list.h: Convert INIT_LIST_HEAD macro to inline function.

[paraslash.git] / net.c

/* Copyright (C) 2005 Andre Noll <maan@tuebingen.mpg.de>, see file COPYING. */

/** \file net.c Networking-related helper functions. */

#include "para.h"

#include <netinet/in.h>
#include <arpa/inet.h>
#include <sys/un.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netdb.h>
#include <regex.h>

#include "error.h"
#include "net.h"
#include "string.h"
#include "list.h"
#include "fd.h"

/**
 * Parse and validate IPv4 address/netmask string.
 *
 * \param cidr    Address in CIDR notation
 * \param addr    Copy of the IPv4 address part of \a cidr
 * \param addrlen Size of \a addr in bytes
 * \param netmask Value of the netmask part in \a cidr or the
 *                default of 32 if not specified.
 *
 * \return Pointer to \a addr if successful, NULL on error.
 * \sa RFC 4632.
 */
char *parse_cidr(const char *cidr,
                 char    *addr, ssize_t addrlen,
                 int32_t *netmask)
{
        const char *o = cidr;
        char *c = addr, *end = c + (addrlen - 1);

        *netmask = 0x20;

        if (cidr == NULL || addrlen < 1)
                goto failed;

        for (o = cidr; (*c = *o == '/'? '\0' : *o); c++, o++)
                if (c == end)
                        goto failed;

        if (*o == '/')
                if (para_atoi32(++o, netmask) < 0 ||
                    *netmask < 0 || *netmask > 0x20)
                        goto failed;

        if (is_valid_ipv4_address(addr))
                return addr;
failed:
        *addr = '\0';
        return NULL;
}


/**
 * Match string as a candidate IPv4 address.
 *
 * \param address The string to match.
 * \return True if \a address has "dot-quad" format.
 */
static bool is_v4_dot_quad(const char *address)
{
        bool result;
        regex_t r;

        assert(para_regcomp(&r, "^([0-9]+\\.){3}[0-9]+$",
                REG_EXTENDED | REG_NOSUB) >= 0);
        result = regexec(&r, address, 0, NULL, 0) == 0;
        regfree(&r);
        return result;
}

/**
 * Perform basic syntax checking on the host-part of an URL:
 *
 * - Since ':' is invalid in IPv4 addresses and DNS names, the
 *   presence of ':' causes interpretation as IPv6 address;
 * - next the first-match-wins algorithm from RFC 3986 is applied;
 * - else the string is considered as DNS name, to be resolved later.
 *
 * \param host The host string to check.
 * \return True if \a host passes the syntax checks.
 *
 * \sa RFC 3986, 3.2.2; RFC 1123, 2.1; RFC 1034, 3.5.
 */
static bool host_string_ok(const char *host)
{
        if (host == NULL || *host == '\0')
                return false;
        if (strchr(host, ':') != NULL)
                return is_valid_ipv6_address(host);
        if (is_v4_dot_quad(host))
                return is_valid_ipv4_address(host);
        return true;
}

/**
 * Parse and validate URL string.
 *
 * The URL syntax is loosely based on RFC 3986, supporting one of
 * - "["host"]"[:port] for native IPv6 addresses and
 * - host[:port] for IPv4 hostnames and DNS names.
 *
 * Native IPv6 addresses must be enclosed in square brackets, since
 * otherwise there is an ambiguity with the port separator `:'.
 * The 'port' part is always considered to be a number; if absent,
 * it is set to -1, to indicate that a default port is to be used.
 *
 * The following are valid examples:
 * - 10.10.1.1
 * - 10.10.1.2:8000
 * - localhost
 * - localhost:8001
 * - [::1]:8000
 * - [badc0de::1]
 *
 * \param url     The URL string to take apart.
 * \param host    To return the copied host part of \a url.
 * \param hostlen The maximum length of \a host.
 * \param port    To return the port number (if any) of \a url.
 *
 * \return Pointer to \a host, or \p NULL if failed.  If \p NULL is returned,
 * \a host and \a port are undefined. If no port number was present in \a url,
 * \a port is set to -1.
 *
 * \sa RFC 3986, 3.2.2/3.2.3.
 */
char *parse_url(const char *url,
                char    *host, ssize_t hostlen,
                int32_t *port)
{
        const char *o = url;
        char *c = host, *end = c + (hostlen - 1);

        *port = -1;

        if (o == NULL || hostlen < 1)
                goto failed;

        if (*o == '[') {
                for (++o; (*c = *o == ']' ? '\0' : *o); c++, o++)
                        if (c == end)
                                goto failed;

                if (*o++ != ']' || (*o != '\0' && *o != ':'))
                        goto failed;
        } else {
                for (; (*c = *o == ':'? '\0' : *o); c++, o++) {
                        if (c == end && o[1])
                                goto failed;
                }
        }

        if (*o == ':')
                if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff)
                        goto failed;
        if (host_string_ok(host))
                return host;
failed:
        *host = '\0';
        return NULL;
}

/**
 * Pretty-print a host/port pair.
 *
 * \param url NULL, or any string accepted by \ref parse_url().
 * \param default_port Applies if url has no port.
 *
 * If the url argument is NULL, the function returns the string
 * 0.0.0.0:default_port. Otherwise it calls \ref parse_url() to check the
 * syntax of the input string given by url. On errors the string "?" is
 * returned. Otherwise, if url contains a port, a copy of url is returned. If
 * no port was supplied, a colon and the default port are appended to url.
 *
 * \return In all cases the returned string is a allocated with malloc(3) and
 * has to be freed by the caller.
 */
char *format_url(const char *url, int default_port)
{
        char host[MAX_HOSTLEN];
        int url_port;

        if (!url)
                return make_message("0.0.0.0:%d", default_port);
        if (!parse_url(url, host, sizeof(host), &url_port))
                return make_message("?");
        if (url_port < 0)
                return make_message("%s:%d", url, default_port);
        else
                return para_strdup(url);
}

/**
 * Stringify port number, resolve into service name where defined.
 *
 * \param port 2-byte port number, in host-byte-order.
 * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL.
 * \return Pointer to static result buffer.
 *
 * \sa getservent(3), services(5), nsswitch.conf(5).
 */
const char *stringify_port(int port, const char *transport)
{
        static char service[NI_MAXSERV];

        if (port < 0 || port > 0xFFFF) {
                snprintf(service, sizeof(service), "undefined (%d)", port);
        } else {
                struct servent *se = getservbyport(htons(port), transport);

                if (se == NULL)
                        snprintf(service, sizeof(service), "%d", port);
                else
                        snprintf(service, sizeof(service), "%s", se->s_name);
        }
        return service;
}

/**
 * Determine the socket type for a given layer-4 protocol.
 *
 * \param l4type The symbolic name of the transport-layer protocol.
 *
 * \sa ip(7), socket(2).
 */
static inline int sock_type(const unsigned l4type)
{
        switch (l4type) {
        case IPPROTO_UDP:       return SOCK_DGRAM;
        case IPPROTO_TCP:       return SOCK_STREAM;
        case IPPROTO_DCCP:      return SOCK_DCCP;
        }
        return -1;              /* not supported here */
}

/**
 * Pretty-print transport-layer name.
 */
static const char *layer4_name(const unsigned l4type)
{
        switch (l4type) {
        case IPPROTO_UDP:       return "UDP";
        case IPPROTO_TCP:       return "TCP";
        case IPPROTO_DCCP:      return "DCCP";
        }
        return "UNKNOWN PROTOCOL";
}

/**
 * Flowopts: Transport-layer independent encapsulation of socket options.
 *
 * These collect individual socket options into a queue, which is disposed of
 * directly after makesock(). The 'pre_conn_opt' structure is for internal use
 * only and should not be visible elsewhere.
 *
 * \sa setsockopt(2), \ref makesock().
 */
struct pre_conn_opt {
        int             sock_level;     /**< Second argument to setsockopt() */
        int             sock_option;    /**< Third argument to setsockopt()  */
        char            *opt_name;      /**< Stringified \a sock_option      */
        void            *opt_val;       /**< Fourth argument to setsockopt() */
        socklen_t       opt_len;        /**< Fifth argument to setsockopt()  */

        struct list_head node;          /**< FIFO, as sockopt order matters. */
};

/** FIFO list of pre-connection socket options to be set */
struct flowopts {
        struct list_head sockopts;
};

/**
 * Allocate and initialize a flowopt queue.
 *
 * \return A new structure to be passed to \ref flowopt_add(). It is
 * automatically deallocated in \ref makesock().
 */
struct flowopts *flowopt_new(void)
{
        struct flowopts *new = para_malloc(sizeof(*new));

        init_list_head(&new->sockopts);
        return new;
}

/**
 * Append new socket option to flowopt queue.
 *
 * \param fo   The flowopt queue to append to.
 * \param lev  Level at which \a opt resides.
 * \param opt  New option to add.
 * \param name Stringified name of \a opt.
 * \param val  The value to set \a opt to.
 * \param len  Length of \a val.
 *
 *  \sa setsockopt(2).
 */
void flowopt_add(struct flowopts *fo, int lev, int opt,
                const char *name, const void *val, int len)
{
        struct pre_conn_opt *new = para_malloc(sizeof(*new));

        new->sock_option = opt;
        new->sock_level  = lev;
        new->opt_name    = para_strdup(name);

        if (val == NULL) {
                new->opt_val = NULL;
                new->opt_len = 0;
        } else {
                new->opt_val = para_malloc(len);
                new->opt_len = len;
                memcpy(new->opt_val, val, len);
        }

        list_add_tail(&new->node, &fo->sockopts);
}

/** Set the entire bunch of pre-connection options at once. */
static void flowopt_setopts(int sockfd, struct flowopts *fo)
{
        struct pre_conn_opt *pc;

        if (fo == NULL)
                return;

        list_for_each_entry(pc, &fo->sockopts, node)
                if (setsockopt(sockfd, pc->sock_level, pc->sock_option,
                                       pc->opt_val, pc->opt_len) < 0) {
                        PARA_EMERG_LOG("Can not set %s socket option: %s",
                                       pc->opt_name, strerror(errno));
                        exit(EXIT_FAILURE);
                }
}

/**
 * Deallocate all resources of a flowopts structure.
 *
 * \param fo A pointer as returned from flowopt_new().
 *
 * It's OK to pass \p NULL here in which case the function does nothing.
 */
void flowopt_cleanup(struct flowopts *fo)
{
        struct pre_conn_opt *cur, *next;

        if (fo == NULL)
                return;

        list_for_each_entry_safe(cur, next, &fo->sockopts, node) {
                free(cur->opt_name);
                free(cur->opt_val);
                free(cur);
        }
        free(fo);
}

/**
 * Resolve an IPv4/IPv6 address.
 *
 * \param l4type The layer-4 type (\p IPPROTO_xxx).
 * \param passive Whether \p AI_PASSIVE should be included as hint.
 * \param host Remote or local hostname or IPv/6 address string.
 * \param port_number Used to set the port in each returned address structure.
 * \param result addrinfo structures are returned here.
 *
 * The interpretation of \a host depends on the value of \a passive. On a
 * passive socket host is interpreted as an interface IPv4/6 address (can be
 * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6
 * address to connect to.
 *
 * \return Standard.
 *
 * \sa getaddrinfo(3).
 */
int lookup_address(unsigned l4type, bool passive, const char *host,
                int port_number, struct addrinfo **result)
{
        int ret;
        char port[6]; /* port number has at most 5 digits */
        struct addrinfo *addr = NULL, hints;

        *result = NULL;
        sprintf(port, "%d", port_number & 0xffff);
        /* Set up address hint structure */
        memset(&hints, 0, sizeof(hints));
        hints.ai_family = AF_UNSPEC;
        hints.ai_socktype = sock_type(l4type);
        /*
         * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup
         * (and only then) pretend to be UDP.
         */
        if (l4type == IPPROTO_DCCP)
                hints.ai_socktype = SOCK_DGRAM;
        /* only use addresses available on the host */
        hints.ai_flags = AI_ADDRCONFIG;
        if (passive && host == NULL)
                hints.ai_flags |= AI_PASSIVE;
        /* Obtain local/remote address information */
        ret = getaddrinfo(host, port, &hints, &addr);
        if (ret != 0) {
                PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n",
                        layer4_name(l4type),
                        host? host : (passive? "[loopback]" : "[localhost]"),
                        port, gai_strerror(ret));
                return -E_ADDRESS_LOOKUP;
        }
        *result = addr;
        return 1;
}

/**
 * Create an active or passive socket.
 *
 * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP.
 * \param passive Whether to call bind(2) or connect(2).
 * \param ai Address information as obtained from \ref lookup_address().
 * \param fo Socket options to be set before making the connection.
 *
 * bind(2) is called on passive sockets, and connect(2) on active sockets. The
 * algorithm tries all possible address combinations until it succeeds. If \a
 * fo is supplied, options are set but cleanup must be performed in the caller.
 *
 * \return File descriptor on success, \p E_MAKESOCK on errors.
 *
 * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2),
 * connect(2).
 */
int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai,
                struct flowopts *fo)
{
        int ret = -E_MAKESOCK, on = 1;

        for (; ai; ai = ai->ai_next) {
                int fd;
                ret = socket(ai->ai_family, sock_type(l4type), l4type);
                if (ret < 0) {
                        PARA_NOTICE_LOG("socket(): %s\n", strerror(errno));
                        continue;
                }
                fd = ret;
                flowopt_setopts(fd, fo);
                if (!passive) {
                        if (connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
                                PARA_NOTICE_LOG("connect(): %s\n",
                                        strerror(errno));
                                close(fd);
                                continue;
                        }
                        return fd;
                }
                /*
                 * Reuse the address on passive sockets to avoid failure on
                 * restart (protocols using listen()) and when creating
                 * multiple listener instances (UDP multicast).
                 */
                if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on,
                                sizeof(on)) == -1) {
                        PARA_NOTICE_LOG("setsockopt(): %s\n", strerror(errno));
                        close(fd);
                        continue;
                }
                if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
                        PARA_NOTICE_LOG("bind(): %s\n", strerror(errno));
                        close(fd);
                        continue;
                }
                return fd;
        }
        return -E_MAKESOCK;
}

/**
 * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket.
 *
 * \param l4type The layer-4 type (\p IPPROTO_xxx).
 * \param passive Whether this is a passive or active socket.
 * \param host Passed to \ref lookup_address().
 * \param port_number Passed to \ref lookup_address().
 * \param fo Passed to \ref makesock_addrinfo().
 *
 * This creates a ready-made IPv4/v6 socket structure after looking up the
 * necessary parameters. The function first calls \ref lookup_address() and
 * passes the address information to makesock_addrinfo() to create and
 * initialize the socket.
 *
 * \return The newly created file descriptor on success, a negative error code
 * on failure.
 *
 * \sa \ref lookup_address(), \ref makesock_addrinfo().
 */
int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number,
                struct flowopts *fo)
{
        struct addrinfo *ai;
        int ret = lookup_address(l4type, passive, host, port_number, &ai);

        if (ret >= 0)
                ret = makesock_addrinfo(l4type, passive, ai, fo);
        if (ai)
                freeaddrinfo(ai);
        if (ret < 0) {
                PARA_ERROR_LOG("can not create %s socket %s#%d.\n",
                layer4_name(l4type), host? host : (passive?
                "[loopback]" : "[localhost]"), port_number);
        }
        return ret;
}

/**
 * Create a passive / listening socket.
 *
 * \param l4type The transport-layer type (\p IPPROTO_xxx).
 * \param addr Passed to \ref parse_url() if not NULL.
 * \param port Ignored if addr contains a port number.
 *
 * \return Positive integer (socket descriptor) on success, negative value
 * otherwise.
 *
 * \sa \ref makesock(), ip(7), ipv6(7), bind(2), listen(2).
 */
int para_listen(unsigned l4type, const char *addr, uint16_t port)
{
        char host[MAX_HOSTLEN];
        int ret, fd, addr_port;

        if (addr) {
                if (!parse_url(addr, host, sizeof(host), &addr_port))
                        return -ERRNO_TO_PARA_ERROR(EINVAL);
                if (addr_port > 0)
                        port = addr_port;
                addr = host;
        }
        fd = makesock(l4type, true /* passive */, addr, port,
                NULL /* no flowopts */);
        if (fd > 0) {
                ret = listen(fd, BACKLOG);
                if (ret < 0) {
                        ret = errno;
                        close(fd);
                        return -ERRNO_TO_PARA_ERROR(ret);
                }
                PARA_INFO_LOG("listening on %s port %u, fd %d\n",
                        layer4_name(l4type), port, fd);
        }
        return fd;
}

/**
 * Create a socket which listens on all network addresses.
 *
 * \param l4type See \ref para_listen().
 * \param port See \ref para_listen().
 *
 * This is a simple wrapper for \ref para_listen() which passes a NULL pointer
 * as the address information.
 *
 * \return See \ref para_listen().
 */
int para_listen_simple(unsigned l4type, uint16_t port)
{
        return para_listen(l4type, NULL, port);
}

/**
 * Determine IPv4/v6 socket address length.
 * \param sa Container of IPv4 or IPv6 address.
 * \return Address-family dependent address length.
 */
static socklen_t salen(const struct sockaddr *sa)
{
        assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6);

        return sa->sa_family == AF_INET6
                ? sizeof(struct sockaddr_in6)
                : sizeof(struct sockaddr_in);
}

/** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */
static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss)
{
        const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;

        return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr);
}

/**
 * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address.
 * \param ss Container of IPv4/6 address.
 * \return Pointer to normalized address (may be static storage).
 *
 * \sa RFC 3493.
 */
static const struct sockaddr *
normalize_ip_address(const struct sockaddr_storage *ss)
{
        assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6);

        if (SS_IS_ADDR_V4MAPPED(ss)) {
                const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss;
                static struct sockaddr_in ia;

                ia.sin_family = AF_INET;
                ia.sin_port   = ia6->sin6_port;
                memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4);
                return (const struct sockaddr *)&ia;
        }
        return (const struct sockaddr *)ss;
}

/**
 * Generic/fallback MTU values
 *
 * These are taken from RFC 1122, RFC 2460, and RFC 5405.
 * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends
 *   to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown;
 * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes;
 * - RFC 5405, 3.2 recommends that if path MTU discovery is not done,
 *   UDP senders should use the respective minimum values of EMTU_S.
 */
static inline int generic_mtu(const int af_type)
{
        return af_type == AF_INET6 ? 1280 : 576;
}

/** Crude approximation of IP header overhead - neglecting options. */
static inline int estimated_header_overhead(const int af_type)
{
        return af_type == AF_INET6 ? 40 : 20;
}

/**
 * Get the maximum transport-layer message size (MMS_S).
 *
 * \param sockfd The socket file descriptor.
 *
 * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family
 * could not be determined, \p AF_INET is assumed.
 *
 * \return The maximum message size of the address family type.
 */
int generic_max_transport_msg_size(int sockfd)
{
        struct sockaddr_storage ss = {.ss_family = 0};
        socklen_t sslen = sizeof(ss);
        int af_type = AF_INET;

        if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) {
                PARA_ERROR_LOG("can not determine remote address type: %s\n",
                                strerror(errno));
        } else if (!SS_IS_ADDR_V4MAPPED(&ss)) {
                af_type = ss.ss_family;
        }
        return generic_mtu(af_type) - estimated_header_overhead(af_type);
}

/**
 * Look up the remote side of a connected socket structure.
 *
 * \param fd The socket descriptor of the connected socket.
 *
 * \return A static character string identifying hostname and port of the
 * chosen side in numeric host:port format.
 *
 * \sa getsockname(2), getpeername(2), \ref parse_url(), getnameinfo(3),
 * services(5), nsswitch.conf(5).
 */
char *remote_name(int fd)
{
        struct sockaddr_storage ss = {.ss_family = 0};
        const struct sockaddr *sa;
        socklen_t sslen = sizeof(ss);
        char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
        static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
        int ret;

        if (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) {
                PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
                        fd, strerror(errno));
                snprintf(output, sizeof(output), "(unknown)");
                return output;
        }
        sa = normalize_ip_address(&ss);
        ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf,
                sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
        if (ret) {
                PARA_WARNING_LOG("hostname lookup error (%s).\n",
                        gai_strerror(ret));
                snprintf(output, sizeof(output), "(lookup error)");
        } else if (sa->sa_family == AF_INET6)
                snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf);
        else
                snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf);
        return output;
}

/**
 * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage.
 *
 * \param ss Container of IPv4/6 address.
 * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful.
 *
 * \sa RFC 3493.
 */
void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia)
{
        const struct sockaddr *sa = normalize_ip_address(ss);

        memset(ia, 0, sizeof(*ia));
        if (sa->sa_family == AF_INET)
                *ia = ((struct sockaddr_in *)sa)->sin_addr;
}

/**
 * Compare the address part of IPv4/6 addresses.
 *
 * \param sa1 First address.
 * \param sa2 Second address.
 *
 * \return True iff the IP address of \a sa1 and \a sa2 match.
 */
bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
{
        if (!sa1 || !sa2)
                return false;
        if (sa1->sa_family != sa2->sa_family)
                return false;
        if (sa1->sa_family == AF_INET) {
                struct sockaddr_in *a1 = (typeof(a1))sa1,
                        *a2 = (typeof (a2))sa2;
                return a1->sin_addr.s_addr == a2->sin_addr.s_addr;
        } else if (sa1->sa_family == AF_INET6) {
                struct sockaddr_in6 *a1 = (typeof(a1))sa1,
                        *a2 = (typeof (a2))sa2;
                return !memcmp(a1, a2, sizeof(*a1));
        } else
                return false;
}

/**
 * Receive data from a file descriptor.
 *
 * \param fd The file descriptor.
 * \param buf The buffer to write the data to.
 * \param size The size of \a buf.
 *
 * Receive at most \a size bytes from file descriptor \a fd.
 *
 * \return The number of bytes received on success, negative on errors, zero if
 * the peer has performed an orderly shutdown.
 *
 * \sa recv(2).
 */
__must_check int recv_bin_buffer(int fd, char *buf, size_t size)
{
        ssize_t n;

        n = recv(fd, buf, size, 0);
        if (n == -1)
                return -ERRNO_TO_PARA_ERROR(errno);
        return n;
}

/**
 * Receive and write terminating NULL byte.
 *
 * \param fd The file descriptor.
 * \param buf The buffer to write the data to.
 * \param size The size of \a buf.
 *
 * Read at most \a size - 1 bytes from file descriptor \a fd and
 * write a NULL byte at the end of the received data.
 *
 * \return The return value of the underlying call to \a recv_bin_buffer().
 *
 * \sa \ref recv_bin_buffer()
 */
int recv_buffer(int fd, char *buf, size_t size)
{
        int n;

        assert(size);
        n = recv_bin_buffer(fd, buf, size - 1);
        if (n >= 0)
                buf[n] = '\0';
        else
                *buf = '\0';
        return n;
}

/**
 * Wrapper around the accept system call.
 *
 * \param fd The listening socket.
 * \param rfds An optional fd_set pointer.
 * \param addr Structure which is filled in with the address of the peer socket.
 * \param size Should contain the size of the structure pointed to by \a addr.
 * \param new_fd Result pointer.
 *
 * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is
 * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept().
 *
 * \return Negative on errors, zero if no connections are present to be accepted,
 * one otherwise.
 *
 * \sa accept(2).
 */
int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd)
{
        int ret;

        if (rfds && !FD_ISSET(fd, rfds))
                return 0;
        do
                ret = accept(fd, (struct sockaddr *) addr, &size);
        while (ret < 0 && errno == EINTR);

        if (ret >= 0) {
                *new_fd = ret;
                return 1;
        }
        if (errno == EAGAIN || errno == EWOULDBLOCK)
                return 0;
        return -ERRNO_TO_PARA_ERROR(errno);
}

/**
 * Probe the list of DCCP CCIDs configured on this host.
 * \param ccid_array Pointer to return statically allocated array in.
 * \return Number of elements returned in \a ccid_array or error.
 *
 * NB: This feature is only available on Linux > 2.6.30; on older kernels
 * ENOPROTOOPT ("Protocol not available") will be returned.
 */
int dccp_available_ccids(uint8_t **ccid_array)
{
        static uint8_t ccids[DCCP_MAX_HOST_CCIDS];
        socklen_t nccids = sizeof(ccids);
        int ret, fd;

        ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL);
        if (ret < 0)
                return ret;

        if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS,
                                                ccids, &nccids) < 0) {
                ret = errno;
                close(fd);
                PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n",
                                strerror(ret));
                return -ERRNO_TO_PARA_ERROR(ret);
        }

        close(fd);
        *ccid_array = ccids;
        return nccids;
}

/*
 * Prepare a structure for AF_UNIX socket addresses.
 *
 * This just copies name to the sun_path component of u, prepending a zero byte
 * if abstract sockets are supported.
 *
 * The first call to this function tries to bind a socket to the abstract name
 * space. The result of this test is stored in a static variable. Subsequent
 * calls read this variable and create abstract sockets on systems that support
 * them. If a NULL pointer is passed as the name, the function only
 * initializes the static variable.
 */
static int init_unix_addr(struct sockaddr_un *u, const char *name)
{
        static int use_abstract;

        memset(u->sun_path, 0, UNIX_PATH_MAX);
        u->sun_family = PF_UNIX;
        if (use_abstract == 0) { /* executed only once */
                int fd = socket(PF_UNIX, SOCK_STREAM, 0);
                memcpy(u->sun_path, "\0x\0", 3);
                if (bind(fd, (struct sockaddr *)u, sizeof(*u)) == 0)
                        use_abstract = 1; /* yes */
                else
                        use_abstract = -1; /* no */
                close(fd);
                PARA_NOTICE_LOG("%susing abstract socket namespace\n",
                        use_abstract == 1? "" : "not ");
        }
        if (!name)
                return 0;
        if (strlen(name) + 1 >= UNIX_PATH_MAX)
                return -E_NAME_TOO_LONG;
        strcpy(u->sun_path + (use_abstract == 1? 1 : 0), name);
        return 1;
}

/**
 * Create a socket for local communication and listen on it.
 *
 * \param name The socket pathname.
 *
 * This function creates a passive local socket for sequenced, reliable,
 * two-way, connection-based byte streams. The socket file descriptor is set to
 * nonblocking mode and listen(2) is called to prepare the socket for
 * accepting incoming connection requests.
 *
 * \return The file descriptor on success, negative error code on failure.
 *
 * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7).
 */
int create_local_socket(const char *name)
{
        struct sockaddr_un unix_addr;
        int fd, ret;

        ret = init_unix_addr(&unix_addr, name);
        if (ret <= 0) /* error, or name was NULL */
                return ret;
        ret = socket(PF_UNIX, SOCK_STREAM, 0);
        if (ret < 0)
                return -ERRNO_TO_PARA_ERROR(errno);
        fd = ret;
        ret = mark_fd_nonblocking(fd);
        if (ret < 0)
                goto err;
        ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr));
        if (ret < 0) {
                ret = -ERRNO_TO_PARA_ERROR(errno);
                goto err;
        }
        if (unix_addr.sun_path[0] != 0) { /* pathname socket */
                mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
                        | S_IROTH | S_IWOTH;
                ret = -E_CHMOD;
                if (chmod(name, mode) < 0)
                        goto err;
        }
        if (listen(fd , 5) < 0) {
                ret = -ERRNO_TO_PARA_ERROR(errno);
                goto err;
        }
        return fd;
err:
        close(fd);
        return ret;
}

/**
 * Prepare, create, and connect to a Unix domain socket for local communication.
 *
 * \param name The socket pathname.
 *
 * This function creates a local socket for sequenced, reliable, two-way,
 * connection-based byte streams.
 *
 * \return The file descriptor of the connected socket on success, negative on
 * errors.
 *
 * \sa \ref create_local_socket(), unix(7), connect(2).
 */
int connect_local_socket(const char *name)
{
        struct sockaddr_un unix_addr;
        int fd, ret;

        PARA_DEBUG_LOG("connecting to %s\n", name);
        fd = socket(PF_UNIX, SOCK_STREAM, 0);
        if (fd < 0)
                return -ERRNO_TO_PARA_ERROR(errno);
        ret = init_unix_addr(&unix_addr, name);
        if (ret < 0)
                goto err;
        if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1)
                return fd;
        ret = -ERRNO_TO_PARA_ERROR(errno);
err:
        close(fd);
        return ret;
}

#ifndef HAVE_UCRED
ssize_t send_cred_buffer(int sock, char *buf)
{
        return write_buffer(sock, buf);
}
int recv_cred_buffer(int fd, char *buf, size_t size)
{
        return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG;
}
#else /* HAVE_UCRED */

/**
 * Send a buffer and the credentials of the current process to a socket.
 *
 * \param sock The file descriptor of the sending socket.
 * \param buf The zero-terminated buffer to send.
 *
 * \return On success, this call returns the number of bytes sent. On errors,
 * \p -E_SENDMSG is returned.
 *
 * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7).
 */
ssize_t send_cred_buffer(int sock, char *buf)
{
        char control[sizeof(struct cmsghdr) + sizeof(struct ucred)];
        struct msghdr msg;
        struct cmsghdr *cmsg;
        static struct iovec iov;
        struct ucred c;
        int ret;

        /* Response data */
        iov.iov_base = buf;
        iov.iov_len = strlen(buf);
        c.pid = getpid();
        c.uid = getuid();
        c.gid = getgid();
        /* compose the message */
        memset(&msg, 0, sizeof(msg));
        msg.msg_iov = &iov;
        msg.msg_iovlen = 1;
        msg.msg_control = control;
        msg.msg_controllen = sizeof(control);
        /* attach the ucred struct */
        cmsg = CMSG_FIRSTHDR(&msg);
        cmsg->cmsg_level = SOL_SOCKET;
        cmsg->cmsg_type = SCM_CREDENTIALS;
        cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
        *(struct ucred *)CMSG_DATA(cmsg) = c;
        msg.msg_controllen = cmsg->cmsg_len;
        ret = sendmsg(sock, &msg, 0);
        if (ret < 0)
                ret = -E_SENDMSG;
        return ret;
}

static void dispose_fds(int *fds, unsigned num)
{
        int i;

        for (i = 0; i < num; i++)
                close(fds[i]);
}

/**
 * Receive a buffer and the Unix credentials of the sending process.
 *
 * \param fd The file descriptor of the receiving socket.
 * \param buf The buffer to store the received message.
 * \param size The length of \a buf in bytes.
 *
 * \return Negative on errors, the user id of the sending process on success.
 *
 * \sa \ref send_cred_buffer and the references given there.
 */
int recv_cred_buffer(int fd, char *buf, size_t size)
{
        char control[255] __a_aligned(8);
        struct msghdr msg;
        struct cmsghdr *cmsg;
        struct iovec iov;
        int result = 0;
        int yes = 1;
        struct ucred cred;

        setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &yes, sizeof(int));
        memset(&msg, 0, sizeof(msg));
        memset(buf, 0, size);
        iov.iov_base = buf;
        iov.iov_len = size;
        msg.msg_iov = &iov;
        msg.msg_iovlen = 1;
        msg.msg_control = control;
        msg.msg_controllen = sizeof(control);
        if (recvmsg(fd, &msg, 0) < 0)
                return -E_RECVMSG;
        result = -E_SCM_CREDENTIALS;
        cmsg = CMSG_FIRSTHDR(&msg);
        while (cmsg) {
                if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type
                                == SCM_CREDENTIALS) {
                        memcpy(&cred, CMSG_DATA(cmsg), sizeof(struct ucred));
                        result = cred.uid;
                } else
                        if (cmsg->cmsg_level == SOL_SOCKET
                                        && cmsg->cmsg_type == SCM_RIGHTS) {
                                dispose_fds((int *)CMSG_DATA(cmsg),
                                        (cmsg->cmsg_len - CMSG_LEN(0))
                                        / sizeof(int));
                        }
                cmsg = CMSG_NXTHDR(&msg, cmsg);
        }
        return result;
}
#endif /* HAVE_UCRED */