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 * Determine the socket type for a given layer-4 protocol.
230 * \param l4type The symbolic name of the transport-layer protocol.
232 * \sa ip(7), socket(2).
234 static inline int sock_type(const unsigned l4type)
237 case IPPROTO_UDP: return SOCK_DGRAM;
238 case IPPROTO_TCP: return SOCK_STREAM;
239 case IPPROTO_DCCP: return SOCK_DCCP;
241 return -1; /* not supported here */
245 * Pretty-print transport-layer name.
247 static const char *layer4_name(const unsigned l4type)
250 case IPPROTO_UDP: return "UDP";
251 case IPPROTO_TCP: return "TCP";
252 case IPPROTO_DCCP: return "DCCP";
254 return "UNKNOWN PROTOCOL";
258 * Flowopts: Transport-layer independent encapsulation of socket options.
260 * These collect individual socket options into a queue, which is disposed of
261 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
262 * only and should not be visible elsewhere.
264 * \sa setsockopt(2), \ref makesock().
266 struct pre_conn_opt {
267 int sock_level; /**< Second argument to setsockopt() */
268 int sock_option; /**< Third argument to setsockopt() */
269 char *opt_name; /**< Stringified \a sock_option */
270 void *opt_val; /**< Fourth argument to setsockopt() */
271 socklen_t opt_len; /**< Fifth argument to setsockopt() */
273 struct list_head node; /**< FIFO, as sockopt order matters. */
276 /** FIFO list of pre-connection socket options to be set */
278 struct list_head sockopts;
282 * Allocate and initialize a flowopt queue.
284 * \return A new structure to be passed to \ref flowopt_add(). It is
285 * automatically deallocated in \ref makesock().
287 struct flowopts *flowopt_new(void)
289 struct flowopts *new = para_malloc(sizeof(*new));
291 init_list_head(&new->sockopts);
296 * Append new socket option to flowopt queue.
298 * \param fo The flowopt queue to append to.
299 * \param lev Level at which \a opt resides.
300 * \param opt New option to add.
301 * \param name Stringified name of \a opt.
302 * \param val The value to set \a opt to.
303 * \param len Length of \a val.
307 void flowopt_add(struct flowopts *fo, int lev, int opt,
308 const char *name, const void *val, int len)
310 struct pre_conn_opt *new = para_malloc(sizeof(*new));
312 new->sock_option = opt;
313 new->sock_level = lev;
314 new->opt_name = para_strdup(name);
320 new->opt_val = para_malloc(len);
322 memcpy(new->opt_val, val, len);
325 list_add_tail(&new->node, &fo->sockopts);
328 /** Set the entire bunch of pre-connection options at once. */
329 static void flowopt_setopts(int sockfd, struct flowopts *fo)
331 struct pre_conn_opt *pc;
336 list_for_each_entry(pc, &fo->sockopts, node)
337 if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
338 pc->opt_val, pc->opt_len) < 0) {
339 PARA_EMERG_LOG("Can not set %s socket option: %s",
340 pc->opt_name, strerror(errno));
346 * Deallocate all resources of a flowopts structure.
348 * \param fo A pointer as returned from flowopt_new().
350 * It's OK to pass \p NULL here in which case the function does nothing.
352 void flowopt_cleanup(struct flowopts *fo)
354 struct pre_conn_opt *cur, *next;
359 list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
368 * Resolve an IPv4/IPv6 address.
370 * \param l4type The layer-4 type (\p IPPROTO_xxx).
371 * \param passive Whether \p AI_PASSIVE should be included as hint.
372 * \param host Remote or local hostname or IPv/6 address string.
373 * \param port_number Used to set the port in each returned address structure.
374 * \param result addrinfo structures are returned here.
376 * The interpretation of \a host depends on the value of \a passive. On a
377 * passive socket host is interpreted as an interface IPv4/6 address (can be
378 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
379 * address to connect to.
383 * \sa getaddrinfo(3).
385 int lookup_address(unsigned l4type, bool passive, const char *host,
386 int port_number, struct addrinfo **result)
389 char port[6]; /* port number has at most 5 digits */
390 struct addrinfo *addr = NULL, hints;
393 sprintf(port, "%d", port_number & 0xffff);
394 /* Set up address hint structure */
395 memset(&hints, 0, sizeof(hints));
396 hints.ai_family = AF_UNSPEC;
397 hints.ai_socktype = sock_type(l4type);
399 * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
400 * (and only then) pretend to be UDP.
402 if (l4type == IPPROTO_DCCP)
403 hints.ai_socktype = SOCK_DGRAM;
404 /* only use addresses available on the host */
405 hints.ai_flags = AI_ADDRCONFIG;
406 if (passive && host == NULL)
407 hints.ai_flags |= AI_PASSIVE;
408 /* Obtain local/remote address information */
409 ret = getaddrinfo(host, port, &hints, &addr);
411 PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
413 host? host : (passive? "[loopback]" : "[localhost]"),
414 port, gai_strerror(ret));
415 return -E_ADDRESS_LOOKUP;
422 * Create an active or passive socket.
424 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
425 * \param passive Whether to call bind(2) or connect(2).
426 * \param ai Address information as obtained from \ref lookup_address().
427 * \param fo Socket options to be set before making the connection.
429 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
430 * algorithm tries all possible address combinations until it succeeds. If \a
431 * fo is supplied, options are set but cleanup must be performed in the caller.
433 * \return File descriptor on success, \p E_MAKESOCK on errors.
435 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
438 int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
441 int ret = -E_MAKESOCK, on = 1;
443 for (; ai; ai = ai->ai_next) {
445 ret = socket(ai->ai_family, sock_type(l4type), l4type);
447 PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
451 flowopt_setopts(fd, fo);
453 if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
454 PARA_NOTICE_LOG("connect(): %s\n",
462 * Reuse the address on passive sockets to avoid failure on
463 * restart (protocols using listen()) and when creating
464 * multiple listener instances (UDP multicast).
466 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
468 PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
472 if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
473 PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
483 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
485 * \param l4type The layer-4 type (\p IPPROTO_xxx).
486 * \param passive Whether this is a passive or active socket.
487 * \param host Passed to \ref lookup_address().
488 * \param port_number Passed to \ref lookup_address().
489 * \param fo Passed to \ref makesock_addrinfo().
491 * This creates a ready-made IPv4/v6 socket structure after looking up the
492 * necessary parameters. The function first calls \ref lookup_address() and
493 * passes the address information to makesock_addrinfo() to create and
494 * initialize the socket.
496 * \return The newly created file descriptor on success, a negative error code
499 * \sa \ref lookup_address(), \ref makesock_addrinfo().
501 int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
505 int ret = lookup_address(l4type, passive, host, port_number, &ai);
508 ret = makesock_addrinfo(l4type, passive, ai, fo);
512 PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
513 layer4_name(l4type), host? host : (passive?
514 "[loopback]" : "[localhost]"), port_number);
520 * Create a passive / listening socket.
522 * \param l4type The transport-layer type (\p IPPROTO_xxx).
523 * \param addr Passed to \ref parse_url() if not NULL.
524 * \param port Ignored if addr contains a port number.
526 * \return Positive integer (socket descriptor) on success, negative value
529 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
531 int para_listen(unsigned l4type, const char *addr, uint16_t port)
533 char host[MAX_HOSTLEN];
534 int ret, fd, addr_port;
537 if (!parse_url(addr, host, sizeof(host), &addr_port))
538 return -ERRNO_TO_PARA_ERROR(EINVAL);
543 fd = makesock(l4type, true /* passive */, addr, port,
544 NULL /* no flowopts */);
546 ret = listen(fd, BACKLOG);
550 return -ERRNO_TO_PARA_ERROR(ret);
552 PARA_INFO_LOG("listening on %s port %u, fd %d\n",
553 layer4_name(l4type), port, fd);
559 * Create a socket which listens on all network addresses.
561 * \param l4type See \ref para_listen().
562 * \param port See \ref para_listen().
564 * This is a simple wrapper for \ref para_listen() which passes a NULL pointer
565 * as the address information.
567 * \return See \ref para_listen().
569 int para_listen_simple(unsigned l4type, uint16_t port)
571 return para_listen(l4type, NULL, port);
575 * Determine IPv4/v6 socket address length.
576 * \param sa Container of IPv4 or IPv6 address.
577 * \return Address-family dependent address length.
579 static socklen_t salen(const struct sockaddr *sa)
581 assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);
583 return sa->sa_family == AF_INET6
584 ? sizeof(struct sockaddr_in6)
585 : sizeof(struct sockaddr_in);
588 /** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
589 static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
591 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
593 return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
597 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
598 * \param ss Container of IPv4/6 address.
599 * \return Pointer to normalized address (may be static storage).
603 static const struct sockaddr *
604 normalize_ip_address(const struct sockaddr_storage *ss)
606 assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);
608 if (SS_IS_ADDR_V4MAPPED(ss)) {
609 const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
610 static struct sockaddr_in ia;
612 ia.sin_family = AF_INET;
613 ia.sin_port = ia6->sin6_port;
614 memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
615 return (const struct sockaddr *)&ia;
617 return (const struct sockaddr *)ss;
621 * Generic/fallback MTU values
623 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
624 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
625 * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
626 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
627 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
628 * UDP senders should use the respective minimum values of EMTU_S.
630 static inline int generic_mtu(const int af_type)
632 return af_type == AF_INET6 ? 1280 : 576;
635 /** Crude approximation of IP header overhead - neglecting options. */
636 static inline int estimated_header_overhead(const int af_type)
638 return af_type == AF_INET6 ? 40 : 20;
642 * Get the maximum transport-layer message size (MMS_S).
644 * \param sockfd The socket file descriptor.
646 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
647 * could not be determined, \p AF_INET is assumed.
649 * \return The maximum message size of the address family type.
651 int generic_max_transport_msg_size(int sockfd)
653 struct sockaddr_storage ss = {.ss_family = 0};
654 socklen_t sslen = sizeof(ss);
655 int af_type = AF_INET;
657 if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
658 PARA_ERROR_LOG("can not determine remote address type: %s\n",
660 } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
661 af_type = ss.ss_family;
663 return generic_mtu(af_type) - estimated_header_overhead(af_type);
667 * Look up the remote side of a connected socket structure.
669 * \param fd The socket descriptor of the connected socket.
671 * \return A static character string identifying hostname and port of the
672 * chosen side in numeric host:port format.
674 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
675 * services(5), nsswitch.conf(5).
677 char *remote_name(int fd)
679 struct sockaddr_storage ss = {.ss_family = 0};
680 const struct sockaddr *sa;
681 socklen_t sslen = sizeof(ss);
682 char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
683 static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
686 if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
687 PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
688 fd, strerror(errno));
689 snprintf(output, sizeof(output), "(unknown)");
692 sa = normalize_ip_address(&ss);
693 ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
694 sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
696 PARA_WARNING_LOG("hostname lookup error (%s).\n",
698 snprintf(output, sizeof(output), "(lookup error)");
699 } else if (sa->sa_family == AF_INET6)
700 snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
702 snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
707 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
709 * \param ss Container of IPv4/6 address.
710 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
714 void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
716 const struct sockaddr *sa = normalize_ip_address(ss);
718 memset(ia, 0, sizeof(*ia));
719 if (sa->sa_family == AF_INET)
720 *ia = ((struct sockaddr_in *)sa)->sin_addr;
724 * Compare the address part of IPv4/6 addresses.
726 * \param sa1 First address.
727 * \param sa2 Second address.
729 * \return True iff the IP address of \a sa1 and \a sa2 match.
731 bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
735 if (sa1->sa_family != sa2->sa_family)
737 if (sa1->sa_family == AF_INET) {
738 struct sockaddr_in *a1 = (typeof(a1))sa1,
739 *a2 = (typeof (a2))sa2;
740 return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
741 } else if (sa1->sa_family == AF_INET6) {
742 struct sockaddr_in6 *a1 = (typeof(a1))sa1,
743 *a2 = (typeof (a2))sa2;
744 return !memcmp(a1, a2, sizeof(*a1));
750 * Receive data from a file descriptor.
752 * \param fd The file descriptor.
753 * \param buf The buffer to write the data to.
754 * \param size The size of \a buf.
756 * Receive at most \a size bytes from file descriptor \a fd.
758 * \return The number of bytes received on success, negative on errors, zero if
759 * the peer has performed an orderly shutdown.
763 __must_check int recv_bin_buffer(int fd, char *buf, size_t size)
767 n = recv(fd, buf, size, 0);
769 return -ERRNO_TO_PARA_ERROR(errno);
774 * Receive and write terminating NULL byte.
776 * \param fd The file descriptor.
777 * \param buf The buffer to write the data to.
778 * \param size The size of \a buf.
780 * Read at most \a size - 1 bytes from file descriptor \a fd and
781 * write a NULL byte at the end of the received data.
783 * \return The return value of the underlying call to \a recv_bin_buffer().
785 * \sa \ref recv_bin_buffer()
787 int recv_buffer(int fd, char *buf, size_t size)
792 n = recv_bin_buffer(fd, buf, size - 1);
801 * Wrapper around the accept system call.
803 * \param fd The listening socket.
804 * \param rfds An optional fd_set pointer.
805 * \param addr Structure which is filled in with the address of the peer socket.
806 * \param size Should contain the size of the structure pointed to by \a addr.
807 * \param new_fd Result pointer.
809 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
810 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
812 * \return Negative on errors, zero if no connections are present to be accepted,
817 int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
821 if (rfds && !FD_ISSET(fd, rfds))
824 ret = accept(fd, (struct sockaddr *) addr, &size);
825 while (ret < 0 && errno == EINTR);
831 if (errno == EAGAIN || errno == EWOULDBLOCK)
833 return -ERRNO_TO_PARA_ERROR(errno);
837 * Probe the list of DCCP CCIDs configured on this host.
838 * \param ccid_array Pointer to return statically allocated array in.
839 * \return Number of elements returned in \a ccid_array or error.
841 * NB: This feature is only available on Linux > 2.6.30; on older kernels
842 * ENOPROTOOPT ("Protocol not available") will be returned.
844 int dccp_available_ccids(uint8_t **ccid_array)
846 static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
847 socklen_t nccids = sizeof(ccids);
850 ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
854 if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
855 ccids, &nccids) < 0) {
858 PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
860 return -ERRNO_TO_PARA_ERROR(ret);
869 * Prepare a structure for AF_UNIX socket addresses.
871 * This just copies name to the sun_path component of u, prepending a zero byte
872 * if abstract sockets are supported.
874 * The first call to this function tries to bind a socket to the abstract name
875 * space. The result of this test is stored in a static variable. Subsequent
876 * calls read this variable and create abstract sockets on systems that support
877 * them. If a NULL pointer is passed as the name, the function only
878 * initializes the static variable.
880 static int init_unix_addr(struct sockaddr_un *u, const char *name)
882 static int use_abstract;
884 memset(u->sun_path, 0, UNIX_PATH_MAX);
885 u->sun_family = PF_UNIX;
886 if (use_abstract == 0) { /* executed only once */
887 int fd = socket(PF_UNIX, SOCK_STREAM, 0);
888 memcpy(u->sun_path, "\0x\0", 3);
889 if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
890 use_abstract = 1; /* yes */
892 use_abstract = -1; /* no */
894 PARA_NOTICE_LOG("%susing abstract socket namespace\n",
895 use_abstract == 1? "" : "not ");
899 if (strlen(name) + 1 >= UNIX_PATH_MAX)
900 return -E_NAME_TOO_LONG;
901 strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
906 * Create a socket for local communication and listen on it.
908 * \param name The socket pathname.
910 * This function creates a passive local socket for sequenced, reliable,
911 * two-way, connection-based byte streams. The socket file descriptor is set to
912 * nonblocking mode and listen(2) is called to prepare the socket for
913 * accepting incoming connection requests.
915 * \return The file descriptor on success, negative error code on failure.
917 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
919 int create_local_socket(const char *name)
921 struct sockaddr_un unix_addr;
924 ret = init_unix_addr(&unix_addr, name);
925 if (ret <= 0) /* error, or name was NULL */
927 ret = socket(PF_UNIX, SOCK_STREAM, 0);
929 return -ERRNO_TO_PARA_ERROR(errno);
931 ret = mark_fd_nonblocking(fd);
934 ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
936 ret = -ERRNO_TO_PARA_ERROR(errno);
939 if (unix_addr.sun_path[0] != 0) { /* pathname socket */
940 mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
943 if (chmod(name, mode) < 0)
946 if (listen(fd , 5) < 0) {
947 ret = -ERRNO_TO_PARA_ERROR(errno);
957 * Prepare, create, and connect to a Unix domain socket for local communication.
959 * \param name The socket pathname.
961 * This function creates a local socket for sequenced, reliable, two-way,
962 * connection-based byte streams.
964 * \return The file descriptor of the connected socket on success, negative on
967 * \sa \ref create_local_socket(), unix(7), connect(2).
969 int connect_local_socket(const char *name)
971 struct sockaddr_un unix_addr;
974 PARA_DEBUG_LOG("connecting to %s\n", name);
975 fd = socket(PF_UNIX, SOCK_STREAM, 0);
977 return -ERRNO_TO_PARA_ERROR(errno);
978 ret = init_unix_addr(&unix_addr, name);
981 if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
983 ret = -ERRNO_TO_PARA_ERROR(errno);
990 ssize_t send_cred_buffer(int sock, char *buf)
992 return write_buffer(sock, buf);
994 int recv_cred_buffer(int fd, char *buf, size_t size)
996 return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
998 #else /* HAVE_UCRED */
1001 * Send a buffer and the credentials of the current process to a socket.
1003 * \param sock The file descriptor of the sending socket.
1004 * \param buf The zero-terminated buffer to send.
1006 * \return On success, this call returns the number of bytes sent. On errors,
1007 * \p -E_SENDMSG is returned.
1009 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
1011 ssize_t send_cred_buffer(int sock, char *buf)
1013 char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
1015 struct cmsghdr *cmsg;
1016 static struct iovec iov;
1022 iov.iov_len = strlen(buf);
1026 /* compose the message */
1027 memset(&msg, 0, sizeof(msg));
1030 msg.msg_control = control;
1031 msg.msg_controllen = sizeof(control);
1032 /* attach the ucred struct */
1033 cmsg = CMSG_FIRSTHDR(&msg);
1034 cmsg->cmsg_level = SOL_SOCKET;
1035 cmsg->cmsg_type = SCM_CREDENTIALS;
1036 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1037 *(struct ucred *)CMSG_DATA(cmsg) = c;
1038 msg.msg_controllen = cmsg->cmsg_len;
1039 ret = sendmsg(sock, &msg, 0);
1045 static void dispose_fds(int *fds, unsigned num)
1049 for (i = 0; i < num; i++)
1054 * Receive a buffer and the Unix credentials of the sending process.
1056 * \param fd The file descriptor of the receiving socket.
1057 * \param buf The buffer to store the received message.
1058 * \param size The length of \a buf in bytes.
1060 * \return Negative on errors, the user id of the sending process on success.
1062 * \sa \ref send_cred_buffer and the references given there.
1064 int recv_cred_buffer(int fd, char *buf, size_t size)
1066 char control[255] __a_aligned(8);
1068 struct cmsghdr *cmsg;
1074 setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
1075 memset(&msg, 0, sizeof(msg));
1076 memset(buf, 0, size);
1081 msg.msg_control = control;
1082 msg.msg_controllen = sizeof(control);
1083 if (recvmsg(fd, &msg, 0) < 0)
1085 result = -E_SCM_CREDENTIALS;
1086 cmsg = CMSG_FIRSTHDR(&msg);
1088 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
1089 == SCM_CREDENTIALS) {
1090 memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
1093 if (cmsg->cmsg_level == SOL_SOCKET
1094 && cmsg->cmsg_type == SCM_RIGHTS) {
1095 dispose_fds((int *)CMSG_DATA(cmsg),
1096 (cmsg->cmsg_len - CMSG_LEN(0))
1099 cmsg = CMSG_NXTHDR(&msg, cmsg);
1103 #endif /* HAVE_UCRED */