[svn] - fix for 100% cpu use

[irc/rqf/shadowircd.git] / libcharybdis / commio.c

/*
 *  ircd-ratbox: A slightly useful ircd.
 *  commio.c: Network/file related functions
 *
 *  Copyright (C) 1990 Jarkko Oikarinen and University of Oulu, Co Center
 *  Copyright (C) 1996-2002 Hybrid Development Team
 *  Copyright (C) 2002-2005 ircd-ratbox development team
 *
 *  This program 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., 59 Temple Place, Suite 330, Boston, MA  02111-1307
 *  USA
 *
 *  $Id: commio.c 3237 2007-03-05 17:35:17Z nenolod $
 */

#include "libcharybdis.h"

#ifndef IN_LOOPBACKNET
#define IN_LOOPBACKNET        0x7f
#endif

#ifndef INADDR_NONE
#define INADDR_NONE ((unsigned int) 0xffffffff)
#endif

const char *const NONB_ERROR_MSG = "set_non_blocking failed for %s:%s";
const char *const SETBUF_ERROR_MSG = "set_sock_buffers failed for server %s:%s";

static const char *comm_err_str[] = { "Comm OK", "Error during bind()",
        "Error during DNS lookup", "connect timeout",
        "Error during connect()",
        "Comm Error"
};

#define FD_HASH_SIZE 128
static dlink_list fd_table[FD_HASH_SIZE];

static void fdlist_update_biggest(int fd, int opening);

/* Highest FD and number of open FDs .. */
int highest_fd = -1;            /* Its -1 because we haven't started yet -- adrian */
int number_fd = 0;

static void comm_connect_callback(int fd, int status);
static PF comm_connect_timeout;
static void comm_connect_dns_callback(void *vptr, struct DNSReply *reply);
static PF comm_connect_tryconnect;

inline fde_t *
comm_locate_fd(int fd)
{
        int bucket = fd % FD_HASH_SIZE;
        dlink_list *list = &fd_table[bucket];
        dlink_node *n;

        DLINK_FOREACH(n, list->head)
        {
                fde_t *F = (fde_t *) n->data;

                if (F->fd == fd)
                        return F;
        }

        return NULL;
}

inline fde_t *
comm_add_fd(int fd)
{
        fde_t *F = comm_locate_fd(fd);
        dlink_list *list;

        if (F != NULL)
                return F;

        F = calloc(sizeof(fde_t), 1);
        F->fd = fd;
        list = &fd_table[fd % FD_HASH_SIZE];
        dlinkAdd(F, &F->node, list);

        return F;
}

inline void
comm_remove_fd(int fd)
{
        int bucket = fd % FD_HASH_SIZE;
        fde_t *F;
        dlink_list *list = &fd_table[bucket];

        F = comm_locate_fd(fd);
        if (F == NULL)
                return;

        dlinkDelete(&F->node, list);
        MyFree(F);
}

/* 32bit solaris is kinda slow and stdio only supports fds < 256
 * so we got to do this crap below.
 * (BTW Fuck you Sun, I hate your guts and I hope you go bankrupt soon)
 * XXX: this is no longer needed in Solaris 10. --nenolod
 */
#if defined (__SVR4) && defined (__sun) 
static void comm_fd_hack(int *fd)
{
        int newfd;
        if(*fd > 256 || *fd < 0)
                return;
        if((newfd = fcntl(*fd, F_DUPFD, 256)) != -1)
        {
                close(*fd);
                *fd = newfd;
        } 
        return;
}
#else
#define comm_fd_hack(fd) 
#endif


/* close_all_connections() can be used *before* the system come up! */

void
comm_close_all(void)
{
        int i;
#ifndef NDEBUG
        int fd;
#endif

        /* XXX someone tell me why we care about 4 fd's ? */
        /* XXX btw, fd 3 is used for profiler ! */

        for (i = 4; i < MAXCONNECTIONS; ++i)
        {
                fde_t *F = comm_locate_fd(i);

                if(F != NULL && F->flags.open)
                        comm_close(i);
                else
                        close(i);
        }

        /* XXX should his hack be done in all cases? */
#ifndef NDEBUG
        /* fugly hack to reserve fd == 2 */
        (void) close(2);
        fd = open("stderr.log", O_WRONLY | O_CREAT | O_APPEND, 0644);
        if(fd >= 0)
        {
                dup2(fd, 2);
                close(fd);
        }
#endif
}

/*
 * get_sockerr - get the error value from the socket or the current errno
 *
 * Get the *real* error from the socket (well try to anyway..).
 * This may only work when SO_DEBUG is enabled but its worth the
 * gamble anyway.
 */
int
comm_get_sockerr(int fd)
{
        int errtmp = errno;
#ifdef SO_ERROR
        int err = 0;
        socklen_t len = sizeof(err);

        if(-1 < fd && !getsockopt(fd, SOL_SOCKET, SO_ERROR, (char *) &err, (socklen_t *) & len))
        {
                if(err)
                        errtmp = err;
        }
        errno = errtmp;
#endif
        return errtmp;
}

/*
 * set_sock_buffers - set send and receive buffers for socket
 * 
 * inputs       - fd file descriptor
 *              - size to set
 * output       - returns true (1) if successful, false (0) otherwise
 * side effects -
 */
int
comm_set_buffers(int fd, int size)
{
        if(setsockopt
           (fd, SOL_SOCKET, SO_RCVBUF, (char *) &size, sizeof(size))
           || setsockopt(fd, SOL_SOCKET, SO_SNDBUF, (char *) &size, sizeof(size)))
                return 0;
        return 1;
}

/*
 * set_non_blocking - Set the client connection into non-blocking mode. 
 *
 * inputs       - fd to set into non blocking mode
 * output       - 1 if successful 0 if not
 * side effects - use POSIX compliant non blocking and
 *                be done with it.
 */
int
comm_set_nb(int fd)
{
        int nonb = 0;
        int res;
        fde_t *F = comm_locate_fd(fd);

        nonb |= O_NONBLOCK;
        res = fcntl(fd, F_GETFL, 0);
        if(-1 == res || fcntl(fd, F_SETFL, res | nonb) == -1)
                return 0;

        if (F != NULL)
                F->flags.nonblocking = 1;

        return 1;
}


/*
 * stolen from squid - its a neat (but overused! :) routine which we
 * can use to see whether we can ignore this errno or not. It is
 * generally useful for non-blocking network IO related errnos.
 *     -- adrian
 */
int
ignoreErrno(int ierrno)
{
        switch (ierrno)
        {
        case EINPROGRESS:
        case EWOULDBLOCK:
#if EAGAIN != EWOULDBLOCK
        case EAGAIN:
#endif
        case EALREADY:
        case EINTR:
#ifdef ERESTART
        case ERESTART:
#endif
                return 1;
        default:
                return 0;
        }
}


/*
 * comm_settimeout() - set the socket timeout
 *
 * Set the timeout for the fd
 */
void
comm_settimeout(int fd, time_t timeout, PF * callback, void *cbdata)
{
        fde_t *F;
        s_assert(fd >= 0);
        F = comm_locate_fd(fd);
        s_assert(F->flags.open);

        F->timeout = CurrentTime + (timeout / 1000);
        F->timeout_handler = callback;
        F->timeout_data = cbdata;
}


/*
 * comm_setflush() - set a flush function
 *
 * A flush function is simply a function called if found during
 * comm_timeouts(). Its basically a second timeout, except in this case
 * I'm too lazy to implement multiple timeout functions! :-)
 * its kinda nice to have it seperate, since this is designed for
 * flush functions, and when comm_close() is implemented correctly
 * with close functions, we _actually_ don't call comm_close() here ..
 */
void
comm_setflush(int fd, time_t timeout, PF * callback, void *cbdata)
{
        fde_t *F;
        s_assert(fd >= 0);
        F = comm_locate_fd(fd);
        s_assert(F->flags.open);

        F->flush_timeout = CurrentTime + (timeout / 1000);
        F->flush_handler = callback;
        F->flush_data = cbdata;
}


/*
 * comm_checktimeouts() - check the socket timeouts
 *
 * All this routine does is call the given callback/cbdata, without closing
 * down the file descriptor. When close handlers have been implemented,
 * this will happen.
 */
void
comm_checktimeouts(void *notused)
{
        PF *hdl;
        void *data;
        fde_t *F;
        dlink_list *bucket;
        int i;
        dlink_node *n, *n2;

        for (i = 0; i <= FD_HASH_SIZE; i++)
        {
                bucket = &fd_table[i];

                if (dlink_list_length(bucket) <= 0)
                        continue;

                DLINK_FOREACH_SAFE(n, n2, bucket->head)
                {
                        F = (fde_t *) n->data;

                        if(F == NULL)
                                continue;
                        if(!F->flags.open)
                                continue;
                        if(F->flags.closing)
                                continue;

                        /* check flush functions */
                        if(F->flush_handler &&
                           F->flush_timeout > 0 && F->flush_timeout < CurrentTime)
                        {
                                hdl = F->flush_handler;
                                data = F->flush_data;
                                comm_setflush(F->fd, 0, NULL, NULL);
                                hdl(F->fd, data);
                        }

                        /* check timeouts */
                        if(F->timeout_handler &&
                           F->timeout > 0 && F->timeout < CurrentTime)
                        {
                                /* Call timeout handler */
                                hdl = F->timeout_handler;
                                data = F->timeout_data;
                                comm_settimeout(F->fd, 0, NULL, NULL);
                                hdl(F->fd, data);
                        }
                }
        }
}

/*
 * void comm_connect_tcp(int fd, const char *host, u_short port,
 *                       struct sockaddr *clocal, int socklen,
 *                       CNCB *callback, void *data, int aftype, int timeout)
 * Input: An fd to connect with, a host and port to connect to,
 *        a local sockaddr to connect from + length(or NULL to use the
 *        default), a callback, the data to pass into the callback, the
 *        address family.
 * Output: None.
 * Side-effects: A non-blocking connection to the host is started, and
 *               if necessary, set up for selection. The callback given
 *               may be called now, or it may be called later.
 */
void
comm_connect_tcp(int fd, const char *host, u_short port,
                 struct sockaddr *clocal, int socklen, CNCB * callback,
                 void *data, int aftype, int timeout)
{
        void *ipptr = NULL;
        fde_t *F;
        s_assert(fd >= 0);
        F = comm_locate_fd(fd);
        F->flags.called_connect = 1;
        s_assert(callback);
        F->connect.callback = callback;
        F->connect.data = data;

        memset(&F->connect.hostaddr, 0, sizeof(F->connect.hostaddr));
#ifdef IPV6
        if(aftype == AF_INET6)
        {
                struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&F->connect.hostaddr;
                SET_SS_LEN(F->connect.hostaddr, sizeof(struct sockaddr_in6));
                in6->sin6_port = htons(port);
                in6->sin6_family = AF_INET6;
                ipptr = &in6->sin6_addr;
        } else
#endif
        {
                struct sockaddr_in *in = (struct sockaddr_in *)&F->connect.hostaddr;
                SET_SS_LEN(F->connect.hostaddr, sizeof(struct sockaddr_in));
                in->sin_port = htons(port);
                in->sin_family = AF_INET;
                ipptr = &in->sin_addr;
        }

        /* Note that we're using a passed sockaddr here. This is because
         * generally you'll be bind()ing to a sockaddr grabbed from
         * getsockname(), so this makes things easier.
         * XXX If NULL is passed as local, we should later on bind() to the
         * virtual host IP, for completeness.
         *   -- adrian
         */
        if((clocal != NULL) && (bind(F->fd, clocal, socklen) < 0))
        {
                /* Failure, call the callback with COMM_ERR_BIND */
                comm_connect_callback(F->fd, COMM_ERR_BIND);
                /* ... and quit */
                return;
        }

        /* Next, if we have been given an IP, get the addr and skip the
         * DNS check (and head direct to comm_connect_tryconnect().
         */
        if(inetpton(aftype, host, ipptr) <= 0)
        {
                /* Send the DNS request, for the next level */
                F->dns_query = MyMalloc(sizeof(struct DNSQuery));
                F->dns_query->ptr = F;
                F->dns_query->callback = comm_connect_dns_callback;
#ifdef IPV6
                if (aftype == AF_INET6)
                        gethost_byname_type(host, F->dns_query, T_AAAA);
                else
#endif
                        gethost_byname_type(host, F->dns_query, T_A);
        }
        else
        {
                /* We have a valid IP, so we just call tryconnect */
                /* Make sure we actually set the timeout here .. */
                comm_settimeout(F->fd, timeout * 1000, comm_connect_timeout, NULL);
                comm_connect_tryconnect(F->fd, NULL);
        }
}

/*
 * comm_connect_callback() - call the callback, and continue with life
 */
static void
comm_connect_callback(int fd, int status)
{
        CNCB *hdl;
        fde_t *F = comm_locate_fd(fd);

        /* This check is gross..but probably necessary */
        if(F == NULL || F->connect.callback == NULL)
                return;

        /* Clear the connect flag + handler */
        hdl = F->connect.callback;
        F->connect.callback = NULL;
        F->flags.called_connect = 0;

        /* Clear the timeout handler */
        comm_settimeout(F->fd, 0, NULL, NULL);

        /* Call the handler */
        hdl(F->fd, status, F->connect.data);
}


/*
 * comm_connect_timeout() - this gets called when the socket connection
 * times out. This *only* can be called once connect() is initially
 * called ..
 */
static void
comm_connect_timeout(int fd, void *notused)
{
        /* error! */
        comm_connect_callback(fd, COMM_ERR_TIMEOUT);
}


/*
 * comm_connect_dns_callback() - called at the completion of the DNS request
 *
 * The DNS request has completed, so if we've got an error, return it,
 * otherwise we initiate the connect()
 */
static void
comm_connect_dns_callback(void *vptr, struct DNSReply *reply)
{
        fde_t *F = vptr;

        /* Free dns_query now to avoid double reslist free -- jilles */
        MyFree(F->dns_query);
        F->dns_query = NULL;

        if(!reply)
        {
                comm_connect_callback(F->fd, COMM_ERR_DNS);
                return;
        }

        /* No error, set a 10 second timeout */
        comm_settimeout(F->fd, 30 * 1000, comm_connect_timeout, NULL);

        /* Copy over the DNS reply info so we can use it in the connect() */
#ifdef IPV6
        if(reply->addr.ss_family == AF_INET6)
        {
                struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&F->connect.hostaddr;
                memcpy(&in6->sin6_addr, &((struct sockaddr_in6 *)&reply->addr)->sin6_addr, sizeof(struct in6_addr));
        }
        else
#endif
        {
                struct sockaddr_in *in = (struct sockaddr_in *)&F->connect.hostaddr;
                in->sin_addr.s_addr = ((struct sockaddr_in *)&reply->addr)->sin_addr.s_addr;
        }

        /* Now, call the tryconnect() routine to try a connect() */
        comm_connect_tryconnect(F->fd, NULL);
}


/* static void comm_connect_tryconnect(int fd, void *notused)
 * Input: The fd, the handler data(unused).
 * Output: None.
 * Side-effects: Try and connect with pending connect data for the FD. If
 *               we succeed or get a fatal error, call the callback.
 *               Otherwise, it is still blocking or something, so register
 *               to select for a write event on this FD.
 */
static void
comm_connect_tryconnect(int fd, void *notused)
{
        int retval;
        fde_t *F = comm_locate_fd(fd);

        if(F->connect.callback == NULL)
                return;
        /* Try the connect() */
        retval = connect(fd, (struct sockaddr *) &F->connect.hostaddr, 
                               GET_SS_LEN(F->connect.hostaddr));
        /* Error? */
        if(retval < 0)
        {
                /*
                 * If we get EISCONN, then we've already connect()ed the socket,
                 * which is a good thing.
                 *   -- adrian
                 */
                if(errno == EISCONN)
                        comm_connect_callback(F->fd, COMM_OK);
                else if(ignoreErrno(errno))
                        /* Ignore error? Reschedule */
                        comm_setselect(F->fd, FDLIST_SERVER, COMM_SELECT_WRITE|COMM_SELECT_RETRY,
                                       comm_connect_tryconnect, NULL, 0);
                else
                        /* Error? Fail with COMM_ERR_CONNECT */
                        comm_connect_callback(F->fd, COMM_ERR_CONNECT);
                return;
        }
        /* If we get here, we've suceeded, so call with COMM_OK */
        comm_connect_callback(F->fd, COMM_OK);
}

/*
 * comm_error_str() - return an error string for the given error condition
 */
const char *
comm_errstr(int error)
{
        if(error < 0 || error >= COMM_ERR_MAX)
                return "Invalid error number!";
        return comm_err_str[error];
}


/*
 * comm_socket() - open a socket
 *
 * This is a highly highly cut down version of squid's comm_open() which
 * for the most part emulates socket(), *EXCEPT* it fails if we're about
 * to run out of file descriptors.
 */
int
comm_socket(int family, int sock_type, int proto, const char *note)
{
        int fd;
        /* First, make sure we aren't going to run out of file descriptors */
        if(number_fd >= MASTER_MAX)
        {
                errno = ENFILE;
                return -1;
        }

        /*
         * Next, we try to open the socket. We *should* drop the reserved FD
         * limit if/when we get an error, but we can deal with that later.
         * XXX !!! -- adrian
         */
        fd = socket(family, sock_type, proto);
        comm_fd_hack(&fd);
        if(fd < 0)
                return -1;      /* errno will be passed through, yay.. */

#if defined(IPV6) && defined(IPV6_V6ONLY)
        /* 
         * Make sure we can take both IPv4 and IPv6 connections
         * on an AF_INET6 socket
         */
        if(family == AF_INET6)
        {
                int off = 1;
                if(setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &off, sizeof(off)) == -1)
                {
                        libcharybdis_log("comm_socket: Could not set IPV6_V6ONLY option to 1 on FD %d: %s",
                             fd, strerror(errno));
                        close(fd);
                        return -1;
                }
        }
#endif

        /* Set the socket non-blocking, and other wonderful bits */
        if(!comm_set_nb(fd))
        {
                libcharybdis_log("comm_open: Couldn't set FD %d non blocking: %s", fd, strerror(errno));
                close(fd);
                return -1;
        }

        /* Next, update things in our fd tracking */
        comm_open(fd, FD_SOCKET, note);
        return fd;
}


/*
 * comm_accept() - accept an incoming connection
 *
 * This is a simple wrapper for accept() which enforces FD limits like
 * comm_open() does.
 */
int
comm_accept(int fd, struct sockaddr *pn, socklen_t *addrlen)
{
        int newfd;
        if(number_fd >= MASTER_MAX)
        {
                errno = ENFILE;
                return -1;
        }

        /*
         * Next, do the accept(). if we get an error, we should drop the
         * reserved fd limit, but we can deal with that when comm_open()
         * also does it. XXX -- adrian
         */
        newfd = accept(fd, (struct sockaddr *) pn, addrlen);
        comm_fd_hack(&newfd);
                        
        if(newfd < 0)
                return -1;

        /* Set the socket non-blocking, and other wonderful bits */
        if(!comm_set_nb(newfd))
        {
                libcharybdis_log("comm_accept: Couldn't set FD %d non blocking!", newfd);
                close(newfd);
                return -1;
        }

        /* Next, tag the FD as an incoming connection */
        comm_open(newfd, FD_SOCKET, "Incoming connection");

        /* .. and return */
        return newfd;
}

/*
 * If a sockaddr_storage is AF_INET6 but is a mapped IPv4
 * socket manged the sockaddr.
 */
#ifndef mangle_mapped_sockaddr
void
mangle_mapped_sockaddr(struct sockaddr *in)
{
        struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;                                                           

        if(in->sa_family == AF_INET)
                return;

        if(in->sa_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&in6->sin6_addr))
        {
                struct sockaddr_in in4;
                memset(&in4, 0, sizeof(struct sockaddr_in));
                in4.sin_family = AF_INET;
                in4.sin_port = in6->sin6_port;
                in4.sin_addr.s_addr = ((uint32_t *)&in6->sin6_addr)[3];
                memcpy(in, &in4, sizeof(struct sockaddr_in));           
        }       
        return;
}
#endif


static void
fdlist_update_biggest(int fd, int opening)
{
        if(fd < highest_fd)
                return;
        s_assert(fd < MAXCONNECTIONS);

        if(fd > highest_fd)
        {
                /*  
                 * s_assert that we are not closing a FD bigger than
                 * our known biggest FD
                 */
                s_assert(opening);
                highest_fd = fd;
                return;
        }
        /* if we are here, then fd == Biggest_FD */
        /*
         * s_assert that we are closing the biggest FD; we can't be
         * re-opening it
         */
        s_assert(!opening);
        while (highest_fd >= 0 && comm_locate_fd(fd)->flags.open) /* XXX */
                highest_fd--;
}


void
fdlist_init(void)
{
        static int initialized = 0;

        if(!initialized)
        {
                memset(&fd_table, '\0', sizeof(dlink_list) * FD_HASH_SIZE);
                initialized = 1;
        }
}

/* Called to open a given filedescriptor */
void
comm_open(int fd, unsigned int type, const char *desc)
{
        fde_t *F = comm_add_fd(fd);
        s_assert(fd >= 0);

        if(F->flags.open)
        {
                comm_close(fd);
        }
        s_assert(!F->flags.open);
        F->fd = fd;
        F->type = type;
        F->flags.open = 1;
#ifdef NOTYET
        F->defer.until = 0;
        F->defer.n = 0;
        F->defer.handler = NULL;
#endif
        fdlist_update_biggest(fd, 1);
        F->comm_index = -1;
        F->list = FDLIST_NONE;
        if(desc)
                strlcpy(F->desc, desc, sizeof(F->desc));
        number_fd++;
}


/* Called to close a given filedescriptor */
void
comm_close(int fd)
{
        fde_t *F = comm_locate_fd(fd);
        s_assert(F->flags.open);
        /* All disk fd's MUST go through file_close() ! */
        s_assert(F->type != FD_FILE);
        if(F->type == FD_FILE)
        {
                s_assert(F->read_handler == NULL);
                s_assert(F->write_handler == NULL);
        }
        comm_setselect(F->fd, FDLIST_NONE, COMM_SELECT_WRITE | COMM_SELECT_READ, NULL, NULL, 0);
        comm_setflush(F->fd, 0, NULL, NULL);
        
        if (F->dns_query != NULL)
        {
                delete_resolver_queries(F->dns_query);
                MyFree(F->dns_query);
                F->dns_query = NULL;
        }
        
        F->flags.open = 0;
        fdlist_update_biggest(fd, 0);
        number_fd--;
        comm_remove_fd(fd);

        F->timeout = 0;
        /* Unlike squid, we're actually closing the FD here! -- adrian */
        close(fd);
}

/*
 * comm_dump() - dump the list of active filedescriptors
 */
void
comm_dump(struct Client *source_p)
{
        int i;

        for (i = 0; i <= FD_HASH_SIZE; i++)
        {
                dlink_node *n;

                if (dlink_list_length(&fd_table[i]) <= 0)
                        continue;

                DLINK_FOREACH(n, fd_table[i].head)
                {
                        fde_t *F = (fde_t *) n->data;

                        if(F == NULL || !F->flags.open)
                                continue;

                        sendto_one_numeric(source_p, RPL_STATSDEBUG, 
                                           "F :fd %-3d bucket %-3d desc '%s'",
                                           F->fd, i, F->desc);
                }
        }
}

/*
 * comm_note() - set the fd note
 *
 * Note: must be careful not to overflow fd_table[fd].desc when
 *       calling.
 */
void
comm_note(int fd, const char *format, ...)
{
        va_list args;
        fde_t *F = comm_add_fd(fd);     /* XXX: epoll, kqueue. */

        if(format)
        {
                va_start(args, format);
                ircvsnprintf(F->desc, FD_DESC_SZ, format, args);
                va_end(args);
        }
        else
                F->desc[0] = '\0';
}