Merge branch 't/decoder_latency'
[paraslash.git] / net.c
1 /*
2 * Copyright (C) 2005-2010 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 <dirent.h>
29 #include <regex.h>
30 #include <openssl/rc4.h>
31
32 #include "para.h"
33 #include "error.h"
34 #include "crypt.h"
35 #include "net.h"
36 #include "string.h"
37 #include "list.h"
38 #include "fd.h"
39
40 /**
41 * Parse and validate IPv4 address/netmask string.
42 *
43 * \param cidr Address in CIDR notation
44 * \param addr Copy of the IPv4 address part of \a cidr
45 * \param addrlen Size of \a addr in bytes
46 * \param netmask Value of the netmask part in \a cidr or the
47 * default of 32 if not specified.
48 *
49 * \return Pointer to \a addr if succesful, NULL on error.
50 * \sa RFC 4632
51 */
52 char *parse_cidr(const char *cidr,
53 char *addr, ssize_t addrlen,
54 int32_t *netmask)
55 {
56 const char *o = cidr;
57 char *c = addr, *end = c + (addrlen - 1);
58
59 *netmask = 0x20;
60
61 if (cidr == NULL || addrlen < 1)
62 goto failed;
63
64 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
65 if (c == end)
66 goto failed;
67
68 if (*o == '/')
69 if (para_atoi32(++o, netmask) < 0 ||
70 *netmask < 0 || *netmask > 0x20)
71 goto failed;
72
73 if (is_valid_ipv4_address(addr))
74 return addr;
75 failed:
76 *addr = '\0';
77 return NULL;
78 }
79
80
81 /**
82 * Match string as a candidate IPv4 address.
83 *
84 * \param address The string to match.
85 * \return True if \a address has "dot-quad" format.
86 */
87 static bool is_v4_dot_quad(const char *address)
88 {
89 bool result;
90 regex_t r;
91
92 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
93 REG_EXTENDED | REG_NOSUB) >= 0);
94 result = regexec(&r, address, 0, NULL, 0) == 0;
95 regfree(&r);
96 return result;
97 }
98
99 /**
100 * Perform basic syntax checking on the host-part of an URL:
101 *
102 * - Since ':' is invalid in IPv4 addresses and DNS names, the
103 * presence of ':' causes interpretation as IPv6 address;
104 * - next the first-match-wins algorithm from RFC 3986 is applied;
105 * - else the string is considered as DNS name, to be resolved later.
106 *
107 * \param host The host string to check.
108 * \return True if \a host passes the syntax checks.
109 *
110 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5
111 */
112 static bool host_string_ok(const char *host)
113 {
114 if (host == NULL || *host == '\0')
115 return false;
116 if (strchr(host, ':') != NULL)
117 return is_valid_ipv6_address(host);
118 if (is_v4_dot_quad(host))
119 return is_valid_ipv4_address(host);
120 return true;
121 }
122
123 /**
124 * Parse and validate URL string.
125 *
126 * The URL syntax is loosely based on RFC 3986, supporting one of
127 * - "["host"]"[:port] for native IPv6 addresses and
128 * - host[:port] for IPv4 hostnames and DNS names.
129 *
130 * Native IPv6 addresses must be enclosed in square brackets, since
131 * otherwise there is an ambiguity with the port separator `:'.
132 * The 'port' part is always considered to be a number; if absent,
133 * it is set to -1, to indicate that a default port is to be used.
134 *
135 * The following are valid examples:
136 * - 10.10.1.1
137 * - 10.10.1.2:8000
138 * - localhost
139 * - localhost:8001
140 * - [::1]:8000
141 * - [badc0de::1]
142 *
143 * \param url The URL string to take apart.
144 * \param host To return the copied host part of \a url.
145 * \param hostlen The maximum length of \a host.
146 * \param port To return the port number (if any) of \a url.
147 *
148 * \return Pointer to \a host, or NULL if failed.
149 * If NULL is returned, \a host and \a portnum are undefined. If no
150 * port number was present in \a url, \a portnum is set to -1.
151 *
152 * \sa RFC 3986, 3.2.2/3.2.3
153 */
154 char *parse_url(const char *url,
155 char *host, ssize_t hostlen,
156 int32_t *port)
157 {
158 const char *o = url;
159 char *c = host, *end = c + (hostlen - 1);
160
161 *port = -1;
162
163 if (o == NULL || hostlen < 1)
164 goto failed;
165
166 if (*o == '[') {
167 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
168 if (c == end)
169 goto failed;
170
171 if (*o++ != ']' || (*o != '\0' && *o != ':'))
172 goto failed;
173 } else {
174 for (; (*c = *o == ':'? '\0' : *o); c++, o++)
175 if (c == end)
176 goto failed;
177 }
178
179 if (*o == ':')
180 if (para_atoi32(++o, port) < 0 ||
181 *port < 0 || *port > 0xffff)
182 goto failed;
183
184 if (host_string_ok(host))
185 return host;
186 failed:
187 *host = '\0';
188 return NULL;
189 }
190
191 /**
192 * Stringify port number, resolve into service name where defined.
193 * \param port 2-byte port number, in host-byte-order.
194 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
195 * \return Pointer to static result buffer.
196 *
197 * \sa getservent(3), services(5), nsswitch.conf(5)
198 */
199 const char *stringify_port(int port, const char *transport)
200 {
201 static char service[NI_MAXSERV];
202
203 if (port < 0 || port > 0xFFFF) {
204 snprintf(service, sizeof(service), "undefined (%d)", port);
205 } else {
206 struct servent *se = getservbyport(htons(port), transport);
207
208 if (se == NULL)
209 snprintf(service, sizeof(service), "%u", port);
210 else
211 snprintf(service, sizeof(service), "%s", se->s_name);
212 }
213 return service;
214 }
215
216 /**
217 * Determine the socket type for a given layer-4 protocol.
218 *
219 * \param l4type The symbolic name of the transport-layer protocol.
220 *
221 * \sa ip(7), socket(2)
222 */
223 static inline int sock_type(const unsigned l4type)
224 {
225 switch (l4type) {
226 case IPPROTO_UDP: return SOCK_DGRAM;
227 case IPPROTO_TCP: return SOCK_STREAM;
228 case IPPROTO_DCCP: return SOCK_DCCP;
229 }
230 return -1; /* not supported here */
231 }
232
233 /**
234 * Pretty-print transport-layer name.
235 */
236 static const char *layer4_name(const unsigned l4type)
237 {
238 switch (l4type) {
239 case IPPROTO_UDP: return "UDP";
240 case IPPROTO_TCP: return "TCP";
241 case IPPROTO_DCCP: return "DCCP";
242 }
243 return "UNKNOWN PROTOCOL";
244 }
245
246 /**
247 * Flowopts: Transport-layer independent encapsulation of socket options.
248 *
249 * These collect individual socket options into a queue, which is disposed of
250 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
251 * only and should not be visible elsewhere.
252 *
253 * \sa setsockopt(2), makesock()
254 */
255 struct pre_conn_opt {
256 int sock_level; /**< Second argument to setsockopt() */
257 int sock_option; /**< Third argument to setsockopt() */
258 char *opt_name; /**< Stringified \a sock_option */
259 void *opt_val; /**< Fourth argument to setsockopt() */
260 socklen_t opt_len; /**< Fifth argument to setsockopt() */
261
262 struct list_head node; /**< FIFO, as sockopt order matters. */
263 };
264
265 /** FIFO list of pre-connection socket options to be set */
266 struct flowopts {
267 struct list_head sockopts;
268 };
269
270 struct flowopts *flowopt_new(void)
271 {
272 struct flowopts *new = para_malloc(sizeof(*new));
273
274 INIT_LIST_HEAD(&new->sockopts);
275 return new;
276 }
277
278 /**
279 * Append new socket option to flowopt queue.
280 *
281 * \param fo The flowopt queue to append to.
282 * \param lev Level at which \a opt resides.
283 * \param opt New option to add.
284 * \param name Stringified name of \a opt.
285 * \param val The value to set \a opt to.
286 * \param len Length of \a val.
287 *
288 * \sa setsockopt(2)
289 */
290 void flowopt_add(struct flowopts *fo, int lev, int opt,
291 char *name, const void *val, int len)
292 {
293 struct pre_conn_opt *new = para_malloc(sizeof(*new));
294
295 new->sock_option = opt;
296 new->sock_level = lev;
297 new->opt_name = para_strdup(name);
298
299 if (val == NULL) {
300 new->opt_val = NULL;
301 new->opt_len = 0;
302 } else {
303 new->opt_val = para_malloc(len);
304 new->opt_len = len;
305 memcpy(new->opt_val, val, len);
306 }
307
308 list_add_tail(&new->node, &fo->sockopts);
309 }
310
311 void flowopt_add_bool(struct flowopts *fo, int lev, int opt,
312 char *optname, bool on_or_off)
313 {
314 int on = on_or_off; /* kernel takes 'int' */
315
316 flowopt_add(fo, lev, opt, optname, &on, sizeof(on));
317 }
318
319 /** Set the entire bunch of pre-connection options at once. */
320 static void flowopt_setopts(int sockfd, struct flowopts *fo)
321 {
322 struct pre_conn_opt *pc;
323
324 if (fo == NULL)
325 return;
326
327 list_for_each_entry(pc, &fo->sockopts, node)
328 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
329 pc->opt_val, pc->opt_len) < 0) {
330 PARA_EMERG_LOG("Can not set %s socket option: %s",
331 pc->opt_name, strerror(errno));
332 exit(EXIT_FAILURE);
333 }
334 }
335
336 static void flowopt_cleanup(struct flowopts *fo)
337 {
338 struct pre_conn_opt *cur, *next;
339
340 if (fo == NULL)
341 return;
342
343 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
344 free(cur->opt_name);
345 free(cur->opt_val);
346 free(cur);
347 }
348 free(fo);
349 }
350
351 /**
352 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
353 *
354 * \param l4type The layer-4 type (\p IPPROTO_xxx).
355 * \param passive Whether this is a passive (1) or active (0) socket.
356 * \param host Remote or local hostname or IPv/6 address string.
357 * \param port_number Decimal port number.
358 * \param fo Socket options to be set before making the connection.
359 *
360 * This creates a ready-made IPv4/v6 socket structure after looking up the
361 * necessary parameters. The interpretation of \a host depends on the value of
362 * \a passive:
363 * - on a passive socket host is interpreted as an interface IPv4/6 address
364 * (can be left NULL);
365 * - on an active socket, \a host is the peer DNS name or IPv4/6 address
366 * to connect to;
367 * - \a port_number is in either case the numeric port number (not service
368 * string).
369 *
370 * Furthermore, bind(2) is called on passive sockets, and connect(2) on active
371 * sockets. The algorithm tries all possible address combinations until it
372 * succeeds. If \a fo is supplied, options are set and cleanup is performed.
373 *
374 * \return This function returns 1 on success and \a -E_ADDRESS_LOOKUP when no
375 * matching connection could be set up (with details in the error log).
376 *
377 * \sa ipv6(7), getaddrinfo(3), bind(2), connect(2).
378 */
379 int makesock(unsigned l4type, bool passive,
380 const char *host, uint16_t port_number,
381 struct flowopts *fo)
382 {
383 struct addrinfo *local = NULL, *src,
384 *remote = NULL, *dst, hints;
385 unsigned int l3type = AF_UNSPEC;
386 int rc, on = 1, sockfd = -1,
387 socktype = sock_type(l4type);
388 char port[6]; /* port number has at most 5 digits */
389
390 sprintf(port, "%u", port_number);
391 /* Set up address hint structure */
392 memset(&hints, 0, sizeof(hints));
393 hints.ai_family = l3type;
394 hints.ai_socktype = socktype;
395 /*
396 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
397 * (and only then) pretend to be UDP.
398 */
399 if (l4type == IPPROTO_DCCP)
400 hints.ai_socktype = SOCK_DGRAM;
401
402 /* only use addresses available on the host */
403 hints.ai_flags = AI_ADDRCONFIG;
404 if (l3type == AF_INET6)
405 /* use v4-mapped-v6 if no v6 addresses found */
406 hints.ai_flags |= AI_V4MAPPED | AI_ALL;
407
408 if (passive && host == NULL)
409 hints.ai_flags |= AI_PASSIVE;
410
411 /* Obtain local/remote address information */
412 if ((rc = getaddrinfo(host, port, &hints, passive ? &local : &remote))) {
413 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s.\n",
414 layer4_name(l4type),
415 host? host : (passive? "[loopback]" : "[localhost]"),
416 port, gai_strerror(rc));
417 return -E_ADDRESS_LOOKUP;
418 }
419
420 /* Iterate over all src/dst combination, exhausting dst first */
421 for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) {
422 if (src && dst && src->ai_family == AF_INET
423 && dst->ai_family == AF_INET6)
424 goto get_next_dst; /* v4 -> v6 is not possible */
425
426 sockfd = socket(src ? src->ai_family : dst->ai_family,
427 socktype, l4type);
428 if (sockfd < 0)
429 goto get_next_dst;
430
431 /*
432 * Reuse the address on passive sockets to avoid failure on
433 * restart (protocols using listen()) and when creating
434 * multiple listener instances (UDP multicast).
435 */
436 if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
437 &on, sizeof(on)) == -1) {
438 rc = errno;
439 close(sockfd);
440 PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
441 strerror(rc));
442 return -ERRNO_TO_PARA_ERROR(rc);
443 }
444 flowopt_setopts(sockfd, fo);
445
446 if (src) {
447 if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) {
448 close(sockfd);
449 goto get_next_src;
450 }
451 if (!dst) /* bind-only completed successfully */
452 break;
453 }
454
455 if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0)
456 break; /* connection completed successfully */
457 close(sockfd);
458 get_next_dst:
459 if (dst && (dst = dst->ai_next))
460 continue;
461 get_next_src:
462 if (src && (src = src->ai_next)) /* restart inner loop */
463 dst = remote;
464 }
465 if (local)
466 freeaddrinfo(local);
467 if (remote)
468 freeaddrinfo(remote);
469 flowopt_cleanup(fo);
470
471 if (src == NULL && dst == NULL) {
472 rc = errno;
473 PARA_ERROR_LOG("can not create %s socket %s#%s.\n",
474 layer4_name(l4type), host? host : (passive?
475 "[loopback]" : "[localhost]"), port);
476 return -ERRNO_TO_PARA_ERROR(rc);
477 }
478 return sockfd;
479 }
480
481 /**
482 * Create a passive / listening socket.
483 *
484 * \param l4type The transport-layer type (\p IPPROTO_xxx).
485 * \param port The decimal port number to listen on.
486 * \param fo Flowopts (if any) to set before starting to listen.
487 *
488 * \return Positive integer (socket descriptor) on success, negative value
489 * otherwise.
490 *
491 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
492 */
493 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
494 {
495 int ret, fd = makesock(l4type, 1, NULL, port, fo);
496
497 if (fd > 0) {
498 ret = listen(fd, BACKLOG);
499 if (ret < 0) {
500 ret = errno;
501 close(fd);
502 return -ERRNO_TO_PARA_ERROR(ret);
503 }
504 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
505 layer4_name(l4type), port, fd);
506 }
507 return fd;
508 }
509
510 /**
511 * Determine IPv4/v6 socket address length.
512 * \param sa Container of IPv4 or IPv6 address.
513 * \return Address-family dependent address length.
514 */
515 static socklen_t salen(const struct sockaddr *sa)
516 {
517 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
518
519 return sa->sa_family == AF_INET6
520 ? sizeof(struct sockaddr_in6)
521 : sizeof(struct sockaddr_in);
522 }
523
524 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
525 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
526 {
527 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
528
529 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
530 }
531
532 /**
533 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
534 * \param ss Container of IPv4/6 address.
535 * \return Pointer to normalized address (may be static storage).
536 *
537 * \sa RFC 3493
538 */
539 static const struct sockaddr *
540 normalize_ip_address(const struct sockaddr_storage *ss)
541 {
542 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
543
544 if (SS_IS_ADDR_V4MAPPED(ss)) {
545 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
546 static struct sockaddr_in ia;
547
548 ia.sin_family = AF_INET;
549 ia.sin_port = ia6->sin6_port;
550 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
551 return (const struct sockaddr *)&ia;
552 }
553 return (const struct sockaddr *)ss;
554 }
555
556 /**
557 * Generic/fallback MTU values
558 *
559 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
560 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
561 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
562 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
563 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
564 * UDP senders should use the respective minimum values of EMTU_S.
565 */
566 static inline int generic_mtu(const int af_type)
567 {
568 return af_type == AF_INET6 ? 1280 : 576;
569 }
570
571 /** Crude approximation of IP header overhead - neglecting options. */
572 static inline int estimated_header_overhead(const int af_type)
573 {
574 return af_type == AF_INET6 ? 40 : 20;
575 }
576
577 /**
578 * Maximum transport-layer message size (MMS_S) as per RFC 1122, 3.3.3
579 * Socket must be connected.
580 */
581 int generic_max_transport_msg_size(int sockfd)
582 {
583 struct sockaddr_storage ss;
584 socklen_t sslen = sizeof(ss);
585 int af_type = AF_INET;
586
587 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
588 PARA_ERROR_LOG("can not determine remote address type: %s\n",
589 strerror(errno));
590 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
591 af_type = ss.ss_family;
592 }
593 return generic_mtu(af_type) - estimated_header_overhead(af_type);
594 }
595
596 /**
597 * Print numeric host and port number (beware - uses static char).
598 *
599 * \param sa The IPv4/IPv6 socket address to use.
600 *
601 * \sa getnameinfo(3), services(5), nsswitch.conf(5)
602 */
603 static char *host_and_port(const struct sockaddr_storage *ss)
604 {
605 const struct sockaddr *sa = normalize_ip_address(ss);
606 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
607 static char output[sizeof(hbuf) + sizeof(sbuf) + 2];
608 int ret;
609
610 ret = getnameinfo(sa, salen(sa),
611 hbuf, sizeof(hbuf),
612 sbuf, sizeof(sbuf),
613 NI_NUMERICHOST);
614 if (ret == 0) {
615 snprintf(output, sizeof(output), "%s#%s", hbuf, sbuf);
616 } else {
617 snprintf(output, sizeof(output), "(unknown)");
618 PARA_WARNING_LOG("hostname lookup error (%s).\n",
619 gai_strerror(ret));
620 }
621 return output;
622 }
623
624 /**
625 * Look up the local or remote side of a connected socket structure.
626 *
627 * \param fd The socket descriptor of the connected socket.
628 * \param getname Either \p getsockname() for local, or \p getpeername() for
629 * remote side.
630 *
631 * \return A static character string identifying hostname and port of the
632 * chosen side.
633 *
634 * \sa getsockname(2), getpeername(2).
635 */
636 static char *__get_sock_name(int fd, int (*getname)(int, struct sockaddr*,
637 socklen_t *))
638 {
639 struct sockaddr_storage ss;
640 socklen_t sslen = sizeof(ss);
641
642 if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
643 static char *dont_know = "(don't know)";
644 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
645 fd, strerror(errno));
646 return dont_know;
647 }
648 return host_and_port(&ss);
649 }
650
651 /**
652 * Look up the local side of a connected socket structure.
653 *
654 * \param sockfd The file descriptor of the socket.
655 *
656 * \return A pointer to a static buffer containing hostname an port. This
657 * buffer must not be freed by the caller.
658 *
659 * \sa remote_name().
660 */
661 char *local_name(int sockfd)
662 {
663 return __get_sock_name(sockfd, getsockname);
664 }
665
666 /**
667 * Look up the remote side of a connected socket structure.
668 *
669 * \param sockfd The file descriptor of the socket.
670 *
671 * \return Analogous to the return value of \ref local_name() but for the
672 * remote side.
673 *
674 * \sa local_name().
675 */
676 char *remote_name(int sockfd)
677 {
678 return __get_sock_name(sockfd, getpeername);
679 }
680
681 /**
682 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
683 * \param ss Container of IPv4/6 address
684 * \return Extracted IPv4 address (different from 0) or 0 if unsuccessful.
685 *
686 * \sa RFC 3493
687 */
688 struct in_addr extract_v4_addr(const struct sockaddr_storage *ss)
689 {
690 struct in_addr ia = {.s_addr = 0};
691 const struct sockaddr *sa = normalize_ip_address(ss);
692
693 if (sa->sa_family == AF_INET)
694 ia = ((struct sockaddr_in *)sa)->sin_addr;
695 return ia;
696 }
697
698 /**
699 * Send a binary buffer.
700 *
701 * \param fd The file descriptor.
702 * \param buf The buffer to be sent.
703 * \param len The length of \a buf.
704 *
705 * Send out the buffer and try to resend the remaining part in case of short
706 * writes.
707 *
708 * \return Standard.
709 */
710 int send_bin_buffer(int fd, const char *buf, size_t len)
711 {
712 if (!len)
713 PARA_CRIT_LOG("len == 0\n");
714 return write_all(fd, buf, &len);
715 }
716
717 /**
718 * Send a \p NULL-terminated buffer.
719 *
720 * \param fd The file descriptor.
721 * \param buf The null-terminated buffer to be send.
722 *
723 * This is equivalent to send_bin_buffer(fd, buf, strlen(buf)).
724 *
725 * \return Standard.
726 */
727 int send_buffer(int fd, const char *buf)
728 {
729 return send_bin_buffer(fd, buf, strlen(buf));
730 }
731
732 /**
733 * Send a buffer given by a format string.
734 *
735 * \param fd The file descriptor.
736 * \param fmt A format string.
737 *
738 * \return Standard.
739 */
740 __printf_2_3 int send_va_buffer(int fd, const char *fmt, ...)
741 {
742 char *msg;
743 int ret;
744
745 PARA_VSPRINTF(fmt, msg);
746 ret = send_buffer(fd, msg);
747 free(msg);
748 return ret;
749 }
750
751 /**
752 * Receive data from a file descriptor.
753 *
754 * \param fd The file descriptor.
755 * \param buf The buffer to write the data to.
756 * \param size The size of \a buf.
757 *
758 * Receive at most \a size bytes from file descriptor \a fd.
759 *
760 * \return The number of bytes received on success, negative on errors, zero if
761 * the peer has performed an orderly shutdown.
762 *
763 * \sa recv(2).
764 */
765 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
766 {
767 ssize_t n;
768
769 n = recv(fd, buf, size, 0);
770 if (n == -1)
771 return -ERRNO_TO_PARA_ERROR(errno);
772 return n;
773 }
774
775 /**
776 * Receive and write terminating NULL byte.
777 *
778 * \param fd The file descriptor.
779 * \param buf The buffer to write the data to.
780 * \param size The size of \a buf.
781 *
782 * Read at most \a size - 1 bytes from file descriptor \a fd and
783 * write a NULL byte at the end of the received data.
784 *
785 * \return The return value of the underlying call to \a recv_bin_buffer().
786 *
787 * \sa recv_bin_buffer()
788 */
789 int recv_buffer(int fd, char *buf, size_t size)
790 {
791 int n;
792
793 assert(size);
794 n = recv_bin_buffer(fd, buf, size - 1);
795 if (n >= 0)
796 buf[n] = '\0';
797 else
798 *buf = '\0';
799 return n;
800 }
801
802 /**
803 * Wrapper around the accept system call.
804 *
805 * \param fd The listening socket.
806 * \param rfds An optional fd_set pointer.
807 * \param addr Structure which is filled in with the address of the peer socket.
808 * \param size Should contain the size of the structure pointed to by \a addr.
809 * \param new_fd Result pointer.
810 *
811 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
812 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
813 *
814 * \return Negative on errors, zero if no connections are present to be accepted,
815 * one otherwise.
816 *
817 * \sa accept(2).
818 */
819 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
820 {
821 int ret;
822
823 if (rfds && !FD_ISSET(fd, rfds))
824 return 0;
825 do
826 ret = accept(fd, (struct sockaddr *) addr, &size);
827 while (ret < 0 && errno == EINTR);
828
829 if (ret >= 0) {
830 *new_fd = ret;
831 return 1;
832 }
833 if (errno == EAGAIN || errno == EWOULDBLOCK)
834 return 0;
835 return -ERRNO_TO_PARA_ERROR(errno);
836 }
837
838 /**
839 * Probe the list of DCCP CCIDs configured on this host.
840 * \param ccid_array Pointer to return statically allocated array in.
841 * \return Number of elements returned in \a ccid_array or error.
842 *
843 * NB: This feature is only available on Linux > 2.6.30; on older kernels
844 * ENOPROTOOPT ("Protocol not available") will be returned.
845 */
846 int dccp_available_ccids(uint8_t **ccid_array)
847 {
848 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
849 socklen_t nccids = sizeof(ccids);
850 int ret, fd;
851
852 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
853 if (ret < 0)
854 return ret;
855
856 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
857 ccids, &nccids) < 0) {
858 ret = errno;
859 close(fd);
860 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
861 strerror(ret));
862 return -ERRNO_TO_PARA_ERROR(ret);
863 }
864
865 close(fd);
866 *ccid_array = ccids;
867 return nccids;
868 }
869
870 /**
871 * Prepare a structure for \p AF_UNIX socket addresses.
872 *
873 * \param u Pointer to the struct to be prepared.
874 * \param name The socket pathname.
875 *
876 * This just copies \a name to the sun_path component of \a u.
877 *
878 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
879 * than \p UNIX_PATH_MAX.
880 */
881 static int init_unix_addr(struct sockaddr_un *u, const char *name)
882 {
883 if (strlen(name) >= UNIX_PATH_MAX)
884 return -E_NAME_TOO_LONG;
885 memset(u->sun_path, 0, UNIX_PATH_MAX);
886 u->sun_family = PF_UNIX;
887 strcpy(u->sun_path, name);
888 return 1;
889 }
890
891 /**
892 * Prepare, create, and bind a socket for local communication.
893 *
894 * \param name The socket pathname.
895 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
896 * \param mode The desired mode of the socket.
897 *
898 * This function creates a local socket for sequenced, reliable,
899 * two-way, connection-based byte streams.
900 *
901 * \return The file descriptor, on success, negative on errors.
902 *
903 * \sa socket(2)
904 * \sa bind(2)
905 * \sa chmod(2)
906 */
907 int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
908 mode_t mode)
909 {
910 int fd, ret;
911
912 ret = init_unix_addr(unix_addr, name);
913 if (ret < 0)
914 return ret;
915 ret = socket(PF_UNIX, SOCK_STREAM, 0);
916 if (ret < 0)
917 return -ERRNO_TO_PARA_ERROR(errno);
918 fd = ret;
919 ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
920 if (ret < 0) {
921 ret = -ERRNO_TO_PARA_ERROR(errno);
922 goto err;
923 }
924 ret = -E_CHMOD;
925 if (chmod(name, mode) < 0)
926 goto err;
927 return fd;
928 err:
929 close(fd);
930 return ret;
931 }
932
933 /**
934 * Prepare, create, and connect to a Unix domain socket for local communication.
935 *
936 * \param name The socket pathname.
937 *
938 * This function creates a local socket for sequenced, reliable, two-way,
939 * connection-based byte streams.
940 *
941 * \return The file descriptor, on success, negative on errors.
942 *
943 * \sa create_local_socket(), unix(7), connect(2).
944 */
945 int create_remote_socket(const char *name)
946 {
947 struct sockaddr_un unix_addr;
948 int fd, ret;
949
950 ret = init_unix_addr(&unix_addr, name);
951 if (ret < 0)
952 return ret;
953 fd = socket(PF_UNIX, SOCK_STREAM, 0);
954 if (fd < 0)
955 return -ERRNO_TO_PARA_ERROR(errno);
956 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
957 ret = -ERRNO_TO_PARA_ERROR(errno);
958 goto err;
959 }
960 return fd;
961 err:
962 close(fd);
963 return ret;
964 }
965
966 #ifndef HAVE_UCRED
967 ssize_t send_cred_buffer(int sock, char *buf)
968 {
969 return send_buffer(sock, buf);
970 }
971 int recv_cred_buffer(int fd, char *buf, size_t size)
972 {
973 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
974 }
975 #else /* HAVE_UCRED */
976 /**
977 * Send \p NULL-terminated buffer and Unix credentials of the current process.
978 *
979 * \param sock The socket file descriptor.
980 * \param buf The buffer to be sent.
981 *
982 * \return On success, this call returns the number of characters sent. On
983 * error, \p -E_SENDMSG is returned.
984 *
985 * \sa sendmsg(2), okir's Black Hats Manual.
986 */
987 ssize_t send_cred_buffer(int sock, char *buf)
988 {
989 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
990 struct msghdr msg;
991 struct cmsghdr *cmsg;
992 static struct iovec iov;
993 struct ucred c;
994 int ret;
995
996 /* Response data */
997 iov.iov_base = buf;
998 iov.iov_len = strlen(buf);
999 c.pid = getpid();
1000 c.uid = getuid();
1001 c.gid = getgid();
1002 /* compose the message */
1003 memset(&msg, 0, sizeof(msg));
1004 msg.msg_iov = &iov;
1005 msg.msg_iovlen = 1;
1006 msg.msg_control = control;
1007 msg.msg_controllen = sizeof(control);
1008 /* attach the ucred struct */
1009 cmsg = CMSG_FIRSTHDR(&msg);
1010 cmsg->cmsg_level = SOL_SOCKET;
1011 cmsg->cmsg_type = SCM_CREDENTIALS;
1012 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1013 *(struct ucred *)CMSG_DATA(cmsg) = c;
1014 msg.msg_controllen = cmsg->cmsg_len;
1015 ret = sendmsg(sock, &msg, 0);
1016 if (ret < 0)
1017 ret = -E_SENDMSG;
1018 return ret;
1019 }
1020
1021 static void dispose_fds(int *fds, unsigned num)
1022 {
1023 int i;
1024
1025 for (i = 0; i < num; i++)
1026 close(fds[i]);
1027 }
1028
1029 /**
1030 * Receive a buffer and the Unix credentials of the sending process.
1031 *
1032 * \param fd the socket file descriptor.
1033 * \param buf the buffer to store the message.
1034 * \param size the size of \a buffer.
1035 *
1036 * \return negative on errors, the user id on success.
1037 *
1038 * \sa recvmsg(2), okir's Black Hats Manual.
1039 */
1040 int recv_cred_buffer(int fd, char *buf, size_t size)
1041 {
1042 char control[255];
1043 struct msghdr msg;
1044 struct cmsghdr *cmsg;
1045 struct iovec iov;
1046 int result = 0;
1047 int yes = 1;
1048 struct ucred cred;
1049
1050 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1051 memset(&msg, 0, sizeof(msg));
1052 memset(buf, 0, size);
1053 iov.iov_base = buf;
1054 iov.iov_len = size;
1055 msg.msg_iov = &iov;
1056 msg.msg_iovlen = 1;
1057 msg.msg_control = control;
1058 msg.msg_controllen = sizeof(control);
1059 if (recvmsg(fd, &msg, 0) < 0)
1060 return -E_RECVMSG;
1061 result = -E_SCM_CREDENTIALS;
1062 cmsg = CMSG_FIRSTHDR(&msg);
1063 while (cmsg) {
1064 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1065 == SCM_CREDENTIALS) {
1066 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1067 result = cred.uid;
1068 } else
1069 if (cmsg->cmsg_level == SOL_SOCKET
1070 && cmsg->cmsg_type == SCM_RIGHTS) {
1071 dispose_fds((int *) CMSG_DATA(cmsg),
1072 (cmsg->cmsg_len - CMSG_LEN(0))
1073 / sizeof(int));
1074 }
1075 cmsg = CMSG_NXTHDR(&msg, cmsg);
1076 }
1077 return result;
1078 }
1079 #endif /* HAVE_UCRED */