1 /* Copyright (C) 2005 Andre Noll <maan@tuebingen.mpg.de>, see file COPYING. */
3 /** \file net.c Networking-related helper functions. */
7 #include <netinet/in.h>
10 #include <sys/types.h>
11 #include <sys/socket.h>
22 * Parse and validate IPv4 address/netmask string.
24 * \param cidr Address in CIDR notation
25 * \param addr Copy of the IPv4 address part of \a cidr
26 * \param addrlen Size of \a addr in bytes
27 * \param netmask Value of the netmask part in \a cidr or the
28 * default of 32 if not specified.
30 * \return Pointer to \a addr if successful, NULL on error.
33 char *parse_cidr(const char *cidr,
34 char *addr, ssize_t addrlen,
38 char *c = addr, *end = c + (addrlen - 1);
42 if (cidr == NULL || addrlen < 1)
45 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
50 if (para_atoi32(++o, netmask) < 0 ||
51 *netmask < 0 || *netmask > 0x20)
54 if (is_valid_ipv4_address(addr))
63 * Match string as a candidate IPv4 address.
65 * \param address The string to match.
66 * \return True if \a address has "dot-quad" format.
68 static bool is_v4_dot_quad(const char *address)
73 assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
74 REG_EXTENDED | REG_NOSUB) >= 0);
75 result = regexec(&r, address, 0, NULL, 0) == 0;
81 * Perform basic syntax checking on the host-part of an URL:
83 * - Since ':' is invalid in IPv4 addresses and DNS names, the
84 * presence of ':' causes interpretation as IPv6 address;
85 * - next the first-match-wins algorithm from RFC 3986 is applied;
86 * - else the string is considered as DNS name, to be resolved later.
88 * \param host The host string to check.
89 * \return True if \a host passes the syntax checks.
91 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5.
93 static bool host_string_ok(const char *host)
95 if (host == NULL || *host == '\0')
97 if (strchr(host, ':') != NULL)
98 return is_valid_ipv6_address(host);
99 if (is_v4_dot_quad(host))
100 return is_valid_ipv4_address(host);
105 * Parse and validate URL string.
107 * The URL syntax is loosely based on RFC 3986, supporting one of
108 * - "["host"]"[:port] for native IPv6 addresses and
109 * - host[:port] for IPv4 hostnames and DNS names.
111 * Native IPv6 addresses must be enclosed in square brackets, since
112 * otherwise there is an ambiguity with the port separator `:'.
113 * The 'port' part is always considered to be a number; if absent,
114 * it is set to -1, to indicate that a default port is to be used.
116 * The following are valid examples:
124 * \param url The URL string to take apart.
125 * \param host To return the copied host part of \a url.
126 * \param hostlen The maximum length of \a host.
127 * \param port To return the port number (if any) of \a url.
129 * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned,
130 * \a host and \a port are undefined. If no port number was present in \a url,
131 * \a port is set to -1.
133 * \sa RFC 3986, 3.2.2/3.2.3.
135 char *parse_url(const char *url,
136 char *host, ssize_t hostlen,
140 char *c = host, *end = c + (hostlen - 1);
144 if (o == NULL || hostlen < 1)
148 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
152 if (*o++ != ']' || (*o != '\0' && *o != ':'))
155 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
156 if (c == end && o[1])
162 if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff)
164 if (host_string_ok(host))
172 * Pretty-print a host/port pair.
174 * \param url NULL, or any string accepted by \ref parse_url().
175 * \param default_port Applies if url has no port.
177 * If the url argument is NULL, the function returns the string
178 * 0.0.0.0:default_port. Otherwise it calls \ref parse_url() to check the
179 * syntax of the input string given by url. On errors the string "?" is
180 * returned. Otherwise, if url contains a port, a copy of url is returned. If
181 * no port was supplied, a colon and the default port are appended to url.
183 * \return In all cases the returned string is a allocated with malloc(3) and
184 * has to be freed by the caller.
186 char *format_url(const char *url, int default_port)
188 char host[MAX_HOSTLEN];
192 return make_message("0.0.0.0:%d", default_port);
193 if (!parse_url(url, host, sizeof(host), &url_port))
194 return make_message("?");
196 return make_message("%s:%d", url, default_port);
198 return para_strdup(url);
202 * Stringify port number, resolve into service name where defined.
204 * \param port 2-byte port number, in host-byte-order.
205 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
206 * \return Pointer to static result buffer.
208 * \sa getservent(3), services(5), nsswitch.conf(5).
210 const char *stringify_port(int port, const char *transport)
212 static char service[NI_MAXSERV];
214 if (port < 0 || port > 0xFFFF) {
215 snprintf(service, sizeof(service), "undefined (%d)", port);
217 struct servent *se = getservbyport(htons(port), transport);
220 snprintf(service, sizeof(service), "%d", port);
222 snprintf(service, sizeof(service), "%s", se->s_name);
228 #define SOCK_DCCP 6 /**< Linux socket type. */
232 * Determine the socket type for a given layer-4 protocol.
234 * \param l4type The symbolic name of the transport-layer protocol.
236 * \sa ip(7), socket(2).
238 static inline int sock_type(const unsigned l4type)
241 case IPPROTO_UDP: return SOCK_DGRAM;
242 case IPPROTO_TCP: return SOCK_STREAM;
243 case IPPROTO_DCCP: return SOCK_DCCP;
245 return -1; /* not supported here */
249 * Pretty-print transport-layer name.
251 static const char *layer4_name(const unsigned l4type)
254 case IPPROTO_UDP: return "UDP";
255 case IPPROTO_TCP: return "TCP";
256 case IPPROTO_DCCP: return "DCCP";
258 return "UNKNOWN PROTOCOL";
262 * Flowopts: Transport-layer independent encapsulation of socket options.
264 * These collect individual socket options into a queue, which is disposed of
265 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
266 * only and should not be visible elsewhere.
268 * \sa setsockopt(2), \ref makesock().
270 struct pre_conn_opt {
271 int sock_level; /**< Second argument to setsockopt() */
272 int sock_option; /**< Third argument to setsockopt() */
273 char *opt_name; /**< Stringified \a sock_option */
274 void *opt_val; /**< Fourth argument to setsockopt() */
275 socklen_t opt_len; /**< Fifth argument to setsockopt() */
277 struct list_head node; /**< FIFO, as sockopt order matters. */
281 * List of pre-connection socket options to be set.
283 * This list contains transport-layer independent encapsulation of socket
284 * options that need to be registered prior to setting up a connection.
287 struct list_head sockopts;
291 * Allocate and initialize a flowopt queue.
293 * \return A new structure to be passed to \ref flowopt_add(). It is
294 * automatically deallocated in \ref makesock().
296 struct flowopts *flowopt_new(void)
298 struct flowopts *new = alloc(sizeof(*new));
300 init_list_head(&new->sockopts);
305 * Append new socket option to flowopt queue.
307 * \param fo The flowopt queue to append to.
308 * \param lev Level at which \a opt resides.
309 * \param opt New option to add.
310 * \param name Stringified name of \a opt.
311 * \param val The value to set \a opt to.
312 * \param len Length of \a val.
316 void flowopt_add(struct flowopts *fo, int lev, int opt,
317 const char *name, const void *val, int len)
319 struct pre_conn_opt *new = alloc(sizeof(*new));
321 new->sock_option = opt;
322 new->sock_level = lev;
323 new->opt_name = para_strdup(name);
329 new->opt_val = alloc(len);
331 memcpy(new->opt_val, val, len);
334 list_add_tail(&new->node, &fo->sockopts);
337 /** Set the entire bunch of pre-connection options at once. */
338 static void flowopt_setopts(int sockfd, struct flowopts *fo)
340 struct pre_conn_opt *pc;
345 list_for_each_entry(pc, &fo->sockopts, node)
346 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
347 pc->opt_val, pc->opt_len) < 0) {
348 PARA_EMERG_LOG("Can not set %s socket option: %s",
349 pc->opt_name, strerror(errno));
355 * Deallocate all resources of a flowopts structure.
357 * \param fo A pointer as returned from flowopt_new().
359 * It's OK to pass \p NULL here in which case the function does nothing.
361 void flowopt_cleanup(struct flowopts *fo)
363 struct pre_conn_opt *cur, *next;
368 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
377 * Resolve an IPv4/IPv6 address.
379 * \param l4type The layer-4 type (\p IPPROTO_xxx).
380 * \param passive Whether \p AI_PASSIVE should be included as hint.
381 * \param host Remote or local hostname or IPv/6 address string.
382 * \param port_number Used to set the port in each returned address structure.
383 * \param result addrinfo structures are returned here.
385 * The interpretation of \a host depends on the value of \a passive. On a
386 * passive socket host is interpreted as an interface IPv4/6 address (can be
387 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
388 * address to connect to.
392 * \sa getaddrinfo(3).
394 int lookup_address(unsigned l4type, bool passive, const char *host,
395 int port_number, struct addrinfo **result)
398 char port[6]; /* port number has at most 5 digits */
399 struct addrinfo *addr = NULL, hints;
402 sprintf(port, "%d", port_number & 0xffff);
403 /* Set up address hint structure */
404 memset(&hints, 0, sizeof(hints));
405 hints.ai_family = AF_UNSPEC;
406 hints.ai_socktype = sock_type(l4type);
408 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
409 * (and only then) pretend to be UDP.
411 if (l4type == IPPROTO_DCCP)
412 hints.ai_socktype = SOCK_DGRAM;
413 /* only use addresses available on the host */
414 hints.ai_flags = AI_ADDRCONFIG;
415 if (passive && host == NULL)
416 hints.ai_flags |= AI_PASSIVE;
417 /* Obtain local/remote address information */
418 ret = getaddrinfo(host, port, &hints, &addr);
420 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
422 host? host : (passive? "[loopback]" : "[localhost]"),
423 port, gai_strerror(ret));
424 return -E_ADDRESS_LOOKUP;
431 * Create an active or passive socket.
433 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
434 * \param passive Whether to call bind(2) or connect(2).
435 * \param ai Address information as obtained from \ref lookup_address().
436 * \param fo Socket options to be set before making the connection.
438 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
439 * algorithm tries all possible address combinations until it succeeds. If \a
440 * fo is supplied, options are set but cleanup must be performed in the caller.
442 * \return File descriptor on success, \p E_MAKESOCK on errors.
444 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
447 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
450 int ret = -E_MAKESOCK, on = 1;
452 for (; ai; ai = ai->ai_next) {
454 ret = socket(ai->ai_family, sock_type(l4type), l4type);
456 PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
460 flowopt_setopts(fd, fo);
462 if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
463 PARA_NOTICE_LOG("connect(): %s\n",
471 * Reuse the address on passive sockets to avoid failure on
472 * restart (protocols using listen()) and when creating
473 * multiple listener instances (UDP multicast).
475 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
477 PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
481 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
482 PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
492 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
494 * \param l4type The layer-4 type (\p IPPROTO_xxx).
495 * \param passive Whether this is a passive or active socket.
496 * \param host Passed to \ref lookup_address().
497 * \param port_number Passed to \ref lookup_address().
498 * \param fo Passed to \ref makesock_addrinfo().
500 * This creates a ready-made IPv4/v6 socket structure after looking up the
501 * necessary parameters. The function first calls \ref lookup_address() and
502 * passes the address information to makesock_addrinfo() to create and
503 * initialize the socket.
505 * \return The newly created file descriptor on success, a negative error code
508 * \sa \ref lookup_address(), \ref makesock_addrinfo().
510 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
514 int ret = lookup_address(l4type, passive, host, port_number, &ai);
517 ret = makesock_addrinfo(l4type, passive, ai, fo);
521 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
522 layer4_name(l4type), host? host : (passive?
523 "[loopback]" : "[localhost]"), port_number);
529 * Create a passive / listening socket.
531 * \param l4type The transport-layer type (\p IPPROTO_xxx).
532 * \param addr Passed to \ref parse_url() if not NULL.
533 * \param port Ignored if addr contains a port number.
535 * \return Positive integer (socket descriptor) on success, negative value
538 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
540 int para_listen(unsigned l4type, const char *addr, uint16_t port)
542 char host[MAX_HOSTLEN];
543 int ret, fd, addr_port;
546 if (!parse_url(addr, host, sizeof(host), &addr_port))
547 return -ERRNO_TO_PARA_ERROR(EINVAL);
552 fd = makesock(l4type, true /* passive */, addr, port,
553 NULL /* no flowopts */);
555 ret = listen(fd, BACKLOG);
559 return -ERRNO_TO_PARA_ERROR(ret);
561 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
562 layer4_name(l4type), port, fd);
568 * Create a socket which listens on all network addresses.
570 * \param l4type See \ref para_listen().
571 * \param port See \ref para_listen().
573 * This is a simple wrapper for \ref para_listen() which passes a NULL pointer
574 * as the address information.
576 * \return See \ref para_listen().
578 int para_listen_simple(unsigned l4type, uint16_t port)
580 return para_listen(l4type, NULL, port);
584 * Determine IPv4/v6 socket address length.
585 * \param sa Container of IPv4 or IPv6 address.
586 * \return Address-family dependent address length.
588 static socklen_t salen(const struct sockaddr *sa)
590 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
592 return sa->sa_family == AF_INET6
593 ? sizeof(struct sockaddr_in6)
594 : sizeof(struct sockaddr_in);
597 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
598 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
600 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
602 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
606 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
607 * \param ss Container of IPv4/6 address.
608 * \return Pointer to normalized address (may be static storage).
612 static const struct sockaddr *
613 normalize_ip_address(const struct sockaddr_storage *ss)
615 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
617 if (SS_IS_ADDR_V4MAPPED(ss)) {
618 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
619 static struct sockaddr_in ia;
621 ia.sin_family = AF_INET;
622 ia.sin_port = ia6->sin6_port;
623 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
624 return (const struct sockaddr *)&ia;
626 return (const struct sockaddr *)ss;
630 * Generic/fallback MTU values
632 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
633 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
634 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
635 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
636 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
637 * UDP senders should use the respective minimum values of EMTU_S.
639 static inline int generic_mtu(const int af_type)
641 return af_type == AF_INET6 ? 1280 : 576;
644 /** Crude approximation of IP header overhead - neglecting options. */
645 static inline int estimated_header_overhead(const int af_type)
647 return af_type == AF_INET6 ? 40 : 20;
651 * Get the maximum transport-layer message size (MMS_S).
653 * \param sockfd The socket file descriptor.
655 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
656 * could not be determined, \p AF_INET is assumed.
658 * \return The maximum message size of the address family type.
660 int generic_max_transport_msg_size(int sockfd)
662 struct sockaddr_storage ss = {.ss_family = 0};
663 socklen_t sslen = sizeof(ss);
664 int af_type = AF_INET;
666 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
667 PARA_ERROR_LOG("can not determine remote address type: %s\n",
669 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
670 af_type = ss.ss_family;
672 return generic_mtu(af_type) - estimated_header_overhead(af_type);
676 * Look up the remote side of a connected socket structure.
678 * \param fd The socket descriptor of the connected socket.
680 * \return A static character string identifying hostname and port of the
681 * chosen side in numeric host:port format.
683 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
684 * services(5), nsswitch.conf(5).
686 char *remote_name(int fd)
688 struct sockaddr_storage ss = {.ss_family = 0};
689 const struct sockaddr *sa;
690 socklen_t sslen = sizeof(ss);
691 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
692 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
695 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
696 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
697 fd, strerror(errno));
698 snprintf(output, sizeof(output), "(unknown)");
701 sa = normalize_ip_address(&ss);
702 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
703 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
705 PARA_WARNING_LOG("hostname lookup error (%s).\n",
707 snprintf(output, sizeof(output), "(lookup error)");
708 } else if (sa->sa_family == AF_INET6)
709 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
711 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
716 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
718 * \param ss Container of IPv4/6 address.
719 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
723 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
725 const struct sockaddr *sa = normalize_ip_address(ss);
727 memset(ia, 0, sizeof(*ia));
728 if (sa->sa_family == AF_INET)
729 *ia = ((struct sockaddr_in *)sa)->sin_addr;
733 * Compare the address part of IPv4/6 addresses.
735 * \param sa1 First address.
736 * \param sa2 Second address.
738 * \return True iff the IP address of \a sa1 and \a sa2 match.
740 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
744 if (sa1->sa_family != sa2->sa_family)
746 if (sa1->sa_family == AF_INET) {
747 struct sockaddr_in *a1 = (typeof(a1))sa1,
748 *a2 = (typeof (a2))sa2;
749 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
750 } else if (sa1->sa_family == AF_INET6) {
751 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
752 *a2 = (typeof (a2))sa2;
753 return !memcmp(a1, a2, sizeof(*a1));
759 * Receive data from a file descriptor.
761 * \param fd The file descriptor.
762 * \param buf The buffer to write the data to.
763 * \param size The size of \a buf.
765 * Receive at most \a size bytes from file descriptor \a fd.
767 * \return The number of bytes received on success, negative on errors, zero if
768 * the peer has performed an orderly shutdown.
772 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
776 n = recv(fd, buf, size, 0);
778 return -ERRNO_TO_PARA_ERROR(errno);
783 * Receive and write terminating NULL byte.
785 * \param fd The file descriptor.
786 * \param buf The buffer to write the data to.
787 * \param size The size of \a buf.
789 * Read at most \a size - 1 bytes from file descriptor \a fd and
790 * write a NULL byte at the end of the received data.
792 * \return The return value of the underlying call to \a recv_bin_buffer().
794 * \sa \ref recv_bin_buffer()
796 int recv_buffer(int fd, char *buf, size_t size)
801 n = recv_bin_buffer(fd, buf, size - 1);
810 * Wrapper around the accept system call.
812 * \param fd The listening socket.
813 * \param addr Structure which is filled in with the address of the peer socket.
814 * \param size Should contain the size of the structure pointed to by \a addr.
815 * \param new_fd Result pointer.
817 * Accept incoming connections on addr, retry if interrupted.
819 * \return Negative on errors, zero if no connections are present to be accepted,
824 int para_accept(int fd, void *addr, socklen_t size, int *new_fd)
829 ret = accept(fd, (struct sockaddr *) addr, &size);
830 while (ret < 0 && errno == EINTR);
836 if (errno == EAGAIN || errno == EWOULDBLOCK)
838 return -ERRNO_TO_PARA_ERROR(errno);
841 #ifndef DCCP_SOCKOPT_AVAILABLE_CCIDS
842 #define DCCP_SOCKOPT_AVAILABLE_CCIDS 12 /**< List of supported CCIDs. */
846 * Probe the list of DCCP CCIDs configured on this host.
847 * \param ccid_array Pointer to return statically allocated array in.
848 * \return Number of elements returned in \a ccid_array or error.
850 * NB: This feature is only available on Linux > 2.6.30; on older kernels
851 * ENOPROTOOPT ("Protocol not available") will be returned.
853 int dccp_available_ccids(uint8_t **ccid_array)
855 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
856 socklen_t nccids = sizeof(ccids);
859 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
863 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
864 ccids, &nccids) < 0) {
867 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
869 return -ERRNO_TO_PARA_ERROR(ret);
878 * The buffer size of the sun_path component of struct sockaddr_un.
880 * While glibc doesn't define UNIX_PATH_MAX, it documents it has being limited
881 * to 108 bytes. On NetBSD it is only 104 bytes though. We trust UNIX_PATH_MAX
882 * if it is defined and use the size of the ->sun_path member otherwise. This
883 * should be safe everywhere.
885 #ifndef UNIX_PATH_MAX
886 #define UNIX_PATH_MAX (sizeof(((struct sockaddr_un *)0)->sun_path))
890 * Prepare a structure for AF_UNIX socket addresses.
892 * This just copies name to the sun_path component of u, prepending a zero byte
893 * if abstract sockets are supported.
895 * The first call to this function tries to bind a socket to the abstract name
896 * space. The result of this test is stored in a static variable. Subsequent
897 * calls read this variable and create abstract sockets on systems that support
898 * them. If a NULL pointer is passed as the name, the function only
899 * initializes the static variable.
901 static int init_unix_addr(struct sockaddr_un *u, const char *name)
903 static int use_abstract;
905 memset(u->sun_path, 0, UNIX_PATH_MAX);
906 u->sun_family = PF_UNIX;
907 if (use_abstract == 0) { /* executed only once */
908 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
909 memcpy(u->sun_path, "\0x\0", 3);
910 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
911 use_abstract = 1; /* yes */
913 use_abstract = -1; /* no */
915 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
916 use_abstract == 1? "" : "not ");
920 if (strlen(name) + 1 >= UNIX_PATH_MAX)
921 return -E_NAME_TOO_LONG;
922 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
927 * Create a socket for local communication and listen on it.
929 * \param name The socket pathname.
931 * This function creates a passive local socket for sequenced, reliable,
932 * two-way, connection-based byte streams. The socket file descriptor is set to
933 * nonblocking mode and listen(2) is called to prepare the socket for
934 * accepting incoming connection requests.
936 * \return The file descriptor on success, negative error code on failure.
938 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
940 int create_local_socket(const char *name)
942 struct sockaddr_un unix_addr;
945 ret = init_unix_addr(&unix_addr, name);
946 if (ret <= 0) /* error, or name was NULL */
948 ret = socket(PF_UNIX, SOCK_STREAM, 0);
950 return -ERRNO_TO_PARA_ERROR(errno);
952 ret = mark_fd_nonblocking(fd);
955 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
957 ret = -ERRNO_TO_PARA_ERROR(errno);
960 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
961 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
964 if (chmod(name, mode) < 0)
967 if (listen(fd , 5) < 0) {
968 ret = -ERRNO_TO_PARA_ERROR(errno);
978 * Prepare, create, and connect to a Unix domain socket for local communication.
980 * \param name The socket pathname.
982 * This function creates a local socket for sequenced, reliable, two-way,
983 * connection-based byte streams.
985 * \return The file descriptor of the connected socket on success, negative on
988 * \sa \ref create_local_socket(), unix(7), connect(2).
990 int connect_local_socket(const char *name)
992 struct sockaddr_un unix_addr;
995 PARA_DEBUG_LOG("connecting to %s\n", name);
996 fd = socket(PF_UNIX, SOCK_STREAM, 0);
998 return -ERRNO_TO_PARA_ERROR(errno);
999 ret = init_unix_addr(&unix_addr, name);
1002 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
1004 ret = -ERRNO_TO_PARA_ERROR(errno);
1011 ssize_t send_cred_buffer(int sock, char *buf)
1013 return write_buffer(sock, buf);
1015 int recv_cred_buffer(int fd, char *buf, size_t size)
1017 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
1019 #else /* HAVE_UCRED */
1022 * Send a buffer and the credentials of the current process to a socket.
1024 * \param sock The file descriptor of the sending socket.
1025 * \param buf The zero-terminated buffer to send.
1027 * \return On success, this call returns the number of bytes sent. On errors,
1028 * \p -E_SENDMSG is returned.
1030 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
1032 ssize_t send_cred_buffer(int sock, char *buf)
1034 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
1036 struct cmsghdr *cmsg;
1037 static struct iovec iov;
1043 iov.iov_len = strlen(buf);
1047 /* compose the message */
1048 memset(&msg, 0, sizeof(msg));
1051 msg.msg_control = control;
1052 msg.msg_controllen = sizeof(control);
1053 /* attach the ucred struct */
1054 cmsg = CMSG_FIRSTHDR(&msg);
1055 cmsg->cmsg_level = SOL_SOCKET;
1056 cmsg->cmsg_type = SCM_CREDENTIALS;
1057 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1058 *(struct ucred *)CMSG_DATA(cmsg) = c;
1059 msg.msg_controllen = cmsg->cmsg_len;
1060 ret = sendmsg(sock, &msg, 0);
1066 static void dispose_fds(int *fds, unsigned num)
1070 for (i = 0; i < num; i++)
1075 * Receive a buffer and the Unix credentials of the sending process.
1077 * \param fd The file descriptor of the receiving socket.
1078 * \param buf The buffer to store the received message.
1079 * \param size The length of \a buf in bytes.
1081 * \return Negative on errors, the user id of the sending process on success.
1083 * \sa \ref send_cred_buffer and the references given there.
1085 int recv_cred_buffer(int fd, char *buf, size_t size)
1087 char control[255] __a_aligned(8);
1089 struct cmsghdr *cmsg;
1095 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1096 memset(&msg, 0, sizeof(msg));
1097 memset(buf, 0, size);
1102 msg.msg_control = control;
1103 msg.msg_controllen = sizeof(control);
1104 if (recvmsg(fd, &msg, 0) < 0)
1106 result = -E_SCM_CREDENTIALS;
1107 cmsg = CMSG_FIRSTHDR(&msg);
1109 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1110 == SCM_CREDENTIALS) {
1111 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1114 if (cmsg->cmsg_level == SOL_SOCKET
1115 && cmsg->cmsg_type == SCM_RIGHTS) {
1116 dispose_fds((int *)CMSG_DATA(cmsg),
1117 (cmsg->cmsg_len - CMSG_LEN(0))
1120 cmsg = CMSG_NXTHDR(&msg, cmsg);
1124 #endif /* HAVE_UCRED */