| 1 | /************************************************* |
| 2 | * Exim - an Internet mail transport agent * |
| 3 | *************************************************/ |
| 4 | |
| 5 | /* Copyright (c) University of Cambridge 1995 - 2018 */ |
| 6 | /* See the file NOTICE for conditions of use and distribution. */ |
| 7 | |
| 8 | /* Functions for doing things with sockets. With the advent of IPv6 this has |
| 9 | got messier, so that it's worth pulling out the code into separate functions |
| 10 | that other parts of Exim can call, especially as there are now several |
| 11 | different places in the code where sockets are used. */ |
| 12 | |
| 13 | |
| 14 | #include "exim.h" |
| 15 | |
| 16 | |
| 17 | /************************************************* |
| 18 | * Create a socket * |
| 19 | *************************************************/ |
| 20 | |
| 21 | /* Socket creation happens in a number of places so it's packaged here for |
| 22 | convenience. |
| 23 | |
| 24 | Arguments: |
| 25 | type SOCK_DGRAM or SOCK_STREAM |
| 26 | af AF_INET or AF_INET6 |
| 27 | |
| 28 | Returns: socket number or -1 on failure |
| 29 | */ |
| 30 | |
| 31 | int |
| 32 | ip_socket(int type, int af) |
| 33 | { |
| 34 | int sock = socket(af, type, 0); |
| 35 | if (sock < 0) |
| 36 | log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s", |
| 37 | (af == AF_INET6)? '6':'4', strerror(errno)); |
| 38 | return sock; |
| 39 | } |
| 40 | |
| 41 | |
| 42 | |
| 43 | |
| 44 | #if HAVE_IPV6 |
| 45 | /************************************************* |
| 46 | * Convert printing address to numeric * |
| 47 | *************************************************/ |
| 48 | |
| 49 | /* This function converts the textual form of an IP address into a numeric form |
| 50 | in an appropriate structure in an IPv6 environment. The getaddrinfo() function |
| 51 | can (apparently) handle more complicated addresses (e.g. those containing |
| 52 | scopes) than inet_pton() in some environments. We use hints to tell it that the |
| 53 | input must be a numeric address. |
| 54 | |
| 55 | However, apparently some operating systems (or libraries) don't support |
| 56 | getaddrinfo(), so there is a build-time option to revert to inet_pton() (which |
| 57 | does not support scopes). |
| 58 | |
| 59 | Arguments: |
| 60 | address textual form of the address |
| 61 | addr where to copy back the answer |
| 62 | |
| 63 | Returns: nothing - failure provokes a panic-die |
| 64 | */ |
| 65 | |
| 66 | static void |
| 67 | ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr) |
| 68 | { |
| 69 | #ifdef IPV6_USE_INET_PTON |
| 70 | |
| 71 | if (inet_pton(AF_INET6, CCS address, &saddr->sin6_addr) != 1) |
| 72 | log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an " |
| 73 | "IP address", address); |
| 74 | saddr->sin6_family = AF_INET6; |
| 75 | |
| 76 | #else |
| 77 | |
| 78 | int rc; |
| 79 | struct addrinfo hints, *res; |
| 80 | memset(&hints, 0, sizeof(hints)); |
| 81 | hints.ai_family = AF_INET6; |
| 82 | hints.ai_socktype = SOCK_STREAM; |
| 83 | hints.ai_flags = AI_NUMERICHOST; |
| 84 | if ((rc = getaddrinfo(CCS address, NULL, &hints, &res)) != 0 || res == NULL) |
| 85 | log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an " |
| 86 | "IP address: %s", address, |
| 87 | (rc == 0)? "NULL result returned" : gai_strerror(rc)); |
| 88 | memcpy(saddr, res->ai_addr, res->ai_addrlen); |
| 89 | freeaddrinfo(res); |
| 90 | |
| 91 | #endif |
| 92 | } |
| 93 | #endif /* HAVE_IPV6 */ |
| 94 | |
| 95 | |
| 96 | /************************************************* |
| 97 | * Bind socket to interface and port * |
| 98 | *************************************************/ |
| 99 | |
| 100 | int |
| 101 | ip_addr(void * sin_, int af, const uschar * address, int port) |
| 102 | { |
| 103 | union sockaddr_46 * sin = sin_; |
| 104 | memset(sin, 0, sizeof(*sin)); |
| 105 | |
| 106 | /* Setup code when using an IPv6 socket. The wildcard address is ":", to |
| 107 | ensure an IPv6 socket is used. */ |
| 108 | |
| 109 | #if HAVE_IPV6 |
| 110 | if (af == AF_INET6) |
| 111 | { |
| 112 | if (address[0] == ':' && address[1] == 0) |
| 113 | { |
| 114 | sin->v6.sin6_family = AF_INET6; |
| 115 | sin->v6.sin6_addr = in6addr_any; |
| 116 | } |
| 117 | else |
| 118 | ip_addrinfo(address, &sin->v6); /* Panic-dies on error */ |
| 119 | sin->v6.sin6_port = htons(port); |
| 120 | return sizeof(sin->v6); |
| 121 | } |
| 122 | else |
| 123 | #else /* HAVE_IPv6 */ |
| 124 | af = af; /* Avoid compiler warning */ |
| 125 | #endif /* HAVE_IPV6 */ |
| 126 | |
| 127 | /* Setup code when using IPv4 socket. The wildcard address is "". */ |
| 128 | |
| 129 | { |
| 130 | sin->v4.sin_family = AF_INET; |
| 131 | sin->v4.sin_port = htons(port); |
| 132 | sin->v4.sin_addr.s_addr = address[0] == 0 |
| 133 | ? (S_ADDR_TYPE)INADDR_ANY |
| 134 | : (S_ADDR_TYPE)inet_addr(CS address); |
| 135 | return sizeof(sin->v4); |
| 136 | } |
| 137 | } |
| 138 | |
| 139 | |
| 140 | |
| 141 | /* This function binds a socket to a local interface address and port. For a |
| 142 | wildcard IPv6 bind, the address is ":". |
| 143 | |
| 144 | Arguments: |
| 145 | sock the socket |
| 146 | af AF_INET or AF_INET6 - the socket type |
| 147 | address the IP address, in text form |
| 148 | port the IP port (host order) |
| 149 | |
| 150 | Returns: the result of bind() |
| 151 | */ |
| 152 | |
| 153 | int |
| 154 | ip_bind(int sock, int af, uschar *address, int port) |
| 155 | { |
| 156 | union sockaddr_46 sin; |
| 157 | int s_len = ip_addr(&sin, af, address, port); |
| 158 | return bind(sock, (struct sockaddr *)&sin, s_len); |
| 159 | } |
| 160 | |
| 161 | |
| 162 | |
| 163 | /************************************************* |
| 164 | *************************************************/ |
| 165 | |
| 166 | #ifdef EXIM_TFO_PROBE |
| 167 | void |
| 168 | tfo_probe(void) |
| 169 | { |
| 170 | # ifdef TCP_FASTOPEN |
| 171 | int sock, backlog = 5; |
| 172 | |
| 173 | if ( (sock = socket(SOCK_STREAM, AF_INET, 0)) < 0 |
| 174 | && setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &backlog, sizeof(backlog)) |
| 175 | ) |
| 176 | f.tcp_fastopen_ok = TRUE; |
| 177 | close(sock); |
| 178 | # endif |
| 179 | } |
| 180 | #endif |
| 181 | |
| 182 | |
| 183 | /************************************************* |
| 184 | * Connect socket to remote host * |
| 185 | *************************************************/ |
| 186 | |
| 187 | /* This function connects a socket to a remote address and port. The socket may |
| 188 | or may not have previously been bound to a local interface. The socket is not |
| 189 | closed, even in cases of error. It is expected that the calling function, which |
| 190 | created the socket, will be the one that closes it. |
| 191 | |
| 192 | Arguments: |
| 193 | sock the socket |
| 194 | af AF_INET6 or AF_INET for the socket type |
| 195 | address the remote address, in text form |
| 196 | port the remote port |
| 197 | timeout a timeout (zero for indefinite timeout) |
| 198 | fastopen_blob non-null iff TCP_FASTOPEN can be used; may indicate early-data to |
| 199 | be sent in SYN segment |
| 200 | |
| 201 | Returns: 0 on success; -1 on failure, with errno set |
| 202 | */ |
| 203 | |
| 204 | int |
| 205 | ip_connect(int sock, int af, const uschar *address, int port, int timeout, |
| 206 | const blob * fastopen_blob) |
| 207 | { |
| 208 | struct sockaddr_in s_in4; |
| 209 | struct sockaddr *s_ptr; |
| 210 | int s_len, rc, save_errno; |
| 211 | |
| 212 | /* For an IPv6 address, use an IPv6 sockaddr structure. */ |
| 213 | |
| 214 | #if HAVE_IPV6 |
| 215 | struct sockaddr_in6 s_in6; |
| 216 | if (af == AF_INET6) |
| 217 | { |
| 218 | memset(&s_in6, 0, sizeof(s_in6)); |
| 219 | ip_addrinfo(address, &s_in6); /* Panic-dies on error */ |
| 220 | s_in6.sin6_port = htons(port); |
| 221 | s_ptr = (struct sockaddr *)&s_in6; |
| 222 | s_len = sizeof(s_in6); |
| 223 | } |
| 224 | else |
| 225 | #else /* HAVE_IPV6 */ |
| 226 | af = af; /* Avoid compiler warning */ |
| 227 | #endif /* HAVE_IPV6 */ |
| 228 | |
| 229 | /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with |
| 230 | IPv6 support. */ |
| 231 | |
| 232 | { |
| 233 | memset(&s_in4, 0, sizeof(s_in4)); |
| 234 | s_in4.sin_family = AF_INET; |
| 235 | s_in4.sin_port = htons(port); |
| 236 | s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address); |
| 237 | s_ptr = (struct sockaddr *)&s_in4; |
| 238 | s_len = sizeof(s_in4); |
| 239 | } |
| 240 | |
| 241 | /* If no connection timeout is set, just call connect() without setting a |
| 242 | timer, thereby allowing the inbuilt OS timeout to operate. */ |
| 243 | |
| 244 | callout_address = string_sprintf("[%s]:%d", address, port); |
| 245 | sigalrm_seen = FALSE; |
| 246 | if (timeout > 0) alarm(timeout); |
| 247 | |
| 248 | #if defined(TCP_FASTOPEN) && defined(MSG_FASTOPEN) |
| 249 | /* TCP Fast Open, if the system has a cookie from a previous call to |
| 250 | this peer, can send data in the SYN packet. The peer can send data |
| 251 | before it gets our ACK of its SYN,ACK - the latter is useful for |
| 252 | the SMTP banner. Other (than SMTP) cases of TCP connections can |
| 253 | possibly use the data-on-syn, so support that too. |
| 254 | |
| 255 | This is a Linux implementation. It might be useable on FreeBSD; I have |
| 256 | not checked. I think MacOS has a "connectx" call for this purpose, |
| 257 | rather than using "sendto" ? */ |
| 258 | |
| 259 | if (fastopen_blob && f.tcp_fastopen_ok) |
| 260 | { |
| 261 | if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len, |
| 262 | MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0) |
| 263 | /* seen for with-data, experimental TFO option, with-cookie case */ |
| 264 | /* seen for with-data, proper TFO opt, with-cookie case */ |
| 265 | { |
| 266 | DEBUG(D_transport|D_v) |
| 267 | debug_printf("non-TFO mode connection attempt to %s, %lu data\n", |
| 268 | address, (unsigned long)fastopen_blob->len); |
| 269 | /*XXX also seen on successful TFO, sigh */ |
| 270 | tcp_out_fastopen = fastopen_blob->len > 0 ? 2 : 1; |
| 271 | } |
| 272 | else if (errno == EINPROGRESS) /* expected if we had no cookie for peer */ |
| 273 | /* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */ |
| 274 | /* apparently no visibility of the diffference at this point */ |
| 275 | /* seen for with-data, proper TFO opt, cookie-req */ |
| 276 | /* with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */ |
| 277 | /* ? older Experimental TFO option behaviour ? */ |
| 278 | { /* queue unsent data */ |
| 279 | DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n", |
| 280 | fastopen_blob->len > 0 ? "with" : "no"); |
| 281 | if (!fastopen_blob->data) |
| 282 | { |
| 283 | tcp_out_fastopen = 1; /* we tried; unknown if useful yet */ |
| 284 | rc = 0; |
| 285 | } |
| 286 | else |
| 287 | rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0); |
| 288 | } |
| 289 | else if(errno == EOPNOTSUPP) |
| 290 | { |
| 291 | DEBUG(D_transport) |
| 292 | debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n"); |
| 293 | goto legacy_connect; |
| 294 | } |
| 295 | } |
| 296 | else |
| 297 | #endif |
| 298 | { |
| 299 | legacy_connect: |
| 300 | DEBUG(D_transport|D_v) if (fastopen_blob) |
| 301 | debug_printf("non-TFO mode connection attempt to %s, %lu data\n", |
| 302 | address, (unsigned long)fastopen_blob->len); |
| 303 | if ((rc = connect(sock, s_ptr, s_len)) >= 0) |
| 304 | if ( fastopen_blob && fastopen_blob->data && fastopen_blob->len |
| 305 | && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0) |
| 306 | rc = -1; |
| 307 | } |
| 308 | |
| 309 | save_errno = errno; |
| 310 | alarm(0); |
| 311 | |
| 312 | /* There is a testing facility for simulating a connection timeout, as I |
| 313 | can't think of any other way of doing this. It converts a connection refused |
| 314 | into a timeout if the timeout is set to 999999. */ |
| 315 | |
| 316 | if (f.running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999) |
| 317 | { |
| 318 | rc = -1; |
| 319 | save_errno = EINTR; |
| 320 | sigalrm_seen = TRUE; |
| 321 | } |
| 322 | |
| 323 | /* Success */ |
| 324 | |
| 325 | if (rc >= 0) |
| 326 | return 0; |
| 327 | |
| 328 | /* A failure whose error code is "Interrupted system call" is in fact |
| 329 | an externally applied timeout if the signal handler has been run. */ |
| 330 | |
| 331 | errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno; |
| 332 | return -1; |
| 333 | } |
| 334 | |
| 335 | |
| 336 | |
| 337 | /************************************************* |
| 338 | * Create connected socket to remote host * |
| 339 | *************************************************/ |
| 340 | |
| 341 | /* Create a socket and connect to host (name or number, ipv6 ok) |
| 342 | at one of port-range. |
| 343 | |
| 344 | Arguments: |
| 345 | type SOCK_DGRAM or SOCK_STREAM |
| 346 | af AF_INET6 or AF_INET for the socket type |
| 347 | hostname host name, or ip address (as text) |
| 348 | portlo,porthi the remote port range |
| 349 | timeout a timeout |
| 350 | connhost if not NULL, host_item to be filled in with connection details |
| 351 | errstr pointer for allocated string on error |
| 352 | fastopen_blob with SOCK_STREAM, if non-null, request TCP Fast Open. |
| 353 | Additionally, optional early-data to send |
| 354 | |
| 355 | Return: |
| 356 | socket fd, or -1 on failure (having allocated an error string) |
| 357 | */ |
| 358 | int |
| 359 | ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi, |
| 360 | int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob) |
| 361 | { |
| 362 | int namelen, port; |
| 363 | host_item shost; |
| 364 | host_item *h; |
| 365 | int af = 0, fd, fd4 = -1, fd6 = -1; |
| 366 | |
| 367 | shost.next = NULL; |
| 368 | shost.address = NULL; |
| 369 | shost.port = portlo; |
| 370 | shost.mx = -1; |
| 371 | |
| 372 | namelen = Ustrlen(hostname); |
| 373 | |
| 374 | /* Anything enclosed in [] must be an IP address. */ |
| 375 | |
| 376 | if (hostname[0] == '[' && |
| 377 | hostname[namelen - 1] == ']') |
| 378 | { |
| 379 | uschar * host = string_copyn(hostname+1, namelen-2); |
| 380 | if (string_is_ip_address(host, NULL) == 0) |
| 381 | { |
| 382 | *errstr = string_sprintf("malformed IP address \"%s\"", hostname); |
| 383 | return -1; |
| 384 | } |
| 385 | shost.name = shost.address = host; |
| 386 | } |
| 387 | |
| 388 | /* Otherwise check for an unadorned IP address */ |
| 389 | |
| 390 | else if (string_is_ip_address(hostname, NULL) != 0) |
| 391 | shost.name = shost.address = string_copyn(hostname, namelen); |
| 392 | |
| 393 | /* Otherwise lookup IP address(es) from the name */ |
| 394 | |
| 395 | else |
| 396 | { |
| 397 | shost.name = string_copyn(hostname, namelen); |
| 398 | if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, |
| 399 | NULL, FALSE) != HOST_FOUND) |
| 400 | { |
| 401 | *errstr = string_sprintf("no IP address found for host %s", shost.name); |
| 402 | return -1; |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | /* Try to connect to the server - test each IP till one works */ |
| 407 | |
| 408 | for (h = &shost; h; h = h->next) |
| 409 | { |
| 410 | fd = Ustrchr(h->address, ':') != 0 |
| 411 | ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6 |
| 412 | : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4; |
| 413 | |
| 414 | if (fd < 0) |
| 415 | { |
| 416 | *errstr = string_sprintf("failed to create socket: %s", strerror(errno)); |
| 417 | goto bad; |
| 418 | } |
| 419 | |
| 420 | for(port = portlo; port <= porthi; port++) |
| 421 | if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0) |
| 422 | { |
| 423 | if (fd != fd6) close(fd6); |
| 424 | if (fd != fd4) close(fd4); |
| 425 | if (connhost) |
| 426 | { |
| 427 | h->port = port; |
| 428 | *connhost = *h; |
| 429 | connhost->next = NULL; |
| 430 | } |
| 431 | return fd; |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | *errstr = string_sprintf("failed to connect to any address for %s: %s", |
| 436 | hostname, strerror(errno)); |
| 437 | |
| 438 | bad: |
| 439 | close(fd4); close(fd6); return -1; |
| 440 | } |
| 441 | |
| 442 | |
| 443 | /*XXX TFO? */ |
| 444 | int |
| 445 | ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo) |
| 446 | { |
| 447 | int scan; |
| 448 | uschar hostname[256]; |
| 449 | unsigned int portlow, porthigh; |
| 450 | |
| 451 | /* extract host and port part */ |
| 452 | scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh); |
| 453 | if (scan != 3) |
| 454 | { |
| 455 | if (scan != 2) |
| 456 | { |
| 457 | *errstr = string_sprintf("invalid socket '%s'", hostport); |
| 458 | return -1; |
| 459 | } |
| 460 | porthigh = portlow; |
| 461 | } |
| 462 | |
| 463 | return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh, |
| 464 | tmo, NULL, errstr, NULL); |
| 465 | } |
| 466 | |
| 467 | int |
| 468 | ip_unixsocket(const uschar * path, uschar ** errstr) |
| 469 | { |
| 470 | int sock; |
| 471 | struct sockaddr_un server; |
| 472 | |
| 473 | if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) |
| 474 | { |
| 475 | *errstr = US"can't open UNIX socket."; |
| 476 | return -1; |
| 477 | } |
| 478 | |
| 479 | callout_address = string_copy(path); |
| 480 | server.sun_family = AF_UNIX; |
| 481 | Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1); |
| 482 | server.sun_path[sizeof(server.sun_path)-1] = '\0'; |
| 483 | if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) |
| 484 | { |
| 485 | int err = errno; |
| 486 | (void)close(sock); |
| 487 | *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s", |
| 488 | path, strerror(err)); |
| 489 | return -1; |
| 490 | } |
| 491 | return sock; |
| 492 | } |
| 493 | |
| 494 | int |
| 495 | ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo) |
| 496 | { |
| 497 | return *spec == '/' |
| 498 | ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo); |
| 499 | } |
| 500 | |
| 501 | /************************************************* |
| 502 | * Set keepalive on a socket * |
| 503 | *************************************************/ |
| 504 | |
| 505 | /* Can be called for both incoming and outgoing sockets. |
| 506 | |
| 507 | Arguments: |
| 508 | sock the socket |
| 509 | address the remote host address, for failure logging |
| 510 | torf true for outgoing connection, false for incoming |
| 511 | |
| 512 | Returns: nothing |
| 513 | */ |
| 514 | |
| 515 | void |
| 516 | ip_keepalive(int sock, const uschar *address, BOOL torf) |
| 517 | { |
| 518 | int fodder = 1; |
| 519 | if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
| 520 | US (&fodder), sizeof(fodder)) != 0) |
| 521 | log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s " |
| 522 | "failed: %s", torf? "to":"from", address, strerror(errno)); |
| 523 | } |
| 524 | |
| 525 | |
| 526 | |
| 527 | /************************************************* |
| 528 | * Receive from a socket with timeout * |
| 529 | *************************************************/ |
| 530 | |
| 531 | /* |
| 532 | Arguments: |
| 533 | fd the file descriptor |
| 534 | timeout the timeout, seconds |
| 535 | Returns: TRUE => ready for i/o |
| 536 | FALSE => timed out, or other error |
| 537 | */ |
| 538 | BOOL |
| 539 | fd_ready(int fd, int timeout) |
| 540 | { |
| 541 | fd_set select_inset; |
| 542 | time_t start_recv = time(NULL); |
| 543 | int time_left = timeout; |
| 544 | int rc; |
| 545 | |
| 546 | if (time_left <= 0) |
| 547 | { |
| 548 | errno = ETIMEDOUT; |
| 549 | return FALSE; |
| 550 | } |
| 551 | /* Wait until the socket is ready */ |
| 552 | |
| 553 | do |
| 554 | { |
| 555 | struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 }; |
| 556 | FD_ZERO (&select_inset); |
| 557 | FD_SET (fd, &select_inset); |
| 558 | |
| 559 | /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/ |
| 560 | rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv); |
| 561 | |
| 562 | /* If some interrupt arrived, just retry. We presume this to be rare, |
| 563 | but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes |
| 564 | select() to exit). |
| 565 | |
| 566 | Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making |
| 567 | the interrupt not at all rare. Since the timeout is typically more than 2 |
| 568 | minutes, the effect was to block the timeout completely. To prevent this |
| 569 | happening again, we do an explicit time test and adjust the timeout |
| 570 | accordingly */ |
| 571 | |
| 572 | if (rc < 0 && errno == EINTR) |
| 573 | { |
| 574 | DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n"); |
| 575 | |
| 576 | /* Watch out, 'continue' jumps to the condition, not to the loops top */ |
| 577 | time_left = timeout - (time(NULL) - start_recv); |
| 578 | if (time_left > 0) continue; |
| 579 | } |
| 580 | |
| 581 | if (rc <= 0) |
| 582 | { |
| 583 | errno = ETIMEDOUT; |
| 584 | return FALSE; |
| 585 | } |
| 586 | |
| 587 | /* Checking the FD_ISSET is not enough, if we're interrupted, the |
| 588 | select_inset may still contain the 'input'. */ |
| 589 | } |
| 590 | while (rc < 0 || !FD_ISSET(fd, &select_inset)); |
| 591 | return TRUE; |
| 592 | } |
| 593 | |
| 594 | /* The timeout is implemented using select(), and we loop to cover select() |
| 595 | getting interrupted, and the possibility of select() returning with a positive |
| 596 | result but no ready descriptor. Is this in fact possible? |
| 597 | |
| 598 | Arguments: |
| 599 | cctx the connection context (socket fd, possibly TLS context) |
| 600 | buffer to read into |
| 601 | bufsize the buffer size |
| 602 | timeout the timeout |
| 603 | |
| 604 | Returns: > 0 => that much data read |
| 605 | <= 0 on error or EOF; errno set - zero for EOF |
| 606 | */ |
| 607 | |
| 608 | int |
| 609 | ip_recv(client_conn_ctx * cctx, uschar * buffer, int buffsize, int timeout) |
| 610 | { |
| 611 | int rc; |
| 612 | |
| 613 | if (!fd_ready(cctx->sock, timeout)) |
| 614 | return -1; |
| 615 | |
| 616 | /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e. |
| 617 | close down of the connection), set errno to zero; otherwise leave it alone. */ |
| 618 | |
| 619 | #ifdef SUPPORT_TLS |
| 620 | if (cctx->tls_ctx) /* client TLS */ |
| 621 | rc = tls_read(cctx->tls_ctx, buffer, buffsize); |
| 622 | else if (tls_in.active.sock == cctx->sock) /* server TLS */ |
| 623 | rc = tls_read(NULL, buffer, buffsize); |
| 624 | else |
| 625 | #endif |
| 626 | rc = recv(cctx->sock, buffer, buffsize, 0); |
| 627 | |
| 628 | if (rc > 0) return rc; |
| 629 | if (rc == 0) errno = 0; |
| 630 | return -1; |
| 631 | } |
| 632 | |
| 633 | |
| 634 | |
| 635 | |
| 636 | /************************************************* |
| 637 | * Lookup address family of potential socket * |
| 638 | *************************************************/ |
| 639 | |
| 640 | /* Given a file-descriptor, check to see if it's a socket and, if so, |
| 641 | return the address family; detects IPv4 vs IPv6. If not a socket then |
| 642 | return -1. |
| 643 | |
| 644 | The value 0 is typically AF_UNSPEC, which should not be seen on a connected |
| 645 | fd. If the return is -1, the errno will be from getsockname(); probably |
| 646 | ENOTSOCK or ECONNRESET. |
| 647 | |
| 648 | Arguments: socket-or-not fd |
| 649 | Returns: address family or -1 |
| 650 | */ |
| 651 | |
| 652 | int |
| 653 | ip_get_address_family(int fd) |
| 654 | { |
| 655 | struct sockaddr_storage ss; |
| 656 | socklen_t sslen = sizeof(ss); |
| 657 | |
| 658 | if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0) |
| 659 | return -1; |
| 660 | |
| 661 | return (int) ss.ss_family; |
| 662 | } |
| 663 | |
| 664 | |
| 665 | |
| 666 | |
| 667 | /************************************************* |
| 668 | * Lookup DSCP settings for a socket * |
| 669 | *************************************************/ |
| 670 | |
| 671 | struct dscp_name_tableentry { |
| 672 | const uschar *name; |
| 673 | int value; |
| 674 | }; |
| 675 | /* Keep both of these tables sorted! */ |
| 676 | static struct dscp_name_tableentry dscp_table[] = { |
| 677 | #ifdef IPTOS_DSCP_AF11 |
| 678 | { CUS"af11", IPTOS_DSCP_AF11 }, |
| 679 | { CUS"af12", IPTOS_DSCP_AF12 }, |
| 680 | { CUS"af13", IPTOS_DSCP_AF13 }, |
| 681 | { CUS"af21", IPTOS_DSCP_AF21 }, |
| 682 | { CUS"af22", IPTOS_DSCP_AF22 }, |
| 683 | { CUS"af23", IPTOS_DSCP_AF23 }, |
| 684 | { CUS"af31", IPTOS_DSCP_AF31 }, |
| 685 | { CUS"af32", IPTOS_DSCP_AF32 }, |
| 686 | { CUS"af33", IPTOS_DSCP_AF33 }, |
| 687 | { CUS"af41", IPTOS_DSCP_AF41 }, |
| 688 | { CUS"af42", IPTOS_DSCP_AF42 }, |
| 689 | { CUS"af43", IPTOS_DSCP_AF43 }, |
| 690 | { CUS"ef", IPTOS_DSCP_EF }, |
| 691 | #endif |
| 692 | #ifdef IPTOS_LOWCOST |
| 693 | { CUS"lowcost", IPTOS_LOWCOST }, |
| 694 | #endif |
| 695 | { CUS"lowdelay", IPTOS_LOWDELAY }, |
| 696 | #ifdef IPTOS_MINCOST |
| 697 | { CUS"mincost", IPTOS_MINCOST }, |
| 698 | #endif |
| 699 | { CUS"reliability", IPTOS_RELIABILITY }, |
| 700 | { CUS"throughput", IPTOS_THROUGHPUT } |
| 701 | }; |
| 702 | static int dscp_table_size = |
| 703 | sizeof(dscp_table) / sizeof(struct dscp_name_tableentry); |
| 704 | |
| 705 | /* DSCP values change by protocol family, and so do the options used for |
| 706 | setsockopt(); this utility does all the lookups. It takes an unexpanded |
| 707 | option string, expands it, strips off affix whitespace, then checks if it's |
| 708 | a number. If all of what's left is a number, then that's how the option will |
| 709 | be parsed and success/failure is a range check. If it's not all a number, |
| 710 | then it must be a supported keyword. |
| 711 | |
| 712 | Arguments: |
| 713 | dscp_name a string, so far unvalidated |
| 714 | af address_family in use |
| 715 | level setsockopt level to use |
| 716 | optname setsockopt name to use |
| 717 | dscp_value value for dscp_name |
| 718 | |
| 719 | Returns: TRUE if okay to setsockopt(), else FALSE |
| 720 | |
| 721 | *level and *optname may be set even if FALSE is returned |
| 722 | */ |
| 723 | |
| 724 | BOOL |
| 725 | dscp_lookup(const uschar *dscp_name, int af, |
| 726 | int *level, int *optname, int *dscp_value) |
| 727 | { |
| 728 | uschar *dscp_lookup, *p; |
| 729 | int first, last; |
| 730 | long rawlong; |
| 731 | |
| 732 | if (af == AF_INET) |
| 733 | { |
| 734 | *level = IPPROTO_IP; |
| 735 | *optname = IP_TOS; |
| 736 | } |
| 737 | #if HAVE_IPV6 && defined(IPV6_TCLASS) |
| 738 | else if (af == AF_INET6) |
| 739 | { |
| 740 | *level = IPPROTO_IPV6; |
| 741 | *optname = IPV6_TCLASS; |
| 742 | } |
| 743 | #endif |
| 744 | else |
| 745 | { |
| 746 | DEBUG(D_transport) |
| 747 | debug_printf("Unhandled address family %d in dscp_lookup()\n", af); |
| 748 | return FALSE; |
| 749 | } |
| 750 | if (!dscp_name) |
| 751 | { |
| 752 | DEBUG(D_transport) |
| 753 | debug_printf("[empty DSCP]\n"); |
| 754 | return FALSE; |
| 755 | } |
| 756 | dscp_lookup = expand_string(US dscp_name); |
| 757 | if (dscp_lookup == NULL || *dscp_lookup == '\0') |
| 758 | return FALSE; |
| 759 | |
| 760 | p = dscp_lookup + Ustrlen(dscp_lookup) - 1; |
| 761 | while (isspace(*p)) *p-- = '\0'; |
| 762 | while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++; |
| 763 | if (*dscp_lookup == '\0') |
| 764 | return FALSE; |
| 765 | |
| 766 | rawlong = Ustrtol(dscp_lookup, &p, 0); |
| 767 | if (p != dscp_lookup && *p == '\0') |
| 768 | { |
| 769 | /* We have six bits available, which will end up shifted to fit in 0xFC mask. |
| 770 | RFC 2597 defines the values unshifted. */ |
| 771 | if (rawlong < 0 || rawlong > 0x3F) |
| 772 | { |
| 773 | DEBUG(D_transport) |
| 774 | debug_printf("DSCP value %ld out of range, ignored.\n", rawlong); |
| 775 | return FALSE; |
| 776 | } |
| 777 | *dscp_value = rawlong << 2; |
| 778 | return TRUE; |
| 779 | } |
| 780 | |
| 781 | first = 0; |
| 782 | last = dscp_table_size; |
| 783 | while (last > first) |
| 784 | { |
| 785 | int middle = (first + last)/2; |
| 786 | int c = Ustrcmp(dscp_lookup, dscp_table[middle].name); |
| 787 | if (c == 0) |
| 788 | { |
| 789 | *dscp_value = dscp_table[middle].value; |
| 790 | return TRUE; |
| 791 | } |
| 792 | else if (c > 0) |
| 793 | first = middle + 1; |
| 794 | else |
| 795 | last = middle; |
| 796 | } |
| 797 | return FALSE; |
| 798 | } |
| 799 | |
| 800 | void |
| 801 | dscp_list_to_stream(FILE *stream) |
| 802 | { |
| 803 | int i; |
| 804 | for (i=0; i < dscp_table_size; ++i) |
| 805 | fprintf(stream, "%s\n", dscp_table[i].name); |
| 806 | } |
| 807 | |
| 808 | |
| 809 | /* End of ip.c */ |
| 810 | /* vi: aw ai sw=2 |
| 811 | */ |