Merge topic branch t/ff-compat into master
[paraslash.git] / net.c
1 /* Copyright (C) 2005 Andre Noll <maan@tuebingen.mpg.de>, see file COPYING. */
2
3 /** \file net.c Networking-related helper functions. */
4
5 #include "para.h"
6
7 #include <netinet/in.h>
8 #include <arpa/inet.h>
9 #include <sys/un.h>
10 #include <sys/types.h>
11 #include <sys/socket.h>
12 #include <netdb.h>
13 #include <regex.h>
14
15 #include "error.h"
16 #include "net.h"
17 #include "string.h"
18 #include "list.h"
19 #include "fd.h"
20
21 /* Whether the given address conforms to the IPv4 address format. */
22 static inline bool is_valid_ipv4_address(const char *address)
23 {
24 struct in_addr test_it;
25 return inet_pton(AF_INET, address, &test_it) != 0;
26 }
27
28 /**
29 * Parse and validate IPv4 address/netmask string.
30 *
31 * \param cidr Address in CIDR notation
32 * \param addr Copy of the IPv4 address part of \a cidr
33 * \param addrlen Size of \a addr in bytes
34 * \param netmask Value of the netmask part in \a cidr or the
35 * default of 32 if not specified.
36 *
37 * \return Pointer to \a addr if successful, NULL on error.
38 * \sa RFC 4632.
39 */
40 char *parse_cidr(const char *cidr,
41 char *addr, ssize_t addrlen,
42 int32_t *netmask)
43 {
44 const char *o = cidr;
45 char *c = addr, *end = c + (addrlen - 1);
46
47 *netmask = 0x20;
48
49 if (cidr == NULL || addrlen < 1)
50 goto failed;
51
52 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
53 if (c == end)
54 goto failed;
55
56 if (*o == '/')
57 if (para_atoi32(++o, netmask) < 0 ||
58 *netmask < 0 || *netmask > 0x20)
59 goto failed;
60
61 if (is_valid_ipv4_address(addr))
62 return addr;
63 failed:
64 *addr = '\0';
65 return NULL;
66 }
67
68 static bool is_v4_dot_quad(const char *address)
69 {
70 bool result;
71 regex_t r;
72
73 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
74 REG_EXTENDED | REG_NOSUB) >= 0);
75 result = regexec(&r, address, 0, NULL, 0) == 0;
76 regfree(&r);
77 return result;
78 }
79
80 /* Whether a string conforms to IPv6 address format (RFC 4291). */
81 static inline bool is_valid_ipv6_address(const char *address)
82 {
83 struct in6_addr test_it;
84 return inet_pton(AF_INET6, address, &test_it) != 0;
85 }
86
87 /**
88 * Perform basic syntax checking on the host-part of an URL:
89 *
90 * - Since ':' is invalid in IPv4 addresses and DNS names, the
91 * presence of ':' causes interpretation as IPv6 address;
92 * - next the first-match-wins algorithm from RFC 3986 is applied;
93 * - else the string is considered as DNS name, to be resolved later.
94 *
95 * \param host The host string to check.
96 * \return True if \a host passes the syntax checks.
97 *
98 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5.
99 */
100 static bool host_string_ok(const char *host)
101 {
102 if (host == NULL || *host == '\0')
103 return false;
104 if (strchr(host, ':') != NULL)
105 return is_valid_ipv6_address(host);
106 if (is_v4_dot_quad(host))
107 return is_valid_ipv4_address(host);
108 return true;
109 }
110
111 /**
112 * Parse and validate URL string.
113 *
114 * The URL syntax is loosely based on RFC 3986, supporting one of
115 * - "["host"]"[:port] for native IPv6 addresses and
116 * - host[:port] for IPv4 hostnames and DNS names.
117 *
118 * Native IPv6 addresses must be enclosed in square brackets, since
119 * otherwise there is an ambiguity with the port separator `:'.
120 * The 'port' part is always considered to be a number; if absent,
121 * it is set to -1, to indicate that a default port is to be used.
122 *
123 * The following are valid examples:
124 * - 10.10.1.1
125 * - 10.10.1.2:8000
126 * - localhost
127 * - localhost:8001
128 * - [::1]:8000
129 * - [badc0de::1]
130 *
131 * \param url The URL string to take apart.
132 * \param host To return the copied host part of \a url.
133 * \param hostlen The maximum length of \a host.
134 * \param port To return the port number (if any) of \a url.
135 *
136 * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned,
137 * \a host and \a port are undefined. If no port number was present in \a url,
138 * \a port is set to -1.
139 *
140 * \sa RFC 3986, 3.2.2/3.2.3.
141 */
142 char *parse_url(const char *url,
143 char *host, ssize_t hostlen,
144 int32_t *port)
145 {
146 const char *o = url;
147 char *c = host, *end = c + (hostlen - 1);
148
149 *port = -1;
150
151 if (o == NULL || hostlen < 1)
152 goto failed;
153
154 if (*o == '[') {
155 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
156 if (c == end)
157 goto failed;
158
159 if (*o++ != ']' || (*o != '\0' && *o != ':'))
160 goto failed;
161 } else {
162 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
163 if (c == end && o[1])
164 goto failed;
165 }
166 }
167
168 if (*o == ':')
169 if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff)
170 goto failed;
171 if (host_string_ok(host))
172 return host;
173 failed:
174 *host = '\0';
175 return NULL;
176 }
177
178 /**
179 * Pretty-print a host/port pair.
180 *
181 * \param url NULL, or any string accepted by \ref parse_url().
182 * \param default_port Applies if url has no port.
183 *
184 * If the url argument is NULL, the function returns the string
185 * 0.0.0.0:default_port. Otherwise it calls \ref parse_url() to check the
186 * syntax of the input string given by url. On errors the string "?" is
187 * returned. Otherwise, if url contains a port, a copy of url is returned. If
188 * no port was supplied, a colon and the default port are appended to url.
189 *
190 * \return In all cases the returned string is a allocated with malloc(3) and
191 * has to be freed by the caller.
192 */
193 char *format_url(const char *url, int default_port)
194 {
195 char host[MAX_HOSTLEN];
196 int url_port;
197
198 if (!url)
199 return make_message("0.0.0.0:%d", default_port);
200 if (!parse_url(url, host, sizeof(host), &url_port))
201 return make_message("?");
202 if (url_port < 0)
203 return make_message("%s:%d", url, default_port);
204 else
205 return para_strdup(url);
206 }
207
208 /**
209 * Stringify port number, resolve into service name where defined.
210 *
211 * \param port 2-byte port number, in host-byte-order.
212 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
213 * \return Pointer to static result buffer.
214 *
215 * \sa getservbyport(3), services(5), nsswitch.conf(5).
216 */
217 const char *stringify_port(int port, const char *transport)
218 {
219 static char service[NI_MAXSERV];
220
221 if (port < 0 || port > 0xFFFF) {
222 snprintf(service, sizeof(service), "undefined (%d)", port);
223 } else {
224 struct servent *se = getservbyport(htons(port), transport);
225
226 if (se == NULL)
227 snprintf(service, sizeof(service), "%d", port);
228 else
229 snprintf(service, sizeof(service), "%s", se->s_name);
230 }
231 return service;
232 }
233
234 #ifndef SOCK_DCCP
235 #define SOCK_DCCP 6 /**< Linux socket type. */
236 #endif
237
238 /*
239 * Determine the socket type, given the symbolic name of the transport-layer
240 * protocol. See ip(7), socket(2).
241 */
242 static inline int sock_type(const unsigned l4type)
243 {
244 switch (l4type) {
245 case IPPROTO_UDP: return SOCK_DGRAM;
246 case IPPROTO_TCP: return SOCK_STREAM;
247 case IPPROTO_DCCP: return SOCK_DCCP;
248 }
249 return -1; /* not supported here */
250 }
251
252 /* Pretty-print transport-layer name. */
253 static const char *layer4_name(const unsigned l4type)
254 {
255 switch (l4type) {
256 case IPPROTO_UDP: return "UDP";
257 case IPPROTO_TCP: return "TCP";
258 case IPPROTO_DCCP: return "DCCP";
259 }
260 return "UNKNOWN PROTOCOL";
261 }
262
263 /**
264 * Flowopts: Transport-layer independent encapsulation of socket options.
265 *
266 * These collect individual socket options into a queue, which is disposed of
267 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
268 * only and should not be visible elsewhere.
269 *
270 * \sa setsockopt(2), \ref makesock().
271 */
272 struct pre_conn_opt {
273 int sock_level; /**< Second argument to setsockopt() */
274 int sock_option; /**< Third argument to setsockopt() */
275 char *opt_name; /**< Stringified \a sock_option */
276 void *opt_val; /**< Fourth argument to setsockopt() */
277 socklen_t opt_len; /**< Fifth argument to setsockopt() */
278
279 struct list_head node; /**< FIFO, as sockopt order matters. */
280 };
281
282 /**
283 * List of pre-connection socket options to be set.
284 *
285 * This list contains transport-layer independent encapsulation of socket
286 * options that need to be registered prior to setting up a connection.
287 */
288 struct flowopts {
289 struct list_head sockopts;
290 };
291
292 /**
293 * Allocate and initialize a flowopt queue.
294 *
295 * \return A new structure to be passed to \ref flowopt_add(). It is
296 * automatically deallocated in \ref makesock().
297 */
298 struct flowopts *flowopt_new(void)
299 {
300 struct flowopts *new = alloc(sizeof(*new));
301
302 init_list_head(&new->sockopts);
303 return new;
304 }
305
306 /**
307 * Append new socket option to flowopt queue.
308 *
309 * \param fo The flowopt queue to append to.
310 * \param lev Level at which \a opt resides.
311 * \param opt New option to add.
312 * \param name Stringified name of \a opt.
313 * \param val The value to set \a opt to.
314 * \param len Length of \a val.
315 *
316 * \sa setsockopt(2).
317 */
318 void flowopt_add(struct flowopts *fo, int lev, int opt,
319 const char *name, const void *val, int len)
320 {
321 struct pre_conn_opt *new = alloc(sizeof(*new));
322
323 new->sock_option = opt;
324 new->sock_level = lev;
325 new->opt_name = para_strdup(name);
326
327 if (val == NULL) {
328 new->opt_val = NULL;
329 new->opt_len = 0;
330 } else {
331 new->opt_val = alloc(len);
332 new->opt_len = len;
333 memcpy(new->opt_val, val, len);
334 }
335
336 list_add_tail(&new->node, &fo->sockopts);
337 }
338
339 /* Set the entire bunch of pre-connection options at once. */
340 static void flowopt_setopts(int sockfd, struct flowopts *fo)
341 {
342 struct pre_conn_opt *pc;
343
344 if (fo == NULL)
345 return;
346
347 list_for_each_entry(pc, &fo->sockopts, node)
348 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
349 pc->opt_val, pc->opt_len) < 0) {
350 PARA_EMERG_LOG("Can not set %s socket option: %s",
351 pc->opt_name, strerror(errno));
352 exit(EXIT_FAILURE);
353 }
354 }
355
356 /**
357 * Deallocate all resources of a flowopts structure.
358 *
359 * \param fo A pointer as returned from flowopt_new().
360 *
361 * It's OK to pass \p NULL here in which case the function does nothing.
362 */
363 void flowopt_cleanup(struct flowopts *fo)
364 {
365 struct pre_conn_opt *cur, *next;
366
367 if (fo == NULL)
368 return;
369
370 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
371 free(cur->opt_name);
372 free(cur->opt_val);
373 free(cur);
374 }
375 free(fo);
376 }
377
378 /**
379 * Resolve an IPv4/IPv6 address.
380 *
381 * \param l4type The layer-4 type (\p IPPROTO_xxx).
382 * \param passive Whether \p AI_PASSIVE should be included as hint.
383 * \param host Remote or local hostname or IPv/6 address string.
384 * \param port_number Used to set the port in each returned address structure.
385 * \param result addrinfo structures are returned here.
386 *
387 * The interpretation of \a host depends on the value of \a passive. On a
388 * passive socket host is interpreted as an interface IPv4/6 address (can be
389 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
390 * address to connect to.
391 *
392 * \return Standard.
393 *
394 * \sa getaddrinfo(3).
395 */
396 int lookup_address(unsigned l4type, bool passive, const char *host,
397 int port_number, struct addrinfo **result)
398 {
399 int ret;
400 char port[6]; /* port number has at most 5 digits */
401 struct addrinfo *addr = NULL, hints;
402
403 *result = NULL;
404 sprintf(port, "%d", port_number & 0xffff);
405 /* Set up address hint structure */
406 memset(&hints, 0, sizeof(hints));
407 hints.ai_family = AF_UNSPEC;
408 hints.ai_socktype = sock_type(l4type);
409 /*
410 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
411 * (and only then) pretend to be UDP.
412 */
413 if (l4type == IPPROTO_DCCP)
414 hints.ai_socktype = SOCK_DGRAM;
415 /* only use addresses available on the host */
416 hints.ai_flags = AI_ADDRCONFIG;
417 if (passive && host == NULL)
418 hints.ai_flags |= AI_PASSIVE;
419 /* Obtain local/remote address information */
420 ret = getaddrinfo(host, port, &hints, &addr);
421 if (ret != 0) {
422 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
423 layer4_name(l4type),
424 host? host : (passive? "[loopback]" : "[localhost]"),
425 port, gai_strerror(ret));
426 return -E_ADDRESS_LOOKUP;
427 }
428 *result = addr;
429 return 1;
430 }
431
432 /**
433 * Create an active or passive socket.
434 *
435 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
436 * \param passive Whether to call bind(2) or connect(2).
437 * \param ai Address information as obtained from \ref lookup_address().
438 * \param fo Socket options to be set before making the connection.
439 *
440 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
441 * algorithm tries all possible address combinations until it succeeds. If \a
442 * fo is supplied, options are set but cleanup must be performed in the caller.
443 *
444 * \return File descriptor on success, \p E_MAKESOCK on errors.
445 *
446 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
447 * connect(2).
448 */
449 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
450 struct flowopts *fo)
451 {
452 int ret = -E_MAKESOCK, on = 1;
453
454 for (; ai; ai = ai->ai_next) {
455 int fd;
456 ret = socket(ai->ai_family, sock_type(l4type), l4type);
457 if (ret < 0) {
458 PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
459 continue;
460 }
461 fd = ret;
462 flowopt_setopts(fd, fo);
463 if (!passive) {
464 if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
465 PARA_NOTICE_LOG("connect(): %s\n",
466 strerror(errno));
467 close(fd);
468 continue;
469 }
470 return fd;
471 }
472 /*
473 * Reuse the address on passive sockets to avoid failure on
474 * restart (protocols using listen()) and when creating
475 * multiple listener instances (UDP multicast).
476 */
477 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
478 sizeof(on)) == -1) {
479 PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
480 close(fd);
481 continue;
482 }
483 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
484 PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
485 close(fd);
486 continue;
487 }
488 return fd;
489 }
490 return -E_MAKESOCK;
491 }
492
493 /**
494 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
495 *
496 * \param l4type The layer-4 type (\p IPPROTO_xxx).
497 * \param passive Whether this is a passive or active socket.
498 * \param host Passed to \ref lookup_address().
499 * \param port_number Passed to \ref lookup_address().
500 * \param fo Passed to \ref makesock_addrinfo().
501 *
502 * This creates a ready-made IPv4/v6 socket structure after looking up the
503 * necessary parameters. The function first calls \ref lookup_address() and
504 * passes the address information to makesock_addrinfo() to create and
505 * initialize the socket.
506 *
507 * \return The newly created file descriptor on success, a negative error code
508 * on failure.
509 *
510 * \sa \ref lookup_address(), \ref makesock_addrinfo().
511 */
512 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
513 struct flowopts *fo)
514 {
515 struct addrinfo *ai;
516 int ret = lookup_address(l4type, passive, host, port_number, &ai);
517
518 if (ret >= 0)
519 ret = makesock_addrinfo(l4type, passive, ai, fo);
520 if (ai)
521 freeaddrinfo(ai);
522 if (ret < 0) {
523 PARA_NOTICE_LOG("can not create %s socket %s#%d.\n",
524 layer4_name(l4type), host? host : (passive?
525 "[loopback]" : "[localhost]"), port_number);
526 }
527 return ret;
528 }
529
530 /**
531 * Create a passive / listening socket.
532 *
533 * \param l4type The transport-layer type (\p IPPROTO_xxx).
534 * \param addr Passed to \ref parse_url() if not NULL.
535 * \param port Ignored if addr contains a port number.
536 *
537 * \return Positive integer (socket descriptor) on success, negative value
538 * otherwise.
539 *
540 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
541 */
542 int para_listen(unsigned l4type, const char *addr, uint16_t port)
543 {
544 char host[MAX_HOSTLEN];
545 int ret, fd, addr_port;
546
547 if (addr) {
548 if (!parse_url(addr, host, sizeof(host), &addr_port))
549 return -ERRNO_TO_PARA_ERROR(EINVAL);
550 if (addr_port > 0)
551 port = addr_port;
552 addr = host;
553 }
554 fd = makesock(l4type, true /* passive */, addr, port,
555 NULL /* no flowopts */);
556 if (fd > 0) {
557 ret = listen(fd, BACKLOG);
558 if (ret < 0) {
559 ret = errno;
560 close(fd);
561 return -ERRNO_TO_PARA_ERROR(ret);
562 }
563 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
564 layer4_name(l4type), port, fd);
565 }
566 return fd;
567 }
568
569 /**
570 * Create a socket which listens on all network addresses.
571 *
572 * \param l4type See \ref para_listen().
573 * \param port See \ref para_listen().
574 *
575 * This is a simple wrapper for \ref para_listen() which passes a NULL pointer
576 * as the address information.
577 *
578 * \return See \ref para_listen().
579 */
580 int para_listen_simple(unsigned l4type, uint16_t port)
581 {
582 return para_listen(l4type, NULL, port);
583 }
584
585 /* Compute the address-family dependent address length of an IPv4/v6 socket. */
586 static socklen_t salen(const struct sockaddr *sa)
587 {
588 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
589
590 return sa->sa_family == AF_INET6
591 ? sizeof(struct sockaddr_in6)
592 : sizeof(struct sockaddr_in);
593 }
594
595 /* True if ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
596 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
597 {
598 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
599
600 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
601 }
602
603 /*
604 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
605 * ss: Container of IPv4/6 address.
606 * Returns: Pointer to normalized address (may be static storage).
607 *
608 * \sa RFC 3493.
609 */
610 static const struct sockaddr *
611 normalize_ip_address(const struct sockaddr_storage *ss)
612 {
613 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
614
615 if (SS_IS_ADDR_V4MAPPED(ss)) {
616 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
617 static struct sockaddr_in ia;
618
619 ia.sin_family = AF_INET;
620 ia.sin_port = ia6->sin6_port;
621 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
622 return (const struct sockaddr *)&ia;
623 }
624 return (const struct sockaddr *)ss;
625 }
626
627 /*
628 * Generic/fallback MTU values
629 *
630 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
631 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
632 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
633 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
634 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
635 * UDP senders should use the respective minimum values of EMTU_S.
636 */
637 static inline int generic_mtu(const int af_type)
638 {
639 return af_type == AF_INET6 ? 1280 : 576;
640 }
641
642 /* Crude approximation of IP header overhead - neglecting options. */
643 static inline int estimated_header_overhead(const int af_type)
644 {
645 return af_type == AF_INET6 ? 40 : 20;
646 }
647
648 /**
649 * Get the maximum transport-layer message size (MMS_S).
650 *
651 * \param sockfd The socket file descriptor.
652 *
653 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
654 * could not be determined, \p AF_INET is assumed.
655 *
656 * \return The maximum message size of the address family type.
657 */
658 int generic_max_transport_msg_size(int sockfd)
659 {
660 struct sockaddr_storage ss = {.ss_family = 0};
661 socklen_t sslen = sizeof(ss);
662 int af_type = AF_INET;
663
664 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
665 PARA_ERROR_LOG("can not determine remote address type: %s\n",
666 strerror(errno));
667 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
668 af_type = ss.ss_family;
669 }
670 return generic_mtu(af_type) - estimated_header_overhead(af_type);
671 }
672
673 /**
674 * Look up the remote side of a connected socket structure.
675 *
676 * \param fd The socket descriptor of the connected socket.
677 *
678 * \return A static character string identifying hostname and port of the
679 * chosen side in numeric host:port format.
680 *
681 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
682 * services(5), nsswitch.conf(5).
683 */
684 char *remote_name(int fd)
685 {
686 struct sockaddr_storage ss = {.ss_family = 0};
687 const struct sockaddr *sa;
688 socklen_t sslen = sizeof(ss);
689 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
690 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
691 int ret;
692
693 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
694 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
695 fd, strerror(errno));
696 snprintf(output, sizeof(output), "(unknown)");
697 return output;
698 }
699 sa = normalize_ip_address(&ss);
700 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
701 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
702 if (ret) {
703 PARA_WARNING_LOG("hostname lookup error (%s).\n",
704 gai_strerror(ret));
705 snprintf(output, sizeof(output), "(lookup error)");
706 } else if (sa->sa_family == AF_INET6)
707 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
708 else
709 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
710 return output;
711 }
712
713 /**
714 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
715 *
716 * \param ss Container of IPv4/6 address.
717 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
718 *
719 * \sa RFC 3493.
720 */
721 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
722 {
723 const struct sockaddr *sa = normalize_ip_address(ss);
724
725 memset(ia, 0, sizeof(*ia));
726 if (sa->sa_family == AF_INET)
727 *ia = ((struct sockaddr_in *)sa)->sin_addr;
728 }
729
730 /**
731 * Compare the address part of IPv4/6 addresses.
732 *
733 * \param sa1 First address.
734 * \param sa2 Second address.
735 *
736 * \return True iff the IP address of \a sa1 and \a sa2 match.
737 */
738 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
739 {
740 if (!sa1 || !sa2)
741 return false;
742 if (sa1->sa_family != sa2->sa_family)
743 return false;
744 if (sa1->sa_family == AF_INET) {
745 struct sockaddr_in *a1 = (typeof(a1))sa1,
746 *a2 = (typeof (a2))sa2;
747 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
748 } else if (sa1->sa_family == AF_INET6) {
749 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
750 *a2 = (typeof (a2))sa2;
751 return !memcmp(a1, a2, sizeof(*a1));
752 } else
753 return false;
754 }
755
756 /**
757 * Receive data from a file descriptor.
758 *
759 * \param fd The file descriptor.
760 * \param buf The buffer to write the data to.
761 * \param size The size of \a buf.
762 *
763 * Receive at most \a size bytes from file descriptor \a fd.
764 *
765 * \return The number of bytes received on success, negative on errors, zero if
766 * the peer has performed an orderly shutdown.
767 *
768 * \sa recv(2).
769 */
770 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
771 {
772 ssize_t n;
773
774 n = recv(fd, buf, size, 0);
775 if (n == -1)
776 return -ERRNO_TO_PARA_ERROR(errno);
777 return n;
778 }
779
780 /**
781 * Receive and write terminating NULL byte.
782 *
783 * \param fd The file descriptor.
784 * \param buf The buffer to write the data to.
785 * \param size The size of \a buf.
786 *
787 * Read at most \a size - 1 bytes from file descriptor \a fd and
788 * write a NULL byte at the end of the received data.
789 *
790 * \return The return value of the underlying call to \a recv_bin_buffer().
791 *
792 * \sa \ref recv_bin_buffer()
793 */
794 int recv_buffer(int fd, char *buf, size_t size)
795 {
796 int n;
797
798 assert(size);
799 n = recv_bin_buffer(fd, buf, size - 1);
800 if (n >= 0)
801 buf[n] = '\0';
802 else
803 *buf = '\0';
804 return n;
805 }
806
807 /**
808 * Wrapper around the accept system call.
809 *
810 * \param fd The listening socket.
811 * \param addr Structure which is filled in with the address of the peer socket.
812 * \param size Should contain the size of the structure pointed to by \a addr.
813 * \param new_fd Result pointer.
814 *
815 * Accept incoming connections on addr, retry if interrupted.
816 *
817 * \return Negative on errors, zero if no connections are present to be accepted,
818 * one otherwise.
819 *
820 * \sa accept(2).
821 */
822 int para_accept(int fd, void *addr, socklen_t size, int *new_fd)
823 {
824 int ret;
825
826 do
827 ret = accept(fd, (struct sockaddr *) addr, &size);
828 while (ret < 0 && errno == EINTR);
829
830 if (ret >= 0) {
831 *new_fd = ret;
832 return 1;
833 }
834 if (errno == EAGAIN || errno == EWOULDBLOCK)
835 return 0;
836 return -ERRNO_TO_PARA_ERROR(errno);
837 }
838
839 #ifndef DCCP_SOCKOPT_AVAILABLE_CCIDS
840 #define DCCP_SOCKOPT_AVAILABLE_CCIDS 12 /**< List of supported CCIDs. */
841 #endif
842
843 /**
844 * Probe the list of DCCP CCIDs configured on this host.
845 * \param ccid_array Pointer to return statically allocated array in.
846 * \return Number of elements returned in \a ccid_array or error.
847 *
848 * NB: This feature is only available on Linux > 2.6.30; on older kernels
849 * ENOPROTOOPT ("Protocol not available") will be returned.
850 */
851 int dccp_available_ccids(uint8_t **ccid_array)
852 {
853 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
854 socklen_t nccids = sizeof(ccids);
855 int ret, fd;
856
857 ret = fd = makesock(IPPROTO_DCCP, true /* passive */, NULL, 0, NULL);
858 if (ret < 0)
859 return ret;
860
861 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
862 ccids, &nccids) < 0) {
863 ret = errno;
864 close(fd);
865 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
866 strerror(ret));
867 return -ERRNO_TO_PARA_ERROR(ret);
868 }
869
870 close(fd);
871 *ccid_array = ccids;
872 return nccids;
873 }
874
875 /**
876 * The buffer size of the sun_path component of struct sockaddr_un.
877 *
878 * While glibc doesn't define UNIX_PATH_MAX, it documents it has being limited
879 * to 108 bytes. On NetBSD it is only 104 bytes though. We trust UNIX_PATH_MAX
880 * if it is defined and use the size of the ->sun_path member otherwise. This
881 * should be safe everywhere.
882 */
883 #ifndef UNIX_PATH_MAX
884 #define UNIX_PATH_MAX (sizeof(((struct sockaddr_un *)0)->sun_path))
885 #endif
886
887 /*
888 * Prepare a structure for AF_UNIX socket addresses.
889 *
890 * This just copies name to the sun_path component of u, prepending a zero byte
891 * if abstract sockets are supported.
892 *
893 * The first call to this function tries to bind a socket to the abstract name
894 * space. The result of this test is stored in a static variable. Subsequent
895 * calls read this variable and create abstract sockets on systems that support
896 * them. If a NULL pointer is passed as the name, the function only
897 * initializes the static variable.
898 */
899 static int init_unix_addr(struct sockaddr_un *u, const char *name)
900 {
901 static int use_abstract;
902
903 memset(u->sun_path, 0, UNIX_PATH_MAX);
904 u->sun_family = PF_UNIX;
905 if (use_abstract == 0) { /* executed only once */
906 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
907 memcpy(u->sun_path, "\0x\0", 3);
908 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
909 use_abstract = 1; /* yes */
910 else
911 use_abstract = -1; /* no */
912 close(fd);
913 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
914 use_abstract == 1? "" : "not ");
915 }
916 if (!name)
917 return 0;
918 if (strlen(name) + 1 >= UNIX_PATH_MAX)
919 return -E_NAME_TOO_LONG;
920 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
921 return 1;
922 }
923
924 /**
925 * Create a socket for local communication and listen on it.
926 *
927 * \param name The socket pathname.
928 *
929 * This function creates a passive local socket for sequenced, reliable,
930 * two-way, connection-based byte streams. The socket file descriptor is set to
931 * nonblocking mode and listen(2) is called to prepare the socket for
932 * accepting incoming connection requests.
933 *
934 * \return The file descriptor on success, negative error code on failure.
935 *
936 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
937 */
938 int create_local_socket(const char *name)
939 {
940 struct sockaddr_un unix_addr;
941 int fd, ret;
942
943 ret = init_unix_addr(&unix_addr, name);
944 if (ret <= 0) /* error, or name was NULL */
945 return ret;
946 ret = socket(PF_UNIX, SOCK_STREAM, 0);
947 if (ret < 0)
948 return -ERRNO_TO_PARA_ERROR(errno);
949 fd = ret;
950 ret = mark_fd_nonblocking(fd);
951 if (ret < 0)
952 goto err;
953 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
954 if (ret < 0) {
955 ret = -ERRNO_TO_PARA_ERROR(errno);
956 goto err;
957 }
958 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
959 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
960 | S_IROTH | S_IWOTH;
961 ret = -E_CHMOD;
962 if (chmod(name, mode) < 0)
963 goto err;
964 }
965 if (listen(fd , 5) < 0) {
966 ret = -ERRNO_TO_PARA_ERROR(errno);
967 goto err;
968 }
969 return fd;
970 err:
971 close(fd);
972 return ret;
973 }
974
975 /**
976 * Prepare, create, and connect to a Unix domain socket for local communication.
977 *
978 * \param name The socket pathname.
979 *
980 * This function creates a local socket for sequenced, reliable, two-way,
981 * connection-based byte streams.
982 *
983 * \return The file descriptor of the connected socket on success, negative on
984 * errors.
985 *
986 * \sa \ref create_local_socket(), unix(7), connect(2).
987 */
988 int connect_local_socket(const char *name)
989 {
990 struct sockaddr_un unix_addr;
991 int fd, ret;
992
993 PARA_DEBUG_LOG("connecting to %s\n", name);
994 fd = socket(PF_UNIX, SOCK_STREAM, 0);
995 if (fd < 0)
996 return -ERRNO_TO_PARA_ERROR(errno);
997 ret = init_unix_addr(&unix_addr, name);
998 if (ret < 0)
999 goto err;
1000 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
1001 return fd;
1002 ret = -ERRNO_TO_PARA_ERROR(errno);
1003 err:
1004 close(fd);
1005 return ret;
1006 }
1007
1008 #ifndef HAVE_UCRED
1009 ssize_t send_cred_buffer(int sock, char *buf)
1010 {
1011 return write_buffer(sock, buf);
1012 }
1013 int recv_cred_buffer(int fd, char *buf, size_t size)
1014 {
1015 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
1016 }
1017 #else /* HAVE_UCRED */
1018
1019 /**
1020 * Send a buffer and the credentials of the current process to a socket.
1021 *
1022 * \param sock The file descriptor of the sending socket.
1023 * \param buf The zero-terminated buffer to send.
1024 *
1025 * \return On success, this call returns the number of bytes sent. On errors,
1026 * \p -E_SENDMSG is returned.
1027 *
1028 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
1029 */
1030 ssize_t send_cred_buffer(int sock, char *buf)
1031 {
1032 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
1033 struct msghdr msg;
1034 struct cmsghdr *cmsg;
1035 static struct iovec iov;
1036 struct ucred c;
1037 int ret;
1038
1039 /* Response data */
1040 iov.iov_base = buf;
1041 iov.iov_len = strlen(buf);
1042 c.pid = getpid();
1043 c.uid = getuid();
1044 c.gid = getgid();
1045 /* compose the message */
1046 memset(&msg, 0, sizeof(msg));
1047 msg.msg_iov = &iov;
1048 msg.msg_iovlen = 1;
1049 msg.msg_control = control;
1050 msg.msg_controllen = sizeof(control);
1051 /* attach the ucred struct */
1052 cmsg = CMSG_FIRSTHDR(&msg);
1053 cmsg->cmsg_level = SOL_SOCKET;
1054 cmsg->cmsg_type = SCM_CREDENTIALS;
1055 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1056 *(struct ucred *)CMSG_DATA(cmsg) = c;
1057 msg.msg_controllen = cmsg->cmsg_len;
1058 ret = sendmsg(sock, &msg, 0);
1059 if (ret < 0)
1060 ret = -E_SENDMSG;
1061 return ret;
1062 }
1063
1064 static void dispose_fds(int *fds, unsigned num)
1065 {
1066 int i;
1067
1068 for (i = 0; i < num; i++)
1069 close(fds[i]);
1070 }
1071
1072 /**
1073 * Receive a buffer and the Unix credentials of the sending process.
1074 *
1075 * \param fd The file descriptor of the receiving socket.
1076 * \param buf The buffer to store the received message.
1077 * \param size The length of \a buf in bytes.
1078 *
1079 * \return Negative on errors, the user id of the sending process on success.
1080 *
1081 * \sa \ref send_cred_buffer and the references given there.
1082 */
1083 int recv_cred_buffer(int fd, char *buf, size_t size)
1084 {
1085 char control[255] __a_aligned(8);
1086 struct msghdr msg;
1087 struct cmsghdr *cmsg;
1088 struct iovec iov;
1089 int result = 0;
1090 int yes = 1;
1091 struct ucred cred;
1092
1093 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1094 memset(&msg, 0, sizeof(msg));
1095 memset(buf, 0, size);
1096 iov.iov_base = buf;
1097 iov.iov_len = size;
1098 msg.msg_iov = &iov;
1099 msg.msg_iovlen = 1;
1100 msg.msg_control = control;
1101 msg.msg_controllen = sizeof(control);
1102 if (recvmsg(fd, &msg, 0) < 0)
1103 return -E_RECVMSG;
1104 result = -E_SCM_CREDENTIALS;
1105 cmsg = CMSG_FIRSTHDR(&msg);
1106 while (cmsg) {
1107 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1108 == SCM_CREDENTIALS) {
1109 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1110 result = cred.uid;
1111 } else
1112 if (cmsg->cmsg_level == SOL_SOCKET
1113 && cmsg->cmsg_type == SCM_RIGHTS) {
1114 dispose_fds((int *)CMSG_DATA(cmsg),
1115 (cmsg->cmsg_len - CMSG_LEN(0))
1116 / sizeof(int));
1117 }
1118 cmsg = CMSG_NXTHDR(&msg, cmsg);
1119 }
1120 return result;
1121 }
1122 #endif /* HAVE_UCRED */