9c300c744c11ddbec362eb02433a6a3d60db8163
[paraslash.git] / net.c
1 /*
2 * Copyright (C) 2005-2012 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 struct flowopts *flowopt_new(void)
268 {
269 struct flowopts *new = para_malloc(sizeof(*new));
270
271 INIT_LIST_HEAD(&new->sockopts);
272 return new;
273 }
274
275 /**
276 * Append new socket option to flowopt queue.
277 *
278 * \param fo The flowopt queue to append to.
279 * \param lev Level at which \a opt resides.
280 * \param opt New option to add.
281 * \param name Stringified name of \a opt.
282 * \param val The value to set \a opt to.
283 * \param len Length of \a val.
284 *
285 * \sa setsockopt(2)
286 */
287 void flowopt_add(struct flowopts *fo, int lev, int opt,
288 const char *name, const void *val, int len)
289 {
290 struct pre_conn_opt *new = para_malloc(sizeof(*new));
291
292 new->sock_option = opt;
293 new->sock_level = lev;
294 new->opt_name = para_strdup(name);
295
296 if (val == NULL) {
297 new->opt_val = NULL;
298 new->opt_len = 0;
299 } else {
300 new->opt_val = para_malloc(len);
301 new->opt_len = len;
302 memcpy(new->opt_val, val, len);
303 }
304
305 list_add_tail(&new->node, &fo->sockopts);
306 }
307
308 void flowopt_add_bool(struct flowopts *fo, int lev, int opt,
309 const char *optname, bool on_or_off)
310 {
311 int on = on_or_off; /* kernel takes 'int' */
312
313 flowopt_add(fo, lev, opt, optname, &on, sizeof(on));
314 }
315
316 /** Set the entire bunch of pre-connection options at once. */
317 static void flowopt_setopts(int sockfd, struct flowopts *fo)
318 {
319 struct pre_conn_opt *pc;
320
321 if (fo == NULL)
322 return;
323
324 list_for_each_entry(pc, &fo->sockopts, node)
325 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
326 pc->opt_val, pc->opt_len) < 0) {
327 PARA_EMERG_LOG("Can not set %s socket option: %s",
328 pc->opt_name, strerror(errno));
329 exit(EXIT_FAILURE);
330 }
331 }
332
333 static void flowopt_cleanup(struct flowopts *fo)
334 {
335 struct pre_conn_opt *cur, *next;
336
337 if (fo == NULL)
338 return;
339
340 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
341 free(cur->opt_name);
342 free(cur->opt_val);
343 free(cur);
344 }
345 free(fo);
346 }
347
348 /**
349 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
350 *
351 * \param l4type The layer-4 type (\p IPPROTO_xxx).
352 * \param passive Whether this is a passive (1) or active (0) socket.
353 * \param host Remote or local hostname or IPv/6 address string.
354 * \param port_number Decimal port number.
355 * \param fo Socket options to be set before making the connection.
356 *
357 * This creates a ready-made IPv4/v6 socket structure after looking up the
358 * necessary parameters. The interpretation of \a host depends on the value of
359 * \a passive:
360 * - on a passive socket host is interpreted as an interface IPv4/6 address
361 * (can be left NULL);
362 * - on an active socket, \a host is the peer DNS name or IPv4/6 address
363 * to connect to;
364 * - \a port_number is in either case the numeric port number (not service
365 * string).
366 *
367 * Furthermore, bind(2) is called on passive sockets, and connect(2) on active
368 * sockets. The algorithm tries all possible address combinations until it
369 * succeeds. If \a fo is supplied, options are set and cleanup is performed.
370 *
371 * \return This function returns 1 on success and \a -E_ADDRESS_LOOKUP when no
372 * matching connection could be set up (with details in the error log).
373 *
374 * \sa ipv6(7), getaddrinfo(3), bind(2), connect(2).
375 */
376 int makesock(unsigned l4type, bool passive,
377 const char *host, uint16_t port_number,
378 struct flowopts *fo)
379 {
380 struct addrinfo *local = NULL, *src = NULL, *remote = NULL,
381 *dst = NULL, hints;
382 unsigned int l3type = AF_UNSPEC;
383 int rc, on = 1, sockfd = -1,
384 socktype = sock_type(l4type);
385 char port[6]; /* port number has at most 5 digits */
386
387 sprintf(port, "%u", port_number);
388 /* Set up address hint structure */
389 memset(&hints, 0, sizeof(hints));
390 hints.ai_family = l3type;
391 hints.ai_socktype = socktype;
392 /*
393 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
394 * (and only then) pretend to be UDP.
395 */
396 if (l4type == IPPROTO_DCCP)
397 hints.ai_socktype = SOCK_DGRAM;
398
399 /* only use addresses available on the host */
400 hints.ai_flags = AI_ADDRCONFIG;
401 if (l3type == AF_INET6)
402 /* use v4-mapped-v6 if no v6 addresses found */
403 hints.ai_flags |= AI_V4MAPPED | AI_ALL;
404
405 if (passive && host == NULL)
406 hints.ai_flags |= AI_PASSIVE;
407
408 /* Obtain local/remote address information */
409 if ((rc = getaddrinfo(host, port, &hints, passive ? &local : &remote))) {
410 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s.\n",
411 layer4_name(l4type),
412 host? host : (passive? "[loopback]" : "[localhost]"),
413 port, gai_strerror(rc));
414 rc = -E_ADDRESS_LOOKUP;
415 goto out;
416 }
417
418 /* Iterate over all src/dst combination, exhausting dst first */
419 for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) {
420 if (src && dst && src->ai_family == AF_INET
421 && dst->ai_family == AF_INET6)
422 goto get_next_dst; /* v4 -> v6 is not possible */
423
424 sockfd = socket(src ? src->ai_family : dst->ai_family,
425 socktype, l4type);
426 if (sockfd < 0)
427 goto get_next_dst;
428
429 /*
430 * Reuse the address on passive sockets to avoid failure on
431 * restart (protocols using listen()) and when creating
432 * multiple listener instances (UDP multicast).
433 */
434 if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
435 &on, sizeof(on)) == -1) {
436 rc = errno;
437 close(sockfd);
438 PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
439 strerror(rc));
440 rc = -ERRNO_TO_PARA_ERROR(rc);
441 break;
442 }
443 flowopt_setopts(sockfd, fo);
444
445 if (src) {
446 if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) {
447 close(sockfd);
448 goto get_next_src;
449 }
450 if (!dst) /* bind-only completed successfully */
451 break;
452 }
453
454 if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0)
455 break; /* connection completed successfully */
456 close(sockfd);
457 get_next_dst:
458 if (dst && (dst = dst->ai_next))
459 continue;
460 get_next_src:
461 if (src && (src = src->ai_next)) /* restart inner loop */
462 dst = remote;
463 }
464 out:
465 if (local)
466 freeaddrinfo(local);
467 if (remote)
468 freeaddrinfo(remote);
469 flowopt_cleanup(fo);
470
471 if (src == NULL && dst == NULL) {
472 if (rc >= 0)
473 rc = -E_MAKESOCK;
474 PARA_ERROR_LOG("can not create %s socket %s#%s.\n",
475 layer4_name(l4type), host? host : (passive?
476 "[loopback]" : "[localhost]"), port);
477 return rc;
478 }
479 return sockfd;
480 }
481
482 /**
483 * Create a passive / listening socket.
484 *
485 * \param l4type The transport-layer type (\p IPPROTO_xxx).
486 * \param port The decimal port number to listen on.
487 * \param fo Flowopts (if any) to set before starting to listen.
488 *
489 * \return Positive integer (socket descriptor) on success, negative value
490 * otherwise.
491 *
492 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
493 */
494 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
495 {
496 int ret, fd = makesock(l4type, 1, NULL, port, fo);
497
498 if (fd > 0) {
499 ret = listen(fd, BACKLOG);
500 if (ret < 0) {
501 ret = errno;
502 close(fd);
503 return -ERRNO_TO_PARA_ERROR(ret);
504 }
505 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
506 layer4_name(l4type), port, fd);
507 }
508 return fd;
509 }
510
511 /**
512 * Determine IPv4/v6 socket address length.
513 * \param sa Container of IPv4 or IPv6 address.
514 * \return Address-family dependent address length.
515 */
516 static socklen_t salen(const struct sockaddr *sa)
517 {
518 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
519
520 return sa->sa_family == AF_INET6
521 ? sizeof(struct sockaddr_in6)
522 : sizeof(struct sockaddr_in);
523 }
524
525 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
526 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
527 {
528 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
529
530 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
531 }
532
533 /**
534 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
535 * \param ss Container of IPv4/6 address.
536 * \return Pointer to normalized address (may be static storage).
537 *
538 * \sa RFC 3493
539 */
540 static const struct sockaddr *
541 normalize_ip_address(const struct sockaddr_storage *ss)
542 {
543 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
544
545 if (SS_IS_ADDR_V4MAPPED(ss)) {
546 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
547 static struct sockaddr_in ia;
548
549 ia.sin_family = AF_INET;
550 ia.sin_port = ia6->sin6_port;
551 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
552 return (const struct sockaddr *)&ia;
553 }
554 return (const struct sockaddr *)ss;
555 }
556
557 /**
558 * Generic/fallback MTU values
559 *
560 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
561 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
562 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
563 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
564 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
565 * UDP senders should use the respective minimum values of EMTU_S.
566 */
567 static inline int generic_mtu(const int af_type)
568 {
569 return af_type == AF_INET6 ? 1280 : 576;
570 }
571
572 /** Crude approximation of IP header overhead - neglecting options. */
573 static inline int estimated_header_overhead(const int af_type)
574 {
575 return af_type == AF_INET6 ? 40 : 20;
576 }
577
578 /**
579 * Get the maximum transport-layer message size (MMS_S).
580 *
581 * \param sockfd The socket file descriptor.
582 *
583 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol familiy
584 * could not be determined, \p AF_INET is assumed.
585 *
586 * \return The maximum message size of the address family type.
587 */
588 int generic_max_transport_msg_size(int sockfd)
589 {
590 struct sockaddr_storage ss;
591 socklen_t sslen = sizeof(ss);
592 int af_type = AF_INET;
593
594 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
595 PARA_ERROR_LOG("can not determine remote address type: %s\n",
596 strerror(errno));
597 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
598 af_type = ss.ss_family;
599 }
600 return generic_mtu(af_type) - estimated_header_overhead(af_type);
601 }
602
603 /**
604 * Look up the local or remote side of a connected socket structure.
605 *
606 * \param fd The socket descriptor of the connected socket.
607 * \param getname Either \p getsockname() for local, or \p getpeername() for
608 * remote side.
609 *
610 * \return A static character string identifying hostname and port of the
611 * chosen side in numeric host:port format.
612 *
613 * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3),
614 * services(5), nsswitch.conf(5).
615 */
616 static char *__get_sock_name(int fd, typeof(getsockname) getname)
617 {
618 struct sockaddr_storage ss;
619 const struct sockaddr *sa;
620 socklen_t sslen = sizeof(ss);
621 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
622 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
623 int ret;
624
625 if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
626 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
627 fd, strerror(errno));
628 snprintf(output, sizeof(output), "(unknown)");
629 return output;
630 }
631 sa = normalize_ip_address(&ss);
632 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
633 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
634 if (ret) {
635 PARA_WARNING_LOG("hostname lookup error (%s).\n",
636 gai_strerror(ret));
637 snprintf(output, sizeof(output), "(lookup error)");
638 } else if (sa->sa_family == AF_INET6)
639 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
640 else
641 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
642 return output;
643 }
644
645 /**
646 * Look up the local side of a connected socket structure.
647 *
648 * \param sockfd The file descriptor of the socket.
649 *
650 * \return A pointer to a static buffer containing hostname an port. This
651 * buffer must not be freed by the caller.
652 *
653 * \sa remote_name().
654 */
655 char *local_name(int sockfd)
656 {
657 return __get_sock_name(sockfd, getsockname);
658 }
659
660 /**
661 * Look up the remote side of a connected socket structure.
662 *
663 * \param sockfd The file descriptor of the socket.
664 *
665 * \return Analogous to the return value of \ref local_name() but for the
666 * remote side.
667 *
668 * \sa local_name().
669 */
670 char *remote_name(int sockfd)
671 {
672 return __get_sock_name(sockfd, getpeername);
673 }
674
675 /**
676 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
677 * \param ss Container of IPv4/6 address
678 * \return Extracted IPv4 address (different from 0) or 0 if unsuccessful.
679 *
680 * \sa RFC 3493
681 */
682 struct in_addr extract_v4_addr(const struct sockaddr_storage *ss)
683 {
684 struct in_addr ia = {.s_addr = 0};
685 const struct sockaddr *sa = normalize_ip_address(ss);
686
687 if (sa->sa_family == AF_INET)
688 ia = ((struct sockaddr_in *)sa)->sin_addr;
689 return ia;
690 }
691
692 /**
693 * Send a \p NULL-terminated buffer.
694 *
695 * \param fd The file descriptor.
696 * \param buf The null-terminated buffer to be send.
697 *
698 * This is equivalent to write_all(fd, buf, strlen(buf)).
699 *
700 * \return Standard.
701 */
702 int send_buffer(int fd, const char *buf)
703 {
704 return write_all(fd, buf, strlen(buf));
705 }
706
707 /**
708 * Send a buffer given by a format string.
709 *
710 * \param fd The file descriptor.
711 * \param fmt A format string.
712 *
713 * \return Standard.
714 */
715 __printf_2_3 int send_va_buffer(int fd, const char *fmt, ...)
716 {
717 char *msg;
718 int ret;
719
720 PARA_VSPRINTF(fmt, msg);
721 ret = send_buffer(fd, msg);
722 free(msg);
723 return ret;
724 }
725
726 /**
727 * Receive data from a file descriptor.
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 * Receive at most \a size bytes from file descriptor \a fd.
734 *
735 * \return The number of bytes received on success, negative on errors, zero if
736 * the peer has performed an orderly shutdown.
737 *
738 * \sa recv(2).
739 */
740 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
741 {
742 ssize_t n;
743
744 n = recv(fd, buf, size, 0);
745 if (n == -1)
746 return -ERRNO_TO_PARA_ERROR(errno);
747 return n;
748 }
749
750 /**
751 * Receive and write terminating NULL byte.
752 *
753 * \param fd The file descriptor.
754 * \param buf The buffer to write the data to.
755 * \param size The size of \a buf.
756 *
757 * Read at most \a size - 1 bytes from file descriptor \a fd and
758 * write a NULL byte at the end of the received data.
759 *
760 * \return The return value of the underlying call to \a recv_bin_buffer().
761 *
762 * \sa recv_bin_buffer()
763 */
764 int recv_buffer(int fd, char *buf, size_t size)
765 {
766 int n;
767
768 assert(size);
769 n = recv_bin_buffer(fd, buf, size - 1);
770 if (n >= 0)
771 buf[n] = '\0';
772 else
773 *buf = '\0';
774 return n;
775 }
776
777 /**
778 * Wrapper around the accept system call.
779 *
780 * \param fd The listening socket.
781 * \param rfds An optional fd_set pointer.
782 * \param addr Structure which is filled in with the address of the peer socket.
783 * \param size Should contain the size of the structure pointed to by \a addr.
784 * \param new_fd Result pointer.
785 *
786 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
787 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
788 *
789 * \return Negative on errors, zero if no connections are present to be accepted,
790 * one otherwise.
791 *
792 * \sa accept(2).
793 */
794 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
795 {
796 int ret;
797
798 if (rfds && !FD_ISSET(fd, rfds))
799 return 0;
800 do
801 ret = accept(fd, (struct sockaddr *) addr, &size);
802 while (ret < 0 && errno == EINTR);
803
804 if (ret >= 0) {
805 *new_fd = ret;
806 return 1;
807 }
808 if (errno == EAGAIN || errno == EWOULDBLOCK)
809 return 0;
810 return -ERRNO_TO_PARA_ERROR(errno);
811 }
812
813 /**
814 * Probe the list of DCCP CCIDs configured on this host.
815 * \param ccid_array Pointer to return statically allocated array in.
816 * \return Number of elements returned in \a ccid_array or error.
817 *
818 * NB: This feature is only available on Linux > 2.6.30; on older kernels
819 * ENOPROTOOPT ("Protocol not available") will be returned.
820 */
821 int dccp_available_ccids(uint8_t **ccid_array)
822 {
823 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
824 socklen_t nccids = sizeof(ccids);
825 int ret, fd;
826
827 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
828 if (ret < 0)
829 return ret;
830
831 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
832 ccids, &nccids) < 0) {
833 ret = errno;
834 close(fd);
835 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
836 strerror(ret));
837 return -ERRNO_TO_PARA_ERROR(ret);
838 }
839
840 close(fd);
841 *ccid_array = ccids;
842 return nccids;
843 }
844
845 /**
846 * Prepare a structure for \p AF_UNIX socket addresses.
847 *
848 * \param u Pointer to the struct to be prepared.
849 * \param name The socket pathname.
850 *
851 * This just copies \a name to the sun_path component of \a u.
852 *
853 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
854 * than \p UNIX_PATH_MAX.
855 */
856 static int init_unix_addr(struct sockaddr_un *u, const char *name)
857 {
858 if (strlen(name) >= UNIX_PATH_MAX)
859 return -E_NAME_TOO_LONG;
860 memset(u->sun_path, 0, UNIX_PATH_MAX);
861 u->sun_family = PF_UNIX;
862 strcpy(u->sun_path, name);
863 return 1;
864 }
865
866 /**
867 * Prepare, create, and bind a socket for local communication.
868 *
869 * \param name The socket pathname.
870 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
871 * \param mode The desired mode of the socket.
872 *
873 * This function creates a local socket for sequenced, reliable,
874 * two-way, connection-based byte streams.
875 *
876 * \return The file descriptor, on success, negative on errors.
877 *
878 * \sa socket(2)
879 * \sa bind(2)
880 * \sa chmod(2)
881 */
882 int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
883 mode_t mode)
884 {
885 int fd, ret;
886
887 ret = init_unix_addr(unix_addr, name);
888 if (ret < 0)
889 return ret;
890 ret = socket(PF_UNIX, SOCK_STREAM, 0);
891 if (ret < 0)
892 return -ERRNO_TO_PARA_ERROR(errno);
893 fd = ret;
894 ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
895 if (ret < 0) {
896 ret = -ERRNO_TO_PARA_ERROR(errno);
897 goto err;
898 }
899 ret = -E_CHMOD;
900 if (chmod(name, mode) < 0)
901 goto err;
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 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 ret = init_unix_addr(&unix_addr, name);
928 if (ret < 0)
929 return ret;
930 fd = socket(PF_UNIX, SOCK_STREAM, 0);
931 if (fd < 0)
932 return -ERRNO_TO_PARA_ERROR(errno);
933 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
934 ret = -ERRNO_TO_PARA_ERROR(errno);
935 goto err;
936 }
937 return fd;
938 err:
939 close(fd);
940 return ret;
941 }
942
943 #ifndef HAVE_UCRED
944 ssize_t send_cred_buffer(int sock, char *buf)
945 {
946 return send_buffer(sock, buf);
947 }
948 int recv_cred_buffer(int fd, char *buf, size_t size)
949 {
950 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
951 }
952 #else /* HAVE_UCRED */
953 /**
954 * Send \p NULL-terminated buffer and Unix credentials of the current process.
955 *
956 * \param sock The socket file descriptor.
957 * \param buf The buffer to be sent.
958 *
959 * \return On success, this call returns the number of characters sent. On
960 * error, \p -E_SENDMSG is returned.
961 *
962 * \sa sendmsg(2), okir's Black Hats Manual.
963 */
964 ssize_t send_cred_buffer(int sock, char *buf)
965 {
966 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
967 struct msghdr msg;
968 struct cmsghdr *cmsg;
969 static struct iovec iov;
970 struct ucred c;
971 int ret;
972
973 /* Response data */
974 iov.iov_base = buf;
975 iov.iov_len = strlen(buf);
976 c.pid = getpid();
977 c.uid = getuid();
978 c.gid = getgid();
979 /* compose the message */
980 memset(&msg, 0, sizeof(msg));
981 msg.msg_iov = &iov;
982 msg.msg_iovlen = 1;
983 msg.msg_control = control;
984 msg.msg_controllen = sizeof(control);
985 /* attach the ucred struct */
986 cmsg = CMSG_FIRSTHDR(&msg);
987 cmsg->cmsg_level = SOL_SOCKET;
988 cmsg->cmsg_type = SCM_CREDENTIALS;
989 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
990 *(struct ucred *)CMSG_DATA(cmsg) = c;
991 msg.msg_controllen = cmsg->cmsg_len;
992 ret = sendmsg(sock, &msg, 0);
993 if (ret < 0)
994 ret = -E_SENDMSG;
995 return ret;
996 }
997
998 static void dispose_fds(int *fds, unsigned num)
999 {
1000 int i;
1001
1002 for (i = 0; i < num; i++)
1003 close(fds[i]);
1004 }
1005
1006 /**
1007 * Receive a buffer and the Unix credentials of the sending process.
1008 *
1009 * \param fd the socket file descriptor.
1010 * \param buf the buffer to store the message.
1011 * \param size the size of \a buffer.
1012 *
1013 * \return negative on errors, the user id on success.
1014 *
1015 * \sa recvmsg(2), okir's Black Hats Manual.
1016 */
1017 int recv_cred_buffer(int fd, char *buf, size_t size)
1018 {
1019 char control[255];
1020 struct msghdr msg;
1021 struct cmsghdr *cmsg;
1022 struct iovec iov;
1023 int result = 0;
1024 int yes = 1;
1025 struct ucred cred;
1026
1027 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1028 memset(&msg, 0, sizeof(msg));
1029 memset(buf, 0, size);
1030 iov.iov_base = buf;
1031 iov.iov_len = size;
1032 msg.msg_iov = &iov;
1033 msg.msg_iovlen = 1;
1034 msg.msg_control = control;
1035 msg.msg_controllen = sizeof(control);
1036 if (recvmsg(fd, &msg, 0) < 0)
1037 return -E_RECVMSG;
1038 result = -E_SCM_CREDENTIALS;
1039 cmsg = CMSG_FIRSTHDR(&msg);
1040 while (cmsg) {
1041 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1042 == SCM_CREDENTIALS) {
1043 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1044 result = cred.uid;
1045 } else
1046 if (cmsg->cmsg_level == SOL_SOCKET
1047 && cmsg->cmsg_type == SCM_RIGHTS) {
1048 dispose_fds((int *) CMSG_DATA(cmsg),
1049 (cmsg->cmsg_len - CMSG_LEN(0))
1050 / sizeof(int));
1051 }
1052 cmsg = CMSG_NXTHDR(&msg, cmsg);
1053 }
1054 return result;
1055 }
1056 #endif /* HAVE_UCRED */