crypt.c: Combine load_key() and get_private_key().
[paraslash.git] / net.c
1 /*
2 * Copyright (C) 2005 Andre Noll <maan@tuebingen.mpg.de>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file net.c Networking-related helper functions. */
8
9 #include "para.h"
10
11 #include <netinet/in.h>
12 #include <arpa/inet.h>
13 #include <sys/un.h>
14 #include <sys/types.h>
15 #include <sys/socket.h>
16 #include <netdb.h>
17
18 /* At least NetBSD needs these. */
19 #ifndef AI_V4MAPPED
20 #define AI_V4MAPPED 0
21 #endif
22 #ifndef AI_ALL
23 #define AI_ALL 0
24 #endif
25 #ifndef AI_ADDRCONFIG
26 #define AI_ADDRCONFIG 0
27 #endif
28
29 #include <regex.h>
30
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 successful, 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 || *port < 0 || *port > 0xffff)
179 goto failed;
180 if (host_string_ok(host))
181 return host;
182 failed:
183 *host = '\0';
184 return NULL;
185 }
186
187 /**
188 * Stringify port number, resolve into service name where defined.
189 *
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), "%d", 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 /**
332 * Deallocate all resources of a flowopts structure.
333 *
334 * \param fo A pointer as returned from flowopt_new().
335 *
336 * It's OK to pass \p NULL here in which case the function does nothing.
337 */
338 void flowopt_cleanup(struct flowopts *fo)
339 {
340 struct pre_conn_opt *cur, *next;
341
342 if (fo == NULL)
343 return;
344
345 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
346 free(cur->opt_name);
347 free(cur->opt_val);
348 free(cur);
349 }
350 free(fo);
351 }
352
353 /**
354 * Resolve an IPv4/IPv6 address.
355 *
356 * \param l4type The layer-4 type (\p IPPROTO_xxx).
357 * \param passive Whether \p AI_PASSIVE should be included as hint.
358 * \param host Remote or local hostname or IPv/6 address string.
359 * \param port_number Used to set the port in each returned address structure.
360 * \param result addrinfo structures are returned here.
361 *
362 * The interpretation of \a host depends on the value of \a passive. On a
363 * passive socket host is interpreted as an interface IPv4/6 address (can be
364 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
365 * address to connect to.
366 *
367 * \return Standard.
368 *
369 * \sa getaddrinfo(3).
370 */
371 int lookup_address(unsigned l4type, bool passive, const char *host,
372 int port_number, struct addrinfo **result)
373 {
374 int ret;
375 char port[6]; /* port number has at most 5 digits */
376 struct addrinfo *addr = NULL, hints;
377
378 *result = NULL;
379 sprintf(port, "%d", port_number & 0xffff);
380 /* Set up address hint structure */
381 memset(&hints, 0, sizeof(hints));
382 hints.ai_family = AF_UNSPEC;
383 hints.ai_socktype = sock_type(l4type);
384 /*
385 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
386 * (and only then) pretend to be UDP.
387 */
388 if (l4type == IPPROTO_DCCP)
389 hints.ai_socktype = SOCK_DGRAM;
390 /* only use addresses available on the host */
391 hints.ai_flags = AI_ADDRCONFIG;
392 if (passive && host == NULL)
393 hints.ai_flags |= AI_PASSIVE;
394 /* Obtain local/remote address information */
395 ret = getaddrinfo(host, port, &hints, &addr);
396 if (ret != 0) {
397 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
398 layer4_name(l4type),
399 host? host : (passive? "[loopback]" : "[localhost]"),
400 port, gai_strerror(ret));
401 return -E_ADDRESS_LOOKUP;
402 }
403 *result = addr;
404 return 1;
405 }
406
407 /**
408 * Create an active or passive socket.
409 *
410 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
411 * \param passive Whether to call bind(2) or connect(2).
412 * \param ai Address information as obtained from \ref lookup_address().
413 * \param fo Socket options to be set before making the connection.
414 *
415 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
416 * algorithm tries all possible address combinations until it succeeds. If \a
417 * fo is supplied, options are set but cleanup must be performed in the caller.
418 *
419 * \return File descriptor on success, \p E_MAKESOCK on errors.
420 *
421 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
422 * connect(2).
423 */
424 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
425 struct flowopts *fo)
426 {
427 int ret = -E_MAKESOCK, on = 1;
428
429 for (; ai; ai = ai->ai_next) {
430 int fd;
431 ret = socket(ai->ai_family, sock_type(l4type), l4type);
432 if (ret < 0)
433 continue;
434 fd = ret;
435 flowopt_setopts(fd, fo);
436 if (!passive) {
437 if (connect(fd, ai->ai_addr, ai->ai_addrlen) == 0)
438 return fd;
439 close(fd);
440 continue;
441 }
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 (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
448 sizeof(on)) == -1) {
449 close(fd);
450 continue;
451 }
452 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
453 close(fd);
454 continue;
455 }
456 return fd;
457 }
458 return -E_MAKESOCK;
459 }
460
461 /**
462 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
463 *
464 * \param l4type The layer-4 type (\p IPPROTO_xxx).
465 * \param passive Whether this is a passive or active socket.
466 * \param host Passed to \ref lookup_address().
467 * \param port_number Passed to \ref lookup_address().
468 * \param fo Passed to \ref makesock_addrinfo().
469 *
470 * This creates a ready-made IPv4/v6 socket structure after looking up the
471 * necessary parameters. The function first calls \ref lookup_address() and
472 * passes the address information to makesock_addrinfo() to create and
473 * initialize the socket.
474 *
475 * \return The newly created file descriptor on success, a negative error code
476 * on failure.
477 *
478 * \sa \ref lookup_address(), \ref makesock_addrinfo().
479 */
480 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
481 struct flowopts *fo)
482 {
483 struct addrinfo *ai;
484 int ret = lookup_address(l4type, passive, host, port_number, &ai);
485
486 if (ret >= 0)
487 ret = makesock_addrinfo(l4type, passive, ai, fo);
488 if (ai)
489 freeaddrinfo(ai);
490 if (ret < 0) {
491 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
492 layer4_name(l4type), host? host : (passive?
493 "[loopback]" : "[localhost]"), port_number);
494 }
495 return ret;
496 }
497
498 /**
499 * Create a passive / listening socket.
500 *
501 * \param l4type The transport-layer type (\p IPPROTO_xxx).
502 * \param port The decimal port number to listen on.
503 * \param fo Flowopts (if any) to set before starting to listen.
504 *
505 * \return Positive integer (socket descriptor) on success, negative value
506 * otherwise.
507 *
508 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
509 */
510 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
511 {
512 int ret, fd = makesock(l4type, 1, NULL, port, fo);
513
514 if (fd > 0) {
515 ret = listen(fd, BACKLOG);
516 if (ret < 0) {
517 ret = errno;
518 close(fd);
519 return -ERRNO_TO_PARA_ERROR(ret);
520 }
521 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
522 layer4_name(l4type), port, fd);
523 }
524 return fd;
525 }
526
527 /**
528 * Determine IPv4/v6 socket address length.
529 * \param sa Container of IPv4 or IPv6 address.
530 * \return Address-family dependent address length.
531 */
532 static socklen_t salen(const struct sockaddr *sa)
533 {
534 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
535
536 return sa->sa_family == AF_INET6
537 ? sizeof(struct sockaddr_in6)
538 : sizeof(struct sockaddr_in);
539 }
540
541 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
542 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
543 {
544 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
545
546 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
547 }
548
549 /**
550 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
551 * \param ss Container of IPv4/6 address.
552 * \return Pointer to normalized address (may be static storage).
553 *
554 * \sa RFC 3493
555 */
556 static const struct sockaddr *
557 normalize_ip_address(const struct sockaddr_storage *ss)
558 {
559 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
560
561 if (SS_IS_ADDR_V4MAPPED(ss)) {
562 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
563 static struct sockaddr_in ia;
564
565 ia.sin_family = AF_INET;
566 ia.sin_port = ia6->sin6_port;
567 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
568 return (const struct sockaddr *)&ia;
569 }
570 return (const struct sockaddr *)ss;
571 }
572
573 /**
574 * Generic/fallback MTU values
575 *
576 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
577 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
578 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
579 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
580 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
581 * UDP senders should use the respective minimum values of EMTU_S.
582 */
583 static inline int generic_mtu(const int af_type)
584 {
585 return af_type == AF_INET6 ? 1280 : 576;
586 }
587
588 /** Crude approximation of IP header overhead - neglecting options. */
589 static inline int estimated_header_overhead(const int af_type)
590 {
591 return af_type == AF_INET6 ? 40 : 20;
592 }
593
594 /**
595 * Get the maximum transport-layer message size (MMS_S).
596 *
597 * \param sockfd The socket file descriptor.
598 *
599 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
600 * could not be determined, \p AF_INET is assumed.
601 *
602 * \return The maximum message size of the address family type.
603 */
604 int generic_max_transport_msg_size(int sockfd)
605 {
606 struct sockaddr_storage ss;
607 socklen_t sslen = sizeof(ss);
608 int af_type = AF_INET;
609
610 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
611 PARA_ERROR_LOG("can not determine remote address type: %s\n",
612 strerror(errno));
613 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
614 af_type = ss.ss_family;
615 }
616 return generic_mtu(af_type) - estimated_header_overhead(af_type);
617 }
618
619 /**
620 * Look up the remote side of a connected socket structure.
621 *
622 * \param fd The socket descriptor of the connected socket.
623 *
624 * \return A static character string identifying hostname and port of the
625 * chosen side in numeric host:port format.
626 *
627 * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3),
628 * services(5), nsswitch.conf(5).
629 */
630 char *remote_name(int fd)
631 {
632 struct sockaddr_storage ss;
633 const struct sockaddr *sa;
634 socklen_t sslen = sizeof(ss);
635 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
636 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
637 int ret;
638
639 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
640 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
641 fd, strerror(errno));
642 snprintf(output, sizeof(output), "(unknown)");
643 return output;
644 }
645 sa = normalize_ip_address(&ss);
646 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
647 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
648 if (ret) {
649 PARA_WARNING_LOG("hostname lookup error (%s).\n",
650 gai_strerror(ret));
651 snprintf(output, sizeof(output), "(lookup error)");
652 } else if (sa->sa_family == AF_INET6)
653 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
654 else
655 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
656 return output;
657 }
658
659 /**
660 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
661 *
662 * \param ss Container of IPv4/6 address.
663 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
664 *
665 * \sa RFC 3493.
666 */
667 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
668 {
669 const struct sockaddr *sa = normalize_ip_address(ss);
670
671 memset(ia, 0, sizeof(*ia));
672 if (sa->sa_family == AF_INET)
673 *ia = ((struct sockaddr_in *)sa)->sin_addr;
674 }
675
676 /**
677 * Compare the address part of IPv4/6 addresses.
678 *
679 * \param sa1 First address.
680 * \param sa2 Second address.
681 *
682 * \return True iff the IP address of \a sa1 and \a sa2 match.
683 */
684 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
685 {
686 if (!sa1 || !sa2)
687 return false;
688 if (sa1->sa_family != sa2->sa_family)
689 return false;
690 if (sa1->sa_family == AF_INET) {
691 struct sockaddr_in *a1 = (typeof(a1))sa1,
692 *a2 = (typeof (a2))sa2;
693 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
694 } else if (sa1->sa_family == AF_INET6) {
695 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
696 *a2 = (typeof (a2))sa2;
697 return !memcmp(a1, a2, sizeof(*a1));
698 } else
699 return false;
700 }
701
702 /**
703 * Receive data from a file descriptor.
704 *
705 * \param fd The file descriptor.
706 * \param buf The buffer to write the data to.
707 * \param size The size of \a buf.
708 *
709 * Receive at most \a size bytes from file descriptor \a fd.
710 *
711 * \return The number of bytes received on success, negative on errors, zero if
712 * the peer has performed an orderly shutdown.
713 *
714 * \sa recv(2).
715 */
716 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
717 {
718 ssize_t n;
719
720 n = recv(fd, buf, size, 0);
721 if (n == -1)
722 return -ERRNO_TO_PARA_ERROR(errno);
723 return n;
724 }
725
726 /**
727 * Receive and write terminating NULL byte.
728 *
729 * \param fd The file descriptor.
730 * \param buf The buffer to write the data to.
731 * \param size The size of \a buf.
732 *
733 * Read at most \a size - 1 bytes from file descriptor \a fd and
734 * write a NULL byte at the end of the received data.
735 *
736 * \return The return value of the underlying call to \a recv_bin_buffer().
737 *
738 * \sa recv_bin_buffer()
739 */
740 int recv_buffer(int fd, char *buf, size_t size)
741 {
742 int n;
743
744 assert(size);
745 n = recv_bin_buffer(fd, buf, size - 1);
746 if (n >= 0)
747 buf[n] = '\0';
748 else
749 *buf = '\0';
750 return n;
751 }
752
753 /**
754 * Wrapper around the accept system call.
755 *
756 * \param fd The listening socket.
757 * \param rfds An optional fd_set pointer.
758 * \param addr Structure which is filled in with the address of the peer socket.
759 * \param size Should contain the size of the structure pointed to by \a addr.
760 * \param new_fd Result pointer.
761 *
762 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
763 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
764 *
765 * \return Negative on errors, zero if no connections are present to be accepted,
766 * one otherwise.
767 *
768 * \sa accept(2).
769 */
770 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
771 {
772 int ret;
773
774 if (rfds && !FD_ISSET(fd, rfds))
775 return 0;
776 do
777 ret = accept(fd, (struct sockaddr *) addr, &size);
778 while (ret < 0 && errno == EINTR);
779
780 if (ret >= 0) {
781 *new_fd = ret;
782 return 1;
783 }
784 if (errno == EAGAIN || errno == EWOULDBLOCK)
785 return 0;
786 return -ERRNO_TO_PARA_ERROR(errno);
787 }
788
789 /**
790 * Probe the list of DCCP CCIDs configured on this host.
791 * \param ccid_array Pointer to return statically allocated array in.
792 * \return Number of elements returned in \a ccid_array or error.
793 *
794 * NB: This feature is only available on Linux > 2.6.30; on older kernels
795 * ENOPROTOOPT ("Protocol not available") will be returned.
796 */
797 int dccp_available_ccids(uint8_t **ccid_array)
798 {
799 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
800 socklen_t nccids = sizeof(ccids);
801 int ret, fd;
802
803 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
804 if (ret < 0)
805 return ret;
806
807 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
808 ccids, &nccids) < 0) {
809 ret = errno;
810 close(fd);
811 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
812 strerror(ret));
813 return -ERRNO_TO_PARA_ERROR(ret);
814 }
815
816 close(fd);
817 *ccid_array = ccids;
818 return nccids;
819 }
820
821 /**
822 * Prepare a structure for \p AF_UNIX socket addresses.
823 *
824 * \param u Pointer to the struct to be prepared.
825 * \param name The socket pathname.
826 *
827 * This just copies \a name to the sun_path component of \a u.
828 *
829 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
830 * than \p UNIX_PATH_MAX.
831 */
832 static int init_unix_addr(struct sockaddr_un *u, const char *name,
833 bool abstract)
834 {
835 if (strlen(name) + abstract >= UNIX_PATH_MAX)
836 return -E_NAME_TOO_LONG;
837 memset(u->sun_path, 0, UNIX_PATH_MAX);
838 u->sun_family = PF_UNIX;
839 strcpy(u->sun_path + abstract, name);
840 return 1;
841 }
842
843 /**
844 * Create a socket for local communication and listen on it.
845 *
846 * \param name The socket pathname.
847 * \param mode The desired permissions of the socket.
848 *
849 * This function creates a passive local socket for sequenced, reliable,
850 * two-way, connection-based byte streams. The socket file descriptor is set to
851 * nonblocking mode and listen(2) is called to prepare the socket for
852 * accepting incoming connection requests.
853 *
854 * If mode is zero, an abstract socket (a non-portable Linux extension) is
855 * created. In this case the socket name has no connection with filesystem
856 * pathnames.
857 *
858 * \return The file descriptor on success, negative error code on failure.
859 *
860 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
861 */
862 int create_local_socket(const char *name, mode_t mode)
863 {
864 struct sockaddr_un unix_addr;
865 int fd, ret;
866 bool abstract = mode == 0;
867
868 ret = init_unix_addr(&unix_addr, name, abstract);
869 if (ret < 0)
870 return ret;
871 ret = socket(PF_UNIX, SOCK_STREAM, 0);
872 if (ret < 0)
873 return -ERRNO_TO_PARA_ERROR(errno);
874 fd = ret;
875 ret = mark_fd_nonblocking(fd);
876 if (ret < 0)
877 goto err;
878 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
879 if (ret < 0) {
880 ret = -ERRNO_TO_PARA_ERROR(errno);
881 goto err;
882 }
883 if (!abstract) {
884 ret = -E_CHMOD;
885 if (chmod(name, mode) < 0)
886 goto err;
887 }
888 if (listen(fd , 5) < 0) {
889 ret = -ERRNO_TO_PARA_ERROR(errno);
890 goto err;
891 }
892 return fd;
893 err:
894 close(fd);
895 return ret;
896 }
897
898 /**
899 * Prepare, create, and connect to a Unix domain socket for local communication.
900 *
901 * \param name The socket pathname.
902 *
903 * This function creates a local socket for sequenced, reliable, two-way,
904 * connection-based byte streams.
905 *
906 * \return The file descriptor of the connected socket on success, negative on
907 * errors.
908 *
909 * \sa create_local_socket(), unix(7), connect(2).
910 */
911 int connect_local_socket(const char *name)
912 {
913 struct sockaddr_un unix_addr;
914 int fd, ret;
915
916 PARA_DEBUG_LOG("connecting to %s\n", name);
917 fd = socket(PF_UNIX, SOCK_STREAM, 0);
918 if (fd < 0)
919 return -ERRNO_TO_PARA_ERROR(errno);
920 /* first try (linux-only) abstract socket */
921 ret = init_unix_addr(&unix_addr, name, true);
922 if (ret < 0)
923 goto err;
924 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
925 return fd;
926 /* next try pathname socket */
927 ret = init_unix_addr(&unix_addr, name, false);
928 if (ret < 0)
929 goto err;
930 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
931 return fd;
932 ret = -ERRNO_TO_PARA_ERROR(errno);
933 err:
934 close(fd);
935 return ret;
936 }
937
938 #ifndef HAVE_UCRED
939 ssize_t send_cred_buffer(int sock, char *buf)
940 {
941 return write_buffer(sock, buf);
942 }
943 int recv_cred_buffer(int fd, char *buf, size_t size)
944 {
945 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
946 }
947 #else /* HAVE_UCRED */
948
949 /**
950 * Send a buffer and the credentials of the current process to a socket.
951 *
952 * \param sock The file descriptor of the sending socket.
953 * \param buf The zero-terminated buffer to send.
954 *
955 * \return On success, this call returns the number of bytes sent. On errors,
956 * \p -E_SENDMSG is returned.
957 *
958 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7), okir's Black Hats
959 * Manual.
960 */
961 ssize_t send_cred_buffer(int sock, char *buf)
962 {
963 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
964 struct msghdr msg;
965 struct cmsghdr *cmsg;
966 static struct iovec iov;
967 struct ucred c;
968 int ret;
969
970 /* Response data */
971 iov.iov_base = buf;
972 iov.iov_len = strlen(buf);
973 c.pid = getpid();
974 c.uid = getuid();
975 c.gid = getgid();
976 /* compose the message */
977 memset(&msg, 0, sizeof(msg));
978 msg.msg_iov = &iov;
979 msg.msg_iovlen = 1;
980 msg.msg_control = control;
981 msg.msg_controllen = sizeof(control);
982 /* attach the ucred struct */
983 cmsg = CMSG_FIRSTHDR(&msg);
984 cmsg->cmsg_level = SOL_SOCKET;
985 cmsg->cmsg_type = SCM_CREDENTIALS;
986 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
987 *(struct ucred *)CMSG_DATA(cmsg) = c;
988 msg.msg_controllen = cmsg->cmsg_len;
989 ret = sendmsg(sock, &msg, 0);
990 if (ret < 0)
991 ret = -E_SENDMSG;
992 return ret;
993 }
994
995 static void dispose_fds(int *fds, unsigned num)
996 {
997 int i;
998
999 for (i = 0; i < num; i++)
1000 close(fds[i]);
1001 }
1002
1003 /**
1004 * Receive a buffer and the Unix credentials of the sending process.
1005 *
1006 * \param fd The file descriptor of the receiving socket.
1007 * \param buf The buffer to store the received message.
1008 * \param size The length of \a buf in bytes.
1009 *
1010 * \return Negative on errors, the user id of the sending process on success.
1011 *
1012 * \sa \ref send_cred_buffer and the references given there.
1013 */
1014 int recv_cred_buffer(int fd, char *buf, size_t size)
1015 {
1016 char control[255] __a_aligned(8);
1017 struct msghdr msg;
1018 struct cmsghdr *cmsg;
1019 struct iovec iov;
1020 int result = 0;
1021 int yes = 1;
1022 struct ucred cred;
1023
1024 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1025 memset(&msg, 0, sizeof(msg));
1026 memset(buf, 0, size);
1027 iov.iov_base = buf;
1028 iov.iov_len = size;
1029 msg.msg_iov = &iov;
1030 msg.msg_iovlen = 1;
1031 msg.msg_control = control;
1032 msg.msg_controllen = sizeof(control);
1033 if (recvmsg(fd, &msg, 0) < 0)
1034 return -E_RECVMSG;
1035 result = -E_SCM_CREDENTIALS;
1036 cmsg = CMSG_FIRSTHDR(&msg);
1037 while (cmsg) {
1038 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1039 == SCM_CREDENTIALS) {
1040 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1041 result = cred.uid;
1042 } else
1043 if (cmsg->cmsg_level == SOL_SOCKET
1044 && cmsg->cmsg_type == SCM_RIGHTS) {
1045 dispose_fds((int *)CMSG_DATA(cmsg),
1046 (cmsg->cmsg_len - CMSG_LEN(0))
1047 / sizeof(int));
1048 }
1049 cmsg = CMSG_NXTHDR(&msg, cmsg);
1050 }
1051 return result;
1052 }
1053 #endif /* HAVE_UCRED */