Changeset 15890 in vbox for trunk/src/VBox/Devices

Timestamp:

Jan 12, 2009 6:26:43 AM (16 years ago)

Author:

vboxsync

Message:

NAT: 1. wo sync enhancement branch is still functional (was corrupted with using ICMP file handler in select(1))

2. after sending send queue doesn't need to synchronize with NAT thread to free mbuf instead NAT queue used to call freeing slirp routine.
3. no more copying on slirp to guest sent.

Location:

trunk/src/VBox/Devices/Network

Files:

• DrvNAT.cpp (modified) (5 diffs)
• slirp/bootp.c (modified) (2 diffs)
• slirp/if.c (modified) (2 diffs)
• slirp/ip_icmp.c (modified) (1 diff)
• slirp/libslirp.h (modified) (1 diff)
• slirp/main.h (modified) (1 diff)
• slirp/slirp.c (modified) (14 diffs)
• slirp/socket.c (modified) (1 diff)
• slirp/tcp_input.c (modified) (15 diffs)
• slirp/tcp_output.c (modified) (1 diff)

trunk/src/VBox/Devices/Network/DrvNAT.cpp

@@ -102 +102 @@
     PDMQUEUEITEMCORE    Core;
     /** The buffer for output to guest. */
-    const uint8_t            *pu8Buf;
+    const uint8_t       *pu8Buf;
     /* size of buffer */
-    size_t             cb;
+    size_t              cb;
+    void                *mbuf;
 } DRVNATQUEUITEM, *PDRVNATQUEUITEM;
 
@@ -452 +453 @@
  * Function called by slirp to feed incoming data to the network port.
  */
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+void slirp_output(void *pvUser, void *pvArg, const uint8_t *pu8Buf, int cb)
+#else
 void slirp_output(void *pvUser, const uint8_t *pu8Buf, int cb)
+#endif
 {
     PDRVNAT pThis = (PDRVNAT)pvUser;
@@ -476 +481 @@
         pItem->pu8Buf = pu8Buf;
         pItem->cb = cb;
+        pItem->mbuf = pvArg;
         Log2(("pItem:%p %.Rhxd\n", pItem, pItem->pu8Buf));
         PDMQueueInsert(pThis->pSendQueue, &pItem->Core);
@@ -496 +502 @@
     PDRVNAT pThis = PDMINS_2_DATA(pDrvIns, PDRVNAT);
     PDRVNATQUEUITEM pItem = (PDRVNATQUEUITEM)pItemCore;
+    PRTREQ pReq = NULL;
     Log(("drvNATQueueConsumer(pItem:%p, pu8Buf:%p, cb:%d)\n", pItem, pItem->pu8Buf, pItem->cb));
     Log2(("drvNATQueueConsumer: pu8Buf:\n%.Rhxd\n", pItem->pu8Buf));
@@ -502 +509 @@
         return false;
     rc = pThis->pPort->pfnReceive(pThis->pPort, pItem->pu8Buf, pItem->cb);
+
+    rc = RTReqAlloc(pThis->pReqQueue, &pReq, RTREQTYPE_INTERNAL);
+    AssertReleaseRC(rc);
+    pReq->u.Internal.pfn      = (PFNRT)slirp_post_sent;
+    pReq->u.Internal.cArgs    = 2;
+    pReq->u.Internal.aArgs[0] = (uintptr_t)pThis->pNATState;
+    pReq->u.Internal.aArgs[1] = (uintptr_t)pItem->mbuf;
+    pReq->fFlags              = RTREQFLAGS_VOID;
     AssertRC(rc);
-    RTMemFree((void *)pItem->pu8Buf); /* XXX: shouldn't free buffer here */
     return RT_SUCCESS(rc);
 }

trunk/src/VBox/Devices/Network/slirp/bootp.c

@@ -166 +166 @@
     if ((m = m_get(pData)) == NULL)
         return;
-    m->m_data += if_maxlinkhdr;
-    rbp = (struct bootp_t *)m->m_data;
+    m->m_data += if_maxlinkhdr; /*reserve ether header */
+    rbp = mtod(m, struct bootp_t *);
+    memset(rbp, 0, sizeof(struct bootp_t));
+#ifndef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
     m->m_data += sizeof(struct udpiphdr);
-    memset(rbp, 0, sizeof(struct bootp_t));
+#endif
 
     if (dhcp_msg_type == DHCPDISCOVER)
@@ -310 +312 @@
              - sizeof(struct ip)
              - sizeof(struct udphdr);
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+    m->m_data += sizeof(struct udphdr)
+             + sizeof(struct ip);
+#endif
     /* Reply to the broadcast address, as some clients perform paranoid checks. */
     daddr.sin_addr.s_addr = INADDR_BROADCAST;

trunk/src/VBox/Devices/Network/slirp/if.c

@@ -28 +28 @@
 if_init(PNATState pData)
 {
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+    /* 14 for ethernet */
+    if_maxlinkhdr =  14;
+#else
     /* 2 for alignment, 14 for ethernet, 40 for TCP/IP */
-    if_maxlinkhdr = 2 + 14 + 40;
+    if_maxlinkhdr =  2 + 14 + 40;
+#endif
     if_queued = 0;
     if_thresh = 10;
@@ -224 +229 @@
         }
 
-        /* Encapsulate the packet for sending */
-        if_encap(pData, ETH_P_IP, (const uint8_t *)ifm->m_data, ifm->m_len);
-
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+        if_encap(pData, ETH_P_IP, ifm);
+#else
+        if_encap(pData,  mtod(ifm, uint8_t *), ifm->m_len);
         m_free(pData, ifm);
+#endif
 
         if (!if_queued)

trunk/src/VBox/Devices/Network/slirp/ip_icmp.c

@@ -521 +521 @@
     {
         int new_m_size;
+        m->m_data += if_maxlinkhdr;
         new_m_size = sizeof(struct ip) + ICMP_MINLEN + msrc->m_len + ICMP_MAXDATALEN;
         if (new_m_size>m->m_size)

trunk/src/VBox/Devices/Network/slirp/libslirp.h

@@ -47 +47 @@
 /* you must provide the following functions: */
 int slirp_can_output(void * pvUser);
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+void slirp_output(void * pvUser, void *pvArg, const uint8_t *pkt, int pkt_len);
+void slirp_post_sent(PNATState pData, void *pvArg);
+#else
 void slirp_output(void * pvUser, const uint8_t *pkt, int pkt_len);
+#endif
 
 int slirp_redir(PNATState pData, int is_udp, int host_port,

trunk/src/VBox/Devices/Network/slirp/main.h

@@ -10 +10 @@
 #endif
 
-void if_encap(PNATState pData, uint16_t eth_proto, const uint8_t *ip_data, int ip_data_len);
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+void if_encap(PNATState pData, uint16_t eth_proto, struct mbuf *m);
+#else
+void if_encap(PNATState pData, const uint8_t *ip_data, int ip_data_len);
+#endif

trunk/src/VBox/Devices/Network/slirp/slirp.c

@@ -33 +33 @@
 # define DO_WIN_CHECK_FD_SET(so, events, fdset ) 0 /* specific for Windows Winsock API */
 
-# define ICMP_ENGAGE_EVENT(so, fdset)               \
+# ifndef RT_OS_WINDOWS
+#  define ICMP_ENGAGE_EVENT(so, fdset)               \
     do {                                             \
         if (pData->icmp_socket.s != -1)              \
             DO_ENGAGE_EVENT1((so), (fdset), ICMP);   \
     } while (0)
+# else /* !RT_OS_WINDOWS */
+#  define ICMP_ENGAGE_EVENT(so, fdset) do {} while(0)
+#endif /* RT_OS_WINDOWS */
 
 #else /* defined(VBOX_WITH_SIMPLIFIED_SLIRP_SYNC) && defined(RT_OS_WINDOWS) */
@@ -585 +589 @@
             if (so->so_state & SS_ISFCONNECTING)
             {
+                Log2(("connecting %R[natsock] engaged\n",so));
                 STAM_REL_COUNTER_INC(&pData->StatTCPHot);
                 TCP_ENGAGE_EVENT1(so, writefds);
@@ -808 +813 @@
                 if (so->so_state & SS_ISFCONNECTING)
                 {
+                    Log2(("connecting %R[natsock] catched\n", so));
                     /* Connected */
                     so->so_state &= ~SS_ISFCONNECTING;
@@ -960 +966 @@
 
 static
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
 void arp_input(PNATState pData, struct mbuf *m)
-{
-    struct ethhdr *eh = mtod(m, struct ethhdr *);
-    struct arphdr *ah = (struct arphdr *)&eh[1];
-    uint8_t arp_reply[sizeof(struct arphdr)];
+#else
+void arp_input(PNATState pData, const uint8_t *pkt, int pkt_len)
+#endif
+{
+    struct ethhdr *eh;
+    struct ethhdr *reh;
+    struct arphdr *ah;
     struct arphdr *rah;
     int ar_op;
     struct ex_list *ex_ptr;
-    uint32_t htip = ntohl(*(uint32_t*)ah->ar_tip);
-
-    rah = (struct arphdr *)arp_reply;
+    uint32_t htip;
+#ifndef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+    uint8_t arp_reply[sizeof(struct arphdr) + ETH_HLEN];
+    eh = (struct ethhdr *)pkt;
+#else
+    struct mbuf *mr; 
+    eh = mtod(m, struct ethhdr *);
+#endif
+    ah = (struct arphdr *)&eh[1];
+    htip = ntohl(*(uint32_t*)ah->ar_tip);
+
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+    mr = m_get(pData);
+    mr->m_data += if_maxlinkhdr;
+    mr->m_len = sizeof(struct arphdr);
+    rah = mtod(mr, struct arphdr *);
+#else
+    reh = (struct ethhdr *)arp_reply;
+    rah = (struct arphdr *)&reh[1];
+#endif
 
     ar_op = ntohs(ah->ar_op);
@@ -989 +1016 @@
         arp_ok:
 
+#ifndef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+                memcpy(reh->h_dest, eh->h_source, ETH_ALEN);
+                memcpy(reh->h_source, &special_addr, ETH_ALEN);
+                reh->h_source[5] = ah->ar_tip[3];
+                reh->h_proto = htons(ETH_P_ARP);
+#endif
                 rah->ar_hrd = htons(1);
                 rah->ar_pro = htons(ETH_P_IP);
@@ -995 +1028 @@
                 rah->ar_op = htons(ARPOP_REPLY);
                 memcpy(rah->ar_sha, special_ethaddr, ETH_ALEN);
+
                 switch (htip & ~pData->netmask) 
                 {
@@ -1008 +1042 @@
                 memcpy(rah->ar_tip, ah->ar_sip, 4);
 #ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
-                if_encap(pData, ETH_P_ARP, arp_reply, sizeof(struct arphdr));
+                if_encap(pData, ETH_P_ARP, mr);
                 m_free(pData, m);
 #else
@@ -1038 +1072 @@
     /* Note: we add to align the IP header */
 
-    if (M_FREEROOM(m) < pkt_len)
-    {
-       m_inc(m, pkt_len);
-    }
-    m->m_len = pkt_len;
-    memcpy(m->m_data, pkt, pkt_len);
+    if (M_FREEROOM(m) < pkt_len + 2)
+    {
+       m_inc(m, pkt_len + 2);
+    }
+    m->m_len = pkt_len + 2;
+    memcpy(m->m_data + 2, pkt, pkt_len);
 
     proto = ntohs(*(uint16_t *)(pkt + 12));
@@ -1049 +1083 @@
     {
         case ETH_P_ARP:
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
             arp_input(pData, m);
+#else
+            arp_input(pData, pkt, pkt_len);
+            m_free(pData, m);
+#endif
             break;
         case ETH_P_IP:
@@ -1055 +1094 @@
              * the first outgoing connection gets an incorrect timestamp. */
             updtime(pData);
-            m->m_data += ETH_HLEN;
-            m->m_len -= ETH_HLEN;
+            m->m_data += ETH_HLEN + 2;
+            m->m_len -= ETH_HLEN + 2;
             ip_input(pData, m);
             break;
@@ -1067 +1106 @@
 
 /* output the IP packet to the ethernet device */
-void if_encap(PNATState pData, uint16_t eth_proto, const uint8_t *ip_data, int ip_data_len)
-{
 #ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
-    uint8_t *buf = RTMemAlloc(1600); /* XXX:temporal solution */
-    struct ethhdr *eh = (struct ethhdr *)buf;
-
-    if (ip_data_len + ETH_HLEN > 1600)
-        return;
+void if_encap(PNATState pData, uint16_t eth_proto, struct mbuf *m)
+#else
+void if_encap(PNATState pData, uint8_t *ip_data, int ip_data_len)
+#endif
+{
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+    struct ethhdr *eh;
+    m->m_data -= if_maxlinkhdr;
+    m->m_len += ETH_HLEN;
+    eh = mtod(m, struct ethhdr *);
 #else
     uint8_t buf[1600]; 
@@ -1081 +1123 @@
     if (ip_data_len + ETH_HLEN > sizeof(buf))
         return;
+    
+    memcpy(buf + sizeof(struct ethhdr), ip_data, ip_data_len);
 #endif
 
@@ -1088 +1132 @@
     /* XXX: not correct */
     eh->h_source[5] = CTL_ALIAS;
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
     eh->h_proto = htons(eth_proto);
-    memcpy(buf + sizeof(struct ethhdr), ip_data, ip_data_len);
+    slirp_output(pData->pvUser, m, mtod(m, uint8_t *), m->m_len);
+#else
+    eh->h_proto = htons(ETH_P_IP);
     slirp_output(pData->pvUser, buf, ip_data_len + ETH_HLEN);
+#endif
 }
 
@@ -1145 +1193 @@
     return 0;
 }
+
+/*
+ * this function called from NAT thread
+ */
+void slirp_post_sent(PNATState pData, void *pvArg)
+{
+    struct socket *so = 0; 
+    struct tcpcb *tp = 0;
+    struct mbuf *m = (struct mbuf *)pvArg;
+    m_free(pData, m); 
+}

trunk/src/VBox/Devices/Network/slirp/socket.c

@@ -461 +461 @@
             return;
         m->m_data += if_maxlinkhdr;
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+        m->m_data += sizeof(struct udphdr) 
+                    + sizeof(struct ip); /*XXX: no options atm*/
+#endif
 
         /*

trunk/src/VBox/Devices/Network/slirp/tcp_input.c

@@ -323 +323 @@
     {
         tcpstat.tcps_rcvbadsum++;
+        Log2(("checksum is invalid => drop\n"));
         goto drop;
     }
@@ -335 +336 @@
     {
         tcpstat.tcps_rcvbadoff++;
+        Log2(("ti_off(tlen(%d)<%d<(tcphdr(%d))) is invalid =>drop\n", tlen, off, sizeof(struct tcphdr)));
         goto drop;
     }
@@ -447 +449 @@
      * or something else, we nuke it.
      */
-    if (so->so_state & SS_ISFCONNECTING)
+    if (so->so_state & SS_ISFCONNECTING) 
+    {
+        Log2(("so_state(%x) of %R[natsock] is still connecting =>drop\n", so->so_state, so));
         goto drop;
+    }
 
     tp = sototcpcb(so);
@@ -456 +461 @@
         goto dropwithreset;
     if (tp->t_state == TCPS_CLOSED)
+    {
+        Log2(("t_state(%x) is closed =>drop\n", tp->t_state));
         goto drop;
+    }
 
     /* Unscale the window into a 32-bit value. */
@@ -654 +662 @@
         case TCPS_LISTEN:
         {
-            if (tiflags & TH_RST)
+            if (tiflags & TH_RST) {
+                Log2(("RST(%x) is on listen =>drop\n", tiflags));
                 goto drop;
+            }
             if (tiflags & TH_ACK)
                 goto dropwithreset;
-            if ((tiflags & TH_SYN) == 0)
+            if ((tiflags & TH_SYN) == 0) 
+            {
+                Log2(("SYN(%x) is off on listen =>drop\n", tiflags));
                 goto drop;
+            }
 
             /*
@@ -768 +781 @@
                 if (tiflags & TH_ACK)
                     tp = tcp_drop(pData, tp,0); /* XXX Check t_softerror! */
+                Log2(("RST(%x) is on SYN_SENT =>drop\n", tiflags));
                 goto drop;
             }
 
-            if ((tiflags & TH_SYN) == 0)
+            if ((tiflags & TH_SYN) == 0) 
+            {
+                Log2(("SYN(%x) bit is off on SYN_SENT =>drop\n", tiflags));
                 goto drop;
+            }
             if (tiflags & TH_ACK)
             {
@@ -1020 +1037 @@
 /*              so->so_error = ECONNRESET; */
 close:
+                Log2(("closing...=>drop\n", tp->t_state)); 
                 tp->t_state = TCPS_CLOSED;
                 tcpstat.tcps_drops++;
@@ -1028 +1046 @@
             case TCPS_LAST_ACK:
             case TCPS_TIME_WAIT:
+                Log2(("t_state is (%x) sort of close =>drop\n", tp->t_state)); 
                 tp = tcp_close(pData, tp);
                 goto drop;
@@ -1045 +1064 @@
      * If the ACK bit is off we drop the segment and return.
      */
-    if ((tiflags & TH_ACK) == 0)
+    if ((tiflags & TH_ACK) == 0) 
+    {
+        Log2(("ACK(%x) bit is off =>drop\n", tiflags)); 
         goto drop;
+    }
 
     /*
@@ -1154 +1176 @@
                         if (SEQ_GT(onxt, tp->snd_nxt))
                             tp->snd_nxt = onxt;
+                        Log2(("t_dupacks(%d) == tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
                         goto drop;
                     }
@@ -1160 +1183 @@
                         tp->snd_cwnd += tp->t_maxseg;
                         (void) tcp_output(pData, tp);
+                        Log2(("t_dupacks(%d) > tcprexmtthresh(%d)=>drop\n", tp->t_dupacks, tcprexmtthresh));
                         goto drop;
                     }
@@ -1305 +1329 @@
                     if (ourfinisacked)
                     {
+                        Log2(("ourfinisacked=>drop\n"));
                         tp = tcp_close(pData, tp);
                         goto drop;
@@ -1528 +1553 @@
 
 dropafterack:
+    Log2(("drop after ack\n"));
     /*
      * Generate an ACK dropping incoming segment if it occupies
@@ -1541 +1567 @@
 dropwithreset:
     /* reuses m if m!=NULL, m_free() unnecessary */
+    Log2(("drop with reset\n"));
     if (tiflags & TH_ACK)
         tcp_respond(pData, tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
@@ -1556 +1583 @@
      * Drop space held by incoming segment and return.
      */
+    Log2(("drop\n"));
     m_free(pData, m);
 

trunk/src/VBox/Devices/Network/slirp/tcp_output.c

@@ -428 +428 @@
         }
         m->m_data += if_maxlinkhdr;
+#ifdef VBOX_WITH_SIMPLIFIED_SLIRP_SYNC
+        m->m_data += sizeof(struct ip) 
+            + sizeof(struct tcphdr);
+#endif
         m->m_len = hdrlen;
     }