |
| 265 | |
| 266 | tun->nl = nl_socket_alloc(); |
| 267 | if (!tun->nl || nl_connect(tun->nl, NETLINK_ROUTE) < 0) { |
| 268 | LOGTUN(tun, LOGL_ERROR, "Cannot create netlink socket in namespace '%s'\n", |
| 269 | tun->netns_name); |
| 270 | goto err_close; |
| 271 | } |
| 272 | |
| 273 | rc = rtnl_link_get_kernel(tun->nl, 0, tun->devname, &link); |
| 274 | if (rc < 0) { |
| 275 | LOGTUN(tun, LOGL_ERROR, "Cannot get ifindex for netif after create?!?\n"); |
| 276 | goto err_free_nl; |
| 277 | } |
| 278 | tun->ifindex = rtnl_link_get_ifindex(link); |
| 279 | rtnl_link_put(link); |
| 280 | |
| 281 | /* switch back to default namespace before creating new thread */ |
| 282 | if (tun->netns_name) |
| 283 | OSMO_ASSERT(restore_ns(&oldmask) == 0); |
| 284 | |
| 285 | /* bring the network device up */ |
| 286 | rc = netdev_set_link(tun->nl, tun->ifindex, true); |
| 287 | if (rc < 0) |
| 288 | LOGTUN(tun, LOGL_ERROR, "Cannot set interface to 'up'\n"); |
| 289 | |
| 290 | if (tun->netns_name) { |
| 291 | rc = netdev_add_defaultroute(tun->nl, tun->ifindex, AF_INET); |
| 292 | if (rc < 0) |
| 293 | LOGTUN(tun, LOGL_ERROR, "Cannot add IPv4 default route\n"); |
| 294 | else |
| 295 | LOGTUN(tun, LOGL_INFO, "Added IPv4 default route\n"); |
| 296 | |
| 297 | rc = netdev_add_defaultroute(tun->nl, tun->ifindex, AF_INET6); |
| 298 | if (rc < 0) |
| 299 | LOGTUN(tun, LOGL_ERROR, "Cannot add IPv6 default route\n"); |
| 300 | else |
| 301 | LOGTUN(tun, LOGL_INFO, "Added IPv6 default route\n"); |
| 302 | } |
| 303 | |
| 304 | if (pthread_create(&tun->thread, NULL, tun_device_thread, tun)) { |
| 305 | LOGTUN(tun, LOGL_ERROR, "Cannot create TUN thread: %s\n", strerror(errno)); |
| 306 | goto err_free_nl; |
| 307 | } |
| 308 | |
| 309 | LOGTUN(tun, LOGL_INFO, "Created (in netns '%s')\n", tun->netns_name); |
| 310 | llist_add_tail(&tun->list, &d->tun_devices); |
| 311 | |
| 312 | return tun; |
| 313 | |
| 314 | err_free_nl: |
| 315 | nl_socket_free(tun->nl); |
| 316 | err_close: |
| 317 | close(tun->fd); |
| 318 | err_restore_ns: |
| 319 | if (tun->netns_name) |
| 320 | OSMO_ASSERT(restore_ns(&oldmask) == 0); |
| 321 | err_close_ns: |
| 322 | if (tun->netns_name) |
| 323 | close(tun->netns_fd); |
| 324 | err_free: |
| 325 | talloc_free(tun); |
| 326 | return NULL; |
| 327 | } |
| 328 | |
| 329 | struct tun_device * |
| 330 | _tun_device_find(struct gtp_daemon *d, const char *devname) |
| 331 | { |
| 332 | struct tun_device *tun; |
| 333 | |
| 334 | llist_for_each_entry(tun, &d->tun_devices, list) { |
| 335 | if (!strcmp(tun->devname, devname)) |
| 336 | return tun; |
| 337 | } |
| 338 | return NULL; |
| 339 | } |
| 340 | |
Harald Welte | 24557a7 | 2020-04-17 22:08:29 +0200 | [diff] [blame] | 341 | /* find the first tun device within given named netns */ |
| 342 | struct tun_device * |
| 343 | tun_device_find_netns(struct gtp_daemon *d, const char *netns_name) |
| 344 | { |
| 345 | struct tun_device *tun; |
| 346 | |
| 347 | pthread_rwlock_rdlock(&d->rwlock); |
| 348 | llist_for_each_entry(tun, &d->tun_devices, list) { |
| 349 | if (!strcmp(tun->netns_name, netns_name)) { |
| 350 | pthread_rwlock_unlock(&d->rwlock); |
| 351 | return tun; |
| 352 | } |
| 353 | } |
| 354 | pthread_rwlock_unlock(&d->rwlock); |
| 355 | return NULL; |
| 356 | } |
| 357 | |
Harald Welte | f736559 | 2020-04-15 21:48:45 +0200 | [diff] [blame] | 358 | struct tun_device * |
| 359 | tun_device_find_or_create(struct gtp_daemon *d, const char *devname, const char *netns_name) |
| 360 | { |
| 361 | struct tun_device *tun; |
| 362 | |
| 363 | /* talloc is not thread safe, all alloc/free must come from main thread */ |
| 364 | ASSERT_MAIN_THREAD(d); |
| 365 | |
| 366 | pthread_rwlock_wrlock(&d->rwlock); |
| 367 | tun = _tun_device_find(d, devname); |
| 368 | if (tun) |
| 369 | tun->use_count++; |
| 370 | else |
| 371 | tun = _tun_device_create(d, devname, netns_name); |
| 372 | pthread_rwlock_unlock(&d->rwlock); |
| 373 | |
| 374 | return tun; |
| 375 | } |
| 376 | |
| 377 | /* UNLOCKED hard/forced destroy; caller must make sure references are cleaned up */ |
| 378 | static void _tun_device_destroy(struct tun_device *tun) |
| 379 | { |
| 380 | /* talloc is not thread safe, all alloc/free must come from main thread */ |
| 381 | ASSERT_MAIN_THREAD(tun->d); |
| 382 | |
| 383 | pthread_cancel(tun->thread); |
| 384 | llist_del(&tun->list); |
| 385 | if (tun->netns_name) |
| 386 | close(tun->netns_fd); |
| 387 | close(tun->fd); |
| 388 | nl_socket_free(tun->nl); |
| 389 | LOGTUN(tun, LOGL_INFO, "Destroying\n"); |
| 390 | talloc_free(tun); |
| 391 | } |
| 392 | |
| 393 | /* UNLOCKED remove all objects referencing this tun and then destroy */ |
| 394 | void _tun_device_deref_destroy(struct tun_device *tun) |
| 395 | { |
| 396 | struct gtp_daemon *d = tun->d; |
| 397 | char *devname = talloc_strdup(d, tun->devname); |
| 398 | struct gtp_tunnel *t, *t2; |
| 399 | struct tun_device *tun2; |
| 400 | |
| 401 | /* talloc is not thread safe, all alloc/free must come from main thread */ |
| 402 | ASSERT_MAIN_THREAD(tun->d); |
| 403 | |
| 404 | llist_for_each_entry_safe(t, t2, &g_daemon->gtp_tunnels, list) { |
| 405 | if (t->tun_dev == tun) |
| 406 | _gtp_tunnel_destroy(t); |
| 407 | } |
| 408 | /* _tun_device_destroy may already have been called via |
| 409 | * _gtp_tunnel_destroy -> _tun_device_release, so we have to |
| 410 | * check if the tun can still be found in the list */ |
| 411 | tun2 = _tun_device_find(d, devname); |
| 412 | if (tun2 && tun2 == tun) |
| 413 | _tun_device_destroy(tun2); |
| 414 | |
| 415 | talloc_free(devname); |
| 416 | } |
| 417 | |
| 418 | /* UNLOCKED release a reference; destroy if refcount drops to 0 */ |
| 419 | bool _tun_device_release(struct tun_device *tun) |
| 420 | { |
| 421 | bool released = false; |
| 422 | |
| 423 | /* talloc is not thread safe, all alloc/free must come from main thread */ |
| 424 | ASSERT_MAIN_THREAD(tun->d); |
| 425 | |
| 426 | tun->use_count--; |
| 427 | if (tun->use_count == 0) { |
| 428 | _tun_device_destroy(tun); |
| 429 | released = true; |
| 430 | } else |
| 431 | LOGTUN(tun, LOGL_DEBUG, "Release; new use_count=%lu\n", tun->use_count); |
| 432 | |
| 433 | return released; |
| 434 | } |
| 435 | |
| 436 | /* release a reference; destroy if refcount drops to 0 */ |
| 437 | bool tun_device_release(struct tun_device *tun) |
| 438 | { |
| 439 | struct gtp_daemon *d = tun->d; |
| 440 | bool released; |
| 441 | |
| 442 | /* talloc is not thread safe, all alloc/free must come from main thread */ |
| 443 | ASSERT_MAIN_THREAD(tun->d); |
| 444 | |
| 445 | pthread_rwlock_wrlock(&d->rwlock); |
| 446 | released = _tun_device_release(tun); |
| 447 | pthread_rwlock_unlock(&d->rwlock); |
| 448 | |
| 449 | return released; |
| 450 | } |