]> git.proxmox.com Git - qemu.git/blobdiff - slirp/slirp.c
rng-egd: remove redundant free
[qemu.git] / slirp / slirp.c
index 1593be177e4a8facec4db75255020cf0b232a708..bad8dad02e97e5350bf966dfa627ccd862d3edd2 100644 (file)
  * THE SOFTWARE.
  */
 #include "qemu-common.h"
-#include "qemu-timer.h"
-#include "qemu-char.h"
+#include "qemu/timer.h"
+#include "sysemu/char.h"
 #include "slirp.h"
 #include "hw/hw.h"
 
 /* host loopback address */
 struct in_addr loopback_addr;
+/* host loopback network mask */
+unsigned long loopback_mask;
 
 /* emulated hosts use the MAC addr 52:55:IP:IP:IP:IP */
-static const uint8_t special_ethaddr[6] = {
+static const uint8_t special_ethaddr[ETH_ALEN] = {
     0x52, 0x55, 0x00, 0x00, 0x00, 0x00
 };
 
-static const uint8_t zero_ethaddr[6] = { 0, 0, 0, 0, 0, 0 };
-
-/* XXX: suppress those select globals */
-fd_set *global_readfds, *global_writefds, *global_xfds;
+static const uint8_t zero_ethaddr[ETH_ALEN] = { 0, 0, 0, 0, 0, 0 };
 
 u_int curtime;
-static u_int time_fasttimo, last_slowtimo;
-static int do_slowtimo;
 
 static QTAILQ_HEAD(slirp_instances, Slirp) slirp_instances =
     QTAILQ_HEAD_INITIALIZER(slirp_instances);
@@ -50,6 +47,11 @@ static QTAILQ_HEAD(slirp_instances, Slirp) slirp_instances =
 static struct in_addr dns_addr;
 static u_int dns_addr_time;
 
+#define TIMEOUT_FAST 2  /* milliseconds */
+#define TIMEOUT_SLOW 499  /* milliseconds */
+/* for the aging of certain requests like DNS */
+#define TIMEOUT_DEFAULT 1000  /* milliseconds */
+
 #ifdef _WIN32
 
 int get_dns_addr(struct in_addr *pdns_addr)
@@ -60,7 +62,7 @@ int get_dns_addr(struct in_addr *pdns_addr)
     IP_ADDR_STRING *pIPAddr;
     struct in_addr tmp_addr;
 
-    if (dns_addr.s_addr != 0 && (curtime - dns_addr_time) < 1000) {
+    if (dns_addr.s_addr != 0 && (curtime - dns_addr_time) < TIMEOUT_DEFAULT) {
         *pdns_addr = dns_addr;
         return 0;
     }
@@ -116,7 +118,7 @@ int get_dns_addr(struct in_addr *pdns_addr)
 
     if (dns_addr.s_addr != 0) {
         struct stat old_stat;
-        if ((curtime - dns_addr_time) < 1000) {
+        if ((curtime - dns_addr_time) < TIMEOUT_DEFAULT) {
             *pdns_addr = dns_addr;
             return 0;
         }
@@ -191,6 +193,7 @@ static void slirp_init_once(void)
 #endif
 
     loopback_addr.s_addr = htonl(INADDR_LOOPBACK);
+    loopback_mask = htonl(IN_CLASSA_NET);
 }
 
 static void slirp_state_save(QEMUFile *f, void *opaque);
@@ -200,9 +203,10 @@ Slirp *slirp_init(int restricted, struct in_addr vnetwork,
                   struct in_addr vnetmask, struct in_addr vhost,
                   const char *vhostname, const char *tftp_path,
                   const char *bootfile, struct in_addr vdhcp_start,
-                  struct in_addr vnameserver, void *opaque)
+                  struct in_addr vnameserver, const char **vdnssearch,
+                  void *opaque)
 {
-    Slirp *slirp = qemu_mallocz(sizeof(Slirp));
+    Slirp *slirp = g_malloc0(sizeof(Slirp));
 
     slirp_init_once();
 
@@ -221,15 +225,15 @@ Slirp *slirp_init(int restricted, struct in_addr vnetwork,
         pstrcpy(slirp->client_hostname, sizeof(slirp->client_hostname),
                 vhostname);
     }
-    if (tftp_path) {
-        slirp->tftp_prefix = qemu_strdup(tftp_path);
-    }
-    if (bootfile) {
-        slirp->bootp_filename = qemu_strdup(bootfile);
-    }
+    slirp->tftp_prefix = g_strdup(tftp_path);
+    slirp->bootp_filename = g_strdup(bootfile);
     slirp->vdhcp_startaddr = vdhcp_start;
     slirp->vnameserver_addr = vnameserver;
 
+    if (vdnssearch) {
+        translate_dnssearch(slirp, vdnssearch);
+    }
+
     slirp->opaque = opaque;
 
     register_savevm(NULL, "slirp", 0, 3,
@@ -246,140 +250,218 @@ void slirp_cleanup(Slirp *slirp)
 
     unregister_savevm(NULL, "slirp", slirp);
 
-    qemu_free(slirp->tftp_prefix);
-    qemu_free(slirp->bootp_filename);
-    qemu_free(slirp);
+    ip_cleanup(slirp);
+    m_cleanup(slirp);
+
+    g_free(slirp->vdnssearch);
+    g_free(slirp->tftp_prefix);
+    g_free(slirp->bootp_filename);
+    g_free(slirp);
 }
 
 #define CONN_CANFSEND(so) (((so)->so_state & (SS_FCANTSENDMORE|SS_ISFCONNECTED)) == SS_ISFCONNECTED)
 #define CONN_CANFRCV(so) (((so)->so_state & (SS_FCANTRCVMORE|SS_ISFCONNECTED)) == SS_ISFCONNECTED)
-#define UPD_NFDS(x) if (nfds < (x)) nfds = (x)
 
-void slirp_select_fill(int *pnfds,
-                       fd_set *readfds, fd_set *writefds, fd_set *xfds)
+static void slirp_update_timeout(uint32_t *timeout)
+{
+    Slirp *slirp;
+    uint32_t t;
+
+    if (*timeout <= TIMEOUT_FAST) {
+        return;
+    }
+
+    t = MIN(1000, *timeout);
+
+    /* If we have tcp timeout with slirp, then we will fill @timeout with
+     * more precise value.
+     */
+    QTAILQ_FOREACH(slirp, &slirp_instances, entry) {
+        if (slirp->time_fasttimo) {
+            *timeout = TIMEOUT_FAST;
+            return;
+        }
+        if (slirp->do_slowtimo) {
+            t = MIN(TIMEOUT_SLOW, t);
+        }
+    }
+    *timeout = t;
+}
+
+void slirp_pollfds_fill(GArray *pollfds, uint32_t *timeout)
 {
     Slirp *slirp;
     struct socket *so, *so_next;
-    int nfds;
 
     if (QTAILQ_EMPTY(&slirp_instances)) {
         return;
     }
 
-    /* fail safe */
-    global_readfds = NULL;
-    global_writefds = NULL;
-    global_xfds = NULL;
-
-    nfds = *pnfds;
-       /*
-        * First, TCP sockets
-        */
-       do_slowtimo = 0;
-
-       QTAILQ_FOREACH(slirp, &slirp_instances, entry) {
-               /*
-                * *_slowtimo needs calling if there are IP fragments
-                * in the fragment queue, or there are TCP connections active
-                */
-               do_slowtimo |= ((slirp->tcb.so_next != &slirp->tcb) ||
-                   (&slirp->ipq.ip_link != slirp->ipq.ip_link.next));
-
-               for (so = slirp->tcb.so_next; so != &slirp->tcb;
-                    so = so_next) {
-                       so_next = so->so_next;
-
-                       /*
-                        * See if we need a tcp_fasttimo
-                        */
-                       if (time_fasttimo == 0 && so->so_tcpcb->t_flags & TF_DELACK)
-                          time_fasttimo = curtime; /* Flag when we want a fasttimo */
-
-                       /*
-                        * NOFDREF can include still connecting to local-host,
-                        * newly socreated() sockets etc. Don't want to select these.
-                        */
-                       if (so->so_state & SS_NOFDREF || so->s == -1)
-                          continue;
-
-                       /*
-                        * Set for reading sockets which are accepting
-                        */
-                       if (so->so_state & SS_FACCEPTCONN) {
-                                FD_SET(so->s, readfds);
-                               UPD_NFDS(so->s);
-                               continue;
-                       }
-
-                       /*
-                        * Set for writing sockets which are connecting
-                        */
-                       if (so->so_state & SS_ISFCONNECTING) {
-                               FD_SET(so->s, writefds);
-                               UPD_NFDS(so->s);
-                               continue;
-                       }
-
-                       /*
-                        * Set for writing if we are connected, can send more, and
-                        * we have something to send
-                        */
-                       if (CONN_CANFSEND(so) && so->so_rcv.sb_cc) {
-                               FD_SET(so->s, writefds);
-                               UPD_NFDS(so->s);
-                       }
-
-                       /*
-                        * Set for reading (and urgent data) if we are connected, can
-                        * receive more, and we have room for it XXX /2 ?
-                        */
-                       if (CONN_CANFRCV(so) && (so->so_snd.sb_cc < (so->so_snd.sb_datalen/2))) {
-                               FD_SET(so->s, readfds);
-                               FD_SET(so->s, xfds);
-                               UPD_NFDS(so->s);
-                       }
-               }
-
-               /*
-                * UDP sockets
-                */
-               for (so = slirp->udb.so_next; so != &slirp->udb;
-                    so = so_next) {
-                       so_next = so->so_next;
-
-                       /*
-                        * See if it's timed out
-                        */
-                       if (so->so_expire) {
-                               if (so->so_expire <= curtime) {
-                                       udp_detach(so);
-                                       continue;
-                               } else
-                                       do_slowtimo = 1; /* Let socket expire */
-                       }
-
-                       /*
-                        * When UDP packets are received from over the
-                        * link, they're sendto()'d straight away, so
-                        * no need for setting for writing
-                        * Limit the number of packets queued by this session
-                        * to 4.  Note that even though we try and limit this
-                        * to 4 packets, the session could have more queued
-                        * if the packets needed to be fragmented
-                        * (XXX <= 4 ?)
-                        */
-                       if ((so->so_state & SS_ISFCONNECTED) && so->so_queued <= 4) {
-                               FD_SET(so->s, readfds);
-                               UPD_NFDS(so->s);
-                       }
-               }
-       }
-
-        *pnfds = nfds;
+    /*
+     * First, TCP sockets
+     */
+
+    QTAILQ_FOREACH(slirp, &slirp_instances, entry) {
+        /*
+         * *_slowtimo needs calling if there are IP fragments
+         * in the fragment queue, or there are TCP connections active
+         */
+        slirp->do_slowtimo = ((slirp->tcb.so_next != &slirp->tcb) ||
+                (&slirp->ipq.ip_link != slirp->ipq.ip_link.next));
+
+        for (so = slirp->tcb.so_next; so != &slirp->tcb;
+                so = so_next) {
+            int events = 0;
+
+            so_next = so->so_next;
+
+            so->pollfds_idx = -1;
+
+            /*
+             * See if we need a tcp_fasttimo
+             */
+            if (slirp->time_fasttimo == 0 &&
+                so->so_tcpcb->t_flags & TF_DELACK) {
+                slirp->time_fasttimo = curtime; /* Flag when want a fasttimo */
+            }
+
+            /*
+             * NOFDREF can include still connecting to local-host,
+             * newly socreated() sockets etc. Don't want to select these.
+             */
+            if (so->so_state & SS_NOFDREF || so->s == -1) {
+                continue;
+            }
+
+            /*
+             * Set for reading sockets which are accepting
+             */
+            if (so->so_state & SS_FACCEPTCONN) {
+                GPollFD pfd = {
+                    .fd = so->s,
+                    .events = G_IO_IN | G_IO_HUP | G_IO_ERR,
+                };
+                so->pollfds_idx = pollfds->len;
+                g_array_append_val(pollfds, pfd);
+                continue;
+            }
+
+            /*
+             * Set for writing sockets which are connecting
+             */
+            if (so->so_state & SS_ISFCONNECTING) {
+                GPollFD pfd = {
+                    .fd = so->s,
+                    .events = G_IO_OUT | G_IO_ERR,
+                };
+                so->pollfds_idx = pollfds->len;
+                g_array_append_val(pollfds, pfd);
+                continue;
+            }
+
+            /*
+             * Set for writing if we are connected, can send more, and
+             * we have something to send
+             */
+            if (CONN_CANFSEND(so) && so->so_rcv.sb_cc) {
+                events |= G_IO_OUT | G_IO_ERR;
+            }
+
+            /*
+             * Set for reading (and urgent data) if we are connected, can
+             * receive more, and we have room for it XXX /2 ?
+             */
+            if (CONN_CANFRCV(so) &&
+                (so->so_snd.sb_cc < (so->so_snd.sb_datalen/2))) {
+                events |= G_IO_IN | G_IO_HUP | G_IO_ERR | G_IO_PRI;
+            }
+
+            if (events) {
+                GPollFD pfd = {
+                    .fd = so->s,
+                    .events = events,
+                };
+                so->pollfds_idx = pollfds->len;
+                g_array_append_val(pollfds, pfd);
+            }
+        }
+
+        /*
+         * UDP sockets
+         */
+        for (so = slirp->udb.so_next; so != &slirp->udb;
+                so = so_next) {
+            so_next = so->so_next;
+
+            so->pollfds_idx = -1;
+
+            /*
+             * See if it's timed out
+             */
+            if (so->so_expire) {
+                if (so->so_expire <= curtime) {
+                    udp_detach(so);
+                    continue;
+                } else {
+                    slirp->do_slowtimo = true; /* Let socket expire */
+                }
+            }
+
+            /*
+             * When UDP packets are received from over the
+             * link, they're sendto()'d straight away, so
+             * no need for setting for writing
+             * Limit the number of packets queued by this session
+             * to 4.  Note that even though we try and limit this
+             * to 4 packets, the session could have more queued
+             * if the packets needed to be fragmented
+             * (XXX <= 4 ?)
+             */
+            if ((so->so_state & SS_ISFCONNECTED) && so->so_queued <= 4) {
+                GPollFD pfd = {
+                    .fd = so->s,
+                    .events = G_IO_IN | G_IO_HUP | G_IO_ERR,
+                };
+                so->pollfds_idx = pollfds->len;
+                g_array_append_val(pollfds, pfd);
+            }
+        }
+
+        /*
+         * ICMP sockets
+         */
+        for (so = slirp->icmp.so_next; so != &slirp->icmp;
+                so = so_next) {
+            so_next = so->so_next;
+
+            so->pollfds_idx = -1;
+
+            /*
+             * See if it's timed out
+             */
+            if (so->so_expire) {
+                if (so->so_expire <= curtime) {
+                    icmp_detach(so);
+                    continue;
+                } else {
+                    slirp->do_slowtimo = true; /* Let socket expire */
+                }
+            }
+
+            if (so->so_state & SS_ISFCONNECTED) {
+                GPollFD pfd = {
+                    .fd = so->s,
+                    .events = G_IO_IN | G_IO_HUP | G_IO_ERR,
+                };
+                so->pollfds_idx = pollfds->len;
+                g_array_append_val(pollfds, pfd);
+            }
+        }
+    }
+    slirp_update_timeout(timeout);
 }
 
-void slirp_select_poll(fd_set *readfds, fd_set *writefds, fd_set *xfds,
-                       int select_error)
+void slirp_pollfds_poll(GArray *pollfds, int select_error)
 {
     Slirp *slirp;
     struct socket *so, *so_next;
@@ -389,215 +471,208 @@ void slirp_select_poll(fd_set *readfds, fd_set *writefds, fd_set *xfds,
         return;
     }
 
-    global_readfds = readfds;
-    global_writefds = writefds;
-    global_xfds = xfds;
-
-    curtime = qemu_get_clock_ms(rt_clock);
+    curtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 
     QTAILQ_FOREACH(slirp, &slirp_instances, entry) {
-       /*
-        * See if anything has timed out
-        */
-               if (time_fasttimo && ((curtime - time_fasttimo) >= 2)) {
-                       tcp_fasttimo(slirp);
-                       time_fasttimo = 0;
-               }
-               if (do_slowtimo && ((curtime - last_slowtimo) >= 499)) {
-                       ip_slowtimo(slirp);
-                       tcp_slowtimo(slirp);
-                       last_slowtimo = curtime;
-               }
-
-       /*
-        * Check sockets
-        */
-       if (!select_error) {
-               /*
-                * Check TCP sockets
-                */
-               for (so = slirp->tcb.so_next; so != &slirp->tcb;
-                    so = so_next) {
-                       so_next = so->so_next;
-
-                       /*
-                        * FD_ISSET is meaningless on these sockets
-                        * (and they can crash the program)
-                        */
-                       if (so->so_state & SS_NOFDREF || so->s == -1)
-                          continue;
-
-                       /*
-                        * Check for URG data
-                        * This will soread as well, so no need to
-                        * test for readfds below if this succeeds
-                        */
-                       if (FD_ISSET(so->s, xfds))
-                          sorecvoob(so);
-                       /*
-                        * Check sockets for reading
-                        */
-                       else if (FD_ISSET(so->s, readfds)) {
-                               /*
-                                * Check for incoming connections
-                                */
-                               if (so->so_state & SS_FACCEPTCONN) {
-                                       tcp_connect(so);
-                                       continue;
-                               } /* else */
-                               ret = soread(so);
-
-                               /* Output it if we read something */
-                               if (ret > 0)
-                                  tcp_output(sototcpcb(so));
-                       }
-
-                       /*
-                        * Check sockets for writing
-                        */
-                       if (FD_ISSET(so->s, writefds)) {
-                         /*
-                          * Check for non-blocking, still-connecting sockets
-                          */
-                         if (so->so_state & SS_ISFCONNECTING) {
-                           /* Connected */
-                           so->so_state &= ~SS_ISFCONNECTING;
-
-                           ret = send(so->s, (const void *) &ret, 0, 0);
-                           if (ret < 0) {
-                             /* XXXXX Must fix, zero bytes is a NOP */
-                             if (errno == EAGAIN || errno == EWOULDBLOCK ||
-                                 errno == EINPROGRESS || errno == ENOTCONN)
-                               continue;
-
-                             /* else failed */
-                             so->so_state &= SS_PERSISTENT_MASK;
-                             so->so_state |= SS_NOFDREF;
-                           }
-                           /* else so->so_state &= ~SS_ISFCONNECTING; */
-
-                           /*
-                            * Continue tcp_input
-                            */
-                           tcp_input((struct mbuf *)NULL, sizeof(struct ip), so);
-                           /* continue; */
-                         } else
-                           ret = sowrite(so);
-                         /*
-                          * XXXXX If we wrote something (a lot), there
-                          * could be a need for a window update.
-                          * In the worst case, the remote will send
-                          * a window probe to get things going again
-                          */
-                       }
-
-                       /*
-                        * Probe a still-connecting, non-blocking socket
-                        * to check if it's still alive
-                        */
+        /*
+         * See if anything has timed out
+         */
+        if (slirp->time_fasttimo &&
+            ((curtime - slirp->time_fasttimo) >= TIMEOUT_FAST)) {
+            tcp_fasttimo(slirp);
+            slirp->time_fasttimo = 0;
+        }
+        if (slirp->do_slowtimo &&
+            ((curtime - slirp->last_slowtimo) >= TIMEOUT_SLOW)) {
+            ip_slowtimo(slirp);
+            tcp_slowtimo(slirp);
+            slirp->last_slowtimo = curtime;
+        }
+
+        /*
+         * Check sockets
+         */
+        if (!select_error) {
+            /*
+             * Check TCP sockets
+             */
+            for (so = slirp->tcb.so_next; so != &slirp->tcb;
+                    so = so_next) {
+                int revents;
+
+                so_next = so->so_next;
+
+                revents = 0;
+                if (so->pollfds_idx != -1) {
+                    revents = g_array_index(pollfds, GPollFD,
+                                            so->pollfds_idx).revents;
+                }
+
+                if (so->so_state & SS_NOFDREF || so->s == -1) {
+                    continue;
+                }
+
+                /*
+                 * Check for URG data
+                 * This will soread as well, so no need to
+                 * test for G_IO_IN below if this succeeds
+                 */
+                if (revents & G_IO_PRI) {
+                    sorecvoob(so);
+                }
+                /*
+                 * Check sockets for reading
+                 */
+                else if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
+                    /*
+                     * Check for incoming connections
+                     */
+                    if (so->so_state & SS_FACCEPTCONN) {
+                        tcp_connect(so);
+                        continue;
+                    } /* else */
+                    ret = soread(so);
+
+                    /* Output it if we read something */
+                    if (ret > 0) {
+                        tcp_output(sototcpcb(so));
+                    }
+                }
+
+                /*
+                 * Check sockets for writing
+                 */
+                if (!(so->so_state & SS_NOFDREF) &&
+                        (revents & (G_IO_OUT | G_IO_ERR))) {
+                    /*
+                     * Check for non-blocking, still-connecting sockets
+                     */
+                    if (so->so_state & SS_ISFCONNECTING) {
+                        /* Connected */
+                        so->so_state &= ~SS_ISFCONNECTING;
+
+                        ret = send(so->s, (const void *) &ret, 0, 0);
+                        if (ret < 0) {
+                            /* XXXXX Must fix, zero bytes is a NOP */
+                            if (errno == EAGAIN || errno == EWOULDBLOCK ||
+                                errno == EINPROGRESS || errno == ENOTCONN) {
+                                continue;
+                            }
+
+                            /* else failed */
+                            so->so_state &= SS_PERSISTENT_MASK;
+                            so->so_state |= SS_NOFDREF;
+                        }
+                        /* else so->so_state &= ~SS_ISFCONNECTING; */
+
+                        /*
+                         * Continue tcp_input
+                         */
+                        tcp_input((struct mbuf *)NULL, sizeof(struct ip), so);
+                        /* continue; */
+                    } else {
+                        ret = sowrite(so);
+                    }
+                    /*
+                     * XXXXX If we wrote something (a lot), there
+                     * could be a need for a window update.
+                     * In the worst case, the remote will send
+                     * a window probe to get things going again
+                     */
+                }
+
+                /*
+                 * Probe a still-connecting, non-blocking socket
+                 * to check if it's still alive
+                 */
 #ifdef PROBE_CONN
-                       if (so->so_state & SS_ISFCONNECTING) {
-                         ret = recv(so->s, (char *)&ret, 0,0);
-
-                         if (ret < 0) {
-                           /* XXX */
-                           if (errno == EAGAIN || errno == EWOULDBLOCK ||
-                               errno == EINPROGRESS || errno == ENOTCONN)
-                             continue; /* Still connecting, continue */
-
-                           /* else failed */
-                           so->so_state &= SS_PERSISTENT_MASK;
-                           so->so_state |= SS_NOFDREF;
-
-                           /* tcp_input will take care of it */
-                         } else {
-                           ret = send(so->s, &ret, 0,0);
-                           if (ret < 0) {
-                             /* XXX */
-                             if (errno == EAGAIN || errno == EWOULDBLOCK ||
-                                 errno == EINPROGRESS || errno == ENOTCONN)
-                               continue;
-                             /* else failed */
-                             so->so_state &= SS_PERSISTENT_MASK;
-                             so->so_state |= SS_NOFDREF;
-                           } else
-                             so->so_state &= ~SS_ISFCONNECTING;
-
-                         }
-                         tcp_input((struct mbuf *)NULL, sizeof(struct ip),so);
-                       } /* SS_ISFCONNECTING */
-#endif
-               }
-
-               /*
-                * Now UDP sockets.
-                * Incoming packets are sent straight away, they're not buffered.
-                * Incoming UDP data isn't buffered either.
-                */
-               for (so = slirp->udb.so_next; so != &slirp->udb;
-                    so = so_next) {
-                       so_next = so->so_next;
-
-                       if (so->s != -1 && FD_ISSET(so->s, readfds)) {
-                            sorecvfrom(so);
+                if (so->so_state & SS_ISFCONNECTING) {
+                    ret = qemu_recv(so->s, &ret, 0, 0);
+
+                    if (ret < 0) {
+                        /* XXX */
+                        if (errno == EAGAIN || errno == EWOULDBLOCK ||
+                            errno == EINPROGRESS || errno == ENOTCONN) {
+                            continue; /* Still connecting, continue */
                         }
-               }
-       }
-
-       /*
-        * See if we can start outputting
-        */
-       if (slirp->if_queued) {
-           if_start(slirp);
-       }
-    }
-
-       /* clear global file descriptor sets.
-        * these reside on the stack in vl.c
-        * so they're unusable if we're not in
-        * slirp_select_fill or slirp_select_poll.
-        */
-        global_readfds = NULL;
-        global_writefds = NULL;
-        global_xfds = NULL;
-}
 
-#define ETH_ALEN 6
-#define ETH_HLEN 14
+                        /* else failed */
+                        so->so_state &= SS_PERSISTENT_MASK;
+                        so->so_state |= SS_NOFDREF;
+
+                        /* tcp_input will take care of it */
+                    } else {
+                        ret = send(so->s, &ret, 0, 0);
+                        if (ret < 0) {
+                            /* XXX */
+                            if (errno == EAGAIN || errno == EWOULDBLOCK ||
+                                errno == EINPROGRESS || errno == ENOTCONN) {
+                                continue;
+                            }
+                            /* else failed */
+                            so->so_state &= SS_PERSISTENT_MASK;
+                            so->so_state |= SS_NOFDREF;
+                        } else {
+                            so->so_state &= ~SS_ISFCONNECTING;
+                        }
 
-#define ETH_P_IP       0x0800          /* Internet Protocol packet     */
-#define ETH_P_ARP      0x0806          /* Address Resolution packet    */
+                    }
+                    tcp_input((struct mbuf *)NULL, sizeof(struct ip), so);
+                } /* SS_ISFCONNECTING */
+#endif
+            }
 
-#define        ARPOP_REQUEST   1               /* ARP request                  */
-#define        ARPOP_REPLY     2               /* ARP reply                    */
+            /*
+             * Now UDP sockets.
+             * Incoming packets are sent straight away, they're not buffered.
+             * Incoming UDP data isn't buffered either.
+             */
+            for (so = slirp->udb.so_next; so != &slirp->udb;
+                    so = so_next) {
+                int revents;
+
+                so_next = so->so_next;
+
+                revents = 0;
+                if (so->pollfds_idx != -1) {
+                    revents = g_array_index(pollfds, GPollFD,
+                            so->pollfds_idx).revents;
+                }
+
+                if (so->s != -1 &&
+                    (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR))) {
+                    sorecvfrom(so);
+                }
+            }
 
-struct ethhdr
-{
-       unsigned char   h_dest[ETH_ALEN];       /* destination eth addr */
-       unsigned char   h_source[ETH_ALEN];     /* source ether addr    */
-       unsigned short  h_proto;                /* packet type ID field */
-};
+            /*
+             * Check incoming ICMP relies.
+             */
+            for (so = slirp->icmp.so_next; so != &slirp->icmp;
+                    so = so_next) {
+                    int revents;
+
+                    so_next = so->so_next;
+
+                    revents = 0;
+                    if (so->pollfds_idx != -1) {
+                        revents = g_array_index(pollfds, GPollFD,
+                                                so->pollfds_idx).revents;
+                    }
+
+                    if (so->s != -1 &&
+                        (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR))) {
+                    icmp_receive(so);
+                }
+            }
+        }
 
-struct arphdr
-{
-       unsigned short  ar_hrd;         /* format of hardware address   */
-       unsigned short  ar_pro;         /* format of protocol address   */
-       unsigned char   ar_hln;         /* length of hardware address   */
-       unsigned char   ar_pln;         /* length of protocol address   */
-       unsigned short  ar_op;          /* ARP opcode (command)         */
-
-        /*
-         *      Ethernet looks like this : This bit is variable sized however...
-         */
-       unsigned char           ar_sha[ETH_ALEN];       /* sender hardware address      */
-       uint32_t                ar_sip;                 /* sender IP address            */
-       unsigned char           ar_tha[ETH_ALEN];       /* target hardware address      */
-       uint32_t                ar_tip  ;               /* target IP address            */
-} __attribute__((packed));
+        if_start(slirp);
+    }
+}
 
 static void arp_input(Slirp *slirp, const uint8_t *pkt, int pkt_len)
 {
-    struct ethhdr *eh = (struct ethhdr *)pkt;
     struct arphdr *ah = (struct arphdr *)(pkt + ETH_HLEN);
     uint8_t arp_reply[max(ETH_HLEN + sizeof(struct arphdr), 64)];
     struct ethhdr *reh = (struct ethhdr *)arp_reply;
@@ -608,6 +683,12 @@ static void arp_input(Slirp *slirp, const uint8_t *pkt, int pkt_len)
     ar_op = ntohs(ah->ar_op);
     switch(ar_op) {
     case ARPOP_REQUEST:
+        if (ah->ar_tip == ah->ar_sip) {
+            /* Gratuitous ARP */
+            arp_table_add(slirp, ah->ar_sip, ah->ar_sha);
+            return;
+        }
+
         if ((ah->ar_tip & slirp->vnetwork_mask.s_addr) ==
             slirp->vnetwork_addr.s_addr) {
             if (ah->ar_tip == slirp->vnameserver_addr.s_addr ||
@@ -620,8 +701,8 @@ static void arp_input(Slirp *slirp, const uint8_t *pkt, int pkt_len)
             return;
         arp_ok:
             memset(arp_reply, 0, sizeof(arp_reply));
-            /* XXX: make an ARP request to have the client address */
-            memcpy(slirp->client_ethaddr, eh->h_source, ETH_ALEN);
+
+            arp_table_add(slirp, ah->ar_sip, ah->ar_sha);
 
             /* ARP request for alias/dns mac address */
             memcpy(reh->h_dest, pkt + ETH_ALEN, ETH_ALEN);
@@ -642,11 +723,7 @@ static void arp_input(Slirp *slirp, const uint8_t *pkt, int pkt_len)
         }
         break;
     case ARPOP_REPLY:
-        /* reply to request of client mac address ? */
-        if (!memcmp(slirp->client_ethaddr, zero_ethaddr, ETH_ALEN) &&
-            ah->ar_sip == slirp->client_ipaddr.s_addr) {
-            memcpy(slirp->client_ethaddr, ah->ar_sha, ETH_ALEN);
-        }
+        arp_table_add(slirp, ah->ar_sip, ah->ar_sha);
         break;
     default:
         break;
@@ -687,54 +764,66 @@ void slirp_input(Slirp *slirp, const uint8_t *pkt, int pkt_len)
     }
 }
 
-/* output the IP packet to the ethernet device */
-void if_encap(Slirp *slirp, const uint8_t *ip_data, int ip_data_len)
+/* Output the IP packet to the ethernet device. Returns 0 if the packet must be
+ * re-queued.
+ */
+int if_encap(Slirp *slirp, struct mbuf *ifm)
 {
     uint8_t buf[1600];
     struct ethhdr *eh = (struct ethhdr *)buf;
+    uint8_t ethaddr[ETH_ALEN];
+    const struct ip *iph = (const struct ip *)ifm->m_data;
 
-    if (ip_data_len + ETH_HLEN > sizeof(buf))
-        return;
-    
-    if (!memcmp(slirp->client_ethaddr, zero_ethaddr, ETH_ALEN)) {
+    if (ifm->m_len + ETH_HLEN > sizeof(buf)) {
+        return 1;
+    }
+
+    if (!arp_table_search(slirp, iph->ip_dst.s_addr, ethaddr)) {
         uint8_t arp_req[ETH_HLEN + sizeof(struct arphdr)];
         struct ethhdr *reh = (struct ethhdr *)arp_req;
         struct arphdr *rah = (struct arphdr *)(arp_req + ETH_HLEN);
-        const struct ip *iph = (const struct ip *)ip_data;
-
-        /* If the client addr is not known, there is no point in
-           sending the packet to it. Normally the sender should have
-           done an ARP request to get its MAC address. Here we do it
-           in place of sending the packet and we hope that the sender
-           will retry sending its packet. */
-        memset(reh->h_dest, 0xff, ETH_ALEN);
-        memcpy(reh->h_source, special_ethaddr, ETH_ALEN - 4);
-        memcpy(&reh->h_source[2], &slirp->vhost_addr, 4);
-        reh->h_proto = htons(ETH_P_ARP);
-        rah->ar_hrd = htons(1);
-        rah->ar_pro = htons(ETH_P_IP);
-        rah->ar_hln = ETH_ALEN;
-        rah->ar_pln = 4;
-        rah->ar_op = htons(ARPOP_REQUEST);
-        /* source hw addr */
-        memcpy(rah->ar_sha, special_ethaddr, ETH_ALEN - 4);
-        memcpy(&rah->ar_sha[2], &slirp->vhost_addr, 4);
-        /* source IP */
-        rah->ar_sip = slirp->vhost_addr.s_addr;
-        /* target hw addr (none) */
-        memset(rah->ar_tha, 0, ETH_ALEN);
-        /* target IP */
-        rah->ar_tip = iph->ip_dst.s_addr;
-        slirp->client_ipaddr = iph->ip_dst;
-        slirp_output(slirp->opaque, arp_req, sizeof(arp_req));
+
+        if (!ifm->arp_requested) {
+            /* If the client addr is not known, send an ARP request */
+            memset(reh->h_dest, 0xff, ETH_ALEN);
+            memcpy(reh->h_source, special_ethaddr, ETH_ALEN - 4);
+            memcpy(&reh->h_source[2], &slirp->vhost_addr, 4);
+            reh->h_proto = htons(ETH_P_ARP);
+            rah->ar_hrd = htons(1);
+            rah->ar_pro = htons(ETH_P_IP);
+            rah->ar_hln = ETH_ALEN;
+            rah->ar_pln = 4;
+            rah->ar_op = htons(ARPOP_REQUEST);
+
+            /* source hw addr */
+            memcpy(rah->ar_sha, special_ethaddr, ETH_ALEN - 4);
+            memcpy(&rah->ar_sha[2], &slirp->vhost_addr, 4);
+
+            /* source IP */
+            rah->ar_sip = slirp->vhost_addr.s_addr;
+
+            /* target hw addr (none) */
+            memset(rah->ar_tha, 0, ETH_ALEN);
+
+            /* target IP */
+            rah->ar_tip = iph->ip_dst.s_addr;
+            slirp->client_ipaddr = iph->ip_dst;
+            slirp_output(slirp->opaque, arp_req, sizeof(arp_req));
+            ifm->arp_requested = true;
+
+            /* Expire request and drop outgoing packet after 1 second */
+            ifm->expiration_date = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + 1000000000ULL;
+        }
+        return 0;
     } else {
-        memcpy(eh->h_dest, slirp->client_ethaddr, ETH_ALEN);
+        memcpy(eh->h_dest, ethaddr, ETH_ALEN);
         memcpy(eh->h_source, special_ethaddr, ETH_ALEN - 4);
         /* XXX: not correct */
         memcpy(&eh->h_source[2], &slirp->vhost_addr, 4);
         eh->h_proto = htons(ETH_P_IP);
-        memcpy(buf + sizeof(struct ethhdr), ip_data, ip_data_len);
-        slirp_output(slirp->opaque, buf, ip_data_len + ETH_HLEN);
+        memcpy(buf + sizeof(struct ethhdr), ifm->m_data, ifm->m_len);
+        slirp_output(slirp->opaque, buf, ifm->m_len + ETH_HLEN);
+        return 1;
     }
 }
 
@@ -800,12 +889,12 @@ int slirp_add_exec(Slirp *slirp, int do_pty, const void *args,
 
 ssize_t slirp_send(struct socket *so, const void *buf, size_t len, int flags)
 {
-       if (so->s == -1 && so->extra) {
-               qemu_chr_write(so->extra, buf, len);
-               return len;
-       }
+    if (so->s == -1 && so->extra) {
+        qemu_chr_fe_write(so->extra, buf, len);
+        return len;
+    }
 
-       return send(so->s, buf, len, flags);
+    return send(so->s, buf, len, flags);
 }
 
 static struct socket *
@@ -825,18 +914,20 @@ slirp_find_ctl_socket(Slirp *slirp, struct in_addr guest_addr, int guest_port)
 size_t slirp_socket_can_recv(Slirp *slirp, struct in_addr guest_addr,
                              int guest_port)
 {
-       struct iovec iov[2];
-       struct socket *so;
+    struct iovec iov[2];
+    struct socket *so;
 
-       so = slirp_find_ctl_socket(slirp, guest_addr, guest_port);
+    so = slirp_find_ctl_socket(slirp, guest_addr, guest_port);
 
-       if (!so || so->so_state & SS_NOFDREF)
-               return 0;
+    if (!so || so->so_state & SS_NOFDREF) {
+        return 0;
+    }
 
-       if (!CONN_CANFRCV(so) || so->so_snd.sb_cc >= (so->so_snd.sb_datalen/2))
-               return 0;
+    if (!CONN_CANFRCV(so) || so->so_snd.sb_cc >= (so->so_snd.sb_datalen/2)) {
+        return 0;
+    }
 
-       return sopreprbuf(so, iov, NULL);
+    return sopreprbuf(so, iov, NULL);
 }
 
 void slirp_socket_recv(Slirp *slirp, struct in_addr guest_addr, int guest_port,