NEWS update.

[paraslash.git] / net.c

/*
 * Copyright (C) 2005-2011 Andre Noll <maan@systemlinux.org>
 *
 * Licensed under the GPL v2. For licencing details see COPYING.
 */

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

/*
 * Since glibc 2.8, the _GNU_SOURCE feature test macro must be defined in order
 * to obtain the definition of the ucred structure.
 */
#define _GNU_SOURCE

#include <netdb.h>

/* At least NetBSD needs these. */
#ifndef AI_V4MAPPED
#define AI_V4MAPPED 0
#endif
#ifndef AI_ALL
#define AI_ALL 0
#endif
#ifndef AI_ADDRCONFIG
#define AI_ADDRCONFIG 0
#endif

#include <dirent.h>
#include <regex.h>
#include <openssl/rc4.h>

#include "para.h"
#include "error.h"
#include "crypt.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 succesful, 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 NULL if failed.
 * If NULL is returned, \a host and \a portnum are undefined. If no
 * port number was present in \a url, \a portnum 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)
                                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;
}

/**
 * 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), "%u", 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), 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;
};

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,
                 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);
}

void flowopt_add_bool(struct flowopts *fo, int lev, int opt,
                      char *optname, bool on_or_off)
{
        int on = on_or_off;     /* kernel takes 'int' */

        flowopt_add(fo, lev, opt, optname, &on, sizeof(on));
}

/** 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);
                }
}

static 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 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 (1) or active (0) socket.
 * \param host Remote or local hostname or IPv/6 address string.
 * \param port_number Decimal port number.
 * \param fo Socket options to be set before making the connection.
 *
 * This creates a ready-made IPv4/v6 socket structure after looking up the
 * necessary parameters. 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;
 *      - \a port_number is in either case the numeric port number (not service
 *        string).
 *
 * Furthermore, 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 and cleanup is performed.
 *
 * \return This function returns 1 on success and \a -E_ADDRESS_LOOKUP when no
 * matching connection could be set up (with details in the error log).
 *
 *  \sa ipv6(7), getaddrinfo(3), bind(2), connect(2).
 */
int makesock(unsigned l4type, bool passive,
             const char *host, uint16_t port_number,
             struct flowopts *fo)
{
        struct addrinfo *local = NULL, *src = NULL, *remote = NULL,
                *dst = NULL, hints;
        unsigned int    l3type = AF_UNSPEC;
        int             rc, on = 1, sockfd = -1,
                        socktype = sock_type(l4type);
        char port[6]; /* port number has at most 5 digits */

        sprintf(port, "%u", port_number);
        /* Set up address hint structure */
        memset(&hints, 0, sizeof(hints));
        hints.ai_family = l3type;
        hints.ai_socktype = socktype;
        /* 
         * 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 (l3type == AF_INET6)
                /* use v4-mapped-v6 if no v6 addresses found */
                hints.ai_flags |= AI_V4MAPPED | AI_ALL;

        if (passive && host == NULL)
                hints.ai_flags |= AI_PASSIVE;

        /* Obtain local/remote address information */
        if ((rc = getaddrinfo(host, port, &hints, passive ? &local : &remote))) {
                PARA_ERROR_LOG("can not resolve %s address %s#%s: %s.\n",
                                layer4_name(l4type),
                                host? host : (passive? "[loopback]" : "[localhost]"),
                                port, gai_strerror(rc));
                rc = -E_ADDRESS_LOOKUP;
                goto out;
        }

        /* Iterate over all src/dst combination, exhausting dst first */
        for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) {
                if (src && dst && src->ai_family == AF_INET
                                && dst->ai_family == AF_INET6)
                        goto get_next_dst; /* v4 -> v6 is not possible */

                sockfd = socket(src ? src->ai_family : dst->ai_family,
                        socktype, l4type);
                if (sockfd < 0)
                        goto get_next_dst;

                /*
                 * Reuse the address on passive sockets to avoid failure on
                 * restart (protocols using listen()) and when creating
                 * multiple listener instances (UDP multicast).
                 */
                if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
                                                  &on, sizeof(on)) == -1) {
                        rc = errno;
                        close(sockfd);
                        PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n",
                                       strerror(rc));
                        rc = -ERRNO_TO_PARA_ERROR(rc);
                        break;
                }
                flowopt_setopts(sockfd, fo);

                if (src) {
                        if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) {
                                close(sockfd);
                                goto get_next_src;
                        }
                        if (!dst) /* bind-only completed successfully */
                                break;
                }

                if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0)
                        break; /* connection completed successfully */
                close(sockfd);
get_next_dst:
                if (dst && (dst = dst->ai_next))
                        continue;
get_next_src:
                if (src && (src = src->ai_next)) /* restart inner loop */
                        dst = remote;
        }
out:
        if (local)
                freeaddrinfo(local);
        if (remote)
                freeaddrinfo(remote);
        flowopt_cleanup(fo);

        if (src == NULL && dst == NULL) {
                if (rc >= 0)
                        rc = -E_MAKESOCK;
                PARA_ERROR_LOG("can not create %s socket %s#%s.\n",
                        layer4_name(l4type), host? host : (passive?
                        "[loopback]" : "[localhost]"), port);
                return rc;
        }
        return sockfd;
}

/**
 * Create a passive / listening socket.
 *
 * \param l4type The transport-layer type (\p IPPROTO_xxx).
 * \param port The decimal port number to listen on.
 * \param fo Flowopts (if any) to set before starting to listen.
 *
 * \return Positive integer (socket descriptor) on success, negative value
 * otherwise.
 *
 * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2).
 */
int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo)
{
        int ret, fd = makesock(l4type, 1, NULL, port, fo);

        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;
}

/**
 * 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;
}

/**
 * Maximum transport-layer message size (MMS_S) as per RFC 1122, 3.3.3
 * Socket must be connected.
 */
int generic_max_transport_msg_size(int sockfd)
{
        struct sockaddr_storage ss;
        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);
}

/**
 * Print numeric host and port number (beware - uses static char).
 *
 * \param sa The IPv4/IPv6 socket address to use.
 *
 * \return Host string in numeric host:port format, \sa parse_url().
 * \sa getnameinfo(3), services(5), nsswitch.conf(5)
 */
static char *host_and_port(const struct sockaddr_storage *ss)
{
        const struct sockaddr *sa = normalize_ip_address(ss);
        char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
        static char output[sizeof(hbuf) + sizeof(sbuf) + 4];
        int ret;

        ret = getnameinfo(sa, salen(sa),
                          hbuf, sizeof(hbuf),
                          sbuf, sizeof(sbuf),
                          NI_NUMERICHOST | NI_NUMERICSERV);
        if (ret) {
                snprintf(output, sizeof(output), "(unknown)");
                PARA_WARNING_LOG("hostname lookup error (%s).\n",
                                 gai_strerror(ret));
        } 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;
}

/**
 * Look up the local or remote side of a connected socket structure.
 *
 * \param fd The socket descriptor of the connected socket.
 * \param getname Either \p getsockname() for local, or \p getpeername() for
 * remote side.
 *
 * \return A static character string identifying hostname and port of the
 * chosen side.
 *
 * \sa getsockname(2), getpeername(2).
 */
static char *__get_sock_name(int fd, int (*getname)(int, struct sockaddr*,
                socklen_t *))
{
        struct sockaddr_storage ss;
        socklen_t sslen = sizeof(ss);

        if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) {
                static char *dont_know = "(don't know)";
                PARA_ERROR_LOG("can not determine address from fd %d: %s\n",
                        fd, strerror(errno));
                return dont_know;
        }
        return host_and_port(&ss);
}

/**
 * Look up the local side of a connected socket structure.
 *
 * \param sockfd The file descriptor of the socket.
 *
 * \return A pointer to a static buffer containing hostname an port. This
 * buffer must not be freed by the caller.
 *
 * \sa remote_name().
 */
char *local_name(int sockfd)
{
        return __get_sock_name(sockfd, getsockname);
}

/**
 * Look up the remote side of a connected socket structure.
 *
 * \param sockfd The file descriptor of the socket.
 *
 * \return Analogous to the return value of \ref local_name() but for the
 * remote side.
 *
 * \sa local_name().
 */
char *remote_name(int sockfd)
{
        return __get_sock_name(sockfd, getpeername);
}

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

        if (sa->sa_family == AF_INET)
                ia = ((struct sockaddr_in *)sa)->sin_addr;
        return ia;
}

/**
 * Send a binary buffer.
 *
 * \param fd The file descriptor.
 * \param buf The buffer to be sent.
 * \param len The length of \a buf.
 *
 * Send out the buffer and try to resend the remaining part in case of short
 * writes.
 *
 * \return Standard.
 */
int send_bin_buffer(int fd, const char *buf, size_t len)
{
        if (!len)
                PARA_CRIT_LOG("len == 0\n");
        return write_all(fd, buf, &len);
}

/**
 * Send a \p NULL-terminated buffer.
 *
 * \param fd The file descriptor.
 * \param buf The null-terminated buffer to be send.
 *
 * This is equivalent to send_bin_buffer(fd, buf, strlen(buf)).
 *
 * \return Standard.
 */
int send_buffer(int fd, const char *buf)
{
        return send_bin_buffer(fd, buf, strlen(buf));
}

/**
 * Send a buffer given by a format string.
 *
 * \param fd The file descriptor.
 * \param fmt A format string.
 *
 * \return Standard.
 */
__printf_2_3 int send_va_buffer(int fd, const char *fmt, ...)
{
        char *msg;
        int ret;

        PARA_VSPRINTF(fmt, msg);
        ret = send_buffer(fd, msg);
        free(msg);
        return ret;
}

/**
 * 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 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 \p AF_UNIX socket addresses.
 *
 * \param u Pointer to the struct to be prepared.
 * \param name The socket pathname.
 *
 * This just copies \a name to the sun_path component of \a u.
 *
 * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer
 * than \p UNIX_PATH_MAX.
 */
static int init_unix_addr(struct sockaddr_un *u, const char *name)
{
        if (strlen(name) >= UNIX_PATH_MAX)
                return -E_NAME_TOO_LONG;
        memset(u->sun_path, 0, UNIX_PATH_MAX);
        u->sun_family = PF_UNIX;
        strcpy(u->sun_path, name);
        return 1;
}

/**
 * Prepare, create, and bind a socket for local communication.
 *
 * \param name The socket pathname.
 * \param unix_addr Pointer to the \p AF_UNIX socket structure.
 * \param mode The desired mode of the socket.
 *
 * This function creates a local socket for sequenced, reliable,
 * two-way, connection-based byte streams.
 *
 * \return The file descriptor, on success, negative on errors.
 *
 * \sa socket(2)
 * \sa bind(2)
 * \sa chmod(2)
 */
int create_local_socket(const char *name, struct sockaddr_un *unix_addr,
                mode_t mode)
{
        int fd, ret;

        ret = init_unix_addr(unix_addr, name);
        if (ret < 0)
                return ret;
        ret = socket(PF_UNIX, SOCK_STREAM, 0);
        if (ret < 0)
                return -ERRNO_TO_PARA_ERROR(errno);
        fd = ret;
        ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX);
        if (ret < 0) {
                ret = -ERRNO_TO_PARA_ERROR(errno);
                goto err;
        }
        ret = -E_CHMOD;
        if (chmod(name, mode) < 0)
                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 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);
        ret = init_unix_addr(&unix_addr, name);
        if (ret < 0)
                return ret;
        fd = socket(PF_UNIX, SOCK_STREAM, 0);
        if (fd < 0)
                return -ERRNO_TO_PARA_ERROR(errno);
        if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) {
                ret = -ERRNO_TO_PARA_ERROR(errno);
                goto err;
        }
        return fd;
err:
        close(fd);
        return ret;
}

#ifndef HAVE_UCRED
ssize_t send_cred_buffer(int sock, char *buf)
{
        return send_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 \p NULL-terminated buffer and Unix credentials of the current process.
 *
 * \param sock The socket file descriptor.
 * \param buf The buffer to be sent.
 *
 * \return On success, this call returns the number of characters sent.  On
 * error, \p -E_SENDMSG is returned.
 *
 * \sa sendmsg(2), okir's Black Hats Manual.
 */
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 socket file descriptor.
 * \param buf the buffer to store the message.
 * \param size the size of \a buffer.
 *
 * \return negative on errors, the user id on success.
 *
 * \sa recvmsg(2), okir's Black Hats Manual.
 */
int recv_cred_buffer(int fd, char *buf, size_t size)
{
        char control[255];
        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 */