fd_open.c

Go to the documentation of this file.

/*
 *   This program is is free software; you can redistribute it and/or modify
 *   it under the terms of the GNU General Public License as published by
 *   the Free Software Foundation; either version 2 of the License, or (at
 *   your option) any later version.
 *
 *   This program is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *   GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU General Public License
 *   along with this program; if not, write to the Free Software
 *   Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
 */
 
/**
 * $Id: c62780bd899859500431d58b6408ca9e7f39976e $
 * @file lib/bio/fd_open.c
 * @brief BIO abstractions for opening file descriptors
 *
 * @copyright 2024 Network RADIUS SAS (legal@networkradius.com)
 */
 
#include <freeradius-devel/bio/fd_priv.h>
#include <freeradius-devel/util/file.h>
#include <freeradius-devel/util/cap.h>
#include <freeradius-devel/util/rand.h>
 
#include <sys/stat.h>
#include <net/if.h>
#include <fcntl.h>
#include <libgen.h>
#include <netinet/tcp.h>
#include <netinet/in.h>
 
/** Initialize common datagram information
 *
 */
static int fr_bio_fd_common_tcp(int fd, UNUSED fr_socket_t const *sock, fr_bio_fd_config_t const *cfg)
{
        int on = 1;
 
#ifdef SO_KEEPALIVE
        /*
         *      TCP keepalives are always a good idea.  Too many people put firewalls between critical
         *      systems, and then the firewalls drop live TCP streams.
         */
        if (setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on)) < 0) {
                fr_strerror_printf("Failed setting SO_KEEPALIVE: %s", fr_syserror(errno));
                return -1;
        }
#endif
 
#ifdef TCP_NODELAY
        /*
         *      Add some defines for *BSD, and Solaris systems.
         */
#  if !defined(SOL_TCP) && defined(IPPROTO_TCP)
#    define SOL_TCP IPPROTO_TCP
#  endif
 
        /*
         *      Also set TCP_NODELAY, to force the data to be written quickly.
         *
         *      We buffer full packets in memory before we write them, so there's no reason for the kernel to
         *      sit around waiting for more data from us.
         */
        if (!cfg->tcp_delay) {
                if (setsockopt(fd, SOL_TCP, TCP_NODELAY, &on, sizeof(on)) < 0) {
                        fr_strerror_printf("Failed setting TCP_NODELAY: %s", fr_syserror(errno));
                        return -1;
                }
        }
#endif
 
        return 0;
}
 
 
/** Initialize common datagram information
 *
 */
static int fr_bio_fd_common_datagram(int fd, UNUSED fr_socket_t const *sock, fr_bio_fd_config_t const *cfg)
{
        int on = 1;
 
#ifdef SO_TIMESTAMPNS
        /*
         *      Enable receive timestamps, these should reflect
         *      when the packet was received, not when it was read
         *      from the socket.
         */
        if (setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPNS, &on, sizeof(int)) < 0) {
                fr_strerror_printf("Failed setting SO_TIMESTAMPNS: %s", fr_syserror(errno));
                return -1;
        }
 
#elif defined(SO_TIMESTAMP)
        /*
         *      Enable receive timestamps, these should reflect
         *      when the packet was received, not when it was read
         *      from the socket.
         */
        if (setsockopt(fd, SOL_SOCKET, SO_TIMESTAMP, &on, sizeof(int)) < 0) {
                fr_strerror_printf("Failed setting SO_TIMESTAMP: %s", fr_syserror(errno));
                return -1;
        }
#endif
 
 
#ifdef SO_RCVBUF
        if (cfg->recv_buff) {
                int opt = cfg->recv_buff;
 
                /*
                 *      Clamp value to something reasonable.
                 */
                if (opt > (1 << 29)) opt = (1 << 29);
 
                if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &opt, sizeof(opt)) < 0) {
                        fr_strerror_printf("Failed setting SO_RCVBUF: %s", fr_syserror(errno));
                        return -1;
                }
        }
#endif
 
#ifdef SO_SNDBUF
        if (cfg->send_buff) {
                int opt = cfg->send_buff;
 
                /*
                 *      Clamp value to something reasonable.
                 */
                if (opt > (1 << 29)) opt = (1 << 29);
 
                if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &opt, sizeof(opt)) < 0) {
                        fr_strerror_printf("Failed setting SO_SNDBUF: %s", fr_syserror(errno));
                        return -1;
                }
        }
#endif
 
        return 0;
}
 
/** Initialize a UDP socket.
 *
 */
static int fr_bio_fd_common_udp(int fd, fr_socket_t const *sock, fr_bio_fd_config_t const *cfg)
{
#ifdef SO_REUSEPORT
        /*
         *      Servers re-use ports by default.  And clients, too, if they ask nicely.
         */
        if (cfg->reuse_port) {
                int on = 1;
 
                /*
                 *      Set SO_REUSEPORT before bind, so that all sockets can
                 *      listen on the same destination IP address.
                 */
                if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &on, sizeof(on)) < 0) {
                        fr_strerror_printf("Failed setting SO_REUSEPORT: %s", fr_syserror(errno));
                        return -1;
                }
        }
#endif
 
        /*
         *      IPv4 fragments packets.  IPv6 does not.
         *
         *      Many systems set the "don't fragment" bit by default for IPv4.  This setting means that "too
         *      large" packets are silently discarded in the network.
         *
         *      Many local networks support UDP fragmentation, so we just send packets and hope for the best.
         *      In order to support local networks, we disable the "don't fragment" bit.
         *
         *      The wider Internet does not support UDP fragmentation.  This means that fragmented packets are
         *      silently discarded.
         *
         *      @todo - add more code to properly handle EMSGSIZE for PMTUD.  We can then also do
         *      getsockopt(fd,IP_MTU,&integer) to get the current path MTU.  It can only be used after the
         *      socket has been connected.  It should also be called after an EMSGSIZE error is returned.
         *
         *      getsockopt(fd,IP_MTU,&integer) can also be used for unconnected sockets, if we also set
         *      IP_RECVERR.  In which case the errors are put into an error queue.  This is only for Linux.
         */
        if ((sock->af == AF_INET) && cfg->exceed_mtu) {
#if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
                /*
                 *      Disable PMTU discovery.  On Linux, this also makes sure that the "don't
                 *      fragment" flag is zero.
                 */
                {
                        int flag = IP_PMTUDISC_DONT;
 
                        if (setsockopt(fd, IPPROTO_IP, IP_MTU_DISCOVER, &flag, sizeof(flag)) < 0) {
                                fr_strerror_printf("Failed setting IP_MTU_DISCOVER: %s", fr_syserror(errno));
                                return -1;
                        }
                }
#endif
 
#if defined(IP_DONTFRAG)
                /*
                 *      Ensure that the "don't fragment" flag is zero.
                 */
                {
                        int off = 0;
 
                        if (setsockopt(fd, IPPROTO_IP, IP_DONTFRAG, &off, sizeof(off)) < 0) {
                                fr_strerror_printf("Failed setting IP_DONTFRAG: %s", fr_syserror(errno));
                                return -1;
                        }
                }
#endif
        }
 
        return fr_bio_fd_common_datagram(fd, sock, cfg);
}
 
/** Initialize a TCP server socket.
 *
 */
static int fr_bio_fd_server_tcp(int fd, UNUSED fr_socket_t const *sock)
{
        int on = 1;
 
        if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)) < 0) {
                fr_strerror_printf("Failed setting SO_REUSEADDR: %s", fr_syserror(errno));
                return -1;
        }
 
        return 0;
}
 
/** Initialize an IPv4 server socket.
 *
 */
static int fr_bio_fd_server_ipv4(int fd, fr_socket_t const *sock, fr_bio_fd_config_t const *cfg)
{
        /*
         *      And set up any UDP / TCP specific information.
         */
        if (sock->type == SOCK_DGRAM) return fr_bio_fd_common_udp(fd, sock, cfg);
 
        return fr_bio_fd_server_tcp(fd, sock);
}
 
/** Initialize an IPv6 server socket.
 *
 */
static int fr_bio_fd_server_ipv6(int fd, fr_socket_t const *sock, fr_bio_fd_config_t const *cfg)
{
#ifdef IPV6_V6ONLY
        /*
         *      Don't allow v4 packets on v6 connections.
         */
        if (IN6_IS_ADDR_UNSPECIFIED(UNCONST(struct in6_addr *, &sock->inet.src_ipaddr.addr.v6))) {
                int on = 1;
 
                if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, (char *)&on, sizeof(on)) < 0) {
                        fr_strerror_printf("Failed setting IPV6_ONLY: %s", fr_syserror(errno));
                        return -1;
                }
        }
#endif /* IPV6_V6ONLY */
 
        /*
         *      And set up any UDP / TCP specific information.
         */
        if (sock->type == SOCK_DGRAM) return fr_bio_fd_common_udp(fd, sock, cfg);
 
        return fr_bio_fd_server_tcp(fd, sock);
}
 
/** Verify or clean up a pre-existing domain socket.
 *
 */
static int fr_bio_fd_socket_unix_verify(int dirfd, char const *filename, fr_bio_fd_config_t const *cfg)
{
        int fd;
        struct stat buf;
 
        /*
         *      See if the socket exits.  If there's an error opening it, that's an issue.
         *
         *      If it doesn't exist, that's fine.
         */
        if (fstatat(dirfd, filename, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
                if (errno != ENOENT) {
                        fr_strerror_printf("Failed opening domain socket %s: %s", cfg->path, fr_syserror(errno));
                        return -1;
                }
 
                return 0;
        }
 
        /*
         *      If it exists, it must be a socket.
         */
        if (!S_ISSOCK(buf.st_mode)) {
                fr_strerror_printf("Failed open domain socket %s: it is not a socket", filename);
                return -1;
        }
 
        /*
         *      Refuse to open sockets not owned by us.  This prevents configurations from stomping on each
         *      other.
         */
        if (buf.st_uid != cfg->uid) {
                fr_strerror_printf("Failed opening domain socket %s: incorrect UID", cfg->path);
                return -1;
        }
 
        /*
         *      The file exists,and someone is listening.  We can't claim it for ourselves.
         *
         *      Note that this function calls connect(), but connect() always returns immediately for domain
         *      sockets.
         *
         *      @todo - redo that function here, with separate checks for permission errors vs anything else.
         */
        fd = fr_socket_client_unix(cfg->path, false);
        if (fd >= 0) {
                close(fd);
                fr_strerror_printf("Failed creating domain socket %s: It is currently active", cfg->path);
                return -1;
        }
 
        /*
         *      It exists, but no one is listening.  Delete it so that we can re-bind to it.
         */
        if (unlinkat(dirfd, filename, 0) < 0) {
                fr_strerror_printf("Failed removing pre-existing domain socket %s: %s",
                                   cfg->path, fr_syserror(errno));
                return -1;
        }
 
        return 0;
}
 
/*
 *      We normally can't call fchmod() or fchown() on sockets, as they don't really exist in the file system.
 *      Instead, we enforce those permissions on the parent directory of the socket.
 */
static int fr_bio_fd_socket_unix_mkdir(int *dirfd, char const **filename, fr_bio_fd_config_t const *cfg)
{
        mode_t perm;
        int parent_fd, fd;
        char const *path = cfg->path;
        char *p, *dir = NULL;
        char *slashes[2];
 
        perm = S_IREAD | S_IWRITE | S_IEXEC;
        perm |= S_IRGRP | S_IWGRP | S_IXGRP;
 
        /*
         *      The parent directory exists.  Ensure that it has the correct ownership and permissions.
         *
         *      If the parent directory exists, then it enforces access, and we can create the domain socket
         *      within it.
         */
        if (fr_dirfd(dirfd, filename, path) == 0) {
                struct stat buf;
 
                if (fstat(*dirfd, &buf) < 0) {
                        fr_strerror_printf("Failed reading parent directory for file %s: %s", path, fr_syserror(errno));
                fail:
                        talloc_free(dir);
                        close(*dirfd);
                        return -1;
                }
 
                if (buf.st_uid != cfg->uid) {
                        fr_strerror_printf("Failed reading parent directory for file %s: Incorrect UID", path);
                        goto fail;
                }
 
                if (buf.st_gid != cfg->gid) {
                        fr_strerror_printf("Failed reading parent directory for file %s: Incorrect GID", path);
                        goto fail;
                }
 
                /*
                 *      We don't have the correct permissions on the directory, so we fix them.
                 *
                 *      @todo - allow for "other" to read/write if we do authentication on the socket?
                 */
                if (fchmod(*dirfd, perm) < 0) {
                        fr_strerror_printf("Failed setting parent directory permissions for file %s: %s", path, fr_syserror(errno));
                        goto fail;
                }
 
                return 0;
        }
 
        dir = talloc_strdup(NULL, path);
        if (!dir) goto fail;
 
        /*
         *      Find the last two directory separators.
         */
        slashes[0] = slashes[1] = NULL;
        for (p = dir; *p != '\0'; p++) {
                if (*p == '/') {
                        slashes[0] = slashes[1];
                        slashes[1] = p;
                }
        }
 
        /*
         *      There's only one / in the path, we can't do anything.
         *
         *      Opening 'foo/bar.sock' might be useful, but isn't normally a good idea.
         */
        if (!slashes[0]) {
                fr_strerror_printf("Failed parsing filename %s: it is not absolute", path);
                goto fail;
        }
 
        /*
         *      Ensure that the grandparent directory exists.
         *
         *      /var/run/radiusd/foo.sock
         *
         *      slashes[0] points to the slash after 'run'.
         *
         *      slashes[1] points to the slash after 'radiusd', which doesn't exist.
         */
        *slashes[0] = '\0';
 
        /*
         *      If the grandparent doesn't exist, then we don't create it.
         *
         *      These checks minimize the possibility that a misconfiguration by user "radiusd" can cause a
         *      suid-root binary top create a directory in the wrong place.  These checks are only necessary
         *      if the unix domain socket is opened as root.
         */
        parent_fd = open(dir, O_DIRECTORY | O_NOFOLLOW);
        if (parent_fd < 0) {
                fr_strerror_printf("Failed opening directory %s: %s", dir, fr_syserror(errno));
                goto fail;
        }
 
        /*
         *      Create the parent directory.
         */
        *slashes[0] = '/';
        *slashes[1] = '\0';
        if (mkdirat(parent_fd, dir, 0700) < 0) {
                fr_strerror_printf("Failed creating directory %s: %s", dir, fr_syserror(errno));
        close_parent:
                close(parent_fd);
                goto fail;
        }
 
        fd = openat(parent_fd, dir, O_DIRECTORY);
        if (fd < 0) {
                fr_strerror_printf("Failed opening directory %s: %s", dir, fr_syserror(errno));
                goto close_parent;
        }
 
        if (fchmod(fd, perm) < 0) {
                fr_strerror_printf("Failed changing permission for directory %s: %s", dir, fr_syserror(errno));
        close_fd:
                close(fd);
                goto close_parent;
        }
 
        /*
         *      This is a NOOP if we're chowning a file owned by ourselves to our own UID / GID.
         *
         *      Otherwise if we're running as root, it will set ownership to the correct user.
         */
        if (fchown(fd, cfg->uid, cfg->gid) < 0) {
                fr_strerror_printf("Failed changing ownershipt for directory %s: %s", dir, fr_syserror(errno));
                goto close_fd;
        }
 
        talloc_free(dir);
        close(fd);
        close(parent_fd);
 
        return 0;
}
 
static int fr_bio_fd_unix_shutdown(fr_bio_t *bio)
{
        fr_bio_fd_t *my = talloc_get_type_abort(bio, fr_bio_fd_t);
 
        /*
         *      The bio must be open in order to shut it down.
         *
         *      Unix domain sockets are deleted when the bio is closed.
         *
         *      Unix domain sockets are never in the "connecting" state, because connect() always returns
         *      immediately.
         */
        fr_assert(my->info.state == FR_BIO_FD_STATE_OPEN);
 
        /*
         *      Run the user shutdown before we run ours.
         */
        if (my->user_shutdown) {
                int rcode;
 
                rcode = my->user_shutdown(bio);
                if (rcode < 0) return rcode;
        }
 
        if (unlink(my->info.socket.unix.path) < 0) return fr_bio_error(GENERIC);
 
        return 0;
}
 
/** Bind to a Unix domain socket.
 *
 *  @todo - this function only does a tiny bit of what fr_server_domain_socket_peercred() and
 *  fr_server_domain_socket_perm() do.  Those functions do a lot more sanity checks.
 *
 *  The main question is whether or not those checks are useful.  In many cases, fchmod() and fchown() are not
 *  possible on Unix sockets, so we shouldn't bother doing them,
 *
 *  Note that the listeners generally call these functions with wrappers of fr_suid_up() and fr_suid_down().
 *  So these functions are running as "root", and will create files owned as "root".
 */
static int fr_bio_fd_socket_bind_unix(fr_bio_fd_t *my, fr_bio_fd_config_t const *cfg)
{
        int dirfd, rcode;
        char const *filename, *p;
        socklen_t sunlen;
        struct sockaddr_un sun;
 
        if (!cfg->path) {
                fr_strerror_const("Failed to specify path");
                return -1;
        }
 
        /*
         *      The UID and GID should be taken automatically from the "user" and "group" settings in
         *      mainconfig.  There is no reason to set them to anything else.
         */
        if (cfg->uid == (uid_t) -1) {
                fr_strerror_printf("Failed opening domain socket %s: no UID specified", cfg->path);
                return -1;
        }
 
        if (cfg->gid == (gid_t) -1) {
                fr_strerror_printf("Failed opening domain socket %s: no GID specified", cfg->path);
                return -1;
        }
 
        /*
         *      Opening 'foo.sock' is OK.
         */
        p = strrchr(cfg->path, '/');
        if (!p) {
                dirfd = AT_FDCWD;
                filename = cfg->path;
 
        } else if (p == cfg->path) {
                /*
                 *      Opening '/foo.sock' is dumb.
                 */
                fr_strerror_printf("Failed opening domain socket %s: cannot exist at file system root", p);
                return -1;
 
        } else if (fr_bio_fd_socket_unix_mkdir(&dirfd, &filename, cfg) < 0) {
                return -1;
        }
 
        /*
         *      Verify and/or clean up the domain socket.
         */
        if (fr_bio_fd_socket_unix_verify(dirfd, filename, cfg) < 0) {
        fail:
                if (dirfd != AT_FDCWD) close(dirfd);
                return -1;
        }
 
#ifdef HAVE_BINDAT
        /*
         *      The best function to use here is bindat(), but only quite recent versions of FreeBSD actually
         *      have it, and it's definitely not POSIX.
         *
         *      If we use bindat(), we pass a relative pathname.
         */
        if (fr_filename_to_sockaddr(&sun, &sunlen, filename) < 0) goto fail;
 
        rcode = bindat(dirfd, my->info.socket.fd, (struct sockaddr *) &sun, sunlen);
#else
        /*
         *      For bind(), we pass the full path.
         */
        if (fr_filename_to_sockaddr(&sun, &sunlen, cfg->path) < 0) goto fail;
 
        rcode = bind(my->info.socket.fd, (struct sockaddr *) &sun, sunlen);
#endif
        if (rcode < 0) {
                /*
                 *      @todo - if EADDRINUSE, then the socket exists.  Try connect(), and if that fails,
                 *      delete the socket and try again.  This may be simpler than the checks above.
                 */
                fr_strerror_printf("Failed binding to domain socket %s: %s", cfg->path, fr_syserror(errno));
                goto fail;
        }
 
#ifdef __linux__
        /*
         *      Linux supports chown && chmod for sockets.
         */
        if (fchmod(my->info.socket.fd, S_IREAD | S_IWRITE | S_IEXEC | S_IRGRP | S_IWGRP | S_IXGRP) < 0) {
                fr_strerror_printf("Failed changing permission for domain socket %s: %s", cfg->path, fr_syserror(errno));
                goto fail;
        }
 
        /*
         *      This is a NOOP if we're chowning a file owned by ourselves to our own UID / GID.
         *
         *      Otherwise if we're running as root, it will set ownership to the correct user.
         */
        if (fchown(my->info.socket.fd, cfg->uid, cfg->gid) < 0) {
                fr_strerror_printf("Failed changing ownershipt for domain directory %s: %s", cfg->path, fr_syserror(errno));
                goto fail;
        }
 
#endif
 
        /*
         *      Socket is open.  We need to clean it up on shutdown.
         */
        if (my->cb.shutdown) my->user_shutdown = my->cb.shutdown;
        my->cb.shutdown = fr_bio_fd_unix_shutdown;
 
        return 0;
}
 
/*
 *      Use the OSX native versions on OSX.
 */
#ifdef __APPLE__
#undef SO_BINDTODEVICE
#endif
 
#ifdef SO_BINDTODEVICE
/** Linux bind to device by name.
 *
 */
static int fr_bio_fd_socket_bind_to_device(fr_bio_fd_t *my, fr_bio_fd_config_t const *cfg)
{
        /*
         *      ifindex isn't set, do nothing.
         */
        if (!my->info.socket.inet.ifindex) return 0;
 
        if (!cfg->interface) return 0;
 
        /*
         *      The internet hints that CAP_NET_RAW is required to use SO_BINDTODEVICE.
         *
         *      This function also sets fr_strerror() on failure, which will be seen if the bind fails.  If
         *      the bind succeeds, then we don't really care that the capability change has failed.  We must
         *      already have that capability.
         */
#ifdef HAVE_CAPABILITY_H
        (void)fr_cap_enable(CAP_NET_RAW, CAP_EFFECTIVE);
#endif
 
        if (setsockopt(my->info.socket.fd, SOL_SOCKET, SO_BINDTODEVICE, cfg->interface, strlen(cfg->interface)) < 0) {
                fr_strerror_printf("Failed setting SO_BINDTODEVICE for %s: %s", cfg->interface, fr_syserror(errno));
                return -1;
        }
 
        return 0;
}
 
#elif defined(IP_BOUND_IF) || defined(IPV6_BOUND_IF)
/** OSX bind to interface by index.
 *
 */
static int fr_bio_fd_socket_bind_to_device(fr_bio_fd_t *my, UNUSED fr_bio_fd_config_t const *cfg)
{
        int opt, rcode;
 
        if (!my->info.socket.inet.ifindex) return 0;
 
        opt = my->info.socket.inet.ifindex;
 
        switch (my->info.socket.af) {
        case AF_LOCAL:
                rcode = setsockopt(my->info.socket.fd, IPPROTO_IP, IP_BOUND_IF, &opt, sizeof(opt));
                break;
 
        case AF_INET6:
                rcode = setsockopt(my->info.socket.fd, IPPROTO_IPV6, IPV6_BOUND_IF, &opt, sizeof(opt));
                break;
 
        default:
                rcode = -1;
                errno = EAFNOSUPPORT;
                break;
        }
 
        if (rcode < 0) fr_strerror_printf("Failed setting IP_BOUND_IF: %s", fr_syserror(errno));
        return rcode;
}
#else
 
/** This system is missing SO_BINDTODEVICE, IP_BOUND_IF, IPV6_BOUND_IF
 *
 *      @todo - FreeBSD IP_RECVIF and IP_SENDIF
 *
 *      Except that has to be done in recvmsg() and sendmsg().  And it only works on datagram sockets.
 *
 *      cmsg_len = sizeof(struct sockaddr_dl)
 *      cmsg_level = IPPROTO_IP
 *      cmsg_type = IP_RECVIF
 */
static int fr_bio_fd_socket_bind_to_device(fr_bio_fd_t *my, UNUSED fr_bio_fd_config_t const *cfg)
{
        if (!my->info.socket.inet.ifindex) return 0;
 
        fr_strerror_const("Bind to interface is not supported on this platform");
        return -1;
}
 
/* bind to device */
#endif
 
static int fr_bio_fd_socket_bind(fr_bio_fd_t *my, fr_bio_fd_config_t const *cfg)
{
        socklen_t salen;
        struct sockaddr_storage salocal;
 
        fr_assert((my->info.socket.af == AF_INET) || (my->info.socket.af == AF_INET6));
 
#ifdef HAVE_CAPABILITY_H
        /*
         *      If we're binding to a special port as non-root, then
         *      check capabilities.  If we're root, we already have
         *      equivalent capabilities so we don't need to check.
         */
        if ((my->info.socket.inet.src_port < 1024) && (geteuid() != 0)) {
                (void)fr_cap_enable(CAP_NET_BIND_SERVICE, CAP_EFFECTIVE);
        }
#endif
 
        if (fr_bio_fd_socket_bind_to_device(my, cfg) < 0) return -1;
 
        /*
         *      Bind to the IP + interface.
         */
        if (fr_ipaddr_to_sockaddr(&salocal, &salen, &my->info.socket.inet.src_ipaddr, my->info.socket.inet.src_port) < 0) return -1;
 
        /*
         *      If we have a fixed source port, just use that.
         */
        if (my->info.cfg->src_port || !my->info.cfg->src_port_start) {
                if (bind(my->info.socket.fd, (struct sockaddr *) &salocal, salen) < 0) {
                        fr_strerror_printf("Failed binding to socket: %s", fr_syserror(errno));
                        return -1;
                }
        } else {
                uint16_t src_port, current, range;
                struct sockaddr_in *sin = (struct sockaddr_in *) &salocal;
 
                fr_assert(my->info.cfg->src_port_start);
                fr_assert(my->info.cfg->src_port_end);
                fr_assert(my->info.cfg->src_port_end >= my->info.cfg->src_port_start);
 
                range = my->info.cfg->src_port_end - my->info.cfg->src_port_start;
                src_port = fr_rand() % range;
 
                /*
                 *      We only care about sin_port, which is in the same location for sockaddr_in and sockaddr_in6.
                 */
                fr_assert(sin->sin_port == 0);
 
                sin->sin_port = htons(my->info.cfg->src_port_start + src_port);
 
                /*
                 *      We've picked a random port in what is hopefully a large range.  If that works, we're
                 *      done.
                 */
                if (bind(my->info.socket.fd, (struct sockaddr *) &salocal, salen) == 0) goto done;
 
                /*
                 *      Hunt & peck.  Which is horrible.
                 */
                current = src_port;
                while (true) {
                        if (current == range) {
                                current = 0;
                        } else {
                                current++;
                        }
 
                        /*
                         *      We've already checked this, so stop.
                         */
                        if (current == src_port) break;
 
                        sin->sin_port = htons(my->info.cfg->src_port_start + current);
 
                        if (bind(my->info.socket.fd, (struct sockaddr *) &salocal, salen) == 0) goto done;
                }
 
                /*
                 *      The error is a good hint at _why_ we failed to bind.
                 *      We expect errno to be EADDRINUSE, anything else is a surprise.
                 */
                fr_strerror_printf("Failed binding port between 'src_port_start' and 'src_port_end': %s", fr_syserror(errno));
                return -1;
        }
 
        /*
         *      The source IP may have changed, so get the new one.
         */
done:
        return fr_bio_fd_socket_name(my);
}
 
/**  Set the name of an FD BIO
 *
 */
void fr_bio_fd_name(fr_bio_fd_t *my)
{
        fr_bio_fd_config_t const *cfg = my->info.cfg;
 
        switch (my->info.type) {
        case FR_BIO_FD_INVALID:
                return;
 
        case FR_BIO_FD_UNCONNECTED:
                fr_assert(cfg->socket_type == SOCK_DGRAM);
 
                switch (my->info.socket.af) {
                case AF_INET:
                case AF_INET6:
                        my->info.name = fr_asprintf(my, "proto udp local %pV port %u",
                                                    fr_box_ipaddr(my->info.socket.inet.src_ipaddr),
                                                    my->info.socket.inet.src_port);
                        break;
 
                case AF_LOCAL:
                        my->info.name = fr_asprintf(my, "proto unix (datagram) filename %s",
                                                    cfg->path);
                        break;
 
                default:
                        fr_assert(0);
                        my->info.name = "??? invalid BIO ???";
                        break;
                }
                break;
 
        case FR_BIO_FD_CONNECTED:
                switch (my->info.socket.af) {
                case AF_INET:
                case AF_INET6:
                        my->info.name = fr_asprintf(my, "proto %s local %pV port %u remote %pV port %u",
                                                    (cfg->socket_type == SOCK_DGRAM) ? "udp" : "tcp",
                                                    fr_box_ipaddr(my->info.socket.inet.src_ipaddr),
                                                    my->info.socket.inet.src_port,
                                                    fr_box_ipaddr(my->info.socket.inet.dst_ipaddr),
                                                    my->info.socket.inet.dst_port);
                        break;
 
                case AF_LOCAL:
                        my->info.name = fr_asprintf(my, "proto unix %sfilename %s",
                                                    (cfg->socket_type == SOCK_DGRAM) ? "(datagram) " : "",
                                                    cfg->path);
                        break;
 
                case AF_FILE_BIO:
                        fr_assert(cfg->socket_type == SOCK_STREAM);
 
                        if (cfg->flags == O_RDONLY) {
                                my->info.name = fr_asprintf(my, "proto file (read-only) filename %s ",
                                                            cfg->filename);
 
                        } else if (cfg->flags == O_WRONLY) {
                                my->info.name = fr_asprintf(my, "proto file (write-only) filename %s",
                                                            cfg->filename);
                        } else {
                                my->info.name = fr_asprintf(my, "proto file (read-write) filename %s",
                                                            cfg->filename);
                        }
                        break;
 
                default:
                        fr_assert(0);
                        my->info.name = "??? invalid BIO ???";
                        break;
                }
                break;
 
        case FR_BIO_FD_LISTEN:
                switch (my->info.socket.af) {
                case AF_INET:
                case AF_INET6:
                        my->info.name = fr_asprintf(my, "proto %s local %pV port %u",
                                                    (cfg->socket_type == SOCK_DGRAM) ? "udp" : "tcp",
                                                    fr_box_ipaddr(my->info.socket.inet.src_ipaddr),
                                                    my->info.socket.inet.src_port);
                        break;
 
                case AF_LOCAL:
                        my->info.name = fr_asprintf(my, "proto unix filename %s",
                                                    cfg->path);
                        break;
 
                default:
                        fr_assert(0);
                        my->info.name = "??? invalid BIO ???";
                        break;
                }
                break;
        }
}
 
/** Checks the configuration without modifying anything.
 *
 */
int fr_bio_fd_check_config(fr_bio_fd_config_t const *cfg)
{
        /*
         *      Unix sockets and files are OK.
         */
        if (cfg->path || cfg->filename) return 0;
 
        /*
         *      Sanitize the IP addresses.
         *
         */
        switch (cfg->type) {
        case FR_BIO_FD_INVALID:
                fr_strerror_const("No connection type was specified");
                return -1;
 
        case FR_BIO_FD_CONNECTED:
                /*
                 *      Ensure that we have a destination address.
                 */
                if (cfg->dst_ipaddr.af == AF_UNSPEC) {
                        fr_strerror_const("No destination IP address was specified");
                        return -1;
                }
 
                if (!cfg->dst_port) {
                        fr_strerror_const("No destination port was specified");
                        return -1;
                }
 
                /*
                 *      The source IP has to be the same address family as the destination IP.
                 */
                if ((cfg->src_ipaddr.af != AF_UNSPEC) && (cfg->src_ipaddr.af != cfg->dst_ipaddr.af)) {
                        fr_strerror_printf("Source and destination IP addresses are not from the same IP address family");
                        return -1;
                }
                break;
 
        case FR_BIO_FD_LISTEN:
                if (!cfg->src_port) {
                        fr_strerror_const("No source port was specified");
                        return -1;
                }
                FALL_THROUGH;
 
                /*
                 *      Unconnected sockets can use a source port, but don't need one.
                 */
        case FR_BIO_FD_UNCONNECTED:
                if (cfg->path && cfg->filename) {
                        fr_strerror_const("Unconnected sockets cannot be used with Unix sockets or files");
                        return -1;
                }
 
                if (cfg->src_ipaddr.af == AF_UNSPEC) {
                        fr_strerror_const("No source IP address was specified");
                        return -1;
                }
                break;
        }
 
        return 0;
}
 
/** Opens a socket and updates sock->fd
 *
 *  If the socket is asynchronous, it also calls connect()
 */
int fr_bio_fd_open(fr_bio_t *bio, fr_bio_fd_config_t const *cfg)
{
        int fd;
        int rcode;
        fr_bio_fd_t *my = talloc_get_type_abort(bio, fr_bio_fd_t);
 
        fr_strerror_clear();
 
        my->info = (fr_bio_fd_info_t) {
                .socket = {
                        .type = cfg->socket_type,
                },
                .cfg = cfg,
        };
 
        if (!cfg->path && !cfg->filename) {
                int protocol;
 
                my->info.socket.af = cfg->src_ipaddr.af;
                my->info.socket.inet.src_ipaddr = cfg->src_ipaddr;
                my->info.socket.inet.dst_ipaddr = cfg->dst_ipaddr;
                my->info.socket.inet.src_port = cfg->src_port;
                my->info.socket.inet.dst_port = cfg->dst_port;
 
                if (fr_bio_fd_check_config(cfg) < 0) return -1;
 
                /*
                 *      Sanitize the IP addresses.
                 *
                 */
                switch (cfg->type) {
                case FR_BIO_FD_INVALID:
                        return -1;
 
                case FR_BIO_FD_CONNECTED:
                        /*
                         *      No source specified, just bootstrap it from the destination.
                         */
                        if (my->info.socket.inet.src_ipaddr.af == AF_UNSPEC) {
                                my->info.socket.inet.src_ipaddr = (fr_ipaddr_t) {
                                        .af = my->info.socket.inet.dst_ipaddr.af,
                                        .prefix = (my->info.socket.inet.dst_ipaddr.af == AF_INET) ? 32 : 128,
                                };
 
                                /*
                                 *      Set the main socket AF too.
                                 */
                                my->info.socket.af = my->info.socket.inet.dst_ipaddr.af;
                        }
 
                        /*
                         *      The source IP has to be the same address family as the destination IP.
                         */
                        if (my->info.socket.inet.src_ipaddr.af != my->info.socket.inet.dst_ipaddr.af) {
                                fr_strerror_const("Source and destination IP addresses are not from the same IP address family");
                                return -1;
                        }
 
#ifdef SO_REUSEPORT
                        /*
                         *      Connected UDP sockets really only matter when writing packets.  They allow the
                         *      system to use write() instead of sendto().
                         *
                         *      However for reading packets, the kernel does NOT check the source IP.
                         *      Instead, it just delivers any packet with the correct dst IP/port.  Even
                         *      worse, if multiple sockets re-use the same dst IP/port, packets are delivered
                         *      to a _random_ socket.
                         *
                         *      As a result, we cannot use wildcard sockets with SO_REUSEPORT, and UDP.
                         */
                        if (cfg->reuse_port && (cfg->socket_type == SOCK_DGRAM) &&
                            fr_ipaddr_is_inaddr_any(&my->info.socket.inet.dst_ipaddr)) { /* checks AF, so we're OK */
                                fr_strerror_const("Cannot set 'reuse_port' for connected UDP sockets with wildcard IP");
                                return -1;
                        }
#endif
 
                        break;
 
                case FR_BIO_FD_UNCONNECTED:
                case FR_BIO_FD_LISTEN:
                        fr_assert(my->info.socket.inet.src_ipaddr.af != AF_UNSPEC);
                        break;
                }
 
                if (cfg->socket_type == SOCK_STREAM) {
                        protocol = IPPROTO_TCP;
                } else {
                        protocol = IPPROTO_UDP;
                }
 
                if (cfg->interface) {
                        my->info.socket.inet.ifindex = if_nametoindex(cfg->interface);
 
                        if (!my->info.socket.inet.ifindex) {
                                fr_strerror_printf_push("Failed finding interface %s: %s", cfg->interface, fr_syserror(errno));
                                return -1;
                        }
                }
 
                fd = socket(my->info.socket.af, my->info.socket.type, protocol);
                if (fd < 0) {
                        fr_strerror_printf("Failed opening socket: %s", fr_syserror(errno));
                        return fr_bio_error(GENERIC);
                }
 
        } else if (cfg->path) {
                my->info.socket.af = AF_LOCAL;
                my->info.socket.type = SOCK_STREAM;
                my->info.socket.unix.path = cfg->path;
 
                fd = socket(my->info.socket.af, my->info.socket.type, 0);
                if (fd < 0) {
                        fr_strerror_printf("Failed opening domain socket %s: %s", cfg->path, fr_syserror(errno));
                        return fr_bio_error(GENERIC);
                }
 
        } else {
                if (cfg->type != FR_BIO_FD_CONNECTED) {
                        fr_strerror_printf("Can only use connected sockets for file IO");
                        return -1;
                }
 
                /*
                 *      Filenames overload the #fr_socket_t for now.
                 */
                my->info.socket.af = AF_FILE_BIO;
                my->info.socket.type = SOCK_STREAM;
                my->info.socket.unix.path = cfg->filename;
 
                /*
                 *      Allow hacks for stdout and stderr
                 */
                if (strcmp(cfg->filename, "/dev/stdout") == 0) {
                        if (cfg->flags != O_WRONLY) {
                        fail_dev:
                                fr_strerror_printf("Cannot read from %s", cfg->filename);
                                return -1;
                        }
 
                        fd = dup(STDOUT_FILENO);
 
                } else if (strcmp(cfg->filename, "/dev/stderr") == 0) {
                        if (cfg->flags != O_WRONLY) goto fail_dev;
 
                        fd = dup(STDERR_FILENO);
 
                } else if (strcmp(cfg->filename, "/dev/stdin") == 0) {
                        if (cfg->flags != O_RDONLY) {
                                fr_strerror_printf("Cannot write to %s", cfg->filename);
                                return -1;
                        }
 
                        fd = dup(STDIN_FILENO);
 
                } else {
                        /*
                         *      Minor hacks so that we have only _one_ source of open / mkdir
                         */
                        my->info.socket.fd = -1;
 
                        fd = fr_bio_fd_reopen(bio);
                }
                if (fd < 0) {
                        fr_strerror_printf("Failed opening file %s: %s", cfg->filename, fr_syserror(errno));
                        return fr_bio_error(GENERIC);
                }
        }
 
        /*
         *      Set it to be non-blocking if required.
         */
        if (cfg->async && (fr_nonblock(fd) < 0)) {
                fr_strerror_printf("Failed setting O_NONBLOCK: %s", fr_syserror(errno));
                rcode = fr_bio_error(GENERIC);
 
        fail:
                my->info.socket = (fr_socket_t) {
                        .fd = -1,
                };
                my->info.state = FR_BIO_FD_STATE_CLOSED;
                my->info.cfg = NULL;
                close(fd);
                return rcode;
        }
 
#ifdef FD_CLOEXEC
        /*
         *      We don't want child processes inheriting these file descriptors.
         */
        rcode = fcntl(fd, F_GETFD);
        if (rcode >= 0) {
                if (fcntl(fd, F_SETFD, rcode | FD_CLOEXEC) < 0) {
                        fr_strerror_printf("Failed  setting FD_CLOEXEC: %s", fr_syserror(errno));
                        goto fail;
                }
        }
#endif
 
        /*
         *      Initialize the bio information before calling the various setup functions.
         */
        my->info.state = FR_BIO_FD_STATE_CONNECTING;
 
        /*
         *      Set the FD so that the subsequent calls can use it.
         */
        my->info.socket.fd = fd;
 
        /*
         *      Set the type, too.
         */
        my->info.type = cfg->type;
 
        /*
         *      Do sanity checks, bootstrap common socket options, bind to the socket, and initialize the read
         *      / write functions.
         */
        switch (cfg->type) {
        case FR_BIO_FD_INVALID:
                return -1;
 
                /*
                 *      Unconnected UDP or datagram AF_LOCAL server sockets.
                 */
        case FR_BIO_FD_UNCONNECTED:
                if (my->info.socket.type != SOCK_DGRAM) {
                        fr_strerror_const("Failed configuring socket: unconnected sockets must be UDP");
                        return -1;
                }
 
                switch (my->info.socket.af) {
                case AF_INET:
                case AF_INET6:
                        if ((rcode = fr_bio_fd_common_udp(fd, &my->info.socket, cfg)) < 0) goto fail;
                        break;
 
                case AF_LOCAL:
                        if ((rcode = fr_bio_fd_common_datagram(fd, &my->info.socket, cfg)) < 0) goto fail;
                        break;
 
                case AF_FILE_BIO:
                        fr_strerror_const("Filenames must use the connected API");
                        goto fail;
 
                default:
                        fr_strerror_const("Unsupported address family for unconnected sockets");
                        goto fail;
                }
 
                if ((rcode = fr_bio_fd_socket_bind(my, cfg)) < 0) goto fail;
 
                if ((rcode = fr_bio_fd_init_common(my)) < 0) goto fail;
                break;
 
                /*
                 *      A connected client: UDP, TCP, AF_LOCAL, or AF_FILE_BIO
                 */
        case FR_BIO_FD_CONNECTED:
                if (my->info.socket.type == SOCK_DGRAM) {
                        switch (my->info.socket.af) {
                        case AF_INET:
                        case AF_INET6:
                                if ((rcode = fr_bio_fd_common_udp(fd, &my->info.socket, cfg)) < 0) goto fail;
                                break;
                        default:
                                if ((rcode = fr_bio_fd_common_datagram(fd, &my->info.socket, cfg)) < 0) goto fail;
                                break;
                        }
 
                } else if ((my->info.socket.af == AF_INET) || (my->info.socket.af == AF_INET6)) {
                        rcode = fr_bio_fd_common_tcp(fd, &my->info.socket, cfg);
                        if (rcode < 0) goto fail;
                }
 
                switch (my->info.socket.af) {
                case AF_LOCAL:
                        if ((rcode = fr_bio_fd_socket_bind_unix(my, cfg)) < 0) goto fail;
                        break;
 
                case AF_FILE_BIO:
                        break;
 
                case AF_INET:
                case AF_INET6:
                        if ((rcode = fr_bio_fd_socket_bind(my, cfg)) < 0) goto fail;
                        break;
 
                default:
                        return -1;
                }
 
                if ((rcode = fr_bio_fd_init_connected(my)) < 0) goto fail;
                break;
 
                /*
                 *      Server socket which listens for new stream connections
                 */
        case FR_BIO_FD_LISTEN:
                if ((my->info.socket.type == SOCK_DGRAM) && !cfg->reuse_port) {
                        fr_strerror_const("reuseport must be set for datagram sockets");
                        rcode = -1;
                        goto fail;
                }
 
                switch (my->info.socket.af) {
                case AF_INET:
                        if ((rcode = fr_bio_fd_server_ipv4(fd, &my->info.socket, cfg)) < 0) goto fail;
 
                        if ((rcode = fr_bio_fd_socket_bind(my, cfg)) < 0) goto fail;
                        break;
 
                case AF_INET6:
                        if ((rcode = fr_bio_fd_server_ipv6(fd, &my->info.socket, cfg)) < 0) goto fail;
 
                        if ((rcode = fr_bio_fd_socket_bind(my, cfg)) < 0) goto fail;
                        break;
 
                case AF_LOCAL:
                        if ((rcode = fr_bio_fd_socket_bind_unix(my, cfg)) < 0) goto fail;
                        break;
 
                default:
                        fr_strerror_const("Unsupported address family for accept() socket");
                        rcode = -1;
                        goto fail;
                }
 
                if ((rcode = fr_bio_fd_init_listen(my)) < 0) goto fail;
                break;
        }
 
        /*
         *      Set the name of the BIO.
         */
        fr_bio_fd_name(my);
 
        return 0;
}
 
/** Reopen a file BIO
 *
 *  e.g. for log files.
 */
int fr_bio_fd_reopen(fr_bio_t *bio)
{
        fr_bio_fd_t *my = talloc_get_type_abort(bio, fr_bio_fd_t);
        fr_bio_fd_config_t const *cfg = my->info.cfg;
        int fd, flags;
 
        if (my->info.socket.af != AF_FILE_BIO) {
                fr_strerror_const("Cannot reopen a non-file BIO");
                return -1;
        }
 
        /*
         *      Create it if necessary.
         */
        flags = cfg->flags;
        if (flags != O_RDONLY) flags |= O_CREAT;
 
        if (!cfg->mkdir) {
                /*
                 *      Client BIOs writing to a file, and therefore need to create it.
                 */
        do_open:
                fd = open(cfg->filename, flags, cfg->perm);
                if (fd < 0) {
                failed_open:
                        fr_strerror_printf("Failed opening file %s: %s", cfg->filename, fr_syserror(errno));
                        return -1;
                }
 
        } else {
                /*
                 *      We make the parent directory if told to, AND if there's a '/' in the path.
                 */
                char *p = strrchr(cfg->filename, '/');
                int dir_fd;
 
                if (!p) goto do_open;
 
                if (fr_mkdir(&dir_fd, cfg->filename, (size_t) (p - cfg->filename), cfg->perm, fr_mkdir_chown,
                              &(fr_mkdir_chown_t) {
                                      .uid = cfg->uid,
                                      .gid = cfg->gid,
                              }) < 0) {
                        return -1;
                }
 
                fd = openat(dir_fd, p + 1, flags, cfg->perm);
                if (fd < 0) {
                        close(dir_fd);
                        goto failed_open;
                }
        }
 
        /*
         *      We're boot-strapping, just set the new FD and return.
         */
        if (my->info.socket.fd < 0) {
                return fd;
        }
 
        /*
         *      Replace the FD rather than swapping it out with a new one.  This is potentially more
         *      thread-safe.
         */
        if (dup2(fd, my->info.socket.fd) < 0) {
                close(fd);
                fr_strerror_printf("Failed reopening file - %s", fr_syserror(errno));
                return -1;
        }
 
        close(fd);
        return my->info.socket.fd;
}