sender: Deplete ACLs on exit.
[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 continue;
430 fd = ret;
431 flowopt_setopts(fd, fo);
432 if (!passive) {
433 if (connect(fd, ai->ai_addr, ai->ai_addrlen) == 0)
434 return fd;
435 close(fd);
436 continue;
437 }
438 /*
439 * Reuse the address on passive sockets to avoid failure on
440 * restart (protocols using listen()) and when creating
441 * multiple listener instances (UDP multicast).
442 */
443 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
444 sizeof(on)) == -1) {
445 close(fd);
446 continue;
447 }
448 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
449 close(fd);
450 continue;
451 }
452 return fd;
453 }
454 return -E_MAKESOCK;
455 }
456
457 /**
458 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
459 *
460 * \param l4type The layer-4 type (\p IPPROTO_xxx).
461 * \param passive Whether this is a passive or active socket.
462 * \param host Passed to \ref lookup_address().
463 * \param port_number Passed to \ref lookup_address().
464 * \param fo Passed to \ref makesock_addrinfo().
465 *
466 * This creates a ready-made IPv4/v6 socket structure after looking up the
467 * necessary parameters. The function first calls \ref lookup_address() and
468 * passes the address information to makesock_addrinfo() to create and
469 * initialize the socket.
470 *
471 * \return The newly created file descriptor on success, a negative error code
472 * on failure.
473 *
474 * \sa \ref lookup_address(), \ref makesock_addrinfo().
475 */
476 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
477 struct flowopts *fo)
478 {
479 struct addrinfo *ai;
480 int ret = lookup_address(l4type, passive, host, port_number, &ai);
481
482 if (ret >= 0)
483 ret = makesock_addrinfo(l4type, passive, ai, fo);
484 if (ai)
485 freeaddrinfo(ai);
486 if (ret < 0) {
487 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
488 layer4_name(l4type), host? host : (passive?
489 "[loopback]" : "[localhost]"), port_number);
490 }
491 return ret;
492 }
493
494 /**
495 * Create a passive / listening socket.
496 *
497 * \param l4type The transport-layer type (\p IPPROTO_xxx).
498 * \param port The decimal port number to listen on.
499 * \param fo Flowopts (if any) to set before starting to listen.
500 *
501 * \return Positive integer (socket descriptor) on success, negative value
502 * otherwise.
503 *
504 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
505 */
506 int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
507 {
508 int ret, fd = makesock(l4type, 1, NULL, port, fo);
509
510 if (fd > 0) {
511 ret = listen(fd, BACKLOG);
512 if (ret < 0) {
513 ret = errno;
514 close(fd);
515 return -ERRNO_TO_PARA_ERROR(ret);
516 }
517 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
518 layer4_name(l4type), port, fd);
519 }
520 return fd;
521 }
522
523 /**
524 * Determine IPv4/v6 socket address length.
525 * \param sa Container of IPv4 or IPv6 address.
526 * \return Address-family dependent address length.
527 */
528 static socklen_t salen(const struct sockaddr *sa)
529 {
530 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
531
532 return sa->sa_family == AF_INET6
533 ? sizeof(struct sockaddr_in6)
534 : sizeof(struct sockaddr_in);
535 }
536
537 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
538 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
539 {
540 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
541
542 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
543 }
544
545 /**
546 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
547 * \param ss Container of IPv4/6 address.
548 * \return Pointer to normalized address (may be static storage).
549 *
550 * \sa RFC 3493.
551 */
552 static const struct sockaddr *
553 normalize_ip_address(const struct sockaddr_storage *ss)
554 {
555 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
556
557 if (SS_IS_ADDR_V4MAPPED(ss)) {
558 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
559 static struct sockaddr_in ia;
560
561 ia.sin_family = AF_INET;
562 ia.sin_port = ia6->sin6_port;
563 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
564 return (const struct sockaddr *)&ia;
565 }
566 return (const struct sockaddr *)ss;
567 }
568
569 /**
570 * Generic/fallback MTU values
571 *
572 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
573 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
574 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
575 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
576 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
577 * UDP senders should use the respective minimum values of EMTU_S.
578 */
579 static inline int generic_mtu(const int af_type)
580 {
581 return af_type == AF_INET6 ? 1280 : 576;
582 }
583
584 /** Crude approximation of IP header overhead - neglecting options. */
585 static inline int estimated_header_overhead(const int af_type)
586 {
587 return af_type == AF_INET6 ? 40 : 20;
588 }
589
590 /**
591 * Get the maximum transport-layer message size (MMS_S).
592 *
593 * \param sockfd The socket file descriptor.
594 *
595 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
596 * could not be determined, \p AF_INET is assumed.
597 *
598 * \return The maximum message size of the address family type.
599 */
600 int generic_max_transport_msg_size(int sockfd)
601 {
602 struct sockaddr_storage ss = {.ss_family = 0};
603 socklen_t sslen = sizeof(ss);
604 int af_type = AF_INET;
605
606 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
607 PARA_ERROR_LOG("can not determine remote address type: %s\n",
608 strerror(errno));
609 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
610 af_type = ss.ss_family;
611 }
612 return generic_mtu(af_type) - estimated_header_overhead(af_type);
613 }
614
615 /**
616 * Look up the remote side of a connected socket structure.
617 *
618 * \param fd The socket descriptor of the connected socket.
619 *
620 * \return A static character string identifying hostname and port of the
621 * chosen side in numeric host:port format.
622 *
623 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
624 * services(5), nsswitch.conf(5).
625 */
626 char *remote_name(int fd)
627 {
628 struct sockaddr_storage ss = {.ss_family = 0};
629 const struct sockaddr *sa;
630 socklen_t sslen = sizeof(ss);
631 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
632 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
633 int ret;
634
635 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
636 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
637 fd, strerror(errno));
638 snprintf(output, sizeof(output), "(unknown)");
639 return output;
640 }
641 sa = normalize_ip_address(&ss);
642 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
643 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
644 if (ret) {
645 PARA_WARNING_LOG("hostname lookup error (%s).\n",
646 gai_strerror(ret));
647 snprintf(output, sizeof(output), "(lookup error)");
648 } else if (sa->sa_family == AF_INET6)
649 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
650 else
651 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
652 return output;
653 }
654
655 /**
656 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
657 *
658 * \param ss Container of IPv4/6 address.
659 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
660 *
661 * \sa RFC 3493.
662 */
663 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
664 {
665 const struct sockaddr *sa = normalize_ip_address(ss);
666
667 memset(ia, 0, sizeof(*ia));
668 if (sa->sa_family == AF_INET)
669 *ia = ((struct sockaddr_in *)sa)->sin_addr;
670 }
671
672 /**
673 * Compare the address part of IPv4/6 addresses.
674 *
675 * \param sa1 First address.
676 * \param sa2 Second address.
677 *
678 * \return True iff the IP address of \a sa1 and \a sa2 match.
679 */
680 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
681 {
682 if (!sa1 || !sa2)
683 return false;
684 if (sa1->sa_family != sa2->sa_family)
685 return false;
686 if (sa1->sa_family == AF_INET) {
687 struct sockaddr_in *a1 = (typeof(a1))sa1,
688 *a2 = (typeof (a2))sa2;
689 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
690 } else if (sa1->sa_family == AF_INET6) {
691 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
692 *a2 = (typeof (a2))sa2;
693 return !memcmp(a1, a2, sizeof(*a1));
694 } else
695 return false;
696 }
697
698 /**
699 * Receive data from a file descriptor.
700 *
701 * \param fd The file descriptor.
702 * \param buf The buffer to write the data to.
703 * \param size The size of \a buf.
704 *
705 * Receive at most \a size bytes from file descriptor \a fd.
706 *
707 * \return The number of bytes received on success, negative on errors, zero if
708 * the peer has performed an orderly shutdown.
709 *
710 * \sa recv(2).
711 */
712 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
713 {
714 ssize_t n;
715
716 n = recv(fd, buf, size, 0);
717 if (n == -1)
718 return -ERRNO_TO_PARA_ERROR(errno);
719 return n;
720 }
721
722 /**
723 * Receive and write terminating NULL byte.
724 *
725 * \param fd The file descriptor.
726 * \param buf The buffer to write the data to.
727 * \param size The size of \a buf.
728 *
729 * Read at most \a size - 1 bytes from file descriptor \a fd and
730 * write a NULL byte at the end of the received data.
731 *
732 * \return The return value of the underlying call to \a recv_bin_buffer().
733 *
734 * \sa \ref recv_bin_buffer()
735 */
736 int recv_buffer(int fd, char *buf, size_t size)
737 {
738 int n;
739
740 assert(size);
741 n = recv_bin_buffer(fd, buf, size - 1);
742 if (n >= 0)
743 buf[n] = '\0';
744 else
745 *buf = '\0';
746 return n;
747 }
748
749 /**
750 * Wrapper around the accept system call.
751 *
752 * \param fd The listening socket.
753 * \param rfds An optional fd_set pointer.
754 * \param addr Structure which is filled in with the address of the peer socket.
755 * \param size Should contain the size of the structure pointed to by \a addr.
756 * \param new_fd Result pointer.
757 *
758 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
759 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
760 *
761 * \return Negative on errors, zero if no connections are present to be accepted,
762 * one otherwise.
763 *
764 * \sa accept(2).
765 */
766 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
767 {
768 int ret;
769
770 if (rfds && !FD_ISSET(fd, rfds))
771 return 0;
772 do
773 ret = accept(fd, (struct sockaddr *) addr, &size);
774 while (ret < 0 && errno == EINTR);
775
776 if (ret >= 0) {
777 *new_fd = ret;
778 return 1;
779 }
780 if (errno == EAGAIN || errno == EWOULDBLOCK)
781 return 0;
782 return -ERRNO_TO_PARA_ERROR(errno);
783 }
784
785 /**
786 * Probe the list of DCCP CCIDs configured on this host.
787 * \param ccid_array Pointer to return statically allocated array in.
788 * \return Number of elements returned in \a ccid_array or error.
789 *
790 * NB: This feature is only available on Linux > 2.6.30; on older kernels
791 * ENOPROTOOPT ("Protocol not available") will be returned.
792 */
793 int dccp_available_ccids(uint8_t **ccid_array)
794 {
795 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
796 socklen_t nccids = sizeof(ccids);
797 int ret, fd;
798
799 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
800 if (ret < 0)
801 return ret;
802
803 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
804 ccids, &nccids) < 0) {
805 ret = errno;
806 close(fd);
807 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
808 strerror(ret));
809 return -ERRNO_TO_PARA_ERROR(ret);
810 }
811
812 close(fd);
813 *ccid_array = ccids;
814 return nccids;
815 }
816
817 /*
818 * Prepare a structure for AF_UNIX socket addresses.
819 *
820 * This just copies name to the sun_path component of u, prepending a zero byte
821 * if abstract sockets are supported.
822 *
823 * The first call to this function tries to bind a socket to the abstract name
824 * space. The result of this test is stored in a static variable. Subsequent
825 * calls read this variable and create abstract sockets on systems that support
826 * them.
827 */
828 static int init_unix_addr(struct sockaddr_un *u, const char *name)
829 {
830 static int use_abstract;
831
832 if (strlen(name) + 1 >= UNIX_PATH_MAX)
833 return -E_NAME_TOO_LONG;
834 memset(u->sun_path, 0, UNIX_PATH_MAX);
835 u->sun_family = PF_UNIX;
836 if (use_abstract == 0) { /* executed only once */
837 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
838 memcpy(u->sun_path, "\0x\0", 3);
839 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
840 use_abstract = 1; /* yes */
841 else
842 use_abstract = -1; /* no */
843 close(fd);
844 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
845 use_abstract == 1? "" : "not ");
846 }
847 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
848 return 1;
849 }
850
851 /**
852 * Create a socket for local communication and listen on it.
853 *
854 * \param name The socket pathname.
855 *
856 * This function creates a passive local socket for sequenced, reliable,
857 * two-way, connection-based byte streams. The socket file descriptor is set to
858 * nonblocking mode and listen(2) is called to prepare the socket for
859 * accepting incoming connection requests.
860 *
861 * \return The file descriptor on success, negative error code on failure.
862 *
863 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
864 */
865 int create_local_socket(const char *name)
866 {
867 struct sockaddr_un unix_addr;
868 int fd, ret;
869
870 ret = init_unix_addr(&unix_addr, name);
871 if (ret < 0)
872 return ret;
873 ret = socket(PF_UNIX, SOCK_STREAM, 0);
874 if (ret < 0)
875 return -ERRNO_TO_PARA_ERROR(errno);
876 fd = ret;
877 ret = mark_fd_nonblocking(fd);
878 if (ret < 0)
879 goto err;
880 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
881 if (ret < 0) {
882 ret = -ERRNO_TO_PARA_ERROR(errno);
883 goto err;
884 }
885 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
886 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
887 | S_IROTH | S_IWOTH;
888 ret = -E_CHMOD;
889 if (chmod(name, mode) < 0)
890 goto err;
891 }
892 if (listen(fd , 5) < 0) {
893 ret = -ERRNO_TO_PARA_ERROR(errno);
894 goto err;
895 }
896 return fd;
897 err:
898 close(fd);
899 return ret;
900 }
901
902 /**
903 * Prepare, create, and connect to a Unix domain socket for local communication.
904 *
905 * \param name The socket pathname.
906 *
907 * This function creates a local socket for sequenced, reliable, two-way,
908 * connection-based byte streams.
909 *
910 * \return The file descriptor of the connected socket on success, negative on
911 * errors.
912 *
913 * \sa \ref create_local_socket(), unix(7), connect(2).
914 */
915 int connect_local_socket(const char *name)
916 {
917 struct sockaddr_un unix_addr;
918 int fd, ret;
919
920 PARA_DEBUG_LOG("connecting to %s\n", name);
921 fd = socket(PF_UNIX, SOCK_STREAM, 0);
922 if (fd < 0)
923 return -ERRNO_TO_PARA_ERROR(errno);
924 ret = init_unix_addr(&unix_addr, name);
925 if (ret < 0)
926 goto err;
927 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
928 return fd;
929 ret = -ERRNO_TO_PARA_ERROR(errno);
930 err:
931 close(fd);
932 return ret;
933 }
934
935 #ifndef HAVE_UCRED
936 ssize_t send_cred_buffer(int sock, char *buf)
937 {
938 return write_buffer(sock, buf);
939 }
940 int recv_cred_buffer(int fd, char *buf, size_t size)
941 {
942 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
943 }
944 #else /* HAVE_UCRED */
945
946 /**
947 * Send a buffer and the credentials of the current process to a socket.
948 *
949 * \param sock The file descriptor of the sending socket.
950 * \param buf The zero-terminated buffer to send.
951 *
952 * \return On success, this call returns the number of bytes sent. On errors,
953 * \p -E_SENDMSG is returned.
954 *
955 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
956 */
957 ssize_t send_cred_buffer(int sock, char *buf)
958 {
959 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
960 struct msghdr msg;
961 struct cmsghdr *cmsg;
962 static struct iovec iov;
963 struct ucred c;
964 int ret;
965
966 /* Response data */
967 iov.iov_base = buf;
968 iov.iov_len = strlen(buf);
969 c.pid = getpid();
970 c.uid = getuid();
971 c.gid = getgid();
972 /* compose the message */
973 memset(&msg, 0, sizeof(msg));
974 msg.msg_iov = &iov;
975 msg.msg_iovlen = 1;
976 msg.msg_control = control;
977 msg.msg_controllen = sizeof(control);
978 /* attach the ucred struct */
979 cmsg = CMSG_FIRSTHDR(&msg);
980 cmsg->cmsg_level = SOL_SOCKET;
981 cmsg->cmsg_type = SCM_CREDENTIALS;
982 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
983 *(struct ucred *)CMSG_DATA(cmsg) = c;
984 msg.msg_controllen = cmsg->cmsg_len;
985 ret = sendmsg(sock, &msg, 0);
986 if (ret < 0)
987 ret = -E_SENDMSG;
988 return ret;
989 }
990
991 static void dispose_fds(int *fds, unsigned num)
992 {
993 int i;
994
995 for (i = 0; i < num; i++)
996 close(fds[i]);
997 }
998
999 /**
1000 * Receive a buffer and the Unix credentials of the sending process.
1001 *
1002 * \param fd The file descriptor of the receiving socket.
1003 * \param buf The buffer to store the received message.
1004 * \param size The length of \a buf in bytes.
1005 *
1006 * \return Negative on errors, the user id of the sending process on success.
1007 *
1008 * \sa \ref send_cred_buffer and the references given there.
1009 */
1010 int recv_cred_buffer(int fd, char *buf, size_t size)
1011 {
1012 char control[255] __a_aligned(8);
1013 struct msghdr msg;
1014 struct cmsghdr *cmsg;
1015 struct iovec iov;
1016 int result = 0;
1017 int yes = 1;
1018 struct ucred cred;
1019
1020 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1021 memset(&msg, 0, sizeof(msg));
1022 memset(buf, 0, size);
1023 iov.iov_base = buf;
1024 iov.iov_len = size;
1025 msg.msg_iov = &iov;
1026 msg.msg_iovlen = 1;
1027 msg.msg_control = control;
1028 msg.msg_controllen = sizeof(control);
1029 if (recvmsg(fd, &msg, 0) < 0)
1030 return -E_RECVMSG;
1031 result = -E_SCM_CREDENTIALS;
1032 cmsg = CMSG_FIRSTHDR(&msg);
1033 while (cmsg) {
1034 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1035 == SCM_CREDENTIALS) {
1036 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1037 result = cred.uid;
1038 } else
1039 if (cmsg->cmsg_level == SOL_SOCKET
1040 && cmsg->cmsg_type == SCM_RIGHTS) {
1041 dispose_fds((int *)CMSG_DATA(cmsg),
1042 (cmsg->cmsg_len - CMSG_LEN(0))
1043 / sizeof(int));
1044 }
1045 cmsg = CMSG_NXTHDR(&msg, cmsg);
1046 }
1047 return result;
1048 }
1049 #endif /* HAVE_UCRED */