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>
14 /* At least NetBSD needs these. */
22 #define AI_ADDRCONFIG 0
34 * Parse and validate IPv4 address/netmask string.
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.
42 * \return Pointer to \a addr if successful, NULL on error.
45 char *parse_cidr(const char *cidr,
46 char *addr, ssize_t addrlen,
50 char *c = addr, *end = c + (addrlen - 1);
54 if (cidr == NULL || addrlen < 1)
57 for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
62 if (para_atoi32(++o, netmask) < 0 ||
63 *netmask < 0 || *netmask > 0x20)
66 if (is_valid_ipv4_address(addr))
75 * Match string as a candidate IPv4 address.
77 * \param address The string to match.
78 * \return True if \a address has "dot-quad" format.
80 static bool is_v4_dot_quad(const char *address)
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;
93 * Perform basic syntax checking on the host-part of an URL:
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.
100 * \param host The host string to check.
101 * \return True if \a host passes the syntax checks.
103 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5.
105 static bool host_string_ok(const char *host)
107 if (host == NULL || *host == '\0')
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);
117 * Parse and validate URL string.
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.
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.
128 * The following are valid examples:
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.
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.
145 * \sa RFC 3986, 3.2.2/3.2.3.
147 char *parse_url(const char *url,
148 char *host, ssize_t hostlen,
152 char *c = host, *end = c + (hostlen - 1);
156 if (o == NULL || hostlen < 1)
160 for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
164 if (*o++ != ']' || (*o != '\0' && *o != ':'))
167 for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
168 if (c == end && o[1])
174 if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff)
176 if (host_string_ok(host))
184 * Pretty-print a host/port pair.
186 * \param url NULL, or any string accepted by \ref parse_url().
187 * \param default_port Applies if url has no port.
189 * If the url argument is NULL, the function returns the string
190 * 0.0.0.0:default_port. Otherwise it calls \ref parse_url() to check the
191 * syntax of the input string given by url. On errors the string "?" is
192 * returned. Otherwise, if url contains a port, a copy of url is returned. If
193 * no port was supplied, a colon and the default port are appended to url.
195 * \return In all cases the returned string is a allocated with malloc(3) and
196 * has to be freed by the caller.
198 char *format_url(const char *url, int default_port)
200 char host[MAX_HOSTLEN];
204 return make_message("0.0.0.0:%d", default_port);
205 if (!parse_url(url, host, sizeof(host), &url_port))
206 return make_message("?");
208 return make_message("%s:%d", url, default_port);
210 return para_strdup(url);
214 * Stringify port number, resolve into service name where defined.
216 * \param port 2-byte port number, in host-byte-order.
217 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
218 * \return Pointer to static result buffer.
220 * \sa getservent(3), services(5), nsswitch.conf(5).
222 const char *stringify_port(int port, const char *transport)
224 static char service[NI_MAXSERV];
226 if (port < 0 || port > 0xFFFF) {
227 snprintf(service, sizeof(service), "undefined (%d)", port);
229 struct servent *se = getservbyport(htons(port), transport);
232 snprintf(service, sizeof(service), "%d", port);
234 snprintf(service, sizeof(service), "%s", se->s_name);
240 * Determine the socket type for a given layer-4 protocol.
242 * \param l4type The symbolic name of the transport-layer protocol.
244 * \sa ip(7), socket(2).
246 static inline int sock_type(const unsigned l4type)
249 case IPPROTO_UDP: return SOCK_DGRAM;
250 case IPPROTO_TCP: return SOCK_STREAM;
251 case IPPROTO_DCCP: return SOCK_DCCP;
253 return -1; /* not supported here */
257 * Pretty-print transport-layer name.
259 static const char *layer4_name(const unsigned l4type)
262 case IPPROTO_UDP: return "UDP";
263 case IPPROTO_TCP: return "TCP";
264 case IPPROTO_DCCP: return "DCCP";
266 return "UNKNOWN PROTOCOL";
270 * Flowopts: Transport-layer independent encapsulation of socket options.
272 * These collect individual socket options into a queue, which is disposed of
273 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
274 * only and should not be visible elsewhere.
276 * \sa setsockopt(2), \ref makesock().
278 struct pre_conn_opt {
279 int sock_level; /**< Second argument to setsockopt() */
280 int sock_option; /**< Third argument to setsockopt() */
281 char *opt_name; /**< Stringified \a sock_option */
282 void *opt_val; /**< Fourth argument to setsockopt() */
283 socklen_t opt_len; /**< Fifth argument to setsockopt() */
285 struct list_head node; /**< FIFO, as sockopt order matters. */
288 /** FIFO list of pre-connection socket options to be set */
290 struct list_head sockopts;
294 * Allocate and initialize a flowopt queue.
296 * \return A new structure to be passed to \ref flowopt_add(). It is
297 * automatically deallocated in \ref makesock().
299 struct flowopts *flowopt_new(void)
301 struct flowopts *new = para_malloc(sizeof(*new));
303 INIT_LIST_HEAD(&new->sockopts);
308 * Append new socket option to flowopt queue.
310 * \param fo The flowopt queue to append to.
311 * \param lev Level at which \a opt resides.
312 * \param opt New option to add.
313 * \param name Stringified name of \a opt.
314 * \param val The value to set \a opt to.
315 * \param len Length of \a val.
319 void flowopt_add(struct flowopts *fo, int lev, int opt,
320 const char *name, const void *val, int len)
322 struct pre_conn_opt *new = para_malloc(sizeof(*new));
324 new->sock_option = opt;
325 new->sock_level = lev;
326 new->opt_name = para_strdup(name);
332 new->opt_val = para_malloc(len);
334 memcpy(new->opt_val, val, len);
337 list_add_tail(&new->node, &fo->sockopts);
340 /** Set the entire bunch of pre-connection options at once. */
341 static void flowopt_setopts(int sockfd, struct flowopts *fo)
343 struct pre_conn_opt *pc;
348 list_for_each_entry(pc, &fo->sockopts, node)
349 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
350 pc->opt_val, pc->opt_len) < 0) {
351 PARA_EMERG_LOG("Can not set %s socket option: %s",
352 pc->opt_name, strerror(errno));
358 * Deallocate all resources of a flowopts structure.
360 * \param fo A pointer as returned from flowopt_new().
362 * It's OK to pass \p NULL here in which case the function does nothing.
364 void flowopt_cleanup(struct flowopts *fo)
366 struct pre_conn_opt *cur, *next;
371 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
380 * Resolve an IPv4/IPv6 address.
382 * \param l4type The layer-4 type (\p IPPROTO_xxx).
383 * \param passive Whether \p AI_PASSIVE should be included as hint.
384 * \param host Remote or local hostname or IPv/6 address string.
385 * \param port_number Used to set the port in each returned address structure.
386 * \param result addrinfo structures are returned here.
388 * The interpretation of \a host depends on the value of \a passive. On a
389 * passive socket host is interpreted as an interface IPv4/6 address (can be
390 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
391 * address to connect to.
395 * \sa getaddrinfo(3).
397 int lookup_address(unsigned l4type, bool passive, const char *host,
398 int port_number, struct addrinfo **result)
401 char port[6]; /* port number has at most 5 digits */
402 struct addrinfo *addr = NULL, hints;
405 sprintf(port, "%d", port_number & 0xffff);
406 /* Set up address hint structure */
407 memset(&hints, 0, sizeof(hints));
408 hints.ai_family = AF_UNSPEC;
409 hints.ai_socktype = sock_type(l4type);
411 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
412 * (and only then) pretend to be UDP.
414 if (l4type == IPPROTO_DCCP)
415 hints.ai_socktype = SOCK_DGRAM;
416 /* only use addresses available on the host */
417 hints.ai_flags = AI_ADDRCONFIG;
418 if (passive && host == NULL)
419 hints.ai_flags |= AI_PASSIVE;
420 /* Obtain local/remote address information */
421 ret = getaddrinfo(host, port, &hints, &addr);
423 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
425 host? host : (passive? "[loopback]" : "[localhost]"),
426 port, gai_strerror(ret));
427 return -E_ADDRESS_LOOKUP;
434 * Create an active or passive socket.
436 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
437 * \param passive Whether to call bind(2) or connect(2).
438 * \param ai Address information as obtained from \ref lookup_address().
439 * \param fo Socket options to be set before making the connection.
441 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
442 * algorithm tries all possible address combinations until it succeeds. If \a
443 * fo is supplied, options are set but cleanup must be performed in the caller.
445 * \return File descriptor on success, \p E_MAKESOCK on errors.
447 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
450 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
453 int ret = -E_MAKESOCK, on = 1;
455 for (; ai; ai = ai->ai_next) {
457 ret = socket(ai->ai_family, sock_type(l4type), l4type);
459 PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
463 flowopt_setopts(fd, fo);
465 if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
466 PARA_NOTICE_LOG("connect(): %s\n",
474 * Reuse the address on passive sockets to avoid failure on
475 * restart (protocols using listen()) and when creating
476 * multiple listener instances (UDP multicast).
478 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
480 PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
484 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
485 PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
495 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
497 * \param l4type The layer-4 type (\p IPPROTO_xxx).
498 * \param passive Whether this is a passive or active socket.
499 * \param host Passed to \ref lookup_address().
500 * \param port_number Passed to \ref lookup_address().
501 * \param fo Passed to \ref makesock_addrinfo().
503 * This creates a ready-made IPv4/v6 socket structure after looking up the
504 * necessary parameters. The function first calls \ref lookup_address() and
505 * passes the address information to makesock_addrinfo() to create and
506 * initialize the socket.
508 * \return The newly created file descriptor on success, a negative error code
511 * \sa \ref lookup_address(), \ref makesock_addrinfo().
513 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
517 int ret = lookup_address(l4type, passive, host, port_number, &ai);
520 ret = makesock_addrinfo(l4type, passive, ai, fo);
524 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
525 layer4_name(l4type), host? host : (passive?
526 "[loopback]" : "[localhost]"), port_number);
532 * Create a passive / listening socket.
534 * \param l4type The transport-layer type (\p IPPROTO_xxx).
535 * \param addr Passed to \ref parse_url() if not NULL.
536 * \param port Ignored if addr contains a port number.
538 * \return Positive integer (socket descriptor) on success, negative value
541 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
543 int para_listen(unsigned l4type, const char *addr, uint16_t port)
545 char host[MAX_HOSTLEN];
546 int ret, fd, addr_port;
549 if (!parse_url(addr, host, sizeof(host), &addr_port))
550 return -ERRNO_TO_PARA_ERROR(EINVAL);
555 fd = makesock(l4type, true /* passive */, addr, port,
556 NULL /* no flowopts */);
558 ret = listen(fd, BACKLOG);
562 return -ERRNO_TO_PARA_ERROR(ret);
564 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
565 layer4_name(l4type), port, fd);
571 * Create a socket which listens on all network addresses.
573 * \param l4type See \ref para_listen().
574 * \param port See \ref para_listen().
576 * This is a simple wrapper for \ref para_listen() which passes a NULL pointer
577 * as the address information.
579 * \return See \ref para_listen().
581 int para_listen_simple(unsigned l4type, uint16_t port)
583 return para_listen(l4type, NULL, port);
587 * Determine IPv4/v6 socket address length.
588 * \param sa Container of IPv4 or IPv6 address.
589 * \return Address-family dependent address length.
591 static socklen_t salen(const struct sockaddr *sa)
593 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
595 return sa->sa_family == AF_INET6
596 ? sizeof(struct sockaddr_in6)
597 : sizeof(struct sockaddr_in);
600 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
601 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
603 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
605 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
609 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
610 * \param ss Container of IPv4/6 address.
611 * \return Pointer to normalized address (may be static storage).
615 static const struct sockaddr *
616 normalize_ip_address(const struct sockaddr_storage *ss)
618 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
620 if (SS_IS_ADDR_V4MAPPED(ss)) {
621 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
622 static struct sockaddr_in ia;
624 ia.sin_family = AF_INET;
625 ia.sin_port = ia6->sin6_port;
626 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
627 return (const struct sockaddr *)&ia;
629 return (const struct sockaddr *)ss;
633 * Generic/fallback MTU values
635 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
636 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
637 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
638 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
639 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
640 * UDP senders should use the respective minimum values of EMTU_S.
642 static inline int generic_mtu(const int af_type)
644 return af_type == AF_INET6 ? 1280 : 576;
647 /** Crude approximation of IP header overhead - neglecting options. */
648 static inline int estimated_header_overhead(const int af_type)
650 return af_type == AF_INET6 ? 40 : 20;
654 * Get the maximum transport-layer message size (MMS_S).
656 * \param sockfd The socket file descriptor.
658 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
659 * could not be determined, \p AF_INET is assumed.
661 * \return The maximum message size of the address family type.
663 int generic_max_transport_msg_size(int sockfd)
665 struct sockaddr_storage ss = {.ss_family = 0};
666 socklen_t sslen = sizeof(ss);
667 int af_type = AF_INET;
669 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
670 PARA_ERROR_LOG("can not determine remote address type: %s\n",
672 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
673 af_type = ss.ss_family;
675 return generic_mtu(af_type) - estimated_header_overhead(af_type);
679 * Look up the remote side of a connected socket structure.
681 * \param fd The socket descriptor of the connected socket.
683 * \return A static character string identifying hostname and port of the
684 * chosen side in numeric host:port format.
686 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
687 * services(5), nsswitch.conf(5).
689 char *remote_name(int fd)
691 struct sockaddr_storage ss = {.ss_family = 0};
692 const struct sockaddr *sa;
693 socklen_t sslen = sizeof(ss);
694 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
695 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
698 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
699 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
700 fd, strerror(errno));
701 snprintf(output, sizeof(output), "(unknown)");
704 sa = normalize_ip_address(&ss);
705 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
706 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
708 PARA_WARNING_LOG("hostname lookup error (%s).\n",
710 snprintf(output, sizeof(output), "(lookup error)");
711 } else if (sa->sa_family == AF_INET6)
712 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
714 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
719 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
721 * \param ss Container of IPv4/6 address.
722 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
726 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
728 const struct sockaddr *sa = normalize_ip_address(ss);
730 memset(ia, 0, sizeof(*ia));
731 if (sa->sa_family == AF_INET)
732 *ia = ((struct sockaddr_in *)sa)->sin_addr;
736 * Compare the address part of IPv4/6 addresses.
738 * \param sa1 First address.
739 * \param sa2 Second address.
741 * \return True iff the IP address of \a sa1 and \a sa2 match.
743 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
747 if (sa1->sa_family != sa2->sa_family)
749 if (sa1->sa_family == AF_INET) {
750 struct sockaddr_in *a1 = (typeof(a1))sa1,
751 *a2 = (typeof (a2))sa2;
752 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
753 } else if (sa1->sa_family == AF_INET6) {
754 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
755 *a2 = (typeof (a2))sa2;
756 return !memcmp(a1, a2, sizeof(*a1));
762 * Receive data from a file descriptor.
764 * \param fd The file descriptor.
765 * \param buf The buffer to write the data to.
766 * \param size The size of \a buf.
768 * Receive at most \a size bytes from file descriptor \a fd.
770 * \return The number of bytes received on success, negative on errors, zero if
771 * the peer has performed an orderly shutdown.
775 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
779 n = recv(fd, buf, size, 0);
781 return -ERRNO_TO_PARA_ERROR(errno);
786 * Receive and write terminating NULL byte.
788 * \param fd The file descriptor.
789 * \param buf The buffer to write the data to.
790 * \param size The size of \a buf.
792 * Read at most \a size - 1 bytes from file descriptor \a fd and
793 * write a NULL byte at the end of the received data.
795 * \return The return value of the underlying call to \a recv_bin_buffer().
797 * \sa \ref recv_bin_buffer()
799 int recv_buffer(int fd, char *buf, size_t size)
804 n = recv_bin_buffer(fd, buf, size - 1);
813 * Wrapper around the accept system call.
815 * \param fd The listening socket.
816 * \param rfds An optional fd_set pointer.
817 * \param addr Structure which is filled in with the address of the peer socket.
818 * \param size Should contain the size of the structure pointed to by \a addr.
819 * \param new_fd Result pointer.
821 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
822 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
824 * \return Negative on errors, zero if no connections are present to be accepted,
829 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
833 if (rfds && !FD_ISSET(fd, rfds))
836 ret = accept(fd, (struct sockaddr *) addr, &size);
837 while (ret < 0 && errno == EINTR);
843 if (errno == EAGAIN || errno == EWOULDBLOCK)
845 return -ERRNO_TO_PARA_ERROR(errno);
849 * Probe the list of DCCP CCIDs configured on this host.
850 * \param ccid_array Pointer to return statically allocated array in.
851 * \return Number of elements returned in \a ccid_array or error.
853 * NB: This feature is only available on Linux > 2.6.30; on older kernels
854 * ENOPROTOOPT ("Protocol not available") will be returned.
856 int dccp_available_ccids(uint8_t **ccid_array)
858 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
859 socklen_t nccids = sizeof(ccids);
862 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
866 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
867 ccids, &nccids) < 0) {
870 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
872 return -ERRNO_TO_PARA_ERROR(ret);
881 * Prepare a structure for AF_UNIX socket addresses.
883 * This just copies name to the sun_path component of u, prepending a zero byte
884 * if abstract sockets are supported.
886 * The first call to this function tries to bind a socket to the abstract name
887 * space. The result of this test is stored in a static variable. Subsequent
888 * calls read this variable and create abstract sockets on systems that support
889 * them. If a NULL pointer is passed as the name, the function only
890 * initializes the static variable.
892 static int init_unix_addr(struct sockaddr_un *u, const char *name)
894 static int use_abstract;
896 memset(u->sun_path, 0, UNIX_PATH_MAX);
897 u->sun_family = PF_UNIX;
898 if (use_abstract == 0) { /* executed only once */
899 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
900 memcpy(u->sun_path, "\0x\0", 3);
901 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
902 use_abstract = 1; /* yes */
904 use_abstract = -1; /* no */
906 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
907 use_abstract == 1? "" : "not ");
911 if (strlen(name) + 1 >= UNIX_PATH_MAX)
912 return -E_NAME_TOO_LONG;
913 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
918 * Create a socket for local communication and listen on it.
920 * \param name The socket pathname.
922 * This function creates a passive local socket for sequenced, reliable,
923 * two-way, connection-based byte streams. The socket file descriptor is set to
924 * nonblocking mode and listen(2) is called to prepare the socket for
925 * accepting incoming connection requests.
927 * \return The file descriptor on success, negative error code on failure.
929 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
931 int create_local_socket(const char *name)
933 struct sockaddr_un unix_addr;
936 ret = init_unix_addr(&unix_addr, name);
937 if (ret <= 0) /* error, or name was NULL */
939 ret = socket(PF_UNIX, SOCK_STREAM, 0);
941 return -ERRNO_TO_PARA_ERROR(errno);
943 ret = mark_fd_nonblocking(fd);
946 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
948 ret = -ERRNO_TO_PARA_ERROR(errno);
951 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
952 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
955 if (chmod(name, mode) < 0)
958 if (listen(fd , 5) < 0) {
959 ret = -ERRNO_TO_PARA_ERROR(errno);
969 * Prepare, create, and connect to a Unix domain socket for local communication.
971 * \param name The socket pathname.
973 * This function creates a local socket for sequenced, reliable, two-way,
974 * connection-based byte streams.
976 * \return The file descriptor of the connected socket on success, negative on
979 * \sa \ref create_local_socket(), unix(7), connect(2).
981 int connect_local_socket(const char *name)
983 struct sockaddr_un unix_addr;
986 PARA_DEBUG_LOG("connecting to %s\n", name);
987 fd = socket(PF_UNIX, SOCK_STREAM, 0);
989 return -ERRNO_TO_PARA_ERROR(errno);
990 ret = init_unix_addr(&unix_addr, name);
993 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
995 ret = -ERRNO_TO_PARA_ERROR(errno);
1002 ssize_t send_cred_buffer(int sock, char *buf)
1004 return write_buffer(sock, buf);
1006 int recv_cred_buffer(int fd, char *buf, size_t size)
1008 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
1010 #else /* HAVE_UCRED */
1013 * Send a buffer and the credentials of the current process to a socket.
1015 * \param sock The file descriptor of the sending socket.
1016 * \param buf The zero-terminated buffer to send.
1018 * \return On success, this call returns the number of bytes sent. On errors,
1019 * \p -E_SENDMSG is returned.
1021 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
1023 ssize_t send_cred_buffer(int sock, char *buf)
1025 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
1027 struct cmsghdr *cmsg;
1028 static struct iovec iov;
1034 iov.iov_len = strlen(buf);
1038 /* compose the message */
1039 memset(&msg, 0, sizeof(msg));
1042 msg.msg_control = control;
1043 msg.msg_controllen = sizeof(control);
1044 /* attach the ucred struct */
1045 cmsg = CMSG_FIRSTHDR(&msg);
1046 cmsg->cmsg_level = SOL_SOCKET;
1047 cmsg->cmsg_type = SCM_CREDENTIALS;
1048 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1049 *(struct ucred *)CMSG_DATA(cmsg) = c;
1050 msg.msg_controllen = cmsg->cmsg_len;
1051 ret = sendmsg(sock, &msg, 0);
1057 static void dispose_fds(int *fds, unsigned num)
1061 for (i = 0; i < num; i++)
1066 * Receive a buffer and the Unix credentials of the sending process.
1068 * \param fd The file descriptor of the receiving socket.
1069 * \param buf The buffer to store the received message.
1070 * \param size The length of \a buf in bytes.
1072 * \return Negative on errors, the user id of the sending process on success.
1074 * \sa \ref send_cred_buffer and the references given there.
1076 int recv_cred_buffer(int fd, char *buf, size_t size)
1078 char control[255] __a_aligned(8);
1080 struct cmsghdr *cmsg;
1086 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1087 memset(&msg, 0, sizeof(msg));
1088 memset(buf, 0, size);
1093 msg.msg_control = control;
1094 msg.msg_controllen = sizeof(control);
1095 if (recvmsg(fd, &msg, 0) < 0)
1097 result = -E_SCM_CREDENTIALS;
1098 cmsg = CMSG_FIRSTHDR(&msg);
1100 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1101 == SCM_CREDENTIALS) {
1102 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1105 if (cmsg->cmsg_level == SOL_SOCKET
1106 && cmsg->cmsg_type == SCM_RIGHTS) {
1107 dispose_fds((int *)CMSG_DATA(cmsg),
1108 (cmsg->cmsg_len - CMSG_LEN(0))
1111 cmsg = CMSG_NXTHDR(&msg, cmsg);
1115 #endif /* HAVE_UCRED */