Change copyright year to 2014.
[paraslash.git] / net.c
1 /*
2 * Copyright (C) 2005-2014 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file net.c Networking-related helper functions. */
8
9 /*
10 * Since glibc 2.8, the _GNU_SOURCE feature test macro must be defined in order
11 * to obtain the definition of the ucred structure.
12 */
13 #define _GNU_SOURCE
14
15 #include <netdb.h>
16
17 /* At least NetBSD needs these. */
18 #ifndef AI_V4MAPPED
19 #define AI_V4MAPPED 0
20 #endif
21 #ifndef AI_ALL
22 #define AI_ALL 0
23 #endif
24 #ifndef AI_ADDRCONFIG
25 #define AI_ADDRCONFIG 0
26 #endif
27
28 #include <regex.h>
29
30 #include "para.h"
31 #include "error.h"
32 #include "net.h"
33 #include "string.h"
34 #include "list.h"
35 #include "fd.h"
36
37 /**
38 * Parse and validate IPv4 address/netmask string.
39 *
40 * \param cidr Address in CIDR notation
41 * \param addr Copy of the IPv4 address part of \a cidr
42 * \param addrlen Size of \a addr in bytes
43 * \param netmask Value of the netmask part in \a cidr or the
44 * default of 32 if not specified.
45 *
46 * \return Pointer to \a addr if succesful, NULL on error.
47 * \sa RFC 4632
48 */
49 char *parse_cidr(const char *cidr,
50 char *addr, ssize_t addrlen,
51 int32_t *netmask)
52 {
53 const char *o = cidr;
54 char *c = addr, *end = c + (addrlen - 1);
55
56 *netmask = 0x20;
57
58 if (cidr == NULL || addrlen < 1)
59 goto failed;
60
61 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
62 if (c == end)
63 goto failed;
64
65 if (*o == '/')
66 if (para_atoi32(++o, netmask) < 0 ||
67 *netmask < 0 || *netmask > 0x20)
68 goto failed;
69
70 if (is_valid_ipv4_address(addr))
71 return addr;
72 failed:
73 *addr = '\0';
74 return NULL;
75 }
76
77
78 /**
79 * Match string as a candidate IPv4 address.
80 *
81 * \param address The string to match.
82 * \return True if \a address has "dot-quad" format.
83 */
84 static bool is_v4_dot_quad(const char *address)
85 {
86 bool result;
87 regex_t r;
88
89 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
90 REG_EXTENDED | REG_NOSUB) >= 0);
91 result = regexec(&r, address, 0, NULL, 0) == 0;
92 regfree(&r);
93 return result;
94 }
95
96 /**
97 * Perform basic syntax checking on the host-part of an URL:
98 *
99 * - Since ':' is invalid in IPv4 addresses and DNS names, the
100 * presence of ':' causes interpretation as IPv6 address;
101 * - next the first-match-wins algorithm from RFC 3986 is applied;
102 * - else the string is considered as DNS name, to be resolved later.
103 *
104 * \param host The host string to check.
105 * \return True if \a host passes the syntax checks.
106 *
107 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5
108 */
109 static bool host_string_ok(const char *host)
110 {
111 if (host == NULL || *host == '\0')
112 return false;
113 if (strchr(host, ':') != NULL)
114 return is_valid_ipv6_address(host);
115 if (is_v4_dot_quad(host))
116 return is_valid_ipv4_address(host);
117 return true;
118 }
119
120 /**
121 * Parse and validate URL string.
122 *
123 * The URL syntax is loosely based on RFC 3986, supporting one of
124 * - "["host"]"[:port] for native IPv6 addresses and
125 * - host[:port] for IPv4 hostnames and DNS names.
126 *
127 * Native IPv6 addresses must be enclosed in square brackets, since
128 * otherwise there is an ambiguity with the port separator `:'.
129 * The 'port' part is always considered to be a number; if absent,
130 * it is set to -1, to indicate that a default port is to be used.
131 *
132 * The following are valid examples:
133 * - 10.10.1.1
134 * - 10.10.1.2:8000
135 * - localhost
136 * - localhost:8001
137 * - [::1]:8000
138 * - [badc0de::1]
139 *
140 * \param url The URL string to take apart.
141 * \param host To return the copied host part of \a url.
142 * \param hostlen The maximum length of \a host.
143 * \param port To return the port number (if any) of \a url.
144 *
145 * \return Pointer to \a host, or NULL if failed.
146 * If NULL is returned, \a host and \a portnum are undefined. If no
147 * port number was present in \a url, \a portnum 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)
173 goto failed;
174 }
175
176 if (*o == ':')
177 if (para_atoi32(++o, port) < 0 ||
178 *port < 0 || *port > 0xffff)
179 goto failed;
180
181 if (host_string_ok(host))
182 return host;
183 failed:
184 *host = '\0';
185 return NULL;
186 }
187
188 /**
189 * Stringify port number, resolve into service name where defined.
190 * \param port 2-byte port number, in host-byte-order.
191 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
192 * \return Pointer to static result buffer.
193 *
194 * \sa getservent(3), services(5), nsswitch.conf(5)
195 */
196 const char *stringify_port(int port, const char *transport)
197 {
198 static char service[NI_MAXSERV];
199
200 if (port < 0 || port > 0xFFFF) {
201 snprintf(service, sizeof(service), "undefined (%d)", port);
202 } else {
203 struct servent *se = getservbyport(htons(port), transport);
204
205 if (se == NULL)
206 snprintf(service, sizeof(service), "%u", port);
207 else
208 snprintf(service, sizeof(service), "%s", se->s_name);
209 }
210 return service;
211 }
212
213 /**
214 * Determine the socket type for a given layer-4 protocol.
215 *
216 * \param l4type The symbolic name of the transport-layer protocol.
217 *
218 * \sa ip(7), socket(2)
219 */
220 static inline int sock_type(const unsigned l4type)
221 {
222 switch (l4type) {
223 case IPPROTO_UDP: return SOCK_DGRAM;
224 case IPPROTO_TCP: return SOCK_STREAM;
225 case IPPROTO_DCCP: return SOCK_DCCP;
226 }
227 return -1; /* not supported here */
228 }
229
230 /**
231 * Pretty-print transport-layer name.
232 */
233 static const char *layer4_name(const unsigned l4type)
234 {
235 switch (l4type) {
236 case IPPROTO_UDP: return "UDP";
237 case IPPROTO_TCP: return "TCP";
238 case IPPROTO_DCCP: return "DCCP";
239 }
240 return "UNKNOWN PROTOCOL";
241 }
242
243 /**
244 * Flowopts: Transport-layer independent encapsulation of socket options.
245 *
246 * These collect individual socket options into a queue, which is disposed of
247 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
248 * only and should not be visible elsewhere.
249 *
250 * \sa setsockopt(2), makesock()
251 */
252 struct pre_conn_opt {
253 int sock_level; /**< Second argument to setsockopt() */
254 int sock_option; /**< Third argument to setsockopt() */
255 char *opt_name; /**< Stringified \a sock_option */
256 void *opt_val; /**< Fourth argument to setsockopt() */
257 socklen_t opt_len; /**< Fifth argument to setsockopt() */
258
259 struct list_head node; /**< FIFO, as sockopt order matters. */
260 };
261
262 /** FIFO list of pre-connection socket options to be set */
263 struct flowopts {
264 struct list_head sockopts;
265 };
266
267 /**
268 * Allocate and initialize a flowopt queue.
269 *
270 * \return A new structure to be passed to \ref flowopt_add(). It is
271 * automatically deallocated in \ref makesock().
272 */
273 struct flowopts *flowopt_new(void)
274 {
275 struct flowopts *new = para_malloc(sizeof(*new));
276
277 INIT_LIST_HEAD(&new->sockopts);
278 return new;
279 }
280
281 /**
282 * Append new socket option to flowopt queue.
283 *
284 * \param fo The flowopt queue to append to.
285 * \param lev Level at which \a opt resides.
286 * \param opt New option to add.
287 * \param name Stringified name of \a opt.
288 * \param val The value to set \a opt to.
289 * \param len Length of \a val.
290 *
291 * \sa setsockopt(2)
292 */
293 void flowopt_add(struct flowopts *fo, int lev, int opt,
294 const char *name, const void *val, int len)
295 {
296 struct pre_conn_opt *new = para_malloc(sizeof(*new));
297
298 new->sock_option = opt;
299 new->sock_level = lev;
300 new->opt_name = para_strdup(name);
301
302 if (val == NULL) {
303 new->opt_val = NULL;
304 new->opt_len = 0;
305 } else {
306 new->opt_val = para_malloc(len);
307 new->opt_len = len;
308 memcpy(new->opt_val, val, len);
309 }
310
311 list_add_tail(&new->node, &fo->sockopts);
312 }
313
314 /** Set the entire bunch of pre-connection options at once. */
315 static void flowopt_setopts(int sockfd, struct flowopts *fo)
316 {
317 struct pre_conn_opt *pc;
318
319 if (fo == NULL)
320 return;
321
322 list_for_each_entry(pc, &fo->sockopts, node)
323 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
324 pc->opt_val, pc->opt_len) < 0) {
325 PARA_EMERG_LOG("Can not set %s socket option: %s",
326 pc->opt_name, strerror(errno));
327 exit(EXIT_FAILURE);
328 }
329 }
330
331 static void flowopt_cleanup(struct flowopts *fo)
332 {
333 struct pre_conn_opt *cur, *next;
334
335 if (fo == NULL)
336 return;
337
338 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
339 free(cur->opt_name);
340 free(cur->opt_val);
341 free(cur);
342 }
343 free(fo);
344 }
345
346 /**
347 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
348 *
349 * \param l4type The layer-4 type (\p IPPROTO_xxx).
350 * \param passive Whether this is a passive (1) or active (0) socket.
351 * \param host Remote or local hostname or IPv/6 address string.
352 * \param port_number Decimal port number.
353 * \param fo Socket options to be set before making the connection.
354 *
355 * This creates a ready-made IPv4/v6 socket structure after looking up the
356 * necessary parameters. The interpretation of \a host depends on the value of
357 * \a passive:
358 * - on a passive socket host is interpreted as an interface IPv4/6 address
359 * (can be left NULL);
360 * - on an active socket, \a host is the peer DNS name or IPv4/6 address
361 * to connect to;
362 * - \a port_number is in either case the numeric port number (not service
363 * string).
364 *
365 * Furthermore, bind(2) is called on passive sockets, and connect(2) on active
366 * sockets. The algorithm tries all possible address combinations until it
367 * succeeds. If \a fo is supplied, options are set and cleanup is performed.
368 *
369 * \return This function returns 1 on success and \a -E_ADDRESS_LOOKUP when no
370 * matching connection could be set up (with details in the error log).
371 *
372 * \sa ipv6(7), getaddrinfo(3), bind(2), connect(2).
373 */
374 int makesock(unsigned l4type, bool passive,
375 const char *host, uint16_t port_number,
376 struct flowopts *fo)
377 {
378 struct addrinfo *local = NULL, *src = NULL, *remote = NULL,
379 *dst = NULL, hints;
380 unsigned int l3type = AF_UNSPEC;
381 int rc, on = 1, sockfd = -1,
382 socktype = sock_type(l4type);
383 char port[6]; /* port number has at most 5 digits */
384
385 sprintf(port, "%u", port_number);
386 /* Set up address hint structure */
387 memset(&hints, 0, sizeof(hints));
388 hints.ai_family = l3type;
389 hints.ai_socktype = socktype;
390 /*
391 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
392 * (and only then) pretend to be UDP.
393 */
394 if (l4type == IPPROTO_DCCP)
395 hints.ai_socktype = SOCK_DGRAM;
396
397 /* only use addresses available on the host */
398 hints.ai_flags = AI_ADDRCONFIG;
399 if (l3type == AF_INET6)
400 /* use v4-mapped-v6 if no v6 addresses found */
401 hints.ai_flags |= AI_V4MAPPED | AI_ALL;
402
403 if (passive && host == NULL)
404 hints.ai_flags |= AI_PASSIVE;
405
406 /* Obtain local/remote address information */
407 if ((rc = getaddrinfo(host, port, &hints, passive ? &local : &remote))) {
408 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s.\n",
409 layer4_name(l4type),
410 host? host : (passive? "[loopback]" : "[localhost]"),
411 port, gai_strerror(rc));
412 rc = -E_ADDRESS_LOOKUP;
413 goto out;
414 }
415
416 /* Iterate over all src/dst combination, exhausting dst first */
417 for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) {
418 if (src && dst && src->ai_family == AF_INET
419 && dst->ai_family == AF_INET6)
420 goto get_next_dst; /* v4 -> v6 is not possible */
421
422 sockfd = socket(src ? src->ai_family : dst->ai_family,
423 socktype, l4type);
424 if (sockfd < 0)
425 goto get_next_dst;
426
427 /*
428 * Reuse the address on passive sockets to avoid failure on
429 * restart (protocols using listen()) and when creating
430 * multiple listener instances (UDP multicast).
431 */
432 if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
433 &on, sizeof(on)) == -1) {
434 rc = errno;
435 close(sockfd);
436 PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
437 strerror(rc));
438 rc = -ERRNO_TO_PARA_ERROR(rc);
439 break;
440 }
441 flowopt_setopts(sockfd, fo);
442
443 if (src) {
444 if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) {
445 close(sockfd);
446 goto get_next_src;
447 }
448 if (!dst) /* bind-only completed successfully */
449 break;
450 }
451
452 if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0)
453 break; /* connection completed successfully */
454 close(sockfd);
455 get_next_dst:
456 if (dst && (dst = dst->ai_next))
457 continue;
458 get_next_src:
459 if (src && (src = src->ai_next)) /* restart inner loop */
460 dst = remote;
461 }
462 out:
463 if (local)
464 freeaddrinfo(local);
465 if (remote)
466 freeaddrinfo(remote);
467 flowopt_cleanup(fo);
468
469 if (src == NULL && dst == NULL) {
470 if (rc >= 0)
471 rc = -E_MAKESOCK;
472 PARA_ERROR_LOG("can not create %s socket %s#%s.\n",
473 layer4_name(l4type), host? host : (passive?
474 "[loopback]" : "[localhost]"), port);
475 return rc;
476 }
477 return sockfd;
478 }
479
480 /**
481 * Create a passive / listening socket.
482 *
483 * \param l4type The transport-layer type (\p IPPROTO_xxx).
484 * \param port The decimal port number to listen on.
485 * \param fo Flowopts (if any) to set before starting to listen.
486 *
487 * \return Positive integer (socket descriptor) on success, negative value
488 * otherwise.
489 *
490 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
491 */
492 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
493 {
494 int ret, fd = makesock(l4type, 1, NULL, port, fo);
495
496 if (fd > 0) {
497 ret = listen(fd, BACKLOG);
498 if (ret < 0) {
499 ret = errno;
500 close(fd);
501 return -ERRNO_TO_PARA_ERROR(ret);
502 }
503 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
504 layer4_name(l4type), port, fd);
505 }
506 return fd;
507 }
508
509 /**
510 * Determine IPv4/v6 socket address length.
511 * \param sa Container of IPv4 or IPv6 address.
512 * \return Address-family dependent address length.
513 */
514 static socklen_t salen(const struct sockaddr *sa)
515 {
516 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
517
518 return sa->sa_family == AF_INET6
519 ? sizeof(struct sockaddr_in6)
520 : sizeof(struct sockaddr_in);
521 }
522
523 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
524 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
525 {
526 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
527
528 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
529 }
530
531 /**
532 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
533 * \param ss Container of IPv4/6 address.
534 * \return Pointer to normalized address (may be static storage).
535 *
536 * \sa RFC 3493
537 */
538 static const struct sockaddr *
539 normalize_ip_address(const struct sockaddr_storage *ss)
540 {
541 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
542
543 if (SS_IS_ADDR_V4MAPPED(ss)) {
544 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
545 static struct sockaddr_in ia;
546
547 ia.sin_family = AF_INET;
548 ia.sin_port = ia6->sin6_port;
549 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
550 return (const struct sockaddr *)&ia;
551 }
552 return (const struct sockaddr *)ss;
553 }
554
555 /**
556 * Generic/fallback MTU values
557 *
558 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
559 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
560 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
561 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
562 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
563 * UDP senders should use the respective minimum values of EMTU_S.
564 */
565 static inline int generic_mtu(const int af_type)
566 {
567 return af_type == AF_INET6 ? 1280 : 576;
568 }
569
570 /** Crude approximation of IP header overhead - neglecting options. */
571 static inline int estimated_header_overhead(const int af_type)
572 {
573 return af_type == AF_INET6 ? 40 : 20;
574 }
575
576 /**
577 * Get the maximum transport-layer message size (MMS_S).
578 *
579 * \param sockfd The socket file descriptor.
580 *
581 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol familiy
582 * could not be determined, \p AF_INET is assumed.
583 *
584 * \return The maximum message size of the address family type.
585 */
586 int generic_max_transport_msg_size(int sockfd)
587 {
588 struct sockaddr_storage ss;
589 socklen_t sslen = sizeof(ss);
590 int af_type = AF_INET;
591
592 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
593 PARA_ERROR_LOG("can not determine remote address type: %s\n",
594 strerror(errno));
595 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
596 af_type = ss.ss_family;
597 }
598 return generic_mtu(af_type) - estimated_header_overhead(af_type);
599 }
600
601 /**
602 * Look up the local or remote side of a connected socket structure.
603 *
604 * \param fd The socket descriptor of the connected socket.
605 * \param getname Either \p getsockname() for local, or \p getpeername() for
606 * remote side.
607 *
608 * \return A static character string identifying hostname and port of the
609 * chosen side in numeric host:port format.
610 *
611 * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3),
612 * services(5), nsswitch.conf(5).
613 */
614 static char *__get_sock_name(int fd, typeof(getsockname) getname)
615 {
616 struct sockaddr_storage ss;
617 const struct sockaddr *sa;
618 socklen_t sslen = sizeof(ss);
619 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
620 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
621 int ret;
622
623 if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
624 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
625 fd, strerror(errno));
626 snprintf(output, sizeof(output), "(unknown)");
627 return output;
628 }
629 sa = normalize_ip_address(&ss);
630 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
631 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
632 if (ret) {
633 PARA_WARNING_LOG("hostname lookup error (%s).\n",
634 gai_strerror(ret));
635 snprintf(output, sizeof(output), "(lookup error)");
636 } else if (sa->sa_family == AF_INET6)
637 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
638 else
639 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
640 return output;
641 }
642
643 /**
644 * Look up the local side of a connected socket structure.
645 *
646 * \param sockfd The file descriptor of the socket.
647 *
648 * \return A pointer to a static buffer containing hostname an port. This
649 * buffer must not be freed by the caller.
650 *
651 * \sa remote_name().
652 */
653 char *local_name(int sockfd)
654 {
655 return __get_sock_name(sockfd, getsockname);
656 }
657
658 /**
659 * Look up the remote side of a connected socket structure.
660 *
661 * \param sockfd The file descriptor of the socket.
662 *
663 * \return Analogous to the return value of \ref local_name() but for the
664 * remote side.
665 *
666 * \sa local_name().
667 */
668 char *remote_name(int sockfd)
669 {
670 return __get_sock_name(sockfd, getpeername);
671 }
672
673 /**
674 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
675 *
676 * \param ss Container of IPv4/6 address.
677 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
678 *
679 * \sa RFC 3493.
680 */
681 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
682 {
683 const struct sockaddr *sa = normalize_ip_address(ss);
684
685 memset(ia, 0, sizeof(*ia));
686 if (sa->sa_family == AF_INET)
687 *ia = ((struct sockaddr_in *)sa)->sin_addr;
688 }
689
690 /**
691 * Receive data from a file descriptor.
692 *
693 * \param fd The file descriptor.
694 * \param buf The buffer to write the data to.
695 * \param size The size of \a buf.
696 *
697 * Receive at most \a size bytes from file descriptor \a fd.
698 *
699 * \return The number of bytes received on success, negative on errors, zero if
700 * the peer has performed an orderly shutdown.
701 *
702 * \sa recv(2).
703 */
704 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
705 {
706 ssize_t n;
707
708 n = recv(fd, buf, size, 0);
709 if (n == -1)
710 return -ERRNO_TO_PARA_ERROR(errno);
711 return n;
712 }
713
714 /**
715 * Receive and write terminating NULL byte.
716 *
717 * \param fd The file descriptor.
718 * \param buf The buffer to write the data to.
719 * \param size The size of \a buf.
720 *
721 * Read at most \a size - 1 bytes from file descriptor \a fd and
722 * write a NULL byte at the end of the received data.
723 *
724 * \return The return value of the underlying call to \a recv_bin_buffer().
725 *
726 * \sa recv_bin_buffer()
727 */
728 int recv_buffer(int fd, char *buf, size_t size)
729 {
730 int n;
731
732 assert(size);
733 n = recv_bin_buffer(fd, buf, size - 1);
734 if (n >= 0)
735 buf[n] = '\0';
736 else
737 *buf = '\0';
738 return n;
739 }
740
741 /**
742 * Wrapper around the accept system call.
743 *
744 * \param fd The listening socket.
745 * \param rfds An optional fd_set pointer.
746 * \param addr Structure which is filled in with the address of the peer socket.
747 * \param size Should contain the size of the structure pointed to by \a addr.
748 * \param new_fd Result pointer.
749 *
750 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
751 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
752 *
753 * \return Negative on errors, zero if no connections are present to be accepted,
754 * one otherwise.
755 *
756 * \sa accept(2).
757 */
758 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
759 {
760 int ret;
761
762 if (rfds && !FD_ISSET(fd, rfds))
763 return 0;
764 do
765 ret = accept(fd, (struct sockaddr *) addr, &size);
766 while (ret < 0 && errno == EINTR);
767
768 if (ret >= 0) {
769 *new_fd = ret;
770 return 1;
771 }
772 if (errno == EAGAIN || errno == EWOULDBLOCK)
773 return 0;
774 return -ERRNO_TO_PARA_ERROR(errno);
775 }
776
777 /**
778 * Probe the list of DCCP CCIDs configured on this host.
779 * \param ccid_array Pointer to return statically allocated array in.
780 * \return Number of elements returned in \a ccid_array or error.
781 *
782 * NB: This feature is only available on Linux > 2.6.30; on older kernels
783 * ENOPROTOOPT ("Protocol not available") will be returned.
784 */
785 int dccp_available_ccids(uint8_t **ccid_array)
786 {
787 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
788 socklen_t nccids = sizeof(ccids);
789 int ret, fd;
790
791 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
792 if (ret < 0)
793 return ret;
794
795 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
796 ccids, &nccids) < 0) {
797 ret = errno;
798 close(fd);
799 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
800 strerror(ret));
801 return -ERRNO_TO_PARA_ERROR(ret);
802 }
803
804 close(fd);
805 *ccid_array = ccids;
806 return nccids;
807 }
808
809 /**
810 * Prepare a structure for \p AF_UNIX socket addresses.
811 *
812 * \param u Pointer to the struct to be prepared.
813 * \param name The socket pathname.
814 *
815 * This just copies \a name to the sun_path component of \a u.
816 *
817 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
818 * than \p UNIX_PATH_MAX.
819 */
820 static int init_unix_addr(struct sockaddr_un *u, const char *name)
821 {
822 if (strlen(name) >= UNIX_PATH_MAX)
823 return -E_NAME_TOO_LONG;
824 memset(u->sun_path, 0, UNIX_PATH_MAX);
825 u->sun_family = PF_UNIX;
826 strcpy(u->sun_path, name);
827 return 1;
828 }
829
830 /**
831 * Prepare, create, and bind a socket for local communication.
832 *
833 * \param name The socket pathname.
834 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
835 * \param mode The desired mode of the socket.
836 *
837 * This function creates a local socket for sequenced, reliable,
838 * two-way, connection-based byte streams.
839 *
840 * \return The file descriptor, on success, negative on errors.
841 *
842 * \sa socket(2)
843 * \sa bind(2)
844 * \sa chmod(2)
845 */
846 int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
847 mode_t mode)
848 {
849 int fd, ret;
850
851 ret = init_unix_addr(unix_addr, name);
852 if (ret < 0)
853 return ret;
854 ret = socket(PF_UNIX, SOCK_STREAM, 0);
855 if (ret < 0)
856 return -ERRNO_TO_PARA_ERROR(errno);
857 fd = ret;
858 ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
859 if (ret < 0) {
860 ret = -ERRNO_TO_PARA_ERROR(errno);
861 goto err;
862 }
863 ret = -E_CHMOD;
864 if (chmod(name, mode) < 0)
865 goto err;
866 return fd;
867 err:
868 close(fd);
869 return ret;
870 }
871
872 /**
873 * Prepare, create, and connect to a Unix domain socket for local communication.
874 *
875 * \param name The socket pathname.
876 *
877 * This function creates a local socket for sequenced, reliable, two-way,
878 * connection-based byte streams.
879 *
880 * \return The file descriptor of the connected socket on success, negative on
881 * errors.
882 *
883 * \sa create_local_socket(), unix(7), connect(2).
884 */
885 int connect_local_socket(const char *name)
886 {
887 struct sockaddr_un unix_addr;
888 int fd, ret;
889
890 PARA_DEBUG_LOG("connecting to %s\n", name);
891 ret = init_unix_addr(&unix_addr, name);
892 if (ret < 0)
893 return ret;
894 fd = socket(PF_UNIX, SOCK_STREAM, 0);
895 if (fd < 0)
896 return -ERRNO_TO_PARA_ERROR(errno);
897 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
898 ret = -ERRNO_TO_PARA_ERROR(errno);
899 goto err;
900 }
901 return fd;
902 err:
903 close(fd);
904 return ret;
905 }
906
907 #ifndef HAVE_UCRED
908 ssize_t send_cred_buffer(int sock, char *buf)
909 {
910 return write_buffer(sock, buf);
911 }
912 int recv_cred_buffer(int fd, char *buf, size_t size)
913 {
914 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
915 }
916 #else /* HAVE_UCRED */
917 /**
918 * Send \p NULL-terminated buffer and Unix credentials of the current process.
919 *
920 * \param sock The socket file descriptor.
921 * \param buf The buffer to be sent.
922 *
923 * \return On success, this call returns the number of characters sent. On
924 * error, \p -E_SENDMSG is returned.
925 *
926 * \sa sendmsg(2), okir's Black Hats Manual.
927 */
928 ssize_t send_cred_buffer(int sock, char *buf)
929 {
930 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
931 struct msghdr msg;
932 struct cmsghdr *cmsg;
933 static struct iovec iov;
934 struct ucred c;
935 int ret;
936
937 /* Response data */
938 iov.iov_base = buf;
939 iov.iov_len = strlen(buf);
940 c.pid = getpid();
941 c.uid = getuid();
942 c.gid = getgid();
943 /* compose the message */
944 memset(&msg, 0, sizeof(msg));
945 msg.msg_iov = &iov;
946 msg.msg_iovlen = 1;
947 msg.msg_control = control;
948 msg.msg_controllen = sizeof(control);
949 /* attach the ucred struct */
950 cmsg = CMSG_FIRSTHDR(&msg);
951 cmsg->cmsg_level = SOL_SOCKET;
952 cmsg->cmsg_type = SCM_CREDENTIALS;
953 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
954 *(struct ucred *)CMSG_DATA(cmsg) = c;
955 msg.msg_controllen = cmsg->cmsg_len;
956 ret = sendmsg(sock, &msg, 0);
957 if (ret < 0)
958 ret = -E_SENDMSG;
959 return ret;
960 }
961
962 static void dispose_fds(int *fds, unsigned num)
963 {
964 int i;
965
966 for (i = 0; i < num; i++)
967 close(fds[i]);
968 }
969
970 /**
971 * Receive a buffer and the Unix credentials of the sending process.
972 *
973 * \param fd the socket file descriptor.
974 * \param buf the buffer to store the message.
975 * \param size the size of \a buffer.
976 *
977 * \return negative on errors, the user id on success.
978 *
979 * \sa recvmsg(2), okir's Black Hats Manual.
980 */
981 int recv_cred_buffer(int fd, char *buf, size_t size)
982 {
983 char control[255];
984 struct msghdr msg;
985 struct cmsghdr *cmsg;
986 struct iovec iov;
987 int result = 0;
988 int yes = 1;
989 struct ucred cred;
990
991 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
992 memset(&msg, 0, sizeof(msg));
993 memset(buf, 0, size);
994 iov.iov_base = buf;
995 iov.iov_len = size;
996 msg.msg_iov = &iov;
997 msg.msg_iovlen = 1;
998 msg.msg_control = control;
999 msg.msg_controllen = sizeof(control);
1000 if (recvmsg(fd, &msg, 0) < 0)
1001 return -E_RECVMSG;
1002 result = -E_SCM_CREDENTIALS;
1003 cmsg = CMSG_FIRSTHDR(&msg);
1004 while (cmsg) {
1005 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1006 == SCM_CREDENTIALS) {
1007 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1008 result = cred.uid;
1009 } else
1010 if (cmsg->cmsg_level == SOL_SOCKET
1011 && cmsg->cmsg_type == SCM_RIGHTS) {
1012 dispose_fds((int *) CMSG_DATA(cmsg),
1013 (cmsg->cmsg_len - CMSG_LEN(0))
1014 / sizeof(int));
1015 }
1016 cmsg = CMSG_NXTHDR(&msg, cmsg);
1017 }
1018 return result;
1019 }
1020 #endif /* HAVE_UCRED */