740dfa182d4a162f2b493ac77ecebc8140c6f0ae
[paraslash.git] / net.c
1 /*
2 * Copyright (C) 2005-2013 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file net.c Networking-related helper functions. */
8
9 /*
10 * Since glibc 2.8, the _GNU_SOURCE feature test macro must be defined in order
11 * to obtain the definition of the ucred structure.
12 */
13 #define _GNU_SOURCE
14
15 #include <netdb.h>
16
17 /* At least NetBSD needs these. */
18 #ifndef AI_V4MAPPED
19 #define AI_V4MAPPED 0
20 #endif
21 #ifndef AI_ALL
22 #define AI_ALL 0
23 #endif
24 #ifndef AI_ADDRCONFIG
25 #define AI_ADDRCONFIG 0
26 #endif
27
28 #include <regex.h>
29
30 #include "para.h"
31 #include "error.h"
32 #include "net.h"
33 #include "string.h"
34 #include "list.h"
35 #include "fd.h"
36
37 /**
38 * Parse and validate IPv4 address/netmask string.
39 *
40 * \param cidr Address in CIDR notation
41 * \param addr Copy of the IPv4 address part of \a cidr
42 * \param addrlen Size of \a addr in bytes
43 * \param netmask Value of the netmask part in \a cidr or the
44 * default of 32 if not specified.
45 *
46 * \return Pointer to \a addr if succesful, NULL on error.
47 * \sa RFC 4632
48 */
49 char *parse_cidr(const char *cidr,
50 char *addr, ssize_t addrlen,
51 int32_t *netmask)
52 {
53 const char *o = cidr;
54 char *c = addr, *end = c + (addrlen - 1);
55
56 *netmask = 0x20;
57
58 if (cidr == NULL || addrlen < 1)
59 goto failed;
60
61 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
62 if (c == end)
63 goto failed;
64
65 if (*o == '/')
66 if (para_atoi32(++o, netmask) < 0 ||
67 *netmask < 0 || *netmask > 0x20)
68 goto failed;
69
70 if (is_valid_ipv4_address(addr))
71 return addr;
72 failed:
73 *addr = '\0';
74 return NULL;
75 }
76
77
78 /**
79 * Match string as a candidate IPv4 address.
80 *
81 * \param address The string to match.
82 * \return True if \a address has "dot-quad" format.
83 */
84 static bool is_v4_dot_quad(const char *address)
85 {
86 bool result;
87 regex_t r;
88
89 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
90 REG_EXTENDED | REG_NOSUB) >= 0);
91 result = regexec(&r, address, 0, NULL, 0) == 0;
92 regfree(&r);
93 return result;
94 }
95
96 /**
97 * Perform basic syntax checking on the host-part of an URL:
98 *
99 * - Since ':' is invalid in IPv4 addresses and DNS names, the
100 * presence of ':' causes interpretation as IPv6 address;
101 * - next the first-match-wins algorithm from RFC 3986 is applied;
102 * - else the string is considered as DNS name, to be resolved later.
103 *
104 * \param host The host string to check.
105 * \return True if \a host passes the syntax checks.
106 *
107 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5
108 */
109 static bool host_string_ok(const char *host)
110 {
111 if (host == NULL || *host == '\0')
112 return false;
113 if (strchr(host, ':') != NULL)
114 return is_valid_ipv6_address(host);
115 if (is_v4_dot_quad(host))
116 return is_valid_ipv4_address(host);
117 return true;
118 }
119
120 /**
121 * Parse and validate URL string.
122 *
123 * The URL syntax is loosely based on RFC 3986, supporting one of
124 * - "["host"]"[:port] for native IPv6 addresses and
125 * - host[:port] for IPv4 hostnames and DNS names.
126 *
127 * Native IPv6 addresses must be enclosed in square brackets, since
128 * otherwise there is an ambiguity with the port separator `:'.
129 * The 'port' part is always considered to be a number; if absent,
130 * it is set to -1, to indicate that a default port is to be used.
131 *
132 * The following are valid examples:
133 * - 10.10.1.1
134 * - 10.10.1.2:8000
135 * - localhost
136 * - localhost:8001
137 * - [::1]:8000
138 * - [badc0de::1]
139 *
140 * \param url The URL string to take apart.
141 * \param host To return the copied host part of \a url.
142 * \param hostlen The maximum length of \a host.
143 * \param port To return the port number (if any) of \a url.
144 *
145 * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned,
146 * \a host and \a port are undefined. If no port number was present in \a url,
147 * \a port is set to -1.
148 *
149 * \sa RFC 3986, 3.2.2/3.2.3
150 */
151 char *parse_url(const char *url,
152 char *host, ssize_t hostlen,
153 int32_t *port)
154 {
155 const char *o = url;
156 char *c = host, *end = c + (hostlen - 1);
157
158 *port = -1;
159
160 if (o == NULL || hostlen < 1)
161 goto failed;
162
163 if (*o == '[') {
164 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
165 if (c == end)
166 goto failed;
167
168 if (*o++ != ']' || (*o != '\0' && *o != ':'))
169 goto failed;
170 } else {
171 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
172 if (c == end && o[1])
173 goto failed;
174 }
175 }
176
177 if (*o == ':')
178 if (para_atoi32(++o, port) < 0 ||
179 *port < 0 || *port > 0xffff)
180 goto failed;
181 if (host_string_ok(host))
182 return host;
183 failed:
184 *host = '\0';
185 return NULL;
186 }
187
188 /**
189 * Stringify port number, resolve into service name where defined.
190 * \param port 2-byte port number, in host-byte-order.
191 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
192 * \return Pointer to static result buffer.
193 *
194 * \sa getservent(3), services(5), nsswitch.conf(5)
195 */
196 const char *stringify_port(int port, const char *transport)
197 {
198 static char service[NI_MAXSERV];
199
200 if (port < 0 || port > 0xFFFF) {
201 snprintf(service, sizeof(service), "undefined (%d)", port);
202 } else {
203 struct servent *se = getservbyport(htons(port), transport);
204
205 if (se == NULL)
206 snprintf(service, sizeof(service), "%u", port);
207 else
208 snprintf(service, sizeof(service), "%s", se->s_name);
209 }
210 return service;
211 }
212
213 /**
214 * Determine the socket type for a given layer-4 protocol.
215 *
216 * \param l4type The symbolic name of the transport-layer protocol.
217 *
218 * \sa ip(7), socket(2)
219 */
220 static inline int sock_type(const unsigned l4type)
221 {
222 switch (l4type) {
223 case IPPROTO_UDP: return SOCK_DGRAM;
224 case IPPROTO_TCP: return SOCK_STREAM;
225 case IPPROTO_DCCP: return SOCK_DCCP;
226 }
227 return -1; /* not supported here */
228 }
229
230 /**
231 * Pretty-print transport-layer name.
232 */
233 static const char *layer4_name(const unsigned l4type)
234 {
235 switch (l4type) {
236 case IPPROTO_UDP: return "UDP";
237 case IPPROTO_TCP: return "TCP";
238 case IPPROTO_DCCP: return "DCCP";
239 }
240 return "UNKNOWN PROTOCOL";
241 }
242
243 /**
244 * Flowopts: Transport-layer independent encapsulation of socket options.
245 *
246 * These collect individual socket options into a queue, which is disposed of
247 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
248 * only and should not be visible elsewhere.
249 *
250 * \sa setsockopt(2), makesock()
251 */
252 struct pre_conn_opt {
253 int sock_level; /**< Second argument to setsockopt() */
254 int sock_option; /**< Third argument to setsockopt() */
255 char *opt_name; /**< Stringified \a sock_option */
256 void *opt_val; /**< Fourth argument to setsockopt() */
257 socklen_t opt_len; /**< Fifth argument to setsockopt() */
258
259 struct list_head node; /**< FIFO, as sockopt order matters. */
260 };
261
262 /** FIFO list of pre-connection socket options to be set */
263 struct flowopts {
264 struct list_head sockopts;
265 };
266
267 /**
268 * Allocate and initialize a flowopt queue.
269 *
270 * \return A new structure to be passed to \ref flowopt_add(). It is
271 * automatically deallocated in \ref makesock().
272 */
273 struct flowopts *flowopt_new(void)
274 {
275 struct flowopts *new = para_malloc(sizeof(*new));
276
277 INIT_LIST_HEAD(&new->sockopts);
278 return new;
279 }
280
281 /**
282 * Append new socket option to flowopt queue.
283 *
284 * \param fo The flowopt queue to append to.
285 * \param lev Level at which \a opt resides.
286 * \param opt New option to add.
287 * \param name Stringified name of \a opt.
288 * \param val The value to set \a opt to.
289 * \param len Length of \a val.
290 *
291 * \sa setsockopt(2)
292 */
293 void flowopt_add(struct flowopts *fo, int lev, int opt,
294 const char *name, const void *val, int len)
295 {
296 struct pre_conn_opt *new = para_malloc(sizeof(*new));
297
298 new->sock_option = opt;
299 new->sock_level = lev;
300 new->opt_name = para_strdup(name);
301
302 if (val == NULL) {
303 new->opt_val = NULL;
304 new->opt_len = 0;
305 } else {
306 new->opt_val = para_malloc(len);
307 new->opt_len = len;
308 memcpy(new->opt_val, val, len);
309 }
310
311 list_add_tail(&new->node, &fo->sockopts);
312 }
313
314 /** Set the entire bunch of pre-connection options at once. */
315 static void flowopt_setopts(int sockfd, struct flowopts *fo)
316 {
317 struct pre_conn_opt *pc;
318
319 if (fo == NULL)
320 return;
321
322 list_for_each_entry(pc, &fo->sockopts, node)
323 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
324 pc->opt_val, pc->opt_len) < 0) {
325 PARA_EMERG_LOG("Can not set %s socket option: %s",
326 pc->opt_name, strerror(errno));
327 exit(EXIT_FAILURE);
328 }
329 }
330
331 static void flowopt_cleanup(struct flowopts *fo)
332 {
333 struct pre_conn_opt *cur, *next;
334
335 if (fo == NULL)
336 return;
337
338 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
339 free(cur->opt_name);
340 free(cur->opt_val);
341 free(cur);
342 }
343 free(fo);
344 }
345
346 /*
347 * Resolve an IPv4/IPv6 address.
348 *
349 * \param l4type The layer-4 type (\p IPPROTO_xxx).
350 * \param passive Whether \p AI_PASSIVE should be included as hint.
351 * \param host Remote or local hostname or IPv/6 address string.
352 * \param port_number Used to set the port in each returned address structure.
353 * \param result addrinfo structures are returned here.
354 *
355 * The interpretation of \a host depends on the value of \a passive. On a
356 * passive socket host is interpreted as an interface IPv4/6 address (can be
357 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
358 * address to connect to.
359 *
360 * \return Standard.
361 *
362 * \sa getaddrinfo(3).
363 */
364 static int lookup_address(unsigned l4type, bool passive, const char *host,
365 int port_number, struct addrinfo **result)
366 {
367 int ret;
368 char port[6]; /* port number has at most 5 digits */
369 struct addrinfo *addr = NULL, hints;
370
371 *result = NULL;
372 sprintf(port, "%u", port_number & 0xffff);
373 /* Set up address hint structure */
374 memset(&hints, 0, sizeof(hints));
375 hints.ai_family = AF_UNSPEC;
376 hints.ai_socktype = sock_type(l4type);
377 /*
378 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
379 * (and only then) pretend to be UDP.
380 */
381 if (l4type == IPPROTO_DCCP)
382 hints.ai_socktype = SOCK_DGRAM;
383 /* only use addresses available on the host */
384 hints.ai_flags = AI_ADDRCONFIG;
385 if (passive && host == NULL)
386 hints.ai_flags |= AI_PASSIVE;
387 /* Obtain local/remote address information */
388 ret = getaddrinfo(host, port, &hints, &addr);
389 if (ret != 0) {
390 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
391 layer4_name(l4type),
392 host? host : (passive? "[loopback]" : "[localhost]"),
393 port, gai_strerror(ret));
394 return -E_ADDRESS_LOOKUP;
395 }
396 *result = addr;
397 return 1;
398 }
399
400 /*
401 * Create an active or passive socket.
402 *
403 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
404 * \param passive Whether to call bind(2) or connect(2).
405 * \param ai Address information as obtained from \ref lookup_address().
406 * \param fo Socket options to be set before making the connection.
407 *
408 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
409 * algorithm tries all possible address combinations until it succeeds. If \a
410 * fo is supplied, options are set and cleanup is performed.
411 *
412 * \return File descriptor on success, \p E_MAKESOCK on errors.
413 *
414 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
415 * connect(2).
416 */
417 static int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
418 struct flowopts *fo)
419 {
420 struct addrinfo *local, *remote, *src = NULL, *dst = NULL;
421 int ret = -E_MAKESOCK, on = 1, sockfd = -1;
422
423 if (passive) {
424 local = ai;
425 remote = NULL;
426 } else {
427 local = NULL;
428 remote = ai;
429 }
430
431 /* Iterate over all src/dst combination, exhausting dst first */
432 for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) {
433 if (src && dst && src->ai_family == AF_INET
434 && dst->ai_family == AF_INET6)
435 goto get_next_dst; /* v4 -> v6 is not possible */
436
437 ret = socket(src ? src->ai_family : dst->ai_family,
438 sock_type(l4type), l4type);
439 if (ret < 0)
440 goto get_next_dst;
441 sockfd = ret;
442 /*
443 * Reuse the address on passive sockets to avoid failure on
444 * restart (protocols using listen()) and when creating
445 * multiple listener instances (UDP multicast).
446 */
447 if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
448 &on, sizeof(on)) == -1) {
449 ret = -ERRNO_TO_PARA_ERROR(errno);
450 close(sockfd);
451 PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
452 para_strerror(-ret));
453 break;
454 }
455 flowopt_setopts(sockfd, fo);
456
457 if (src) {
458 if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) {
459 close(sockfd);
460 goto get_next_src;
461 }
462 if (!dst) /* bind-only completed successfully */
463 break;
464 }
465
466 if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0)
467 break; /* connection completed successfully */
468 close(sockfd);
469 get_next_dst:
470 if (dst && (dst = dst->ai_next))
471 continue;
472 get_next_src:
473 if (src && (src = src->ai_next)) /* restart inner loop */
474 dst = remote;
475 }
476 if (src == NULL && dst == NULL)
477 return ret < 0? ret : -E_MAKESOCK;
478 return sockfd;
479 }
480
481 /**
482 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
483 *
484 * \param l4type The layer-4 type (\p IPPROTO_xxx).
485 * \param passive Whether this is a passive or active socket.
486 * \param host Passed to \ref lookup_address().
487 * \param port_number Passed to \ref lookup_address().
488 * \param fo Passed to \ref makesock_addrinfo().
489 *
490 * This creates a ready-made IPv4/v6 socket structure after looking up the
491 * necessary parameters. The function first calls \ref lookup_address() and
492 * passes the address information to makesock_addrinfo() to create and
493 * initialize the socket.
494 *
495 * \return The newly created file descriptor on success, a negative error code
496 * on failure.
497 *
498 * \sa \ref lookup_address(), \ref makesock_addrinfo().
499 */
500 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
501 struct flowopts *fo)
502 {
503 struct addrinfo *ai;
504 int ret = lookup_address(l4type, passive, host, port_number, &ai);
505
506 if (ret >= 0)
507 ret = makesock_addrinfo(l4type, passive, ai, fo);
508 if (ai)
509 freeaddrinfo(ai);
510 flowopt_cleanup(fo);
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 port The decimal port number to listen on.
524 * \param fo Flowopts (if any) to set before starting to listen.
525 *
526 * \return Positive integer (socket descriptor) on success, negative value
527 * otherwise.
528 *
529 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
530 */
531 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
532 {
533 int ret, fd = makesock(l4type, 1, NULL, port, fo);
534
535 if (fd > 0) {
536 ret = listen(fd, BACKLOG);
537 if (ret < 0) {
538 ret = errno;
539 close(fd);
540 return -ERRNO_TO_PARA_ERROR(ret);
541 }
542 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
543 layer4_name(l4type), port, fd);
544 }
545 return fd;
546 }
547
548 /**
549 * Determine IPv4/v6 socket address length.
550 * \param sa Container of IPv4 or IPv6 address.
551 * \return Address-family dependent address length.
552 */
553 static socklen_t salen(const struct sockaddr *sa)
554 {
555 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
556
557 return sa->sa_family == AF_INET6
558 ? sizeof(struct sockaddr_in6)
559 : sizeof(struct sockaddr_in);
560 }
561
562 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
563 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
564 {
565 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
566
567 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
568 }
569
570 /**
571 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
572 * \param ss Container of IPv4/6 address.
573 * \return Pointer to normalized address (may be static storage).
574 *
575 * \sa RFC 3493
576 */
577 static const struct sockaddr *
578 normalize_ip_address(const struct sockaddr_storage *ss)
579 {
580 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
581
582 if (SS_IS_ADDR_V4MAPPED(ss)) {
583 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
584 static struct sockaddr_in ia;
585
586 ia.sin_family = AF_INET;
587 ia.sin_port = ia6->sin6_port;
588 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
589 return (const struct sockaddr *)&ia;
590 }
591 return (const struct sockaddr *)ss;
592 }
593
594 /**
595 * Generic/fallback MTU values
596 *
597 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
598 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
599 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
600 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
601 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
602 * UDP senders should use the respective minimum values of EMTU_S.
603 */
604 static inline int generic_mtu(const int af_type)
605 {
606 return af_type == AF_INET6 ? 1280 : 576;
607 }
608
609 /** Crude approximation of IP header overhead - neglecting options. */
610 static inline int estimated_header_overhead(const int af_type)
611 {
612 return af_type == AF_INET6 ? 40 : 20;
613 }
614
615 /**
616 * Get the maximum transport-layer message size (MMS_S).
617 *
618 * \param sockfd The socket file descriptor.
619 *
620 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol familiy
621 * could not be determined, \p AF_INET is assumed.
622 *
623 * \return The maximum message size of the address family type.
624 */
625 int generic_max_transport_msg_size(int sockfd)
626 {
627 struct sockaddr_storage ss;
628 socklen_t sslen = sizeof(ss);
629 int af_type = AF_INET;
630
631 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
632 PARA_ERROR_LOG("can not determine remote address type: %s\n",
633 strerror(errno));
634 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
635 af_type = ss.ss_family;
636 }
637 return generic_mtu(af_type) - estimated_header_overhead(af_type);
638 }
639
640 /**
641 * Look up the local or remote side of a connected socket structure.
642 *
643 * \param fd The socket descriptor of the connected socket.
644 * \param getname Either \p getsockname() for local, or \p getpeername() for
645 * remote side.
646 *
647 * \return A static character string identifying hostname and port of the
648 * chosen side in numeric host:port format.
649 *
650 * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3),
651 * services(5), nsswitch.conf(5).
652 */
653 static char *__get_sock_name(int fd, typeof(getsockname) getname)
654 {
655 struct sockaddr_storage ss;
656 const struct sockaddr *sa;
657 socklen_t sslen = sizeof(ss);
658 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
659 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
660 int ret;
661
662 if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
663 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
664 fd, strerror(errno));
665 snprintf(output, sizeof(output), "(unknown)");
666 return output;
667 }
668 sa = normalize_ip_address(&ss);
669 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
670 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
671 if (ret) {
672 PARA_WARNING_LOG("hostname lookup error (%s).\n",
673 gai_strerror(ret));
674 snprintf(output, sizeof(output), "(lookup error)");
675 } else if (sa->sa_family == AF_INET6)
676 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
677 else
678 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
679 return output;
680 }
681
682 /**
683 * Look up the local side of a connected socket structure.
684 *
685 * \param sockfd The file descriptor of the socket.
686 *
687 * \return A pointer to a static buffer containing hostname an port. This
688 * buffer must not be freed by the caller.
689 *
690 * \sa remote_name().
691 */
692 char *local_name(int sockfd)
693 {
694 return __get_sock_name(sockfd, getsockname);
695 }
696
697 /**
698 * Look up the remote side of a connected socket structure.
699 *
700 * \param sockfd The file descriptor of the socket.
701 *
702 * \return Analogous to the return value of \ref local_name() but for the
703 * remote side.
704 *
705 * \sa local_name().
706 */
707 char *remote_name(int sockfd)
708 {
709 return __get_sock_name(sockfd, getpeername);
710 }
711
712 /**
713 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
714 *
715 * \param ss Container of IPv4/6 address.
716 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
717 *
718 * \sa RFC 3493.
719 */
720 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
721 {
722 const struct sockaddr *sa = normalize_ip_address(ss);
723
724 memset(ia, 0, sizeof(*ia));
725 if (sa->sa_family == AF_INET)
726 *ia = ((struct sockaddr_in *)sa)->sin_addr;
727 }
728
729 /**
730 * Receive data from a file descriptor.
731 *
732 * \param fd The file descriptor.
733 * \param buf The buffer to write the data to.
734 * \param size The size of \a buf.
735 *
736 * Receive at most \a size bytes from file descriptor \a fd.
737 *
738 * \return The number of bytes received on success, negative on errors, zero if
739 * the peer has performed an orderly shutdown.
740 *
741 * \sa recv(2).
742 */
743 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
744 {
745 ssize_t n;
746
747 n = recv(fd, buf, size, 0);
748 if (n == -1)
749 return -ERRNO_TO_PARA_ERROR(errno);
750 return n;
751 }
752
753 /**
754 * Receive and write terminating NULL byte.
755 *
756 * \param fd The file descriptor.
757 * \param buf The buffer to write the data to.
758 * \param size The size of \a buf.
759 *
760 * Read at most \a size - 1 bytes from file descriptor \a fd and
761 * write a NULL byte at the end of the received data.
762 *
763 * \return The return value of the underlying call to \a recv_bin_buffer().
764 *
765 * \sa recv_bin_buffer()
766 */
767 int recv_buffer(int fd, char *buf, size_t size)
768 {
769 int n;
770
771 assert(size);
772 n = recv_bin_buffer(fd, buf, size - 1);
773 if (n >= 0)
774 buf[n] = '\0';
775 else
776 *buf = '\0';
777 return n;
778 }
779
780 /**
781 * Wrapper around the accept system call.
782 *
783 * \param fd The listening socket.
784 * \param rfds An optional fd_set pointer.
785 * \param addr Structure which is filled in with the address of the peer socket.
786 * \param size Should contain the size of the structure pointed to by \a addr.
787 * \param new_fd Result pointer.
788 *
789 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
790 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
791 *
792 * \return Negative on errors, zero if no connections are present to be accepted,
793 * one otherwise.
794 *
795 * \sa accept(2).
796 */
797 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
798 {
799 int ret;
800
801 if (rfds && !FD_ISSET(fd, rfds))
802 return 0;
803 do
804 ret = accept(fd, (struct sockaddr *) addr, &size);
805 while (ret < 0 && errno == EINTR);
806
807 if (ret >= 0) {
808 *new_fd = ret;
809 return 1;
810 }
811 if (errno == EAGAIN || errno == EWOULDBLOCK)
812 return 0;
813 return -ERRNO_TO_PARA_ERROR(errno);
814 }
815
816 /**
817 * Probe the list of DCCP CCIDs configured on this host.
818 * \param ccid_array Pointer to return statically allocated array in.
819 * \return Number of elements returned in \a ccid_array or error.
820 *
821 * NB: This feature is only available on Linux > 2.6.30; on older kernels
822 * ENOPROTOOPT ("Protocol not available") will be returned.
823 */
824 int dccp_available_ccids(uint8_t **ccid_array)
825 {
826 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
827 socklen_t nccids = sizeof(ccids);
828 int ret, fd;
829
830 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
831 if (ret < 0)
832 return ret;
833
834 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
835 ccids, &nccids) < 0) {
836 ret = errno;
837 close(fd);
838 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
839 strerror(ret));
840 return -ERRNO_TO_PARA_ERROR(ret);
841 }
842
843 close(fd);
844 *ccid_array = ccids;
845 return nccids;
846 }
847
848 /**
849 * Prepare a structure for \p AF_UNIX socket addresses.
850 *
851 * \param u Pointer to the struct to be prepared.
852 * \param name The socket pathname.
853 *
854 * This just copies \a name to the sun_path component of \a u.
855 *
856 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
857 * than \p UNIX_PATH_MAX.
858 */
859 static int init_unix_addr(struct sockaddr_un *u, const char *name)
860 {
861 if (strlen(name) >= UNIX_PATH_MAX)
862 return -E_NAME_TOO_LONG;
863 memset(u->sun_path, 0, UNIX_PATH_MAX);
864 u->sun_family = PF_UNIX;
865 strcpy(u->sun_path, name);
866 return 1;
867 }
868
869 /**
870 * Prepare, create, and bind a socket for local communication.
871 *
872 * \param name The socket pathname.
873 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
874 * \param mode The desired mode of the socket.
875 *
876 * This function creates a local socket for sequenced, reliable,
877 * two-way, connection-based byte streams.
878 *
879 * \return The file descriptor, on success, negative on errors.
880 *
881 * \sa socket(2)
882 * \sa bind(2)
883 * \sa chmod(2)
884 */
885 int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
886 mode_t mode)
887 {
888 int fd, ret;
889
890 ret = init_unix_addr(unix_addr, name);
891 if (ret < 0)
892 return ret;
893 ret = socket(PF_UNIX, SOCK_STREAM, 0);
894 if (ret < 0)
895 return -ERRNO_TO_PARA_ERROR(errno);
896 fd = ret;
897 ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
898 if (ret < 0) {
899 ret = -ERRNO_TO_PARA_ERROR(errno);
900 goto err;
901 }
902 ret = -E_CHMOD;
903 if (chmod(name, mode) < 0)
904 goto err;
905 return fd;
906 err:
907 close(fd);
908 return ret;
909 }
910
911 /**
912 * Prepare, create, and connect to a Unix domain socket for local communication.
913 *
914 * \param name The socket pathname.
915 *
916 * This function creates a local socket for sequenced, reliable, two-way,
917 * connection-based byte streams.
918 *
919 * \return The file descriptor of the connected socket on success, negative on
920 * errors.
921 *
922 * \sa create_local_socket(), unix(7), connect(2).
923 */
924 int connect_local_socket(const char *name)
925 {
926 struct sockaddr_un unix_addr;
927 int fd, ret;
928
929 PARA_DEBUG_LOG("connecting to %s\n", name);
930 ret = init_unix_addr(&unix_addr, name);
931 if (ret < 0)
932 return ret;
933 fd = socket(PF_UNIX, SOCK_STREAM, 0);
934 if (fd < 0)
935 return -ERRNO_TO_PARA_ERROR(errno);
936 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
937 ret = -ERRNO_TO_PARA_ERROR(errno);
938 goto err;
939 }
940 return fd;
941 err:
942 close(fd);
943 return ret;
944 }
945
946 #ifndef HAVE_UCRED
947 ssize_t send_cred_buffer(int sock, char *buf)
948 {
949 return write_buffer(sock, buf);
950 }
951 int recv_cred_buffer(int fd, char *buf, size_t size)
952 {
953 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
954 }
955 #else /* HAVE_UCRED */
956 /**
957 * Send \p NULL-terminated buffer and Unix credentials of the current process.
958 *
959 * \param sock The socket file descriptor.
960 * \param buf The buffer to be sent.
961 *
962 * \return On success, this call returns the number of characters sent. On
963 * error, \p -E_SENDMSG is returned.
964 *
965 * \sa sendmsg(2), okir's Black Hats Manual.
966 */
967 ssize_t send_cred_buffer(int sock, char *buf)
968 {
969 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
970 struct msghdr msg;
971 struct cmsghdr *cmsg;
972 static struct iovec iov;
973 struct ucred c;
974 int ret;
975
976 /* Response data */
977 iov.iov_base = buf;
978 iov.iov_len = strlen(buf);
979 c.pid = getpid();
980 c.uid = getuid();
981 c.gid = getgid();
982 /* compose the message */
983 memset(&msg, 0, sizeof(msg));
984 msg.msg_iov = &iov;
985 msg.msg_iovlen = 1;
986 msg.msg_control = control;
987 msg.msg_controllen = sizeof(control);
988 /* attach the ucred struct */
989 cmsg = CMSG_FIRSTHDR(&msg);
990 cmsg->cmsg_level = SOL_SOCKET;
991 cmsg->cmsg_type = SCM_CREDENTIALS;
992 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
993 *(struct ucred *)CMSG_DATA(cmsg) = c;
994 msg.msg_controllen = cmsg->cmsg_len;
995 ret = sendmsg(sock, &msg, 0);
996 if (ret < 0)
997 ret = -E_SENDMSG;
998 return ret;
999 }
1000
1001 static void dispose_fds(int *fds, unsigned num)
1002 {
1003 int i;
1004
1005 for (i = 0; i < num; i++)
1006 close(fds[i]);
1007 }
1008
1009 /**
1010 * Receive a buffer and the Unix credentials of the sending process.
1011 *
1012 * \param fd the socket file descriptor.
1013 * \param buf the buffer to store the message.
1014 * \param size the size of \a buffer.
1015 *
1016 * \return negative on errors, the user id on success.
1017 *
1018 * \sa recvmsg(2), okir's Black Hats Manual.
1019 */
1020 int recv_cred_buffer(int fd, char *buf, size_t size)
1021 {
1022 char control[255];
1023 struct msghdr msg;
1024 struct cmsghdr *cmsg;
1025 struct iovec iov;
1026 int result = 0;
1027 int yes = 1;
1028 struct ucred cred;
1029
1030 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1031 memset(&msg, 0, sizeof(msg));
1032 memset(buf, 0, size);
1033 iov.iov_base = buf;
1034 iov.iov_len = size;
1035 msg.msg_iov = &iov;
1036 msg.msg_iovlen = 1;
1037 msg.msg_control = control;
1038 msg.msg_controllen = sizeof(control);
1039 if (recvmsg(fd, &msg, 0) < 0)
1040 return -E_RECVMSG;
1041 result = -E_SCM_CREDENTIALS;
1042 cmsg = CMSG_FIRSTHDR(&msg);
1043 while (cmsg) {
1044 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1045 == SCM_CREDENTIALS) {
1046 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1047 result = cred.uid;
1048 } else
1049 if (cmsg->cmsg_level == SOL_SOCKET
1050 && cmsg->cmsg_type == SCM_RIGHTS) {
1051 dispose_fds((int *) CMSG_DATA(cmsg),
1052 (cmsg->cmsg_len - CMSG_LEN(0))
1053 / sizeof(int));
1054 }
1055 cmsg = CMSG_NXTHDR(&msg, cmsg);
1056 }
1057 return result;
1058 }
1059 #endif /* HAVE_UCRED */