Index: /trunk/src/VBox/Devices/Network/DrvNAT.cpp =================================================================== --- /trunk/src/VBox/Devices/Network/DrvNAT.cpp (revision 53398) +++ /trunk/src/VBox/Devices/Network/DrvNAT.cpp (revision 53399) @@ -833,7 +833,7 @@ DWORD dwEvent = WSAWaitForMultipleEvents(nFDs, phEvents, FALSE, slirp_get_timeout_ms(pThis->pNATState), - FALSE); + /* :fAlertable */ TRUE); if ( (dwEvent < WSA_WAIT_EVENT_0 || dwEvent > WSA_WAIT_EVENT_0 + nFDs - 1) - && dwEvent != WSA_WAIT_TIMEOUT) + && dwEvent != WSA_WAIT_TIMEOUT && dwEvent != WSA_WAIT_IO_COMPLETION) { int error = WSAGetLastError(); @@ -845,10 +845,10 @@ { /* only check for slow/fast timers */ - slirp_select_poll(pThis->pNATState, /* fTimeout=*/true, /*fIcmp=*/false); + slirp_select_poll(pThis->pNATState, /* fTimeout=*/true); continue; } /* poll the sockets in any case */ Log2(("%s: poll\n", __FUNCTION__)); - slirp_select_poll(pThis->pNATState, /* fTimeout=*/false, /* fIcmp=*/(dwEvent == WSA_WAIT_EVENT_0)); + slirp_select_poll(pThis->pNATState, /* fTimeout=*/false); /* process _all_ outstanding requests but don't wait */ RTReqQueueProcess(pThis->hSlirpReqQueue, 0); Index: /trunk/src/VBox/Devices/Network/slirp/ip_icmp.c =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/ip_icmp.c (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/ip_icmp.c (revision 53399) @@ -56,9 +56,4 @@ #include "slirp.h" #include "ip_icmp.h" -#ifdef RT_OS_WINDOWS -# include -# include -# include -#endif /* The message sent when emulating PING */ @@ -97,4 +92,6 @@ pData->icmp_socket.so_type = IPPROTO_ICMP; pData->icmp_socket.so_state = SS_ISFCONNECTED; + +#ifndef RT_OS_WINDOWS if (iIcmpCacheLimit < 0) { @@ -103,5 +100,4 @@ } pData->iIcmpCacheLimit = iIcmpCacheLimit; -#ifndef RT_OS_WINDOWS # ifndef RT_OS_DARWIN pData->icmp_socket.s = socket(PF_INET, SOCK_RAW, IPPROTO_ICMP); @@ -118,4 +114,6 @@ NSOCK_INC(); + LIST_INIT(&pData->icmp_msg_head); + #else /* RT_OS_WINDOWS */ if (icmpwin_init(pData) != 0) @@ -123,5 +121,4 @@ #endif /* RT_OS_WINDOWS */ - LIST_INIT(&pData->icmp_msg_head); return 0; } @@ -133,12 +130,13 @@ icmp_finit(PNATState pData) { - icmp_cache_clean(pData, -1); #ifdef RT_OS_WINDOWS icmpwin_finit(pData); #else + icmp_cache_clean(pData, -1); closesocket(pData->icmp_socket.s); #endif } +#if !defined(RT_OS_WINDOWS) /* * ip here is ip header + 64bytes readed from ICMP packet @@ -339,4 +337,5 @@ return 0; } +#endif /* !RT_OS_WINDOWS */ /* @@ -350,9 +349,5 @@ uint8_t icmp_type; void *icp_buf = NULL; - int status; uint32_t dst; -#if !defined(RT_OS_WINDOWS) - int ttl; -#endif /* int code; */ @@ -409,12 +404,16 @@ goto done; } - else + +#ifdef RT_OS_WINDOWS + { + icmpwin_ping(pData, m, hlen); + break; /* free mbuf */ + } +#else { struct icmp *icp; struct sockaddr_in addr; -#ifdef RT_OS_WINDOWS - IP_OPTION_INFORMATION ipopt; - int error; -#endif + + /* XXX: FIXME: this is bogus, see CTL_CHECKs above */ addr.sin_family = AF_INET; if ((ip->ip_dst.s_addr & RT_H2N_U32(pData->netmask)) == pData->special_addr.s_addr) @@ -447,9 +446,11 @@ icp = (struct icmp *)(mtod(m, char *) + hlen); -#ifndef RT_OS_WINDOWS if (pData->icmp_socket.s != -1) { + static bool fIcmpSocketErrorReported; + int ttl; + int status; ssize_t rc; - static bool fIcmpSocketErrorReported; + ttl = ip->ip_ttl; Log(("NAT/ICMP: try to set TTL(%d)\n", ttl)); @@ -478,52 +479,6 @@ icmp_error(pData, m, ICMP_UNREACH, ICMP_UNREACH_NET, 0, strerror(errno)); } -#else /* RT_OS_WINDOWS */ - pData->icmp_socket.so_laddr.s_addr = ip->ip_src.s_addr; /* XXX: hack*/ - pData->icmp_socket.so_icmp_id = icp->icmp_id; - pData->icmp_socket.so_icmp_seq = icp->icmp_seq; - memset(&ipopt, 0, sizeof(IP_OPTION_INFORMATION)); - ipopt.Ttl = ip->ip_ttl; - status = IcmpSendEcho2(pData->icmp_socket.sh /*=handle*/, - pData->phEvents[VBOX_ICMP_EVENT_INDEX] /*=Event*/, - NULL /*=ApcRoutine*/, - NULL /*=ApcContext*/, - addr.sin_addr.s_addr /*=DestinationAddress*/, - icp->icmp_data /*=RequestData*/, - icmplen - ICMP_MINLEN /*=RequestSize*/, - &ipopt /*=RequestOptions*/, - pData->pvIcmpBuffer /*=ReplyBuffer*/, - pData->cbIcmpBuffer /*=ReplySize*/, - 1 /*=Timeout in ms*/); - error = GetLastError(); - if ( status != 0 - || error == ERROR_IO_PENDING) - { - /* no error! */ - m->m_so = &pData->icmp_socket; - icmp_attach(pData, m); - /* don't let m_freem at the end free atached buffer */ - goto done; - } - Log(("NAT: Error (%d) occurred while sending ICMP (", error)); - switch (error) - { - case ERROR_INVALID_PARAMETER: - Log(("icmp_socket:%lx is invalid)\n", pData->icmp_socket.s)); - break; - case ERROR_NOT_SUPPORTED: - Log(("operation is unsupported)\n")); - break; - case ERROR_NOT_ENOUGH_MEMORY: - Log(("OOM!!!)\n")); - break; - case IP_BUF_TOO_SMALL: - Log(("Buffer too small)\n")); - break; - default: - Log(("Other error!!!)\n")); - break; - } -#endif /* RT_OS_WINDOWS */ - } /* if ip->ip_dst.s_addr == alias_addr.s_addr */ + } +#endif /* !RT_OS_WINDOWS */ break; case ICMP_UNREACH: Index: /trunk/src/VBox/Devices/Network/slirp/ip_icmp.h =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/ip_icmp.h (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/ip_icmp.h (revision 53399) @@ -208,4 +208,5 @@ void icmpwin_finit (PNATState); void icmpwin_ping(PNATState, struct mbuf *, int); +void icmpwin_process(PNATState); #endif Index: /trunk/src/VBox/Devices/Network/slirp/ip_icmpwin.c =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/ip_icmpwin.c (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/ip_icmpwin.c (revision 53399) @@ -41,4 +41,6 @@ PNATState pData; + TAILQ_ENTRY(pong) queue_entry; + struct ip reqiph; struct icmp_echo reqicmph; @@ -53,4 +55,5 @@ static void icmpwin_callback(struct pong *pong); +static void icmpwin_pong(struct pong *pong); static struct mbuf *icmpwin_get_error(struct pong *pong, int type, int code); @@ -91,8 +94,9 @@ } + TAILQ_INIT(&pData->pongs_expected); + TAILQ_INIT(&pData->pongs_received); + pData->icmp_socket.sh = IcmpCreateFile(); pData->phEvents[VBOX_ICMP_EVENT_INDEX] = CreateEvent(NULL, FALSE, FALSE, NULL); - pData->cbIcmpBuffer = sizeof(ICMP_ECHO_REPLY) * 10; - pData->pvIcmpBuffer = RTMemAlloc(pData->cbIcmpBuffer); return 0; @@ -104,5 +108,17 @@ { IcmpCloseHandle(pData->icmp_socket.sh); - RTMemFree(pData->pvIcmpBuffer); + + while (!TAILQ_EMPTY(&pData->pongs_received)) { + struct pong *pong = TAILQ_FIRST(&pData->pongs_received); + TAILQ_REMOVE(&pData->pongs_received, pong, queue_entry); + RTMemFree(pong); + } + + /* this should be empty */ + while (!TAILQ_EMPTY(&pData->pongs_expected)) { + struct pong *pong = TAILQ_FIRST(&pData->pongs_expected); + TAILQ_REMOVE(&pData->pongs_expected, pong, queue_entry); + pong->pData = NULL; + } } @@ -115,5 +131,5 @@ { struct ip *ip = mtod(m, struct ip *); - size_t reqsize; + size_t reqsize, pongsize; uint8_t ttl; size_t bufsize; @@ -137,5 +153,9 @@ bufsize += 16; /* whatever that is; empirically at least XP needs it */ - pong = RTMemAlloc(RT_OFFSETOF(struct pong, buf) + bufsize); + pongsize = RT_OFFSETOF(struct pong, buf) + bufsize; + if (pData->cbIcmpPending + pongsize > 1024 * 1024) + return; + + pong = RTMemAlloc(pongsize); if (RT_UNLIKELY(pong == NULL)) return; @@ -209,9 +229,12 @@ else /* success */ { - Log2(("NAT: pong %p for ping %RTnaipv4 id 0x%04x seq %d len %d\n", + Log2(("NAT: pong %p for ping %RTnaipv4 id 0x%04x seq %d len %zu (%zu)\n", pong, dst, RT_N2H_U16(pong->reqicmph.icmp_echo_id), RT_N2H_U16(pong->reqicmph.icmp_echo_seq), - reqsize)); + pongsize, reqsize)); + + pData->cbIcmpPending += pongsize; + TAILQ_INSERT_TAIL(&pData->pongs_expected, pong, queue_entry); pong = NULL; /* callback owns it now */ } @@ -226,10 +249,6 @@ { struct pong *pong = (struct pong *)ctx; - if (pong != NULL) - { icmpwin_callback(pong); - RTMemFree(pong); - } } @@ -239,10 +258,6 @@ { struct pong *pong = (struct pong *)ctx; - if (pong != NULL) - { icmpwin_callback(pong); - RTMemFree(pong); - } } @@ -251,7 +266,76 @@ * Actual callback code for IcmpSendEcho2(). OS version specific * trampoline will free "pong" argument for us. + * + * Since async callback can be called anytime the thread is alertable, + * it's not safe to do any processing here. Instead queue it and + * notify the main loop. */ static void icmpwin_callback(struct pong *pong) +{ + PNATState pData = pong->pData; + + if (pData == NULL) + { + RTMemFree(pong); + return; + } + +#ifdef DEBUG + { + struct pong *expected, *already; + + TAILQ_FOREACH(expected, &pData->pongs_expected, queue_entry) + { + if (expected == pong) + break; + } + Assert(expected); + + TAILQ_FOREACH(already, &pData->pongs_received, queue_entry) + { + if (already == pong) + break; + } + Assert(!already); + } +#endif + + TAILQ_REMOVE(&pData->pongs_expected, pong, queue_entry); + TAILQ_INSERT_TAIL(&pData->pongs_received, pong, queue_entry); + + WSASetEvent(pData->phEvents[VBOX_ICMP_EVENT_INDEX]); +} + + +void +icmpwin_process(PNATState pData) +{ + struct pong_tailq pongs; + + if (TAILQ_EMPTY(&pData->pongs_received)) + return; + + TAILQ_INIT(&pongs); + TAILQ_CONCAT(&pongs, &pData->pongs_received, queue_entry); + + while (!TAILQ_EMPTY(&pongs)) { + struct pong *pong = TAILQ_FIRST(&pongs); + size_t sz; + + sz = RT_OFFSETOF(struct pong, buf) + pong->bufsize; + Assert(pData->cbIcmpPending >= sz); + pData->cbIcmpPending -= sz; + + icmpwin_pong(pong); + + TAILQ_REMOVE(&pongs, pong, queue_entry); + RTMemFree(pong); + } +} + + +void +icmpwin_pong(struct pong *pong) { PNATState pData; Index: /trunk/src/VBox/Devices/Network/slirp/libslirp.h =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/libslirp.h (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/libslirp.h (revision 53399) @@ -56,5 +56,5 @@ void slirp_select_fill(PNATState pData, int *pndfs); -void slirp_select_poll(PNATState pData, int fTimeout, int fIcmp); +void slirp_select_poll(PNATState pData, int fTimeout); #else /* RT_OS_WINDOWS */ void slirp_select_fill(PNATState pData, int *pnfds, struct pollfd *polls); Index: /trunk/src/VBox/Devices/Network/slirp/slirp.c =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/slirp.c (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/slirp.c (revision 53399) @@ -892,5 +892,5 @@ #if defined(RT_OS_WINDOWS) -void slirp_select_poll(PNATState pData, int fTimeout, int fIcmp) +void slirp_select_poll(PNATState pData, int fTimeout) #else /* RT_OS_WINDOWS */ void slirp_select_poll(PNATState pData, struct pollfd *polls, int ndfs) @@ -942,9 +942,5 @@ goto done; #if defined(RT_OS_WINDOWS) - /*XXX: before renaming please make see define - * fIcmp in slirp_state.h - */ - if (fIcmp) - sorecvfrom(pData, &pData->icmp_socket); + icmpwin_process(pData); #else if ( (pData->icmp_socket.s != -1) Index: /trunk/src/VBox/Devices/Network/slirp/slirp_state.h =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/slirp_state.h (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/slirp_state.h (revision 53399) @@ -96,4 +96,8 @@ +#ifdef RT_OS_WINDOWS +struct pong; +TAILQ_HEAD(pong_tailq, pong); +#endif /* forward declaration */ @@ -199,11 +203,14 @@ struct socket icmp_socket; +# if !defined(RT_OS_WINDOWS) struct icmp_storage icmp_msg_head; int cIcmpCacheSize; int iIcmpCacheLimit; -# ifdef RT_OS_WINDOWS - void *pvIcmpBuffer; - uint32_t cbIcmpBuffer; +# else + struct pong_tailq pongs_expected; + struct pong_tailq pongs_received; + size_t cbIcmpPending; # endif + #if defined(RT_OS_WINDOWS) # define VBOX_SOCKET_EVENT (pData->phEvents[VBOX_SOCKET_EVENT_INDEX]) Index: /trunk/src/VBox/Devices/Network/slirp/socket.c =================================================================== --- /trunk/src/VBox/Devices/Network/slirp/socket.c (revision 53398) +++ /trunk/src/VBox/Devices/Network/slirp/socket.c (revision 53399) @@ -128,8 +128,7 @@ } #endif /* !VBOX_WITH_NAT_SEND2HOME */ + +#if !defined(RT_OS_WINDOWS) static void send_icmp_to_guest(PNATState, char *, size_t, const struct sockaddr_in *); -#ifdef RT_OS_WINDOWS -static void sorecvfrom_icmp_win(PNATState, struct socket *); -#else /* RT_OS_WINDOWS */ static void sorecvfrom_icmp_unix(PNATState, struct socket *); #endif /* !RT_OS_WINDOWS */ @@ -811,16 +810,17 @@ { LogFlowFunc(("sorecvfrom: so = %lx\n", (long)so)); - + +#ifdef RT_OS_WINDOWS + /* ping is handled with ICMP API in ip_icmpwin.c */ + Assert(so->so_type == IPPROTO_UDP); +#else if (so->so_type == IPPROTO_ICMP) { /* This is a "ping" reply */ -#ifdef RT_OS_WINDOWS - sorecvfrom_icmp_win(pData, so); -#else /* RT_OS_WINDOWS */ sorecvfrom_icmp_unix(pData, so); + udp_detach(pData, so); + } + else #endif /* !RT_OS_WINDOWS */ - udp_detach(pData, so); - } - else { static uint8_t au8Buf[64 * 1024]; @@ -1272,4 +1272,5 @@ } +#if !defined(RT_OS_WINDOWS) static void send_icmp_to_guest(PNATState pData, char *buff, size_t len, const struct sockaddr_in *addr) @@ -1442,153 +1443,4 @@ } -#ifdef RT_OS_WINDOWS -static void -sorecvfrom_icmp_win(PNATState pData, struct socket *so) -{ - int len; - int i; - struct ip *ip; - struct mbuf *m; - struct icmp *icp; - struct icmp_msg *icm; - struct ip *ip_broken; /* ICMP returns header + 64 bit of packet */ - uint32_t src; - ICMP_ECHO_REPLY *icr; - int hlen = 0; - int nbytes = 0; - u_char code = ~0; - int out_len; - int size; - - len = IcmpParseReplies(pData->pvIcmpBuffer, pData->cbIcmpBuffer); - if (len < 0) - { - LogRel(("NAT: Error (%d) occurred on ICMP receiving\n", GetLastError())); - return; - } - if (len == 0) - return; /* no error */ - - icr = (ICMP_ECHO_REPLY *)pData->pvIcmpBuffer; - for (i = 0; i < len; ++i) - { - LogFunc(("icr[%d] Data:%p, DataSize:%d\n", - i, icr[i].Data, icr[i].DataSize)); - switch(icr[i].Status) - { - case IP_DEST_HOST_UNREACHABLE: - code = (code != ~0 ? code : ICMP_UNREACH_HOST); - case IP_DEST_NET_UNREACHABLE: - code = (code != ~0 ? code : ICMP_UNREACH_NET); - case IP_DEST_PROT_UNREACHABLE: - code = (code != ~0 ? code : ICMP_UNREACH_PROTOCOL); - /* UNREACH error inject here */ - case IP_DEST_PORT_UNREACHABLE: - code = (code != ~0 ? code : ICMP_UNREACH_PORT); - icmp_error(pData, so->so_m, ICMP_UNREACH, code, 0, "Error occurred!!!"); - so->so_m = NULL; - break; - case IP_SUCCESS: /* echo replied */ - out_len = ETH_HLEN + sizeof(struct ip) + 8; - size; - size = MCLBYTES; - if (out_len < MSIZE) - size = MCLBYTES; - else if (out_len < MCLBYTES) - size = MCLBYTES; - else if (out_len < MJUM9BYTES) - size = MJUM9BYTES; - else if (out_len < MJUM16BYTES) - size = MJUM16BYTES; - else - AssertMsgFailed(("Unsupported size")); - - m = m_getjcl(pData, M_NOWAIT, MT_HEADER, M_PKTHDR, size); - LogFunc(("m_getjcl returns m: %p\n", m)); - if (m == NULL) - return; - m->m_len = 0; - m->m_data += if_maxlinkhdr; - m->m_pkthdr.header = mtod(m, void *); - - ip = mtod(m, struct ip *); - ip->ip_src.s_addr = icr[i].Address; - ip->ip_p = IPPROTO_ICMP; - ip->ip_dst.s_addr = so->so_laddr.s_addr; /*XXX: still the hack*/ - ip->ip_hl = sizeof(struct ip) >> 2; /* requiered for icmp_reflect, no IP options */ - ip->ip_ttl = icr[i].Options.Ttl; - - icp = (struct icmp *)&ip[1]; /* no options */ - icp->icmp_type = ICMP_ECHOREPLY; - icp->icmp_code = 0; - icp->icmp_id = so->so_icmp_id; - icp->icmp_seq = so->so_icmp_seq; - - icm = icmp_find_original_mbuf(pData, ip); - if (icm) - { - /* on this branch we don't need stored variant */ - m_freem(pData, icm->im_m); - LIST_REMOVE(icm, im_list); - pData->cIcmpCacheSize--; - RTMemFree(icm); - } - - - hlen = (ip->ip_hl << 2); - Assert((hlen >= sizeof(struct ip))); - - m->m_data += hlen + ICMP_MINLEN; - if (!RT_VALID_PTR(icr[i].Data)) - { - m_freem(pData, m); - break; - } - m_copyback(pData, m, 0, icr[i].DataSize, icr[i].Data); - m->m_data -= hlen + ICMP_MINLEN; - m->m_len += hlen + ICMP_MINLEN; - - - ip->ip_len = m_length(m, NULL); - Assert((ip->ip_len == hlen + ICMP_MINLEN + icr[i].DataSize)); - - icmp_reflect(pData, m); - break; - case IP_TTL_EXPIRED_TRANSIT: /* TTL expired */ - - ip_broken = icr[i].Data; - icm = icmp_find_original_mbuf(pData, ip_broken); - if (icm == NULL) { - Log(("ICMP: can't find original package (first double word %x)\n", *(uint32_t *)ip_broken)); - return; - } - m = icm->im_m; - ip = mtod(m, struct ip *); - Assert(((ip_broken->ip_hl >> 2) >= sizeof(struct ip))); - ip->ip_ttl = icr[i].Options.Ttl; - src = ip->ip_src.s_addr; - ip->ip_dst.s_addr = src; - ip->ip_dst.s_addr = icr[i].Address; - - hlen = (ip->ip_hl << 2); - icp = (struct icmp *)((char *)ip + hlen); - ip_broken->ip_src.s_addr = src; /*it packet sent from host not from guest*/ - - m->m_len = (ip_broken->ip_hl << 2) + 64; - m->m_pkthdr.header = mtod(m, void *); - m_copyback(pData, m, ip->ip_hl >> 2, icr[i].DataSize, icr[i].Data); - icmp_reflect(pData, m); - /* Here is different situation from Unix world, where we can receive icmp in response on TCP/UDP */ - LIST_REMOVE(icm, im_list); - pData->cIcmpCacheSize--; - RTMemFree(icm); - break; - default: - Log(("ICMP(default): message with Status: %x was received from %x\n", icr[i].Status, icr[i].Address)); - break; - } - } -} -#else /* !RT_OS_WINDOWS */ static void sorecvfrom_icmp_unix(PNATState pData, struct socket *so) {