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