net: Combine para_listen() and para_listen_simple().
[paraslash.git] / net.c
1 /* Copyright (C) 2005 Andre Noll <maan@tuebingen.mpg.de>, see file COPYING. */
2
3 /** \file net.c Networking-related helper functions. */
4
5 #include "para.h"
6
7 #include <netinet/in.h>
8 #include <arpa/inet.h>
9 #include <sys/un.h>
10 #include <sys/types.h>
11 #include <sys/socket.h>
12 #include <netdb.h>
13
14 /* At least NetBSD needs these. */
15 #ifndef AI_V4MAPPED
16 #define AI_V4MAPPED 0
17 #endif
18 #ifndef AI_ALL
19 #define AI_ALL 0
20 #endif
21 #ifndef AI_ADDRCONFIG
22 #define AI_ADDRCONFIG 0
23 #endif
24
25 #include <regex.h>
26
27 #include "error.h"
28 #include "net.h"
29 #include "string.h"
30 #include "list.h"
31 #include "fd.h"
32
33 /**
34 * Parse and validate IPv4 address/netmask string.
35 *
36 * \param cidr Address in CIDR notation
37 * \param addr Copy of the IPv4 address part of \a cidr
38 * \param addrlen Size of \a addr in bytes
39 * \param netmask Value of the netmask part in \a cidr or the
40 * default of 32 if not specified.
41 *
42 * \return Pointer to \a addr if successful, NULL on error.
43 * \sa RFC 4632.
44 */
45 char *parse_cidr(const char *cidr,
46 char *addr, ssize_t addrlen,
47 int32_t *netmask)
48 {
49 const char *o = cidr;
50 char *c = addr, *end = c + (addrlen - 1);
51
52 *netmask = 0x20;
53
54 if (cidr == NULL || addrlen < 1)
55 goto failed;
56
57 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
58 if (c == end)
59 goto failed;
60
61 if (*o == '/')
62 if (para_atoi32(++o, netmask) < 0 ||
63 *netmask < 0 || *netmask > 0x20)
64 goto failed;
65
66 if (is_valid_ipv4_address(addr))
67 return addr;
68 failed:
69 *addr = '\0';
70 return NULL;
71 }
72
73
74 /**
75 * Match string as a candidate IPv4 address.
76 *
77 * \param address The string to match.
78 * \return True if \a address has "dot-quad" format.
79 */
80 static bool is_v4_dot_quad(const char *address)
81 {
82 bool result;
83 regex_t r;
84
85 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
86 REG_EXTENDED | REG_NOSUB) >= 0);
87 result = regexec(&r, address, 0, NULL, 0) == 0;
88 regfree(&r);
89 return result;
90 }
91
92 /**
93 * Perform basic syntax checking on the host-part of an URL:
94 *
95 * - Since ':' is invalid in IPv4 addresses and DNS names, the
96 * presence of ':' causes interpretation as IPv6 address;
97 * - next the first-match-wins algorithm from RFC 3986 is applied;
98 * - else the string is considered as DNS name, to be resolved later.
99 *
100 * \param host The host string to check.
101 * \return True if \a host passes the syntax checks.
102 *
103 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5.
104 */
105 static bool host_string_ok(const char *host)
106 {
107 if (host == NULL || *host == '\0')
108 return false;
109 if (strchr(host, ':') != NULL)
110 return is_valid_ipv6_address(host);
111 if (is_v4_dot_quad(host))
112 return is_valid_ipv4_address(host);
113 return true;
114 }
115
116 /**
117 * Parse and validate URL string.
118 *
119 * The URL syntax is loosely based on RFC 3986, supporting one of
120 * - "["host"]"[:port] for native IPv6 addresses and
121 * - host[:port] for IPv4 hostnames and DNS names.
122 *
123 * Native IPv6 addresses must be enclosed in square brackets, since
124 * otherwise there is an ambiguity with the port separator `:'.
125 * The 'port' part is always considered to be a number; if absent,
126 * it is set to -1, to indicate that a default port is to be used.
127 *
128 * The following are valid examples:
129 * - 10.10.1.1
130 * - 10.10.1.2:8000
131 * - localhost
132 * - localhost:8001
133 * - [::1]:8000
134 * - [badc0de::1]
135 *
136 * \param url The URL string to take apart.
137 * \param host To return the copied host part of \a url.
138 * \param hostlen The maximum length of \a host.
139 * \param port To return the port number (if any) of \a url.
140 *
141 * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned,
142 * \a host and \a port are undefined. If no port number was present in \a url,
143 * \a port is set to -1.
144 *
145 * \sa RFC 3986, 3.2.2/3.2.3.
146 */
147 char *parse_url(const char *url,
148 char *host, ssize_t hostlen,
149 int32_t *port)
150 {
151 const char *o = url;
152 char *c = host, *end = c + (hostlen - 1);
153
154 *port = -1;
155
156 if (o == NULL || hostlen < 1)
157 goto failed;
158
159 if (*o == '[') {
160 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
161 if (c == end)
162 goto failed;
163
164 if (*o++ != ']' || (*o != '\0' && *o != ':'))
165 goto failed;
166 } else {
167 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
168 if (c == end && o[1])
169 goto failed;
170 }
171 }
172
173 if (*o == ':')
174 if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff)
175 goto failed;
176 if (host_string_ok(host))
177 return host;
178 failed:
179 *host = '\0';
180 return NULL;
181 }
182
183 /**
184 * Stringify port number, resolve into service name where defined.
185 *
186 * \param port 2-byte port number, in host-byte-order.
187 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
188 * \return Pointer to static result buffer.
189 *
190 * \sa getservent(3), services(5), nsswitch.conf(5).
191 */
192 const char *stringify_port(int port, const char *transport)
193 {
194 static char service[NI_MAXSERV];
195
196 if (port < 0 || port > 0xFFFF) {
197 snprintf(service, sizeof(service), "undefined (%d)", port);
198 } else {
199 struct servent *se = getservbyport(htons(port), transport);
200
201 if (se == NULL)
202 snprintf(service, sizeof(service), "%d", port);
203 else
204 snprintf(service, sizeof(service), "%s", se->s_name);
205 }
206 return service;
207 }
208
209 /**
210 * Determine the socket type for a given layer-4 protocol.
211 *
212 * \param l4type The symbolic name of the transport-layer protocol.
213 *
214 * \sa ip(7), socket(2).
215 */
216 static inline int sock_type(const unsigned l4type)
217 {
218 switch (l4type) {
219 case IPPROTO_UDP: return SOCK_DGRAM;
220 case IPPROTO_TCP: return SOCK_STREAM;
221 case IPPROTO_DCCP: return SOCK_DCCP;
222 }
223 return -1; /* not supported here */
224 }
225
226 /**
227 * Pretty-print transport-layer name.
228 */
229 static const char *layer4_name(const unsigned l4type)
230 {
231 switch (l4type) {
232 case IPPROTO_UDP: return "UDP";
233 case IPPROTO_TCP: return "TCP";
234 case IPPROTO_DCCP: return "DCCP";
235 }
236 return "UNKNOWN PROTOCOL";
237 }
238
239 /**
240 * Flowopts: Transport-layer independent encapsulation of socket options.
241 *
242 * These collect individual socket options into a queue, which is disposed of
243 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
244 * only and should not be visible elsewhere.
245 *
246 * \sa setsockopt(2), \ref makesock().
247 */
248 struct pre_conn_opt {
249 int sock_level; /**< Second argument to setsockopt() */
250 int sock_option; /**< Third argument to setsockopt() */
251 char *opt_name; /**< Stringified \a sock_option */
252 void *opt_val; /**< Fourth argument to setsockopt() */
253 socklen_t opt_len; /**< Fifth argument to setsockopt() */
254
255 struct list_head node; /**< FIFO, as sockopt order matters. */
256 };
257
258 /** FIFO list of pre-connection socket options to be set */
259 struct flowopts {
260 struct list_head sockopts;
261 };
262
263 /**
264 * Allocate and initialize a flowopt queue.
265 *
266 * \return A new structure to be passed to \ref flowopt_add(). It is
267 * automatically deallocated in \ref makesock().
268 */
269 struct flowopts *flowopt_new(void)
270 {
271 struct flowopts *new = para_malloc(sizeof(*new));
272
273 INIT_LIST_HEAD(&new->sockopts);
274 return new;
275 }
276
277 /**
278 * Append new socket option to flowopt queue.
279 *
280 * \param fo The flowopt queue to append to.
281 * \param lev Level at which \a opt resides.
282 * \param opt New option to add.
283 * \param name Stringified name of \a opt.
284 * \param val The value to set \a opt to.
285 * \param len Length of \a val.
286 *
287 * \sa setsockopt(2).
288 */
289 void flowopt_add(struct flowopts *fo, int lev, int opt,
290 const char *name, const void *val, int len)
291 {
292 struct pre_conn_opt *new = para_malloc(sizeof(*new));
293
294 new->sock_option = opt;
295 new->sock_level = lev;
296 new->opt_name = para_strdup(name);
297
298 if (val == NULL) {
299 new->opt_val = NULL;
300 new->opt_len = 0;
301 } else {
302 new->opt_val = para_malloc(len);
303 new->opt_len = len;
304 memcpy(new->opt_val, val, len);
305 }
306
307 list_add_tail(&new->node, &fo->sockopts);
308 }
309
310 /** Set the entire bunch of pre-connection options at once. */
311 static void flowopt_setopts(int sockfd, struct flowopts *fo)
312 {
313 struct pre_conn_opt *pc;
314
315 if (fo == NULL)
316 return;
317
318 list_for_each_entry(pc, &fo->sockopts, node)
319 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
320 pc->opt_val, pc->opt_len) < 0) {
321 PARA_EMERG_LOG("Can not set %s socket option: %s",
322 pc->opt_name, strerror(errno));
323 exit(EXIT_FAILURE);
324 }
325 }
326
327 /**
328 * Deallocate all resources of a flowopts structure.
329 *
330 * \param fo A pointer as returned from flowopt_new().
331 *
332 * It's OK to pass \p NULL here in which case the function does nothing.
333 */
334 void flowopt_cleanup(struct flowopts *fo)
335 {
336 struct pre_conn_opt *cur, *next;
337
338 if (fo == NULL)
339 return;
340
341 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
342 free(cur->opt_name);
343 free(cur->opt_val);
344 free(cur);
345 }
346 free(fo);
347 }
348
349 /**
350 * Resolve an IPv4/IPv6 address.
351 *
352 * \param l4type The layer-4 type (\p IPPROTO_xxx).
353 * \param passive Whether \p AI_PASSIVE should be included as hint.
354 * \param host Remote or local hostname or IPv/6 address string.
355 * \param port_number Used to set the port in each returned address structure.
356 * \param result addrinfo structures are returned here.
357 *
358 * The interpretation of \a host depends on the value of \a passive. On a
359 * passive socket host is interpreted as an interface IPv4/6 address (can be
360 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
361 * address to connect to.
362 *
363 * \return Standard.
364 *
365 * \sa getaddrinfo(3).
366 */
367 int lookup_address(unsigned l4type, bool passive, const char *host,
368 int port_number, struct addrinfo **result)
369 {
370 int ret;
371 char port[6]; /* port number has at most 5 digits */
372 struct addrinfo *addr = NULL, hints;
373
374 *result = NULL;
375 sprintf(port, "%d", port_number & 0xffff);
376 /* Set up address hint structure */
377 memset(&hints, 0, sizeof(hints));
378 hints.ai_family = AF_UNSPEC;
379 hints.ai_socktype = sock_type(l4type);
380 /*
381 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
382 * (and only then) pretend to be UDP.
383 */
384 if (l4type == IPPROTO_DCCP)
385 hints.ai_socktype = SOCK_DGRAM;
386 /* only use addresses available on the host */
387 hints.ai_flags = AI_ADDRCONFIG;
388 if (passive && host == NULL)
389 hints.ai_flags |= AI_PASSIVE;
390 /* Obtain local/remote address information */
391 ret = getaddrinfo(host, port, &hints, &addr);
392 if (ret != 0) {
393 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
394 layer4_name(l4type),
395 host? host : (passive? "[loopback]" : "[localhost]"),
396 port, gai_strerror(ret));
397 return -E_ADDRESS_LOOKUP;
398 }
399 *result = addr;
400 return 1;
401 }
402
403 /**
404 * Create an active or passive socket.
405 *
406 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
407 * \param passive Whether to call bind(2) or connect(2).
408 * \param ai Address information as obtained from \ref lookup_address().
409 * \param fo Socket options to be set before making the connection.
410 *
411 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
412 * algorithm tries all possible address combinations until it succeeds. If \a
413 * fo is supplied, options are set but cleanup must be performed in the caller.
414 *
415 * \return File descriptor on success, \p E_MAKESOCK on errors.
416 *
417 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
418 * connect(2).
419 */
420 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
421 struct flowopts *fo)
422 {
423 int ret = -E_MAKESOCK, on = 1;
424
425 for (; ai; ai = ai->ai_next) {
426 int fd;
427 ret = socket(ai->ai_family, sock_type(l4type), l4type);
428 if (ret < 0) {
429 PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
430 continue;
431 }
432 fd = ret;
433 flowopt_setopts(fd, fo);
434 if (!passive) {
435 if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
436 PARA_NOTICE_LOG("connect(): %s\n",
437 strerror(errno));
438 close(fd);
439 continue;
440 }
441 return fd;
442 }
443 /*
444 * Reuse the address on passive sockets to avoid failure on
445 * restart (protocols using listen()) and when creating
446 * multiple listener instances (UDP multicast).
447 */
448 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
449 sizeof(on)) == -1) {
450 PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
451 close(fd);
452 continue;
453 }
454 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
455 PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
456 close(fd);
457 continue;
458 }
459 return fd;
460 }
461 return -E_MAKESOCK;
462 }
463
464 /**
465 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
466 *
467 * \param l4type The layer-4 type (\p IPPROTO_xxx).
468 * \param passive Whether this is a passive or active socket.
469 * \param host Passed to \ref lookup_address().
470 * \param port_number Passed to \ref lookup_address().
471 * \param fo Passed to \ref makesock_addrinfo().
472 *
473 * This creates a ready-made IPv4/v6 socket structure after looking up the
474 * necessary parameters. The function first calls \ref lookup_address() and
475 * passes the address information to makesock_addrinfo() to create and
476 * initialize the socket.
477 *
478 * \return The newly created file descriptor on success, a negative error code
479 * on failure.
480 *
481 * \sa \ref lookup_address(), \ref makesock_addrinfo().
482 */
483 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
484 struct flowopts *fo)
485 {
486 struct addrinfo *ai;
487 int ret = lookup_address(l4type, passive, host, port_number, &ai);
488
489 if (ret >= 0)
490 ret = makesock_addrinfo(l4type, passive, ai, fo);
491 if (ai)
492 freeaddrinfo(ai);
493 if (ret < 0) {
494 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
495 layer4_name(l4type), host? host : (passive?
496 "[loopback]" : "[localhost]"), port_number);
497 }
498 return ret;
499 }
500
501 /**
502 * Create a passive / listening socket.
503 *
504 * \param l4type The transport-layer type (\p IPPROTO_xxx).
505 * \param port The decimal port number to listen on.
506 *
507 * \return Positive integer (socket descriptor) on success, negative value
508 * otherwise.
509 *
510 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
511 */
512 int para_listen_simple(unsigned l4type, uint16_t port)
513 {
514 int ret, fd = makesock(l4type, 1, NULL, port, NULL);
515
516 if (fd > 0) {
517 ret = listen(fd, BACKLOG);
518 if (ret < 0) {
519 ret = errno;
520 close(fd);
521 return -ERRNO_TO_PARA_ERROR(ret);
522 }
523 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
524 layer4_name(l4type), port, fd);
525 }
526 return fd;
527 }
528
529 /**
530 * Determine IPv4/v6 socket address length.
531 * \param sa Container of IPv4 or IPv6 address.
532 * \return Address-family dependent address length.
533 */
534 static socklen_t salen(const struct sockaddr *sa)
535 {
536 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
537
538 return sa->sa_family == AF_INET6
539 ? sizeof(struct sockaddr_in6)
540 : sizeof(struct sockaddr_in);
541 }
542
543 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
544 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
545 {
546 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
547
548 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
549 }
550
551 /**
552 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
553 * \param ss Container of IPv4/6 address.
554 * \return Pointer to normalized address (may be static storage).
555 *
556 * \sa RFC 3493.
557 */
558 static const struct sockaddr *
559 normalize_ip_address(const struct sockaddr_storage *ss)
560 {
561 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
562
563 if (SS_IS_ADDR_V4MAPPED(ss)) {
564 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
565 static struct sockaddr_in ia;
566
567 ia.sin_family = AF_INET;
568 ia.sin_port = ia6->sin6_port;
569 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
570 return (const struct sockaddr *)&ia;
571 }
572 return (const struct sockaddr *)ss;
573 }
574
575 /**
576 * Generic/fallback MTU values
577 *
578 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
579 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
580 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
581 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
582 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
583 * UDP senders should use the respective minimum values of EMTU_S.
584 */
585 static inline int generic_mtu(const int af_type)
586 {
587 return af_type == AF_INET6 ? 1280 : 576;
588 }
589
590 /** Crude approximation of IP header overhead - neglecting options. */
591 static inline int estimated_header_overhead(const int af_type)
592 {
593 return af_type == AF_INET6 ? 40 : 20;
594 }
595
596 /**
597 * Get the maximum transport-layer message size (MMS_S).
598 *
599 * \param sockfd The socket file descriptor.
600 *
601 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
602 * could not be determined, \p AF_INET is assumed.
603 *
604 * \return The maximum message size of the address family type.
605 */
606 int generic_max_transport_msg_size(int sockfd)
607 {
608 struct sockaddr_storage ss = {.ss_family = 0};
609 socklen_t sslen = sizeof(ss);
610 int af_type = AF_INET;
611
612 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
613 PARA_ERROR_LOG("can not determine remote address type: %s\n",
614 strerror(errno));
615 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
616 af_type = ss.ss_family;
617 }
618 return generic_mtu(af_type) - estimated_header_overhead(af_type);
619 }
620
621 /**
622 * Look up the remote side of a connected socket structure.
623 *
624 * \param fd The socket descriptor of the connected socket.
625 *
626 * \return A static character string identifying hostname and port of the
627 * chosen side in numeric host:port format.
628 *
629 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
630 * services(5), nsswitch.conf(5).
631 */
632 char *remote_name(int fd)
633 {
634 struct sockaddr_storage ss = {.ss_family = 0};
635 const struct sockaddr *sa;
636 socklen_t sslen = sizeof(ss);
637 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
638 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
639 int ret;
640
641 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
642 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
643 fd, strerror(errno));
644 snprintf(output, sizeof(output), "(unknown)");
645 return output;
646 }
647 sa = normalize_ip_address(&ss);
648 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
649 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
650 if (ret) {
651 PARA_WARNING_LOG("hostname lookup error (%s).\n",
652 gai_strerror(ret));
653 snprintf(output, sizeof(output), "(lookup error)");
654 } else if (sa->sa_family == AF_INET6)
655 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
656 else
657 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
658 return output;
659 }
660
661 /**
662 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
663 *
664 * \param ss Container of IPv4/6 address.
665 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
666 *
667 * \sa RFC 3493.
668 */
669 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
670 {
671 const struct sockaddr *sa = normalize_ip_address(ss);
672
673 memset(ia, 0, sizeof(*ia));
674 if (sa->sa_family == AF_INET)
675 *ia = ((struct sockaddr_in *)sa)->sin_addr;
676 }
677
678 /**
679 * Compare the address part of IPv4/6 addresses.
680 *
681 * \param sa1 First address.
682 * \param sa2 Second address.
683 *
684 * \return True iff the IP address of \a sa1 and \a sa2 match.
685 */
686 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
687 {
688 if (!sa1 || !sa2)
689 return false;
690 if (sa1->sa_family != sa2->sa_family)
691 return false;
692 if (sa1->sa_family == AF_INET) {
693 struct sockaddr_in *a1 = (typeof(a1))sa1,
694 *a2 = (typeof (a2))sa2;
695 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
696 } else if (sa1->sa_family == AF_INET6) {
697 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
698 *a2 = (typeof (a2))sa2;
699 return !memcmp(a1, a2, sizeof(*a1));
700 } else
701 return false;
702 }
703
704 /**
705 * Receive data from a file descriptor.
706 *
707 * \param fd The file descriptor.
708 * \param buf The buffer to write the data to.
709 * \param size The size of \a buf.
710 *
711 * Receive at most \a size bytes from file descriptor \a fd.
712 *
713 * \return The number of bytes received on success, negative on errors, zero if
714 * the peer has performed an orderly shutdown.
715 *
716 * \sa recv(2).
717 */
718 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
719 {
720 ssize_t n;
721
722 n = recv(fd, buf, size, 0);
723 if (n == -1)
724 return -ERRNO_TO_PARA_ERROR(errno);
725 return n;
726 }
727
728 /**
729 * Receive and write terminating NULL byte.
730 *
731 * \param fd The file descriptor.
732 * \param buf The buffer to write the data to.
733 * \param size The size of \a buf.
734 *
735 * Read at most \a size - 1 bytes from file descriptor \a fd and
736 * write a NULL byte at the end of the received data.
737 *
738 * \return The return value of the underlying call to \a recv_bin_buffer().
739 *
740 * \sa \ref recv_bin_buffer()
741 */
742 int recv_buffer(int fd, char *buf, size_t size)
743 {
744 int n;
745
746 assert(size);
747 n = recv_bin_buffer(fd, buf, size - 1);
748 if (n >= 0)
749 buf[n] = '\0';
750 else
751 *buf = '\0';
752 return n;
753 }
754
755 /**
756 * Wrapper around the accept system call.
757 *
758 * \param fd The listening socket.
759 * \param rfds An optional fd_set pointer.
760 * \param addr Structure which is filled in with the address of the peer socket.
761 * \param size Should contain the size of the structure pointed to by \a addr.
762 * \param new_fd Result pointer.
763 *
764 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
765 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
766 *
767 * \return Negative on errors, zero if no connections are present to be accepted,
768 * one otherwise.
769 *
770 * \sa accept(2).
771 */
772 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
773 {
774 int ret;
775
776 if (rfds && !FD_ISSET(fd, rfds))
777 return 0;
778 do
779 ret = accept(fd, (struct sockaddr *) addr, &size);
780 while (ret < 0 && errno == EINTR);
781
782 if (ret >= 0) {
783 *new_fd = ret;
784 return 1;
785 }
786 if (errno == EAGAIN || errno == EWOULDBLOCK)
787 return 0;
788 return -ERRNO_TO_PARA_ERROR(errno);
789 }
790
791 /**
792 * Probe the list of DCCP CCIDs configured on this host.
793 * \param ccid_array Pointer to return statically allocated array in.
794 * \return Number of elements returned in \a ccid_array or error.
795 *
796 * NB: This feature is only available on Linux > 2.6.30; on older kernels
797 * ENOPROTOOPT ("Protocol not available") will be returned.
798 */
799 int dccp_available_ccids(uint8_t **ccid_array)
800 {
801 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
802 socklen_t nccids = sizeof(ccids);
803 int ret, fd;
804
805 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
806 if (ret < 0)
807 return ret;
808
809 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
810 ccids, &nccids) < 0) {
811 ret = errno;
812 close(fd);
813 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
814 strerror(ret));
815 return -ERRNO_TO_PARA_ERROR(ret);
816 }
817
818 close(fd);
819 *ccid_array = ccids;
820 return nccids;
821 }
822
823 /*
824 * Prepare a structure for AF_UNIX socket addresses.
825 *
826 * This just copies name to the sun_path component of u, prepending a zero byte
827 * if abstract sockets are supported.
828 *
829 * The first call to this function tries to bind a socket to the abstract name
830 * space. The result of this test is stored in a static variable. Subsequent
831 * calls read this variable and create abstract sockets on systems that support
832 * them.
833 */
834 static int init_unix_addr(struct sockaddr_un *u, const char *name)
835 {
836 static int use_abstract;
837
838 if (strlen(name) + 1 >= UNIX_PATH_MAX)
839 return -E_NAME_TOO_LONG;
840 memset(u->sun_path, 0, UNIX_PATH_MAX);
841 u->sun_family = PF_UNIX;
842 if (use_abstract == 0) { /* executed only once */
843 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
844 memcpy(u->sun_path, "\0x\0", 3);
845 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
846 use_abstract = 1; /* yes */
847 else
848 use_abstract = -1; /* no */
849 close(fd);
850 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
851 use_abstract == 1? "" : "not ");
852 }
853 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
854 return 1;
855 }
856
857 /**
858 * Create a socket for local communication and listen on it.
859 *
860 * \param name The socket pathname.
861 *
862 * This function creates a passive local socket for sequenced, reliable,
863 * two-way, connection-based byte streams. The socket file descriptor is set to
864 * nonblocking mode and listen(2) is called to prepare the socket for
865 * accepting incoming connection requests.
866 *
867 * \return The file descriptor on success, negative error code on failure.
868 *
869 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
870 */
871 int create_local_socket(const char *name)
872 {
873 struct sockaddr_un unix_addr;
874 int fd, ret;
875
876 ret = init_unix_addr(&unix_addr, name);
877 if (ret < 0)
878 return ret;
879 ret = socket(PF_UNIX, SOCK_STREAM, 0);
880 if (ret < 0)
881 return -ERRNO_TO_PARA_ERROR(errno);
882 fd = ret;
883 ret = mark_fd_nonblocking(fd);
884 if (ret < 0)
885 goto err;
886 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
887 if (ret < 0) {
888 ret = -ERRNO_TO_PARA_ERROR(errno);
889 goto err;
890 }
891 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
892 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
893 | S_IROTH | S_IWOTH;
894 ret = -E_CHMOD;
895 if (chmod(name, mode) < 0)
896 goto err;
897 }
898 if (listen(fd , 5) < 0) {
899 ret = -ERRNO_TO_PARA_ERROR(errno);
900 goto err;
901 }
902 return fd;
903 err:
904 close(fd);
905 return ret;
906 }
907
908 /**
909 * Prepare, create, and connect to a Unix domain socket for local communication.
910 *
911 * \param name The socket pathname.
912 *
913 * This function creates a local socket for sequenced, reliable, two-way,
914 * connection-based byte streams.
915 *
916 * \return The file descriptor of the connected socket on success, negative on
917 * errors.
918 *
919 * \sa \ref create_local_socket(), unix(7), connect(2).
920 */
921 int connect_local_socket(const char *name)
922 {
923 struct sockaddr_un unix_addr;
924 int fd, ret;
925
926 PARA_DEBUG_LOG("connecting to %s\n", name);
927 fd = socket(PF_UNIX, SOCK_STREAM, 0);
928 if (fd < 0)
929 return -ERRNO_TO_PARA_ERROR(errno);
930 ret = init_unix_addr(&unix_addr, name);
931 if (ret < 0)
932 goto err;
933 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
934 return fd;
935 ret = -ERRNO_TO_PARA_ERROR(errno);
936 err:
937 close(fd);
938 return ret;
939 }
940
941 #ifndef HAVE_UCRED
942 ssize_t send_cred_buffer(int sock, char *buf)
943 {
944 return write_buffer(sock, buf);
945 }
946 int recv_cred_buffer(int fd, char *buf, size_t size)
947 {
948 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
949 }
950 #else /* HAVE_UCRED */
951
952 /**
953 * Send a buffer and the credentials of the current process to a socket.
954 *
955 * \param sock The file descriptor of the sending socket.
956 * \param buf The zero-terminated buffer to send.
957 *
958 * \return On success, this call returns the number of bytes sent. On errors,
959 * \p -E_SENDMSG is returned.
960 *
961 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
962 */
963 ssize_t send_cred_buffer(int sock, char *buf)
964 {
965 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
966 struct msghdr msg;
967 struct cmsghdr *cmsg;
968 static struct iovec iov;
969 struct ucred c;
970 int ret;
971
972 /* Response data */
973 iov.iov_base = buf;
974 iov.iov_len = strlen(buf);
975 c.pid = getpid();
976 c.uid = getuid();
977 c.gid = getgid();
978 /* compose the message */
979 memset(&msg, 0, sizeof(msg));
980 msg.msg_iov = &iov;
981 msg.msg_iovlen = 1;
982 msg.msg_control = control;
983 msg.msg_controllen = sizeof(control);
984 /* attach the ucred struct */
985 cmsg = CMSG_FIRSTHDR(&msg);
986 cmsg->cmsg_level = SOL_SOCKET;
987 cmsg->cmsg_type = SCM_CREDENTIALS;
988 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
989 *(struct ucred *)CMSG_DATA(cmsg) = c;
990 msg.msg_controllen = cmsg->cmsg_len;
991 ret = sendmsg(sock, &msg, 0);
992 if (ret < 0)
993 ret = -E_SENDMSG;
994 return ret;
995 }
996
997 static void dispose_fds(int *fds, unsigned num)
998 {
999 int i;
1000
1001 for (i = 0; i < num; i++)
1002 close(fds[i]);
1003 }
1004
1005 /**
1006 * Receive a buffer and the Unix credentials of the sending process.
1007 *
1008 * \param fd The file descriptor of the receiving socket.
1009 * \param buf The buffer to store the received message.
1010 * \param size The length of \a buf in bytes.
1011 *
1012 * \return Negative on errors, the user id of the sending process on success.
1013 *
1014 * \sa \ref send_cred_buffer and the references given there.
1015 */
1016 int recv_cred_buffer(int fd, char *buf, size_t size)
1017 {
1018 char control[255] __a_aligned(8);
1019 struct msghdr msg;
1020 struct cmsghdr *cmsg;
1021 struct iovec iov;
1022 int result = 0;
1023 int yes = 1;
1024 struct ucred cred;
1025
1026 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1027 memset(&msg, 0, sizeof(msg));
1028 memset(buf, 0, size);
1029 iov.iov_base = buf;
1030 iov.iov_len = size;
1031 msg.msg_iov = &iov;
1032 msg.msg_iovlen = 1;
1033 msg.msg_control = control;
1034 msg.msg_controllen = sizeof(control);
1035 if (recvmsg(fd, &msg, 0) < 0)
1036 return -E_RECVMSG;
1037 result = -E_SCM_CREDENTIALS;
1038 cmsg = CMSG_FIRSTHDR(&msg);
1039 while (cmsg) {
1040 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1041 == SCM_CREDENTIALS) {
1042 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1043 result = cred.uid;
1044 } else
1045 if (cmsg->cmsg_level == SOL_SOCKET
1046 && cmsg->cmsg_type == SCM_RIGHTS) {
1047 dispose_fds((int *)CMSG_DATA(cmsg),
1048 (cmsg->cmsg_len - CMSG_LEN(0))
1049 / sizeof(int));
1050 }
1051 cmsg = CMSG_NXTHDR(&msg, cmsg);
1052 }
1053 return result;
1054 }
1055 #endif /* HAVE_UCRED */