94610a84e2a3aa6b42ad9d5c11dbc03bc69cb306
[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 <netinet/in.h>
16 #include <arpa/inet.h>
17 #include <sys/un.h>
18 #include <sys/types.h>
19 #include <sys/socket.h>
20 #include <netdb.h>
21
22 /* At least NetBSD needs these. */
23 #ifndef AI_V4MAPPED
24 #define AI_V4MAPPED 0
25 #endif
26 #ifndef AI_ALL
27 #define AI_ALL 0
28 #endif
29 #ifndef AI_ADDRCONFIG
30 #define AI_ADDRCONFIG 0
31 #endif
32
33 #include <regex.h>
34
35 #include "para.h"
36 #include "error.h"
37 #include "net.h"
38 #include "string.h"
39 #include "list.h"
40 #include "fd.h"
41
42 /**
43 * Parse and validate IPv4 address/netmask string.
44 *
45 * \param cidr Address in CIDR notation
46 * \param addr Copy of the IPv4 address part of \a cidr
47 * \param addrlen Size of \a addr in bytes
48 * \param netmask Value of the netmask part in \a cidr or the
49 * default of 32 if not specified.
50 *
51 * \return Pointer to \a addr if succesful, NULL on error.
52 * \sa RFC 4632
53 */
54 char *parse_cidr(const char *cidr,
55 char *addr, ssize_t addrlen,
56 int32_t *netmask)
57 {
58 const char *o = cidr;
59 char *c = addr, *end = c + (addrlen - 1);
60
61 *netmask = 0x20;
62
63 if (cidr == NULL || addrlen < 1)
64 goto failed;
65
66 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
67 if (c == end)
68 goto failed;
69
70 if (*o == '/')
71 if (para_atoi32(++o, netmask) < 0 ||
72 *netmask < 0 || *netmask > 0x20)
73 goto failed;
74
75 if (is_valid_ipv4_address(addr))
76 return addr;
77 failed:
78 *addr = '\0';
79 return NULL;
80 }
81
82
83 /**
84 * Match string as a candidate IPv4 address.
85 *
86 * \param address The string to match.
87 * \return True if \a address has "dot-quad" format.
88 */
89 static bool is_v4_dot_quad(const char *address)
90 {
91 bool result;
92 regex_t r;
93
94 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
95 REG_EXTENDED | REG_NOSUB) >= 0);
96 result = regexec(&r, address, 0, NULL, 0) == 0;
97 regfree(&r);
98 return result;
99 }
100
101 /**
102 * Perform basic syntax checking on the host-part of an URL:
103 *
104 * - Since ':' is invalid in IPv4 addresses and DNS names, the
105 * presence of ':' causes interpretation as IPv6 address;
106 * - next the first-match-wins algorithm from RFC 3986 is applied;
107 * - else the string is considered as DNS name, to be resolved later.
108 *
109 * \param host The host string to check.
110 * \return True if \a host passes the syntax checks.
111 *
112 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5
113 */
114 static bool host_string_ok(const char *host)
115 {
116 if (host == NULL || *host == '\0')
117 return false;
118 if (strchr(host, ':') != NULL)
119 return is_valid_ipv6_address(host);
120 if (is_v4_dot_quad(host))
121 return is_valid_ipv4_address(host);
122 return true;
123 }
124
125 /**
126 * Parse and validate URL string.
127 *
128 * The URL syntax is loosely based on RFC 3986, supporting one of
129 * - "["host"]"[:port] for native IPv6 addresses and
130 * - host[:port] for IPv4 hostnames and DNS names.
131 *
132 * Native IPv6 addresses must be enclosed in square brackets, since
133 * otherwise there is an ambiguity with the port separator `:'.
134 * The 'port' part is always considered to be a number; if absent,
135 * it is set to -1, to indicate that a default port is to be used.
136 *
137 * The following are valid examples:
138 * - 10.10.1.1
139 * - 10.10.1.2:8000
140 * - localhost
141 * - localhost:8001
142 * - [::1]:8000
143 * - [badc0de::1]
144 *
145 * \param url The URL string to take apart.
146 * \param host To return the copied host part of \a url.
147 * \param hostlen The maximum length of \a host.
148 * \param port To return the port number (if any) of \a url.
149 *
150 * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned,
151 * \a host and \a port are undefined. If no port number was present in \a url,
152 * \a port is set to -1.
153 *
154 * \sa RFC 3986, 3.2.2/3.2.3
155 */
156 char *parse_url(const char *url,
157 char *host, ssize_t hostlen,
158 int32_t *port)
159 {
160 const char *o = url;
161 char *c = host, *end = c + (hostlen - 1);
162
163 *port = -1;
164
165 if (o == NULL || hostlen < 1)
166 goto failed;
167
168 if (*o == '[') {
169 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
170 if (c == end)
171 goto failed;
172
173 if (*o++ != ']' || (*o != '\0' && *o != ':'))
174 goto failed;
175 } else {
176 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
177 if (c == end && o[1])
178 goto failed;
179 }
180 }
181
182 if (*o == ':')
183 if (para_atoi32(++o, port) < 0 ||
184 *port < 0 || *port > 0xffff)
185 goto failed;
186 if (host_string_ok(host))
187 return host;
188 failed:
189 *host = '\0';
190 return NULL;
191 }
192
193 /**
194 * Stringify port number, resolve into service name where defined.
195 * \param port 2-byte port number, in host-byte-order.
196 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
197 * \return Pointer to static result buffer.
198 *
199 * \sa getservent(3), services(5), nsswitch.conf(5)
200 */
201 const char *stringify_port(int port, const char *transport)
202 {
203 static char service[NI_MAXSERV];
204
205 if (port < 0 || port > 0xFFFF) {
206 snprintf(service, sizeof(service), "undefined (%d)", port);
207 } else {
208 struct servent *se = getservbyport(htons(port), transport);
209
210 if (se == NULL)
211 snprintf(service, sizeof(service), "%u", port);
212 else
213 snprintf(service, sizeof(service), "%s", se->s_name);
214 }
215 return service;
216 }
217
218 /**
219 * Determine the socket type for a given layer-4 protocol.
220 *
221 * \param l4type The symbolic name of the transport-layer protocol.
222 *
223 * \sa ip(7), socket(2)
224 */
225 static inline int sock_type(const unsigned l4type)
226 {
227 switch (l4type) {
228 case IPPROTO_UDP: return SOCK_DGRAM;
229 case IPPROTO_TCP: return SOCK_STREAM;
230 case IPPROTO_DCCP: return SOCK_DCCP;
231 }
232 return -1; /* not supported here */
233 }
234
235 /**
236 * Pretty-print transport-layer name.
237 */
238 static const char *layer4_name(const unsigned l4type)
239 {
240 switch (l4type) {
241 case IPPROTO_UDP: return "UDP";
242 case IPPROTO_TCP: return "TCP";
243 case IPPROTO_DCCP: return "DCCP";
244 }
245 return "UNKNOWN PROTOCOL";
246 }
247
248 /**
249 * Flowopts: Transport-layer independent encapsulation of socket options.
250 *
251 * These collect individual socket options into a queue, which is disposed of
252 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
253 * only and should not be visible elsewhere.
254 *
255 * \sa setsockopt(2), makesock()
256 */
257 struct pre_conn_opt {
258 int sock_level; /**< Second argument to setsockopt() */
259 int sock_option; /**< Third argument to setsockopt() */
260 char *opt_name; /**< Stringified \a sock_option */
261 void *opt_val; /**< Fourth argument to setsockopt() */
262 socklen_t opt_len; /**< Fifth argument to setsockopt() */
263
264 struct list_head node; /**< FIFO, as sockopt order matters. */
265 };
266
267 /** FIFO list of pre-connection socket options to be set */
268 struct flowopts {
269 struct list_head sockopts;
270 };
271
272 /**
273 * Allocate and initialize a flowopt queue.
274 *
275 * \return A new structure to be passed to \ref flowopt_add(). It is
276 * automatically deallocated in \ref makesock().
277 */
278 struct flowopts *flowopt_new(void)
279 {
280 struct flowopts *new = para_malloc(sizeof(*new));
281
282 INIT_LIST_HEAD(&new->sockopts);
283 return new;
284 }
285
286 /**
287 * Append new socket option to flowopt queue.
288 *
289 * \param fo The flowopt queue to append to.
290 * \param lev Level at which \a opt resides.
291 * \param opt New option to add.
292 * \param name Stringified name of \a opt.
293 * \param val The value to set \a opt to.
294 * \param len Length of \a val.
295 *
296 * \sa setsockopt(2)
297 */
298 void flowopt_add(struct flowopts *fo, int lev, int opt,
299 const char *name, const void *val, int len)
300 {
301 struct pre_conn_opt *new = para_malloc(sizeof(*new));
302
303 new->sock_option = opt;
304 new->sock_level = lev;
305 new->opt_name = para_strdup(name);
306
307 if (val == NULL) {
308 new->opt_val = NULL;
309 new->opt_len = 0;
310 } else {
311 new->opt_val = para_malloc(len);
312 new->opt_len = len;
313 memcpy(new->opt_val, val, len);
314 }
315
316 list_add_tail(&new->node, &fo->sockopts);
317 }
318
319 /** Set the entire bunch of pre-connection options at once. */
320 static void flowopt_setopts(int sockfd, struct flowopts *fo)
321 {
322 struct pre_conn_opt *pc;
323
324 if (fo == NULL)
325 return;
326
327 list_for_each_entry(pc, &fo->sockopts, node)
328 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
329 pc->opt_val, pc->opt_len) < 0) {
330 PARA_EMERG_LOG("Can not set %s socket option: %s",
331 pc->opt_name, strerror(errno));
332 exit(EXIT_FAILURE);
333 }
334 }
335
336 static void flowopt_cleanup(struct flowopts *fo)
337 {
338 struct pre_conn_opt *cur, *next;
339
340 if (fo == NULL)
341 return;
342
343 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
344 free(cur->opt_name);
345 free(cur->opt_val);
346 free(cur);
347 }
348 free(fo);
349 }
350
351 /*
352 * Resolve an IPv4/IPv6 address.
353 *
354 * \param l4type The layer-4 type (\p IPPROTO_xxx).
355 * \param passive Whether \p AI_PASSIVE should be included as hint.
356 * \param host Remote or local hostname or IPv/6 address string.
357 * \param port_number Used to set the port in each returned address structure.
358 * \param result addrinfo structures are returned here.
359 *
360 * The interpretation of \a host depends on the value of \a passive. On a
361 * passive socket host is interpreted as an interface IPv4/6 address (can be
362 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
363 * address to connect to.
364 *
365 * \return Standard.
366 *
367 * \sa getaddrinfo(3).
368 */
369 static int lookup_address(unsigned l4type, bool passive, const char *host,
370 int port_number, struct addrinfo **result)
371 {
372 int ret;
373 char port[6]; /* port number has at most 5 digits */
374 struct addrinfo *addr = NULL, hints;
375
376 *result = NULL;
377 sprintf(port, "%u", port_number & 0xffff);
378 /* Set up address hint structure */
379 memset(&hints, 0, sizeof(hints));
380 hints.ai_family = AF_UNSPEC;
381 hints.ai_socktype = sock_type(l4type);
382 /*
383 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
384 * (and only then) pretend to be UDP.
385 */
386 if (l4type == IPPROTO_DCCP)
387 hints.ai_socktype = SOCK_DGRAM;
388 /* only use addresses available on the host */
389 hints.ai_flags = AI_ADDRCONFIG;
390 if (passive && host == NULL)
391 hints.ai_flags |= AI_PASSIVE;
392 /* Obtain local/remote address information */
393 ret = getaddrinfo(host, port, &hints, &addr);
394 if (ret != 0) {
395 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
396 layer4_name(l4type),
397 host? host : (passive? "[loopback]" : "[localhost]"),
398 port, gai_strerror(ret));
399 return -E_ADDRESS_LOOKUP;
400 }
401 *result = addr;
402 return 1;
403 }
404
405 /*
406 * Create an active or passive socket.
407 *
408 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
409 * \param passive Whether to call bind(2) or connect(2).
410 * \param ai Address information as obtained from \ref lookup_address().
411 * \param fo Socket options to be set before making the connection.
412 *
413 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
414 * algorithm tries all possible address combinations until it succeeds. If \a
415 * fo is supplied, options are set and cleanup is performed.
416 *
417 * \return File descriptor on success, \p E_MAKESOCK on errors.
418 *
419 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
420 * connect(2).
421 */
422 static int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
423 struct flowopts *fo)
424 {
425 int ret = -E_MAKESOCK, on = 1, sockfd = -1;
426
427 for (; ai; ai = ai->ai_next) {
428 ret = socket(ai->ai_family, sock_type(l4type), l4type);
429 if (ret < 0)
430 continue;
431 sockfd = ret;
432 flowopt_setopts(sockfd, fo);
433 if (passive) {
434 /*
435 * Reuse the address on passive sockets to avoid
436 * failure on restart (protocols using listen()) and
437 * when creating multiple listener instances (UDP
438 * multicast).
439 */
440 if (setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &on,
441 sizeof(on)) == -1) {
442 ret = -ERRNO_TO_PARA_ERROR(errno);
443 close(sockfd);
444 PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
445 para_strerror(-ret));
446 break;
447 }
448 if (bind(sockfd, ai->ai_addr, ai->ai_addrlen) < 0) {
449 close(sockfd);
450 continue;
451 }
452 /* bind completed successfully */
453 break;
454 } else {
455 if (connect(sockfd, ai->ai_addr, ai->ai_addrlen) == 0)
456 break; /* connection completed successfully */
457 }
458 close(sockfd);
459 }
460 if (!ai)
461 return ret < 0? ret : -E_MAKESOCK;
462 return sockfd;
463 }
464
465 /**
466 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
467 *
468 * \param l4type The layer-4 type (\p IPPROTO_xxx).
469 * \param passive Whether this is a passive or active socket.
470 * \param host Passed to \ref lookup_address().
471 * \param port_number Passed to \ref lookup_address().
472 * \param fo Passed to \ref makesock_addrinfo().
473 *
474 * This creates a ready-made IPv4/v6 socket structure after looking up the
475 * necessary parameters. The function first calls \ref lookup_address() and
476 * passes the address information to makesock_addrinfo() to create and
477 * initialize the socket.
478 *
479 * \return The newly created file descriptor on success, a negative error code
480 * on failure.
481 *
482 * \sa \ref lookup_address(), \ref makesock_addrinfo().
483 */
484 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
485 struct flowopts *fo)
486 {
487 struct addrinfo *ai;
488 int ret = lookup_address(l4type, passive, host, port_number, &ai);
489
490 if (ret >= 0)
491 ret = makesock_addrinfo(l4type, passive, ai, fo);
492 if (ai)
493 freeaddrinfo(ai);
494 flowopt_cleanup(fo);
495 if (ret < 0) {
496 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
497 layer4_name(l4type), host? host : (passive?
498 "[loopback]" : "[localhost]"), port_number);
499 }
500 return ret;
501 }
502
503 /**
504 * Create a passive / listening socket.
505 *
506 * \param l4type The transport-layer type (\p IPPROTO_xxx).
507 * \param port The decimal port number to listen on.
508 * \param fo Flowopts (if any) to set before starting to listen.
509 *
510 * \return Positive integer (socket descriptor) on success, negative value
511 * otherwise.
512 *
513 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
514 */
515 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
516 {
517 int ret, fd = makesock(l4type, 1, NULL, port, fo);
518
519 if (fd > 0) {
520 ret = listen(fd, BACKLOG);
521 if (ret < 0) {
522 ret = errno;
523 close(fd);
524 return -ERRNO_TO_PARA_ERROR(ret);
525 }
526 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
527 layer4_name(l4type), port, fd);
528 }
529 return fd;
530 }
531
532 /**
533 * Determine IPv4/v6 socket address length.
534 * \param sa Container of IPv4 or IPv6 address.
535 * \return Address-family dependent address length.
536 */
537 static socklen_t salen(const struct sockaddr *sa)
538 {
539 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
540
541 return sa->sa_family == AF_INET6
542 ? sizeof(struct sockaddr_in6)
543 : sizeof(struct sockaddr_in);
544 }
545
546 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
547 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
548 {
549 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
550
551 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
552 }
553
554 /**
555 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
556 * \param ss Container of IPv4/6 address.
557 * \return Pointer to normalized address (may be static storage).
558 *
559 * \sa RFC 3493
560 */
561 static const struct sockaddr *
562 normalize_ip_address(const struct sockaddr_storage *ss)
563 {
564 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
565
566 if (SS_IS_ADDR_V4MAPPED(ss)) {
567 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
568 static struct sockaddr_in ia;
569
570 ia.sin_family = AF_INET;
571 ia.sin_port = ia6->sin6_port;
572 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
573 return (const struct sockaddr *)&ia;
574 }
575 return (const struct sockaddr *)ss;
576 }
577
578 /**
579 * Generic/fallback MTU values
580 *
581 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
582 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
583 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
584 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
585 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
586 * UDP senders should use the respective minimum values of EMTU_S.
587 */
588 static inline int generic_mtu(const int af_type)
589 {
590 return af_type == AF_INET6 ? 1280 : 576;
591 }
592
593 /** Crude approximation of IP header overhead - neglecting options. */
594 static inline int estimated_header_overhead(const int af_type)
595 {
596 return af_type == AF_INET6 ? 40 : 20;
597 }
598
599 /**
600 * Get the maximum transport-layer message size (MMS_S).
601 *
602 * \param sockfd The socket file descriptor.
603 *
604 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol familiy
605 * could not be determined, \p AF_INET is assumed.
606 *
607 * \return The maximum message size of the address family type.
608 */
609 int generic_max_transport_msg_size(int sockfd)
610 {
611 struct sockaddr_storage ss;
612 socklen_t sslen = sizeof(ss);
613 int af_type = AF_INET;
614
615 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
616 PARA_ERROR_LOG("can not determine remote address type: %s\n",
617 strerror(errno));
618 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
619 af_type = ss.ss_family;
620 }
621 return generic_mtu(af_type) - estimated_header_overhead(af_type);
622 }
623
624 /**
625 * Look up the local or remote side of a connected socket structure.
626 *
627 * \param fd The socket descriptor of the connected socket.
628 * \param getname Either \p getsockname() for local, or \p getpeername() for
629 * remote side.
630 *
631 * \return A static character string identifying hostname and port of the
632 * chosen side in numeric host:port format.
633 *
634 * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3),
635 * services(5), nsswitch.conf(5).
636 */
637 static char *__get_sock_name(int fd, typeof(getsockname) getname)
638 {
639 struct sockaddr_storage ss;
640 const struct sockaddr *sa;
641 socklen_t sslen = sizeof(ss);
642 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
643 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
644 int ret;
645
646 if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
647 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
648 fd, strerror(errno));
649 snprintf(output, sizeof(output), "(unknown)");
650 return output;
651 }
652 sa = normalize_ip_address(&ss);
653 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
654 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
655 if (ret) {
656 PARA_WARNING_LOG("hostname lookup error (%s).\n",
657 gai_strerror(ret));
658 snprintf(output, sizeof(output), "(lookup error)");
659 } else if (sa->sa_family == AF_INET6)
660 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
661 else
662 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
663 return output;
664 }
665
666 /**
667 * Look up the local side of a connected socket structure.
668 *
669 * \param sockfd The file descriptor of the socket.
670 *
671 * \return A pointer to a static buffer containing hostname an port. This
672 * buffer must not be freed by the caller.
673 *
674 * \sa remote_name().
675 */
676 char *local_name(int sockfd)
677 {
678 return __get_sock_name(sockfd, getsockname);
679 }
680
681 /**
682 * Look up the remote side of a connected socket structure.
683 *
684 * \param sockfd The file descriptor of the socket.
685 *
686 * \return Analogous to the return value of \ref local_name() but for the
687 * remote side.
688 *
689 * \sa local_name().
690 */
691 char *remote_name(int sockfd)
692 {
693 return __get_sock_name(sockfd, getpeername);
694 }
695
696 /**
697 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
698 *
699 * \param ss Container of IPv4/6 address.
700 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
701 *
702 * \sa RFC 3493.
703 */
704 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
705 {
706 const struct sockaddr *sa = normalize_ip_address(ss);
707
708 memset(ia, 0, sizeof(*ia));
709 if (sa->sa_family == AF_INET)
710 *ia = ((struct sockaddr_in *)sa)->sin_addr;
711 }
712
713 /**
714 * Receive data from a file descriptor.
715 *
716 * \param fd The file descriptor.
717 * \param buf The buffer to write the data to.
718 * \param size The size of \a buf.
719 *
720 * Receive at most \a size bytes from file descriptor \a fd.
721 *
722 * \return The number of bytes received on success, negative on errors, zero if
723 * the peer has performed an orderly shutdown.
724 *
725 * \sa recv(2).
726 */
727 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
728 {
729 ssize_t n;
730
731 n = recv(fd, buf, size, 0);
732 if (n == -1)
733 return -ERRNO_TO_PARA_ERROR(errno);
734 return n;
735 }
736
737 /**
738 * Receive and write terminating NULL byte.
739 *
740 * \param fd The file descriptor.
741 * \param buf The buffer to write the data to.
742 * \param size The size of \a buf.
743 *
744 * Read at most \a size - 1 bytes from file descriptor \a fd and
745 * write a NULL byte at the end of the received data.
746 *
747 * \return The return value of the underlying call to \a recv_bin_buffer().
748 *
749 * \sa recv_bin_buffer()
750 */
751 int recv_buffer(int fd, char *buf, size_t size)
752 {
753 int n;
754
755 assert(size);
756 n = recv_bin_buffer(fd, buf, size - 1);
757 if (n >= 0)
758 buf[n] = '\0';
759 else
760 *buf = '\0';
761 return n;
762 }
763
764 /**
765 * Wrapper around the accept system call.
766 *
767 * \param fd The listening socket.
768 * \param rfds An optional fd_set pointer.
769 * \param addr Structure which is filled in with the address of the peer socket.
770 * \param size Should contain the size of the structure pointed to by \a addr.
771 * \param new_fd Result pointer.
772 *
773 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
774 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
775 *
776 * \return Negative on errors, zero if no connections are present to be accepted,
777 * one otherwise.
778 *
779 * \sa accept(2).
780 */
781 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
782 {
783 int ret;
784
785 if (rfds && !FD_ISSET(fd, rfds))
786 return 0;
787 do
788 ret = accept(fd, (struct sockaddr *) addr, &size);
789 while (ret < 0 && errno == EINTR);
790
791 if (ret >= 0) {
792 *new_fd = ret;
793 return 1;
794 }
795 if (errno == EAGAIN || errno == EWOULDBLOCK)
796 return 0;
797 return -ERRNO_TO_PARA_ERROR(errno);
798 }
799
800 /**
801 * Probe the list of DCCP CCIDs configured on this host.
802 * \param ccid_array Pointer to return statically allocated array in.
803 * \return Number of elements returned in \a ccid_array or error.
804 *
805 * NB: This feature is only available on Linux > 2.6.30; on older kernels
806 * ENOPROTOOPT ("Protocol not available") will be returned.
807 */
808 int dccp_available_ccids(uint8_t **ccid_array)
809 {
810 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
811 socklen_t nccids = sizeof(ccids);
812 int ret, fd;
813
814 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
815 if (ret < 0)
816 return ret;
817
818 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
819 ccids, &nccids) < 0) {
820 ret = errno;
821 close(fd);
822 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
823 strerror(ret));
824 return -ERRNO_TO_PARA_ERROR(ret);
825 }
826
827 close(fd);
828 *ccid_array = ccids;
829 return nccids;
830 }
831
832 /**
833 * Prepare a structure for \p AF_UNIX socket addresses.
834 *
835 * \param u Pointer to the struct to be prepared.
836 * \param name The socket pathname.
837 *
838 * This just copies \a name to the sun_path component of \a u.
839 *
840 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
841 * than \p UNIX_PATH_MAX.
842 */
843 static int init_unix_addr(struct sockaddr_un *u, const char *name)
844 {
845 if (strlen(name) >= UNIX_PATH_MAX)
846 return -E_NAME_TOO_LONG;
847 memset(u->sun_path, 0, UNIX_PATH_MAX);
848 u->sun_family = PF_UNIX;
849 strcpy(u->sun_path, name);
850 return 1;
851 }
852
853 /**
854 * Prepare, create, and bind a socket for local communication.
855 *
856 * \param name The socket pathname.
857 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
858 * \param mode The desired mode of the socket.
859 *
860 * This function creates a local socket for sequenced, reliable,
861 * two-way, connection-based byte streams.
862 *
863 * \return The file descriptor, on success, negative on errors.
864 *
865 * \sa socket(2)
866 * \sa bind(2)
867 * \sa chmod(2)
868 */
869 int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
870 mode_t mode)
871 {
872 int fd, ret;
873
874 ret = init_unix_addr(unix_addr, name);
875 if (ret < 0)
876 return ret;
877 ret = socket(PF_UNIX, SOCK_STREAM, 0);
878 if (ret < 0)
879 return -ERRNO_TO_PARA_ERROR(errno);
880 fd = ret;
881 ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
882 if (ret < 0) {
883 ret = -ERRNO_TO_PARA_ERROR(errno);
884 goto err;
885 }
886 ret = -E_CHMOD;
887 if (chmod(name, mode) < 0)
888 goto err;
889 return fd;
890 err:
891 close(fd);
892 return ret;
893 }
894
895 /**
896 * Prepare, create, and connect to a Unix domain socket for local communication.
897 *
898 * \param name The socket pathname.
899 *
900 * This function creates a local socket for sequenced, reliable, two-way,
901 * connection-based byte streams.
902 *
903 * \return The file descriptor of the connected socket on success, negative on
904 * errors.
905 *
906 * \sa create_local_socket(), unix(7), connect(2).
907 */
908 int connect_local_socket(const char *name)
909 {
910 struct sockaddr_un unix_addr;
911 int fd, ret;
912
913 PARA_DEBUG_LOG("connecting to %s\n", name);
914 ret = init_unix_addr(&unix_addr, name);
915 if (ret < 0)
916 return ret;
917 fd = socket(PF_UNIX, SOCK_STREAM, 0);
918 if (fd < 0)
919 return -ERRNO_TO_PARA_ERROR(errno);
920 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
921 ret = -ERRNO_TO_PARA_ERROR(errno);
922 goto err;
923 }
924 return fd;
925 err:
926 close(fd);
927 return ret;
928 }
929
930 #ifndef HAVE_UCRED
931 ssize_t send_cred_buffer(int sock, char *buf)
932 {
933 return write_buffer(sock, buf);
934 }
935 int recv_cred_buffer(int fd, char *buf, size_t size)
936 {
937 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
938 }
939 #else /* HAVE_UCRED */
940 /**
941 * Send \p NULL-terminated buffer and Unix credentials of the current process.
942 *
943 * \param sock The socket file descriptor.
944 * \param buf The buffer to be sent.
945 *
946 * \return On success, this call returns the number of characters sent. On
947 * error, \p -E_SENDMSG is returned.
948 *
949 * \sa sendmsg(2), okir's Black Hats Manual.
950 */
951 ssize_t send_cred_buffer(int sock, char *buf)
952 {
953 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
954 struct msghdr msg;
955 struct cmsghdr *cmsg;
956 static struct iovec iov;
957 struct ucred c;
958 int ret;
959
960 /* Response data */
961 iov.iov_base = buf;
962 iov.iov_len = strlen(buf);
963 c.pid = getpid();
964 c.uid = getuid();
965 c.gid = getgid();
966 /* compose the message */
967 memset(&msg, 0, sizeof(msg));
968 msg.msg_iov = &iov;
969 msg.msg_iovlen = 1;
970 msg.msg_control = control;
971 msg.msg_controllen = sizeof(control);
972 /* attach the ucred struct */
973 cmsg = CMSG_FIRSTHDR(&msg);
974 cmsg->cmsg_level = SOL_SOCKET;
975 cmsg->cmsg_type = SCM_CREDENTIALS;
976 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
977 *(struct ucred *)CMSG_DATA(cmsg) = c;
978 msg.msg_controllen = cmsg->cmsg_len;
979 ret = sendmsg(sock, &msg, 0);
980 if (ret < 0)
981 ret = -E_SENDMSG;
982 return ret;
983 }
984
985 static void dispose_fds(int *fds, unsigned num)
986 {
987 int i;
988
989 for (i = 0; i < num; i++)
990 close(fds[i]);
991 }
992
993 /**
994 * Receive a buffer and the Unix credentials of the sending process.
995 *
996 * \param fd the socket file descriptor.
997 * \param buf the buffer to store the message.
998 * \param size the size of \a buffer.
999 *
1000 * \return negative on errors, the user id on success.
1001 *
1002 * \sa recvmsg(2), okir's Black Hats Manual.
1003 */
1004 int recv_cred_buffer(int fd, char *buf, size_t size)
1005 {
1006 char control[255];
1007 struct msghdr msg;
1008 struct cmsghdr *cmsg;
1009 struct iovec iov;
1010 int result = 0;
1011 int yes = 1;
1012 struct ucred cred;
1013
1014 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1015 memset(&msg, 0, sizeof(msg));
1016 memset(buf, 0, size);
1017 iov.iov_base = buf;
1018 iov.iov_len = size;
1019 msg.msg_iov = &iov;
1020 msg.msg_iovlen = 1;
1021 msg.msg_control = control;
1022 msg.msg_controllen = sizeof(control);
1023 if (recvmsg(fd, &msg, 0) < 0)
1024 return -E_RECVMSG;
1025 result = -E_SCM_CREDENTIALS;
1026 cmsg = CMSG_FIRSTHDR(&msg);
1027 while (cmsg) {
1028 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1029 == SCM_CREDENTIALS) {
1030 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1031 result = cred.uid;
1032 } else
1033 if (cmsg->cmsg_level == SOL_SOCKET
1034 && cmsg->cmsg_type == SCM_RIGHTS) {
1035 dispose_fds((int *) CMSG_DATA(cmsg),
1036 (cmsg->cmsg_len - CMSG_LEN(0))
1037 / sizeof(int));
1038 }
1039 cmsg = CMSG_NXTHDR(&msg, cmsg);
1040 }
1041 return result;
1042 }
1043 #endif /* HAVE_UCRED */