| /* -*- Mode: C; tab-width: 4; c-file-style: "bsd"; c-basic-offset: 4; fill-column: 108; indent-tabs-mode: nil; -*- |
| * |
| * Copyright (c) 2002-2022 Apple Inc. All rights reserved. |
| * |
| * Licensed under the Apache License, Version 2.0 (the "License"); |
| * you may not use this file except in compliance with the License. |
| * You may obtain a copy of the License at |
| * |
| * https://www.apache.org/licenses/LICENSE-2.0 |
| * |
| * Unless required by applicable law or agreed to in writing, software |
| * distributed under the License is distributed on an "AS IS" BASIS, |
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| * See the License for the specific language governing permissions and |
| * limitations under the License. |
| * |
| */ |
| |
| #include "mDNSEmbeddedAPI.h" // Defines the interface provided to the client layer above |
| #include "DNSCommon.h" |
| #include "mDNSPosix.h" // Defines the specific types needed to run mDNS on this platform |
| #include "PlatformCommon.h" |
| #include "dns_sd.h" |
| |
| #include <assert.h> |
| #include <stdio.h> |
| #include <stdlib.h> |
| #include <errno.h> |
| #include <string.h> |
| #include <unistd.h> |
| #include <syslog.h> |
| #include <stdarg.h> |
| #include <fcntl.h> |
| #include <sys/types.h> |
| #include <sys/time.h> |
| #include <sys/socket.h> |
| #include <sys/uio.h> |
| #include <sys/select.h> |
| #include <netinet/in.h> |
| #include <arpa/inet.h> |
| #include <time.h> // platform support for UTC time |
| #include <ifaddrs.h> |
| |
| #if USES_NETLINK |
| #include <asm/types.h> |
| #include <linux/netlink.h> |
| #include <linux/rtnetlink.h> |
| #else // USES_NETLINK |
| #include <net/route.h> |
| #include <net/if.h> |
| #endif // USES_NETLINK |
| #if defined(TARGET_OS_MAC) && TARGET_OS_MAC |
| #include <netinet/in_var.h> |
| #include <net/if_dl.h> |
| #endif |
| #if defined(TARGET_OS_LINUX) && TARGET_OS_LINUX |
| #include <net/if_arp.h> |
| #include <sys/ioctl.h> |
| #include <linux/sockios.h> |
| #endif |
| |
| #include "mDNSUNP.h" |
| #include "GenLinkedList.h" |
| #include "mdns_strict.h" |
| |
| // *************************************************************************** |
| // Structures |
| |
| // Context record for interface change callback |
| struct IfChangeRec |
| { |
| int NotifySD; |
| mDNS *mDNS; |
| }; |
| typedef struct IfChangeRec IfChangeRec; |
| |
| // Note that static data is initialized to zero in (modern) C. |
| static PosixEventSource *gEventSources; // linked list of PosixEventSource's |
| static sigset_t gEventSignalSet; // Signals which event loop listens for |
| static sigset_t gEventSignals; // Signals which were received while inside loop |
| |
| static PosixNetworkInterface *gRecentInterfaces; |
| |
| // *************************************************************************** |
| // Globals (for debugging) |
| |
| static int num_registered_interfaces = 0; |
| static int num_pkts_accepted = 0; |
| static int num_pkts_rejected = 0; |
| |
| // *************************************************************************** |
| // Locals |
| mDNSlocal void requestReadEvents(PosixEventSource *eventSource, |
| const char *taskName, mDNSPosixEventCallback callback, void *context); |
| mDNSlocal mStatus stopReadOrWriteEvents(int fd, mDNSBool freeSource, mDNSBool removeSource, int flags); |
| mDNSlocal void requestWriteEvents(PosixEventSource *eventSource, |
| const char *taskName, mDNSPosixEventCallback callback, void *context); |
| mDNSlocal void UDPReadCallback(int fd, void *context); |
| mDNSlocal int SetupIPv4Socket(int fd); |
| mDNSlocal int SetupIPv6Socket(int fd); |
| |
| // *************************************************************************** |
| // Constants |
| |
| static const int kOn = 1; |
| static const int kIntTwoFiveFive = 255; |
| static const unsigned char kByteTwoFiveFive = 255; |
| |
| // *************************************************************************** |
| // Functions |
| |
| #if MDNS_MALLOC_DEBUGGING |
| mDNSexport void mDNSPlatformValidateLists(void) |
| { |
| // This should validate gEventSources and any other Posix-specific stuff that gets allocated. |
| } |
| #endif |
| |
| int gMDNSPlatformPosixVerboseLevel = 0; |
| |
| #define PosixErrorToStatus(errNum) ((errNum) == 0 ? mStatus_NoError : mStatus_UnknownErr) |
| |
| mDNSlocal void SockAddrTomDNSAddr(const struct sockaddr *const sa, mDNSAddr *ipAddr, mDNSIPPort *ipPort) |
| { |
| switch (sa->sa_family) |
| { |
| case AF_INET: |
| { |
| struct sockaddr_in *sin = (struct sockaddr_in*)sa; |
| ipAddr->type = mDNSAddrType_IPv4; |
| ipAddr->ip.v4.NotAnInteger = sin->sin_addr.s_addr; |
| if (ipPort) ipPort->NotAnInteger = sin->sin_port; |
| break; |
| } |
| |
| #if HAVE_IPV6 |
| case AF_INET6: |
| { |
| struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)sa; |
| #ifndef NOT_HAVE_SA_LEN |
| assert(sin6->sin6_len == sizeof(*sin6)); |
| #endif |
| ipAddr->type = mDNSAddrType_IPv6; |
| ipAddr->ip.v6 = *(mDNSv6Addr*)&sin6->sin6_addr; |
| if (ipPort) ipPort->NotAnInteger = sin6->sin6_port; |
| break; |
| } |
| #endif |
| |
| default: |
| verbosedebugf("SockAddrTomDNSAddr: Uknown address family %d\n", sa->sa_family); |
| ipAddr->type = mDNSAddrType_None; |
| if (ipPort) ipPort->NotAnInteger = 0; |
| break; |
| } |
| } |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark ***** Send and Receive |
| #endif |
| |
| // mDNS core calls this routine when it needs to send a packet. |
| mDNSexport mStatus mDNSPlatformSendUDP(const mDNS *const m, const void *const msg, const mDNSu8 *const end, |
| mDNSInterfaceID InterfaceID, UDPSocket *src, const mDNSAddr *dst, |
| mDNSIPPort dstPort, mDNSBool useBackgroundTrafficClass) |
| { |
| int err = 0; |
| struct sockaddr_storage to; |
| PosixNetworkInterface * thisIntf = (PosixNetworkInterface *)(InterfaceID); |
| int sendingsocket = -1; |
| struct sockaddr *sa = (struct sockaddr *)&to; |
| |
| (void) useBackgroundTrafficClass; |
| |
| assert(m != NULL); |
| assert(msg != NULL); |
| assert(end != NULL); |
| assert((((char *) end) - ((char *) msg)) > 0); |
| |
| if (dstPort.NotAnInteger == 0) |
| { |
| LogMsg("mDNSPlatformSendUDP: Invalid argument -dstPort is set to 0"); |
| return PosixErrorToStatus(EINVAL); |
| } |
| if (dst->type == mDNSAddrType_IPv4) |
| { |
| struct sockaddr_in *sin = (struct sockaddr_in*)&to; |
| #ifndef NOT_HAVE_SA_LEN |
| sin->sin_len = sizeof(*sin); |
| #endif |
| sin->sin_family = AF_INET; |
| sin->sin_port = dstPort.NotAnInteger; |
| sin->sin_addr.s_addr = dst->ip.v4.NotAnInteger; |
| sendingsocket = thisIntf ? thisIntf->multicastSocket4 : m->p->unicastSocket4; |
| } |
| |
| #if HAVE_IPV6 |
| else if (dst->type == mDNSAddrType_IPv6) |
| { |
| struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)&to; |
| mDNSPlatformMemZero(sin6, sizeof(*sin6)); |
| #ifndef NOT_HAVE_SA_LEN |
| sin6->sin6_len = sizeof(*sin6); |
| #endif |
| sin6->sin6_family = AF_INET6; |
| sin6->sin6_port = dstPort.NotAnInteger; |
| sin6->sin6_addr = *(struct in6_addr*)&dst->ip.v6; |
| sendingsocket = thisIntf ? thisIntf->multicastSocket6 : m->p->unicastSocket6; |
| } |
| #endif |
| // In case we get some other address family, return an error, since it's not supported. |
| else |
| { |
| return kDNSServiceErr_BadParam; |
| } |
| |
| // We don't open the socket until we get a send, because we don't know whether it's IPv4 or IPv6. |
| if (src) |
| { |
| if (src->events.fd == -1) |
| { |
| int sock = socket(sa->sa_family, SOCK_DGRAM, IPPROTO_UDP); |
| struct sockaddr_storage from; |
| socklen_t fromlen; |
| int times = 0; |
| uint16_t *pport; |
| |
| if (sock < 0) |
| { |
| LogMsg("Can't create UDP socket: %s", strerror(errno)); |
| return PosixErrorToStatus(errno); |
| } |
| |
| // Randomize the port. |
| if (src->randomizePort) |
| { |
| memset(&from, 0, sizeof from); |
| if (sa->sa_family == AF_INET) |
| { |
| ((struct sockaddr_in *)&from)->sin_family = AF_INET; |
| fromlen = sizeof (struct sockaddr_in); |
| pport = &((struct sockaddr_in *)&from)->sin_port; |
| err = SetupIPv4Socket(sock); |
| if (err) { return err; } |
| } |
| else |
| { |
| ((struct sockaddr_in6 *)&from)->sin6_family = AF_INET6; |
| fromlen = sizeof (struct sockaddr_in6); |
| pport = &((struct sockaddr_in6 *)&from)->sin6_port; |
| err = SetupIPv6Socket(sock); |
| if (err) { return err; } |
| } |
| #ifndef NOT_HAVE_SA_LEN |
| ((struct sockaddr *)&from)->sa_len = fromlen; |
| #endif |
| |
| while (times++ < 1000) |
| { |
| *pport = 0xC000 + mDNSRandom(0x3FFF); |
| if (bind(sock, (struct sockaddr *)&from, fromlen) >= 0) |
| { |
| src->port.NotAnInteger = *pport; |
| src->events.fd = sock; |
| break; |
| } |
| if (errno != EADDRINUSE) |
| { |
| LogMsg("Can't get randomized port: %s", strerror(errno)); |
| return PosixErrorToStatus(errno); |
| } |
| } |
| if (src->events.fd == -1) |
| { |
| LogMsg("Unable to get random port: too many tries."); |
| return PosixErrorToStatus(EADDRINUSE); |
| } |
| requestReadEvents(&src->events, "mDNSPosix::UDPReadCallback", UDPReadCallback, src); |
| } |
| } |
| sendingsocket = src->events.fd; |
| } |
| |
| if (sendingsocket >= 0) |
| err = sendto(sendingsocket, msg, (char*)end - (char*)msg, 0, (struct sockaddr *)&to, GET_SA_LEN(to)); |
| |
| if (err > 0) err = 0; |
| else if (err < 0) |
| { |
| static int MessageCount = 0; |
| // Don't report EHOSTDOWN (i.e. ARP failure), ENETDOWN, or no route to host for unicast destinations |
| if (!mDNSAddressIsAllDNSLinkGroup(dst)) |
| if (errno == EHOSTDOWN || errno == ENETDOWN || errno == EHOSTUNREACH || errno == ENETUNREACH) return(mStatus_TransientErr); |
| |
| if (MessageCount < 1000) |
| { |
| MessageCount++; |
| if (thisIntf) |
| LogMsg("mDNSPlatformSendUDP got error %d (%s) sending packet to %#a on interface %#a/%s/%d", |
| errno, strerror(errno), dst, &thisIntf->coreIntf.ip, thisIntf->intfName, thisIntf->index); |
| else |
| LogMsg("mDNSPlatformSendUDP got error %d (%s) sending packet to %#a", errno, strerror(errno), dst); |
| } |
| } |
| |
| return PosixErrorToStatus(err); |
| } |
| |
| mDNSlocal void TCPReadCallback(int fd, void *context) |
| { |
| TCPSocket *sock = context; |
| (void)fd; |
| |
| // TLS reading is handled in mDNSPlatformTCPRead(). |
| sock->callback(sock, sock->context, mDNSfalse, sock->err); |
| } |
| |
| mDNSlocal void tcpConnectCallback(int fd, void *context) |
| { |
| TCPSocket *sock = context; |
| mDNSBool c = !sock->connected; |
| int result; |
| socklen_t len = sizeof result; |
| |
| sock->connected = mDNStrue; |
| |
| if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &result, &len) < 0) |
| { |
| LogInfo("ERROR: TCPConnectCallback - unable to get connect error: socket %d: Error %d (%s)", |
| sock->events.fd, result, strerror(result)); |
| sock->err = mStatus_ConnFailed; |
| } |
| else |
| { |
| if (result != 0) |
| { |
| sock->err = mStatus_ConnFailed; |
| if (result == EHOSTUNREACH || result == EADDRNOTAVAIL || result == ENETDOWN) |
| { |
| LogInfo("ERROR: TCPConnectCallback - connect failed: socket %d: Error %d (%s)", |
| sock->events.fd, result, strerror(result)); |
| } |
| else |
| { |
| LogMsg("ERROR: TCPConnectCallback - connect failed: socket %d: Error %d (%s)", |
| sock->events.fd, result, strerror(result)); |
| } |
| } |
| else |
| { |
| if (sock->flags & kTCPSocketFlags_UseTLS) { |
| #ifdef POSIX_HAS_TLS |
| sock->tls = mDNSPosixTLSClientStateCreate(sock); |
| if (sock->tls == mDNSNULL) { |
| LogMsg("ERROR: TCPConnectCallback: TLS context state create failed"); |
| sock->err = mStatus_NoMemoryErr; |
| } else { |
| if (!mDNSPosixTLSStart(sock)) { |
| LogMsg("ERROR: TCPConnectCallback: TLS start failed"); |
| sock->err = mStatus_ConnFailed; |
| } |
| } |
| #else |
| // We shouldn't ever get here, because we should have already gotten an error when we created the |
| // socket. |
| LogMsg("Error: TCPSocketConnectCallback reached on TLS socket with no TLS support."); |
| sock->err = mStatus_ConnFailed; |
| #endif |
| } |
| if (sock->err == 0) { |
| // The connection succeeded. |
| sock->connected = mDNStrue; |
| // Select for read events. |
| sock->events.fd = fd; |
| requestReadEvents(&sock->events, "mDNSPosix::tcpConnectCallback", TCPReadCallback, sock); |
| } |
| } |
| } |
| |
| if (sock->callback) |
| { |
| sock->callback(sock, sock->context, c, sock->err); |
| // Here sock must be assumed to be invalid, in case the callback freed it. |
| return; |
| } |
| } |
| |
| // This routine is called when the main loop detects that data is available on a socket. |
| mDNSlocal void SocketDataReady(mDNS *const m, PosixNetworkInterface *intf, int skt, UDPSocket *sock) |
| { |
| mDNSAddr senderAddr, destAddr; |
| mDNSIPPort senderPort, destPort; |
| ssize_t packetLen; |
| DNSMessage packet; |
| struct my_in_pktinfo packetInfo; |
| struct sockaddr_storage from; |
| socklen_t fromLen; |
| int flags; |
| mDNSu8 ttl; |
| mDNSBool reject; |
| const mDNSInterfaceID InterfaceID = intf ? intf->coreIntf.InterfaceID : NULL; |
| |
| assert(m != NULL); |
| assert(skt >= 0); |
| |
| fromLen = sizeof(from); |
| flags = 0; |
| packetLen = recvfrom_flags(skt, &packet, sizeof(packet), &flags, (struct sockaddr *) &from, &fromLen, &packetInfo, &ttl); |
| |
| if (packetLen >= 0) |
| { |
| SockAddrTomDNSAddr((struct sockaddr*)&from, &senderAddr, &senderPort); |
| SockAddrTomDNSAddr((struct sockaddr*)&packetInfo.ipi_addr, &destAddr, &destPort); |
| |
| // If we have broken IP_RECVDSTADDR functionality (so far |
| // I've only seen this on OpenBSD) then apply a hack to |
| // convince mDNS Core that this isn't a spoof packet. |
| // Basically what we do is check to see whether the |
| // packet arrived as a multicast and, if so, set its |
| // destAddr to the mDNS address. |
| // |
| // I must admit that I could just be doing something |
| // wrong on OpenBSD and hence triggering this problem |
| // but I'm at a loss as to how. |
| // |
| // If this platform doesn't have IP_PKTINFO or IP_RECVDSTADDR, then we have |
| // no way to tell the destination address or interface this packet arrived on, |
| // so all we can do is just assume it's a multicast |
| |
| #if HAVE_BROKEN_RECVDSTADDR || (!defined(IP_PKTINFO) && !defined(IP_RECVDSTADDR)) |
| if ((destAddr.NotAnInteger == 0) && (flags & MSG_MCAST)) |
| { |
| destAddr.type = senderAddr.type; |
| if (senderAddr.type == mDNSAddrType_IPv4) destAddr.ip.v4 = AllDNSLinkGroup_v4.ip.v4; |
| else if (senderAddr.type == mDNSAddrType_IPv6) destAddr.ip.v6 = AllDNSLinkGroup_v6.ip.v6; |
| } |
| #endif |
| |
| // We only accept the packet if the interface on which it came |
| // in matches the interface associated with this socket. |
| // We do this match by name or by index, depending on which |
| // information is available. recvfrom_flags sets the name |
| // to "" if the name isn't available, or the index to -1 |
| // if the index is available. This accomodates the various |
| // different capabilities of our target platforms. |
| |
| reject = mDNSfalse; |
| if (!intf) |
| { |
| // Ignore multicasts accidentally delivered to our unicast receiving socket |
| if (mDNSAddrIsDNSMulticast(&destAddr)) packetLen = -1; |
| } |
| else |
| { |
| if (packetInfo.ipi_ifname[0] != 0) reject = (strcmp(packetInfo.ipi_ifname, intf->intfName) != 0); |
| else if (packetInfo.ipi_ifindex != -1) reject = (packetInfo.ipi_ifindex != intf->index); |
| |
| if (reject) |
| { |
| verbosedebugf("SocketDataReady ignored a packet from %#a to %#a on interface %s/%d expecting %#a/%s/%d/%d", |
| &senderAddr, &destAddr, packetInfo.ipi_ifname, packetInfo.ipi_ifindex, |
| &intf->coreIntf.ip, intf->intfName, intf->index, skt); |
| packetLen = -1; |
| num_pkts_rejected++; |
| if (num_pkts_rejected > (num_pkts_accepted + 1) * (num_registered_interfaces + 1) * 2) |
| { |
| fprintf(stderr, |
| "*** WARNING: Received %d packets; Accepted %d packets; Rejected %d packets because of interface mismatch\n", |
| num_pkts_accepted + num_pkts_rejected, num_pkts_accepted, num_pkts_rejected); |
| num_pkts_accepted = 0; |
| num_pkts_rejected = 0; |
| } |
| } |
| else |
| { |
| verbosedebugf("SocketDataReady got a packet from %#a to %#a on interface %#a/%s/%d/%d", |
| &senderAddr, &destAddr, &intf->coreIntf.ip, intf->intfName, intf->index, skt); |
| num_pkts_accepted++; |
| } |
| } |
| } |
| |
| if (packetLen >= 0) |
| mDNSCoreReceive(m, &packet, (mDNSu8 *)&packet + packetLen, |
| &senderAddr, senderPort, &destAddr, sock == mDNSNULL ? MulticastDNSPort : sock->port, InterfaceID); |
| } |
| |
| mDNSlocal void UDPReadCallback(int fd, void *context) |
| { |
| extern mDNS mDNSStorage; |
| SocketDataReady(&mDNSStorage, NULL, fd, (UDPSocket *)context); |
| } |
| |
| mDNSexport TCPSocket *mDNSPlatformTCPSocket(TCPSocketFlags flags, mDNSAddr_Type addrType, mDNSIPPort * port, |
| domainname *hostname, mDNSBool useBackgroundTrafficClass) |
| { |
| TCPSocket *sock; |
| int len = sizeof (TCPSocket); |
| |
| (void)useBackgroundTrafficClass; |
| |
| if (hostname) |
| { |
| len += sizeof (domainname); |
| } |
| sock = mdns_malloc(len); |
| |
| if (sock == NULL) |
| { |
| LogMsg("mDNSPlatformTCPSocket: no memory for socket"); |
| return NULL; |
| } |
| memset(sock, 0, sizeof *sock); |
| |
| if (hostname) |
| { |
| sock->hostname = (domainname *)(sock + 1); |
| LogMsg("mDNSPlatformTCPSocket: hostname %##s", hostname->c); |
| AssignDomainName(sock->hostname, hostname); |
| } |
| |
| sock->events.fd = -1; |
| if (!mDNSPosixTCPSocketSetup(&sock->events.fd, addrType, port, &sock->port)) |
| { |
| if (sock->events.fd != -1) close(sock->events.fd); |
| mdns_free(sock); |
| return mDNSNULL; |
| } |
| |
| // Set up the other fields in the structure. |
| sock->flags = flags; |
| sock->err = mStatus_NoError; |
| sock->setup = mDNSfalse; |
| sock->connected = mDNSfalse; |
| return sock; |
| } |
| |
| mDNSexport mStatus mDNSPlatformTCPSocketSetCallback(TCPSocket *sock, TCPConnectionCallback callback, void *context) |
| { |
| sock->callback = callback; |
| sock->context = context; |
| return mStatus_NoError; |
| } |
| |
| mDNSexport TCPSocket *mDNSPlatformTCPAccept(TCPSocketFlags flags, int fd) |
| { |
| TCPSocket *sock; |
| |
| // In order to receive a TLS connection, use mDNSPlatformTCPListen(). |
| if (flags & kTCPSocketFlags_UseTLS) |
| { |
| return mDNSNULL; |
| } |
| |
| sock = mDNSPlatformMemAllocateClear(sizeof(*sock)); |
| if (!sock) |
| { |
| return mDNSNULL; |
| } |
| |
| sock->events.fd = fd; |
| sock->flags = flags; |
| sock->connected = mDNStrue; |
| |
| return sock; |
| } |
| |
| |
| mDNSlocal void tcpListenCallback(int fd, void *context) |
| { |
| TCPListener *listener = context; |
| TCPSocket *sock; |
| |
| sock = mDNSPosixDoTCPListenCallback(fd, listener->addressType, listener->socketFlags, |
| listener->callback, listener->context); |
| if (sock != NULL) |
| { |
| requestReadEvents(&sock->events, "mDNSPosix::tcpListenCallback", TCPReadCallback, sock); |
| } |
| } |
| |
| mDNSexport TCPListener *mDNSPlatformTCPListen(mDNSAddr_Type addrType, mDNSIPPort *port, mDNSAddr *addr, |
| TCPSocketFlags socketFlags, mDNSBool reuseAddr, int queueLength, |
| TCPAcceptedCallback callback, void *context) |
| { |
| TCPListener *ret; |
| int fd = -1; |
| |
| if (!mDNSPosixTCPListen(&fd, addrType, port, addr, reuseAddr, queueLength)) |
| { |
| if (fd != -1) |
| { |
| close(fd); |
| } |
| return mDNSNULL; |
| } |
| |
| // Allocate a listener structure |
| ret = (TCPListener *)mDNSPlatformMemAllocateClear(sizeof *ret); |
| if (ret == NULL) |
| { |
| LogMsg("mDNSPlatformTCPListen: no memory for TCPListener struct."); |
| close(fd); |
| return mDNSNULL; |
| } |
| ret->events.fd = fd; |
| ret->callback = callback; |
| ret->context = context; |
| ret->addressType = addrType; |
| ret->socketFlags = socketFlags; |
| |
| // When we get a connection, mDNSPosixListenCallback will be called, and it will invoke the |
| // callback we were passed. |
| requestReadEvents(&ret->events, "tcpListenCallback", tcpListenCallback, ret); |
| return ret; |
| } |
| |
| mDNSexport int mDNSPlatformTCPGetFD(TCPSocket *sock) |
| { |
| return sock->events.fd; |
| } |
| |
| mDNSexport mStatus mDNSPlatformTCPConnect(TCPSocket *sock, const mDNSAddr *dst, mDNSOpaque16 dstport, |
| mDNSInterfaceID InterfaceID, TCPConnectionCallback callback, void *context) |
| { |
| int result; |
| union { |
| struct sockaddr sa; |
| struct sockaddr_in sin; |
| struct sockaddr_in6 sin6; |
| } addr; |
| socklen_t len; |
| |
| sock->callback = callback; |
| sock->context = context; |
| sock->setup = mDNSfalse; |
| sock->connected = mDNSfalse; |
| sock->err = mStatus_NoError; |
| |
| result = fcntl(sock->events.fd, F_GETFL, 0); |
| if (result < 0) |
| { |
| LogMsg("mDNSPlatformTCPConnect: F_GETFL failed: %s", strerror(errno)); |
| return mStatus_UnknownErr; |
| } |
| |
| result = fcntl(sock->events.fd, F_SETFL, result | O_NONBLOCK); |
| if (result < 0) |
| { |
| LogMsg("mDNSPlatformTCPConnect: F_SETFL failed: %s", strerror(errno)); |
| return mStatus_UnknownErr; |
| } |
| |
| // If we've been asked to bind to a single interface, do it. See comment in mDNSMacOSX.c for more info. |
| if (InterfaceID) |
| { |
| PosixNetworkInterface *iface = (PosixNetworkInterface *)InterfaceID; |
| #if defined(SO_BINDTODEVICE) |
| result = setsockopt(sock->events.fd, |
| SOL_SOCKET, SO_BINDTODEVICE, iface->intfName, strlen(iface->intfName)); |
| if (result < 0) |
| { |
| LogMsg("mDNSPlatformTCPConnect: SO_BINDTODEVICE failed on %s: %s", iface->intfName, strerror(errno)); |
| return mStatus_BadParamErr; |
| } |
| #else |
| if (dst->type == mDNSAddrType_IPv4) |
| { |
| #if defined(IP_BOUND_IF) |
| result = setsockopt(sock->events.fd, IPPROTO_IP, IP_BOUND_IF, &iface->index, sizeof iface->index); |
| if (result < 0) |
| { |
| LogMsg("mDNSPlatformTCPConnect: IP_BOUND_IF failed on %s (%d): %s", |
| iface->intfName, iface->index, strerror(errno)); |
| return mStatus_BadParamErr; |
| } |
| #else |
| (void)iface; |
| #endif // IP_BOUND_IF |
| } |
| else |
| { // IPv6 |
| #if defined(IPV6_BOUND_IF) |
| result = setsockopt(sock->events.fd, IPPROTO_IPV6, IPV6_BOUND_IF, &iface->index, sizeof iface->index); |
| if (result < 0) |
| { |
| LogMsg("mDNSPlatformTCPConnect: IP_BOUND_IF failed on %s (%d): %s", |
| iface->intfName, iface->index, strerror(errno)); |
| return mStatus_BadParamErr; |
| } |
| #else |
| (void)iface; |
| #endif // IPV6_BOUND_IF |
| } |
| #endif // SO_BINDTODEVICE |
| } |
| |
| memset(&addr, 0, sizeof addr); |
| if (dst->type == mDNSAddrType_IPv4) |
| { |
| addr.sa.sa_family = AF_INET; |
| addr.sin.sin_port = dstport.NotAnInteger; |
| len = sizeof (struct sockaddr_in); |
| addr.sin.sin_addr.s_addr = dst->ip.v4.NotAnInteger; |
| } |
| else |
| { |
| addr.sa.sa_family = AF_INET6; |
| len = sizeof (struct sockaddr_in6); |
| addr.sin6.sin6_port = dstport.NotAnInteger; |
| memcpy(&addr.sin6.sin6_addr.s6_addr, &dst->ip.v6, sizeof addr.sin6.sin6_addr.s6_addr); |
| } |
| #ifndef NOT_HAVE_SA_LEN |
| addr.sa.sa_len = len; |
| #endif |
| |
| result = connect(sock->events.fd, (struct sockaddr *)&addr, len); |
| if (result < 0) |
| { |
| if (errno == EINPROGRESS) |
| { |
| requestWriteEvents(&sock->events, "mDNSPlatformConnect", tcpConnectCallback, sock); |
| return mStatus_ConnPending; |
| } |
| if (errno == EHOSTUNREACH || errno == EADDRNOTAVAIL || errno == ENETDOWN) |
| { |
| LogInfo("ERROR: mDNSPlatformTCPConnect - connect failed: socket %d: Error %d (%s)", |
| sock->events.fd, errno, strerror(errno)); |
| } |
| else |
| { |
| LogMsg("ERROR: mDNSPlatformTCPConnect - connect failed: socket %d: Error %d (%s) length %d", |
| sock->events.fd, errno, strerror(errno), len); |
| } |
| return mStatus_ConnFailed; |
| } |
| |
| LogMsg("NOTE: mDNSPlatformTCPConnect completed synchronously"); |
| return mStatus_NoError; |
| } |
| |
| mDNSexport void mDNSPlatformTCPCloseConnection(TCPSocket *sock) |
| { |
| if (sock) |
| { // can sock really be NULL when this is called? |
| shutdown(sock->events.fd, SHUT_RDWR); |
| stopReadOrWriteEvents(sock->events.fd, mDNSfalse, mDNStrue, |
| PosixEventFlag_Read | PosixEventFlag_Write); |
| close(sock->events.fd); |
| mdns_free(sock); |
| } |
| } |
| |
| mDNSexport long mDNSPlatformReadTCP(TCPSocket *sock, void *buf, unsigned long buflen, mDNSBool *closed) |
| { |
| ssize_t nread; |
| |
| *closed = mDNSfalse; |
| if (sock->flags & kTCPSocketFlags_UseTLS) |
| { |
| #ifdef POSIX_HAS_TLS |
| nread = mDNSPosixTLSRead(sock, buf, buflen, closed); |
| #else |
| nread = mStatus_ConnFailed; |
| *closed = mDNStrue; |
| #endif |
| } else { |
| nread = mDNSPosixReadTCP(sock->events.fd, buf, buflen, closed); |
| } |
| return nread; |
| } |
| |
| mDNSexport mDNSBool mDNSPlatformTCPWritable(TCPSocket *sock) |
| { |
| fd_set w; |
| int nfds = sock->events.fd + 1; |
| int count; |
| struct timeval tv; |
| |
| if (nfds > FD_SETSIZE) |
| { |
| LogMsg("ERROR: mDNSPlatformTCPWritable called on an fd that won't fit in an fd_set."); |
| return mDNStrue; // hope for the best? |
| } |
| FD_SET(sock->events.fd, &w); |
| tv.tv_sec = tv.tv_usec = 0; |
| count = select(nfds, NULL, &w, NULL, &tv); |
| if (count > 0) |
| { |
| return mDNStrue; |
| } |
| return mDNSfalse; |
| } |
| |
| mDNSexport long mDNSPlatformWriteTCP(TCPSocket *sock, const char *msg, unsigned long len) |
| { |
| if (sock->flags & kTCPSocketFlags_UseTLS) |
| { |
| #ifdef POSIX_HAS_TLS |
| return mDNSPosixTLSWrite(sock, msg, len); |
| #else |
| return mStatus_ConnFailed; |
| #endif |
| } |
| else |
| { |
| return mDNSPosixWriteTCP(sock->events.fd, msg, len); |
| } |
| } |
| |
| mDNSexport UDPSocket *mDNSPlatformUDPSocket(mDNSIPPort port) |
| { |
| mDNSBool randomizePort = mDNSIPPortIsZero(port); |
| UDPSocket *p = callocL("UDPSocket", sizeof(UDPSocket)); |
| if (!p) { LogMsg("mDNSPlatformUDPSocket: memory exhausted"); return(mDNSNULL); } |
| p->randomizePort = randomizePort; |
| p->port = port; |
| p->events.fd = -1; |
| return(p); |
| } |
| |
| mDNSexport void mDNSPlatformUDPClose(UDPSocket *sock) |
| { |
| if (sock && sock->events.fd != -1) |
| { |
| stopReadOrWriteEvents(sock->events.fd, mDNSfalse, mDNStrue, |
| PosixEventFlag_Read | PosixEventFlag_Write); |
| close(sock->events.fd); |
| mdns_free(sock); |
| } |
| } |
| |
| mDNSexport void mDNSPlatformUpdateProxyList(const mDNSInterfaceID InterfaceID) |
| { |
| (void)InterfaceID; // Unused |
| } |
| |
| mDNSexport void mDNSPlatformSendRawPacket(const void *const msg, const mDNSu8 *const end, mDNSInterfaceID InterfaceID) |
| { |
| (void)msg; // Unused |
| (void)end; // Unused |
| (void)InterfaceID; // Unused |
| } |
| |
| mDNSexport void mDNSPlatformSetLocalAddressCacheEntry(const mDNSAddr *const tpa, const mDNSEthAddr *const tha, mDNSInterfaceID InterfaceID) |
| { |
| (void)tpa; // Unused |
| (void)tha; // Unused |
| (void)InterfaceID; // Unused |
| } |
| |
| mDNSexport mStatus mDNSPlatformTLSSetupCerts(void) |
| { |
| return(mStatus_UnsupportedErr); |
| } |
| |
| mDNSexport void mDNSPlatformTLSTearDownCerts(void) |
| { |
| } |
| |
| mDNSexport void mDNSPlatformSetAllowSleep(mDNSBool allowSleep, const char *reason) |
| { |
| (void) allowSleep; |
| (void) reason; |
| } |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark - |
| #pragma mark - /etc/hosts support |
| #endif |
| |
| mDNSexport void FreeEtcHosts(mDNS *const m, AuthRecord *const rr, mStatus result) |
| { |
| (void)m; // unused |
| (void)rr; |
| (void)result; |
| } |
| |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark ***** DDNS Config Platform Functions |
| #endif |
| |
| mDNSexport mDNSBool mDNSPlatformSetDNSConfig(mDNSBool setservers, mDNSBool setsearch, domainname *const fqdn, DNameListElem **RegDomains, |
| DNameListElem **BrowseDomains, mDNSBool ackConfig) |
| { |
| (void) setservers; |
| (void) setsearch; |
| (void) ackConfig; |
| |
| if (fqdn ) fqdn->c[0] = 0; |
| if (RegDomains ) *RegDomains = NULL; |
| if (BrowseDomains) *BrowseDomains = NULL; |
| |
| return mDNStrue; |
| } |
| |
| mDNSexport mStatus mDNSPlatformGetPrimaryInterface(mDNSAddr * v4, mDNSAddr * v6, mDNSAddr * router) |
| { |
| (void) v4; |
| (void) v6; |
| (void) router; |
| |
| return mStatus_UnsupportedErr; |
| } |
| |
| mDNSexport void mDNSPlatformDynDNSHostNameStatusChanged(const domainname *const dname, const mStatus status) |
| { |
| (void) dname; |
| (void) status; |
| } |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark ***** Init and Term |
| #endif |
| |
| // This gets the current hostname, truncating it at the first dot if necessary |
| mDNSlocal void GetUserSpecifiedRFC1034ComputerName(domainlabel *const namelabel) |
| { |
| int len = 0; |
| gethostname((char *)(&namelabel->c[1]), MAX_DOMAIN_LABEL); |
| while (len < MAX_DOMAIN_LABEL && namelabel->c[len+1] && namelabel->c[len+1] != '.') len++; |
| namelabel->c[0] = len; |
| } |
| |
| // On OS X this gets the text of the field labelled "Computer Name" in the Sharing Prefs Control Panel |
| // Other platforms can either get the information from the appropriate place, |
| // or they can alternatively just require all registering services to provide an explicit name |
| mDNSlocal void GetUserSpecifiedFriendlyComputerName(domainlabel *const namelabel) |
| { |
| // On Unix we have no better name than the host name, so we just use that. |
| GetUserSpecifiedRFC1034ComputerName(namelabel); |
| } |
| |
| mDNSexport int ParseDNSServers(mDNS *m, const char *filePath) |
| { |
| char line[256]; |
| char nameserver[16]; |
| char keyword[11]; |
| int numOfServers = 0; |
| FILE *fp = fopen(filePath, "r"); |
| if (fp == NULL) return -1; |
| while (fgets(line,sizeof(line),fp)) |
| { |
| struct in_addr ina; |
| line[255]='\0'; // just to be safe |
| if (sscanf(line,"%10s %15s", keyword, nameserver) != 2) continue; // it will skip whitespaces |
| if (strncasecmp(keyword,"nameserver",10)) continue; |
| if (inet_aton(nameserver, (struct in_addr *)&ina) != 0) |
| { |
| mDNSAddr DNSAddr; |
| DNSAddr.type = mDNSAddrType_IPv4; |
| DNSAddr.ip.v4.NotAnInteger = ina.s_addr; |
| mDNS_AddDNSServer(m, NULL, mDNSInterface_Any, 0, &DNSAddr, UnicastDNSPort, kScopeNone, 0, mDNSfalse, mDNSfalse, mDNSfalse, mDNSfalse, 0, mDNStrue, mDNStrue, mDNSfalse); |
| numOfServers++; |
| } |
| } |
| fclose(fp); |
| return (numOfServers > 0) ? 0 : -1; |
| } |
| |
| // Searches the interface list looking for the named interface. |
| // Returns a pointer to if it found, or NULL otherwise. |
| mDNSlocal PosixNetworkInterface *SearchForInterfaceByName(mDNS *const m, const char *intfName) |
| { |
| PosixNetworkInterface *intf; |
| |
| assert(m != NULL); |
| assert(intfName != NULL); |
| |
| intf = (PosixNetworkInterface*)(m->HostInterfaces); |
| while ((intf != NULL) && (strcmp(intf->intfName, intfName) != 0)) |
| intf = (PosixNetworkInterface *)(intf->coreIntf.next); |
| |
| return intf; |
| } |
| |
| mDNSexport mDNSInterfaceID mDNSPlatformInterfaceIDfromInterfaceIndex(mDNS *const m, mDNSu32 index) |
| { |
| PosixNetworkInterface *intf; |
| |
| assert(m != NULL); |
| |
| if (index == kDNSServiceInterfaceIndexLocalOnly) return(mDNSInterface_LocalOnly); |
| if (index == kDNSServiceInterfaceIndexP2P ) return(mDNSInterface_P2P); |
| if (index == kDNSServiceInterfaceIndexAny ) return(mDNSInterface_Any); |
| |
| intf = (PosixNetworkInterface*)(m->HostInterfaces); |
| while ((intf != NULL) && (mDNSu32) intf->index != index) |
| intf = (PosixNetworkInterface *)(intf->coreIntf.next); |
| |
| return (mDNSInterfaceID) intf; |
| } |
| |
| mDNSexport mDNSu32 mDNSPlatformInterfaceIndexfromInterfaceID(mDNS *const m, mDNSInterfaceID id, mDNSBool suppressNetworkChange) |
| { |
| PosixNetworkInterface *intf; |
| (void) suppressNetworkChange; // Unused |
| |
| assert(m != NULL); |
| |
| if (id == mDNSInterface_LocalOnly) return(kDNSServiceInterfaceIndexLocalOnly); |
| if (id == mDNSInterface_P2P ) return(kDNSServiceInterfaceIndexP2P); |
| if (id == mDNSInterface_Any ) return(kDNSServiceInterfaceIndexAny); |
| |
| intf = (PosixNetworkInterface*)(m->HostInterfaces); |
| while ((intf != NULL) && (mDNSInterfaceID) intf != id) |
| intf = (PosixNetworkInterface *)(intf->coreIntf.next); |
| |
| if (intf) return intf->index; |
| |
| // If we didn't find the interface, check the RecentInterfaces list as well |
| intf = gRecentInterfaces; |
| while ((intf != NULL) && (mDNSInterfaceID) intf != id) |
| intf = (PosixNetworkInterface *)(intf->coreIntf.next); |
| |
| return intf ? intf->index : 0; |
| } |
| |
| // Frees the specified PosixNetworkInterface structure. The underlying |
| // interface must have already been deregistered with the mDNS core. |
| mDNSlocal void FreePosixNetworkInterface(PosixNetworkInterface *intf) |
| { |
| int rv; |
| assert(intf != NULL); |
| if (intf->intfName != NULL) mdns_free(intf->intfName); |
| if (intf->multicastSocket4 != -1) |
| { |
| rv = close(intf->multicastSocket4); |
| assert(rv == 0); |
| } |
| #if HAVE_IPV6 |
| if (intf->multicastSocket6 != -1) |
| { |
| rv = close(intf->multicastSocket6); |
| assert(rv == 0); |
| } |
| #endif |
| |
| // Move interface to the RecentInterfaces list for a minute |
| intf->LastSeen = mDNSPlatformUTC(); |
| intf->coreIntf.next = &gRecentInterfaces->coreIntf; |
| gRecentInterfaces = intf; |
| } |
| |
| // Grab the first interface, deregister it, free it, and repeat until done. |
| mDNSlocal void ClearInterfaceList(mDNS *const m) |
| { |
| assert(m != NULL); |
| |
| while (m->HostInterfaces) |
| { |
| PosixNetworkInterface *intf = (PosixNetworkInterface*)(m->HostInterfaces); |
| mDNS_DeregisterInterface(m, &intf->coreIntf, NormalActivation); |
| if (gMDNSPlatformPosixVerboseLevel > 0) fprintf(stderr, "Deregistered interface %s\n", intf->intfName); |
| FreePosixNetworkInterface(intf); |
| } |
| num_registered_interfaces = 0; |
| num_pkts_accepted = 0; |
| num_pkts_rejected = 0; |
| } |
| |
| mDNSlocal int SetupIPv6Socket(int fd) |
| { |
| int err; |
| |
| #if defined(IPV6_PKTINFO) |
| err = setsockopt(fd, IPPROTO_IPV6, IPV6_2292_PKTINFO, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_PKTINFO"); } |
| #else |
| #warning This platform has no way to get the destination interface information for IPv6 -- will only work for single-homed hosts |
| #endif |
| return err; |
| } |
| |
| mDNSlocal int SetupIPv4Socket(int fd) |
| { |
| int err; |
| |
| #if defined(IP_PKTINFO) // Linux |
| err = setsockopt(fd, IPPROTO_IP, IP_PKTINFO, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_PKTINFO"); } |
| #elif defined(IP_RECVDSTADDR) || defined(IP_RECVIF) // BSD and Solaris |
| #if defined(IP_RECVDSTADDR) |
| err = setsockopt(fd, IPPROTO_IP, IP_RECVDSTADDR, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_RECVDSTADDR"); } |
| #endif |
| #if defined(IP_RECVIF) |
| if (err == 0) |
| { |
| err = setsockopt(fd, IPPROTO_IP, IP_RECVIF, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_RECVIF"); } |
| } |
| #endif |
| #else |
| #warning This platform has no way to get the destination interface information -- will only work for single-homed hosts |
| #endif |
| return err; |
| } |
| |
| // Sets up a send/receive socket. |
| // If mDNSIPPort port is non-zero, then it's a multicast socket on the specified interface |
| // If mDNSIPPort port is zero, then it's a randomly assigned port number, used for sending unicast queries |
| mDNSlocal int SetupSocket(struct sockaddr *intfAddr, mDNSIPPort port, int interfaceIndex, int *sktPtr) |
| { |
| int err = 0; |
| const mDNSBool JoinMulticastGroup = (port.NotAnInteger != 0); |
| |
| (void) interfaceIndex; // This parameter unused on plaforms that don't have IPv6 |
| assert(intfAddr != NULL); |
| assert(sktPtr != NULL); |
| assert(*sktPtr == -1); |
| |
| // Open the socket... |
| if (intfAddr->sa_family == AF_INET) *sktPtr = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP); |
| #if HAVE_IPV6 |
| else if (intfAddr->sa_family == AF_INET6) *sktPtr = socket(PF_INET6, SOCK_DGRAM, IPPROTO_UDP); |
| #endif |
| else return EINVAL; |
| |
| if (*sktPtr < 0) { err = errno; perror((intfAddr->sa_family == AF_INET) ? "socket AF_INET" : "socket AF_INET6"); } |
| |
| // ... with a shared UDP port, if it's for multicast receiving |
| if (err == 0 && port.NotAnInteger) |
| { |
| // <rdar://problem/20946253> Suggestions from Jonny Törnbom at Axis Communications |
| // We test for SO_REUSEADDR first, as suggested by Jonny Törnbom from Axis Communications |
| // Linux kernel versions 3.9 introduces support for socket option |
| // SO_REUSEPORT, however this is not implemented the same as on *BSD |
| // systems. Linux version implements a "port hijacking" prevention |
| // mechanism, limiting processes wanting to bind to an already existing |
| // addr:port to have the same effective UID as the first who bound it. What |
| // this meant for us was that the daemon ran as one user and when for |
| // instance mDNSClientPosix was executed by another user, it wasn't allowed |
| // to bind to the socket. Our suggestion was to switch the order in which |
| // SO_REUSEPORT and SO_REUSEADDR was tested so that SO_REUSEADDR stays on |
| // top and SO_REUSEPORT to be used only if SO_REUSEADDR doesn't exist. |
| #if defined(SO_REUSEADDR) && !defined(__MAC_OS_X_VERSION_MIN_REQUIRED) |
| err = setsockopt(*sktPtr, SOL_SOCKET, SO_REUSEADDR, &kOn, sizeof(kOn)); |
| #elif defined(SO_REUSEPORT) |
| err = setsockopt(*sktPtr, SOL_SOCKET, SO_REUSEPORT, &kOn, sizeof(kOn)); |
| #else |
| #error This platform has no way to avoid address busy errors on multicast. |
| #endif |
| if (err < 0) { err = errno; perror("setsockopt - SO_REUSExxxx"); } |
| |
| #if TARGET_OS_MAC |
| // Enable inbound packets on IFEF_AWDL interface. |
| // Only done for multicast sockets, since we don't expect unicast socket operations |
| // on the IFEF_AWDL interface. Operation is a no-op for other interface types. |
| #ifndef SO_RECV_ANYIF |
| #define SO_RECV_ANYIF 0x1104 /* unrestricted inbound processing */ |
| #endif |
| if (setsockopt(*sktPtr, SOL_SOCKET, SO_RECV_ANYIF, &kOn, sizeof(kOn)) < 0) perror("setsockopt - SO_RECV_ANYIF"); |
| #endif |
| } |
| |
| // We want to receive destination addresses and interface identifiers. |
| if (intfAddr->sa_family == AF_INET) |
| { |
| struct ip_mreq imr; |
| struct sockaddr_in bindAddr; |
| if (err == 0) |
| { |
| err = SetupIPv4Socket(*sktPtr); |
| } |
| #if defined(IP_RECVTTL) // Linux |
| if (err == 0) |
| { |
| setsockopt(*sktPtr, IPPROTO_IP, IP_RECVTTL, &kOn, sizeof(kOn)); |
| // We no longer depend on being able to get the received TTL, so don't worry if the option fails |
| } |
| #endif |
| |
| // Add multicast group membership on this interface |
| if (err == 0 && JoinMulticastGroup) |
| { |
| imr.imr_multiaddr.s_addr = AllDNSLinkGroup_v4.ip.v4.NotAnInteger; |
| imr.imr_interface = ((struct sockaddr_in*)intfAddr)->sin_addr; |
| err = setsockopt(*sktPtr, IPPROTO_IP, IP_ADD_MEMBERSHIP, &imr, sizeof(imr)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_ADD_MEMBERSHIP"); } |
| } |
| |
| // Specify outgoing interface too |
| if (err == 0 && JoinMulticastGroup) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IP, IP_MULTICAST_IF, &((struct sockaddr_in*)intfAddr)->sin_addr, sizeof(struct in_addr)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_MULTICAST_IF"); } |
| } |
| |
| // Per the mDNS spec, send unicast packets with TTL 255 |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IP, IP_TTL, &kIntTwoFiveFive, sizeof(kIntTwoFiveFive)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_TTL"); } |
| } |
| |
| // and multicast packets with TTL 255 too |
| // There's some debate as to whether IP_MULTICAST_TTL is an int or a byte so we just try both. |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IP, IP_MULTICAST_TTL, &kByteTwoFiveFive, sizeof(kByteTwoFiveFive)); |
| if (err < 0 && errno == EINVAL) |
| err = setsockopt(*sktPtr, IPPROTO_IP, IP_MULTICAST_TTL, &kIntTwoFiveFive, sizeof(kIntTwoFiveFive)); |
| if (err < 0) { err = errno; perror("setsockopt - IP_MULTICAST_TTL"); } |
| } |
| |
| // And start listening for packets |
| if (err == 0) |
| { |
| bindAddr.sin_family = AF_INET; |
| bindAddr.sin_port = port.NotAnInteger; |
| bindAddr.sin_addr.s_addr = INADDR_ANY; // Want to receive multicasts AND unicasts on this socket |
| err = bind(*sktPtr, (struct sockaddr *) &bindAddr, sizeof(bindAddr)); |
| if (err < 0) { err = errno; perror("bind"); fflush(stderr); } |
| } |
| } // endif (intfAddr->sa_family == AF_INET) |
| |
| #if HAVE_IPV6 |
| else if (intfAddr->sa_family == AF_INET6) |
| { |
| struct ipv6_mreq imr6; |
| struct sockaddr_in6 bindAddr6; |
| if (err == 0) { |
| err = SetupIPv6Socket(*sktPtr); |
| } |
| #if defined(IPV6_HOPLIMIT) |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_2292_HOPLIMIT, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_HOPLIMIT"); } |
| } |
| #endif |
| |
| // Add multicast group membership on this interface |
| if (err == 0 && JoinMulticastGroup) |
| { |
| imr6.ipv6mr_multiaddr = *(const struct in6_addr*)&AllDNSLinkGroup_v6.ip.v6; |
| imr6.ipv6mr_interface = interfaceIndex; |
| //LogMsg("Joining %.16a on %d", &imr6.ipv6mr_multiaddr, imr6.ipv6mr_interface); |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_JOIN_GROUP, &imr6, sizeof(imr6)); |
| if (err < 0) |
| { |
| err = errno; |
| verbosedebugf("IPV6_JOIN_GROUP %.16a on %d failed.\n", &imr6.ipv6mr_multiaddr, imr6.ipv6mr_interface); |
| perror("setsockopt - IPV6_JOIN_GROUP"); |
| } |
| } |
| |
| // Specify outgoing interface too |
| if (err == 0 && JoinMulticastGroup) |
| { |
| u_int multicast_if = interfaceIndex; |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_MULTICAST_IF, &multicast_if, sizeof(multicast_if)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_MULTICAST_IF"); } |
| } |
| |
| // We want to receive only IPv6 packets on this socket. |
| // Without this option, we may get IPv4 addresses as mapped addresses. |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_V6ONLY, &kOn, sizeof(kOn)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_V6ONLY"); } |
| } |
| |
| // Per the mDNS spec, send unicast packets with TTL 255 |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &kIntTwoFiveFive, sizeof(kIntTwoFiveFive)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_UNICAST_HOPS"); } |
| } |
| |
| // and multicast packets with TTL 255 too |
| // There's some debate as to whether IPV6_MULTICAST_HOPS is an int or a byte so we just try both. |
| if (err == 0) |
| { |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &kByteTwoFiveFive, sizeof(kByteTwoFiveFive)); |
| if (err < 0 && errno == EINVAL) |
| err = setsockopt(*sktPtr, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &kIntTwoFiveFive, sizeof(kIntTwoFiveFive)); |
| if (err < 0) { err = errno; perror("setsockopt - IPV6_MULTICAST_HOPS"); } |
| } |
| |
| // And start listening for packets |
| if (err == 0) |
| { |
| mDNSPlatformMemZero(&bindAddr6, sizeof(bindAddr6)); |
| #ifndef NOT_HAVE_SA_LEN |
| bindAddr6.sin6_len = sizeof(bindAddr6); |
| #endif |
| bindAddr6.sin6_family = AF_INET6; |
| bindAddr6.sin6_port = port.NotAnInteger; |
| bindAddr6.sin6_flowinfo = 0; |
| bindAddr6.sin6_addr = in6addr_any; // Want to receive multicasts AND unicasts on this socket |
| bindAddr6.sin6_scope_id = 0; |
| err = bind(*sktPtr, (struct sockaddr *) &bindAddr6, sizeof(bindAddr6)); |
| if (err < 0) { err = errno; perror("bind"); fflush(stderr); } |
| } |
| } // endif (intfAddr->sa_family == AF_INET6) |
| #endif |
| |
| // Set the socket to non-blocking. |
| if (err == 0) |
| { |
| err = fcntl(*sktPtr, F_GETFL, 0); |
| if (err < 0) err = errno; |
| else |
| { |
| err = fcntl(*sktPtr, F_SETFL, err | O_NONBLOCK); |
| if (err < 0) err = errno; |
| } |
| } |
| |
| // Clean up |
| if (err != 0 && *sktPtr != -1) |
| { |
| int rv; |
| rv = close(*sktPtr); |
| assert(rv == 0); |
| *sktPtr = -1; |
| } |
| assert((err == 0) == (*sktPtr != -1)); |
| return err; |
| } |
| |
| // Creates a PosixNetworkInterface for the interface whose IP address is |
| // intfAddr and whose name is intfName and registers it with mDNS core. |
| mDNSlocal int SetupOneInterface(mDNS *const m, struct sockaddr *intfAddr, struct sockaddr *intfMask, |
| const mDNSu8 *intfHaddr, mDNSu16 intfHlen, const char *intfName, int intfIndex) |
| { |
| int err = 0; |
| PosixNetworkInterface *intf; |
| PosixNetworkInterface *alias = NULL; |
| |
| assert(m != NULL); |
| assert(intfAddr != NULL); |
| assert(intfName != NULL); |
| assert(intfHaddr != NULL || intfHlen == 0); |
| assert(intfMask != NULL); |
| |
| // Allocate the interface structure itself. |
| intf = (PosixNetworkInterface*)mdns_calloc(1, sizeof(*intf)); |
| if (intf == NULL) { assert(0); err = ENOMEM; } |
| |
| // And make a copy of the intfName. |
| if (err == 0) |
| { |
| #ifdef LINUX |
| char *s; |
| int len; |
| s = strchr(intfName, ':'); |
| if (s != NULL) |
| { |
| len = (s - intfName) + 1; |
| } |
| else |
| { |
| len = strlen(intfName) + 1; |
| } |
| intf->intfName = malloc(len); |
| if (intf->intfName == NULL) { assert(0); err = ENOMEM; } |
| memcpy(intf->intfName, intfName, len - 1); |
| intfName[len - 1] = 0; |
| #else |
| intf->intfName = mdns_strdup(intfName); |
| if (intf->intfName == NULL) { assert(0); err = ENOMEM; } |
| #endif |
| } |
| |
| if (err == 0) |
| { |
| // Set up the fields required by the mDNS core. |
| SockAddrTomDNSAddr(intfAddr, &intf->coreIntf.ip, NULL); |
| SockAddrTomDNSAddr(intfMask, &intf->coreIntf.mask, NULL); |
| if (intfHlen == sizeof(intf->coreIntf.MAC.b)) |
| { |
| mDNSPlatformMemCopy(intf->coreIntf.MAC.b, intfHaddr, sizeof(intf->coreIntf.MAC.b)); |
| |
| // Configure primary MAC address. |
| // Ideally, we would pick the default route interface with the lowest metric (see mDNSWin32). |
| // For now, simply assume the first one that we find is the primary one (see mDNSMacOSX). |
| if (mDNSSameEthAddress(&m->PrimaryMAC, &zeroEthAddr)) |
| mDNSPlatformMemCopy(&m->PrimaryMAC, &intf->coreIntf.MAC, sizeof(m->PrimaryMAC)); |
| } |
| |
| //LogMsg("SetupOneInterface: %#a %#a", &intf->coreIntf.ip, &intf->coreIntf.mask); |
| mDNSPlatformStrLCopy(intf->coreIntf.ifname, intfName, sizeof(intf->coreIntf.ifname)); |
| intf->coreIntf.ifname[sizeof(intf->coreIntf.ifname)-1] = 0; |
| |
| intf->coreIntf.Advertise = m->AdvertiseLocalAddresses; |
| intf->coreIntf.McastTxRx = mDNStrue; |
| |
| // Set up the extra fields in PosixNetworkInterface. |
| assert(intf->intfName != NULL); // intf->intfName already set up above |
| intf->index = intfIndex; |
| intf->multicastSocket4 = -1; |
| #if HAVE_IPV6 |
| intf->multicastSocket6 = -1; |
| #endif |
| alias = SearchForInterfaceByName(m, intf->intfName); |
| if (alias == NULL) alias = intf; |
| intf->coreIntf.InterfaceID = (mDNSInterfaceID)alias; |
| |
| if (alias != intf) |
| debugf("SetupOneInterface: %s %#a is an alias of %#a", intfName, &intf->coreIntf.ip, &alias->coreIntf.ip); |
| } |
| |
| // Set up the multicast socket |
| if (err == 0) |
| { |
| if (alias->multicastSocket4 == -1 && intfAddr->sa_family == AF_INET) |
| err = SetupSocket(intfAddr, MulticastDNSPort, intf->index, &alias->multicastSocket4); |
| #if HAVE_IPV6 |
| else if (alias->multicastSocket6 == -1 && intfAddr->sa_family == AF_INET6) |
| err = SetupSocket(intfAddr, MulticastDNSPort, intf->index, &alias->multicastSocket6); |
| #endif |
| } |
| |
| // If interface is a direct link, address record will be marked as kDNSRecordTypeKnownUnique |
| // and skip the probe phase of the probe/announce packet sequence. |
| intf->coreIntf.DirectLink = mDNSfalse; |
| #ifdef DIRECTLINK_INTERFACE_NAME |
| if (strcmp(intfName, STRINGIFY(DIRECTLINK_INTERFACE_NAME)) == 0) |
| intf->coreIntf.DirectLink = mDNStrue; |
| #endif |
| intf->coreIntf.SupportsUnicastMDNSResponse = mDNStrue; |
| |
| // The interface is all ready to go, let's register it with the mDNS core. |
| if (err == 0) |
| err = mDNS_RegisterInterface(m, &intf->coreIntf, NormalActivation); |
| |
| // Clean up. |
| if (err == 0) |
| { |
| num_registered_interfaces++; |
| debugf("SetupOneInterface: %s %#a Registered", intf->intfName, &intf->coreIntf.ip); |
| if (gMDNSPlatformPosixVerboseLevel > 0) |
| fprintf(stderr, "Registered interface %s\n", intf->intfName); |
| } |
| else |
| { |
| // Use intfName instead of intf->intfName in the next line to avoid dereferencing NULL. |
| debugf("SetupOneInterface: %s %#a failed to register %d", intfName, &intf->coreIntf.ip, err); |
| if (intf) { FreePosixNetworkInterface(intf); intf = NULL; } |
| } |
| |
| assert((err == 0) == (intf != NULL)); |
| |
| return err; |
| } |
| |
| // Call get_ifi_info() to obtain a list of active interfaces and call SetupOneInterface() on each one. |
| mDNSlocal int SetupInterfaceList(mDNS *const m) |
| { |
| mDNSBool foundav4 = mDNSfalse; |
| int err = 0; |
| struct ifaddrs *intfList; |
| struct ifaddrs *firstLoopback = NULL; |
| int firstLoopbackIndex = 0; |
| |
| assert(m != NULL); |
| debugf("SetupInterfaceList"); |
| |
| if (getifaddrs(&intfList) < 0) |
| { |
| err = errno; |
| } |
| if (intfList == NULL) err = ENOENT; |
| |
| if (err == 0) |
| { |
| struct ifaddrs *i = intfList; |
| while (i) |
| { |
| if ( i->ifa_addr != NULL && |
| ((i->ifa_addr->sa_family == AF_INET) |
| #if HAVE_IPV6 |
| || (i->ifa_addr->sa_family == AF_INET6) |
| #endif |
| ) && (i->ifa_flags & IFF_UP) && !(i->ifa_flags & IFF_POINTOPOINT)) |
| { |
| int ifIndex = if_nametoindex(i->ifa_name); |
| if (ifIndex == 0) |
| { |
| continue; |
| } |
| if (i->ifa_flags & IFF_LOOPBACK) |
| { |
| if (firstLoopback == NULL) |
| { |
| firstLoopback = i; |
| firstLoopbackIndex = ifIndex; |
| } |
| } |
| else |
| { |
| #define ethernet_addr_len 6 |
| uint8_t hwaddr[ethernet_addr_len]; |
| int hwaddr_len = 0; |
| |
| #if defined(TARGET_OS_LINUX) && TARGET_OS_LINUX |
| struct ifreq ifr; |
| int sockfd = socket(AF_INET6, SOCK_DGRAM, 0); |
| if (sockfd >= 0) |
| { |
| /* Add hardware address */ |
| memcpy(ifr.ifr_name, i->ifa_name, IFNAMSIZ); |
| if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) != -1) |
| { |
| if (ifr.ifr_hwaddr.sa_family == ARPHRD_ETHER) |
| { |
| memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, ethernet_addr_len); |
| hwaddr_len = ethernet_addr_len; |
| } |
| } |
| close(sockfd); |
| } |
| else |
| { |
| memset(hwaddr, 0, sizeof(hwaddr)); |
| } |
| #endif // TARGET_OS_LINUX |
| |
| #if defined(TARGET_OS_MAC) && TARGET_OS_MAC |
| for (struct ifaddrs *hw_scan = intfList; hw_scan != NULL; hw_scan = hw_scan->ifa_next) |
| { |
| if (hw_scan->ifa_addr != NULL && |
| hw_scan->ifa_addr->sa_family == AF_LINK && !strcmp(hw_scan->ifa_name, i->ifa_name)) |
| { |
| struct sockaddr_dl *sdl = (struct sockaddr_dl *)hw_scan->ifa_addr; |
| if (sdl->sdl_alen == ethernet_addr_len) |
| { |
| hwaddr_len = ethernet_addr_len; |
| memcpy(hwaddr, LLADDR(sdl), hwaddr_len); |
| } |
| break; |
| } |
| } |
| #endif |
| if (SetupOneInterface(m, i->ifa_addr, i->ifa_netmask, |
| hwaddr, hwaddr_len, i->ifa_name, ifIndex) == 0) |
| { |
| if (i->ifa_addr->sa_family == AF_INET) |
| foundav4 = mDNStrue; |
| } |
| } |
| } |
| i = i->ifa_next; |
| } |
| |
| // If we found no normal interfaces but we did find a loopback interface, register the |
| // loopback interface. This allows self-discovery if no interfaces are configured. |
| // Temporary workaround: Multicast loopback on IPv6 interfaces appears not to work. |
| // In the interim, we skip loopback interface only if we found at least one v4 interface to use |
| // if ((m->HostInterfaces == NULL) && (firstLoopback != NULL)) |
| if (!foundav4 && firstLoopback) |
| (void) SetupOneInterface(m, firstLoopback->ifa_addr, firstLoopback->ifa_netmask, |
| NULL, 0, firstLoopback->ifa_name, firstLoopbackIndex); |
| } |
| |
| // Clean up. |
| if (intfList != NULL) freeifaddrs(intfList); |
| |
| // Clean up any interfaces that have been hanging around on the RecentInterfaces list for more than a minute |
| PosixNetworkInterface **ri = &gRecentInterfaces; |
| const mDNSs32 utc = mDNSPlatformUTC(); |
| while (*ri) |
| { |
| PosixNetworkInterface *pi = *ri; |
| if (utc - pi->LastSeen < 60) ri = (PosixNetworkInterface **)&pi->coreIntf.next; |
| else { *ri = (PosixNetworkInterface *)pi->coreIntf.next; mdns_free(pi); } |
| } |
| |
| return err; |
| } |
| |
| #if USES_NETLINK |
| |
| // See <http://www.faqs.org/rfcs/rfc3549.html> for a description of NetLink |
| |
| // Open a socket that will receive interface change notifications |
| mDNSlocal mStatus OpenIfNotifySocket(int *pFD) |
| { |
| mStatus err = mStatus_NoError; |
| struct sockaddr_nl snl; |
| int sock; |
| int ret; |
| |
| sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE); |
| if (sock < 0) |
| return errno; |
| |
| // Configure read to be non-blocking because inbound msg size is not known in advance |
| (void) fcntl(sock, F_SETFL, O_NONBLOCK); |
| |
| /* Subscribe the socket to Link & IP addr notifications. */ |
| mDNSPlatformMemZero(&snl, sizeof snl); |
| snl.nl_family = AF_NETLINK; |
| snl.nl_groups = RTMGRP_LINK | RTMGRP_IPV4_IFADDR | RTMGRP_IPV6_IFADDR; |
| ret = bind(sock, (struct sockaddr *) &snl, sizeof snl); |
| if (0 == ret) |
| *pFD = sock; |
| else |
| err = errno; |
| |
| return err; |
| } |
| |
| #if MDNS_DEBUGMSGS |
| mDNSlocal void PrintNetLinkMsg(const struct nlmsghdr *pNLMsg) |
| { |
| const char *kNLMsgTypes[] = { "", "NLMSG_NOOP", "NLMSG_ERROR", "NLMSG_DONE", "NLMSG_OVERRUN" }; |
| const char *kNLRtMsgTypes[] = { "RTM_NEWLINK", "RTM_DELLINK", "RTM_GETLINK", "RTM_NEWADDR", "RTM_DELADDR", "RTM_GETADDR" }; |
| |
| printf("nlmsghdr len=%d, type=%s, flags=0x%x\n", pNLMsg->nlmsg_len, |
| pNLMsg->nlmsg_type < RTM_BASE ? kNLMsgTypes[pNLMsg->nlmsg_type] : kNLRtMsgTypes[pNLMsg->nlmsg_type - RTM_BASE], |
| pNLMsg->nlmsg_flags); |
| |
| if (RTM_NEWLINK <= pNLMsg->nlmsg_type && pNLMsg->nlmsg_type <= RTM_GETLINK) |
| { |
| struct ifinfomsg *pIfInfo = (struct ifinfomsg*) NLMSG_DATA(pNLMsg); |
| printf("ifinfomsg family=%d, type=%d, index=%d, flags=0x%x, change=0x%x\n", pIfInfo->ifi_family, |
| pIfInfo->ifi_type, pIfInfo->ifi_index, pIfInfo->ifi_flags, pIfInfo->ifi_change); |
| |
| } |
| else if (RTM_NEWADDR <= pNLMsg->nlmsg_type && pNLMsg->nlmsg_type <= RTM_GETADDR) |
| { |
| struct ifaddrmsg *pIfAddr = (struct ifaddrmsg*) NLMSG_DATA(pNLMsg); |
| printf("ifaddrmsg family=%d, index=%d, flags=0x%x\n", pIfAddr->ifa_family, |
| pIfAddr->ifa_index, pIfAddr->ifa_flags); |
| } |
| printf("\n"); |
| } |
| #endif |
| |
| mDNSlocal mDNSu32 ProcessRoutingNotification(int sd) |
| // Read through the messages on sd and if any indicate that any interface records should |
| // be torn down and rebuilt, return affected indices as a bitmask. Otherwise return 0. |
| { |
| ssize_t readCount; |
| char buff[4096]; |
| struct nlmsghdr *pNLMsg = (struct nlmsghdr*) buff; |
| mDNSu32 result = 0; |
| |
| // The structure here is more complex than it really ought to be because, |
| // unfortunately, there's no good way to size a buffer in advance large |
| // enough to hold all pending data and so avoid message fragmentation. |
| // (Note that FIONREAD is not supported on AF_NETLINK.) |
| |
| readCount = read(sd, buff, sizeof buff); |
| while (1) |
| { |
| // Make sure we've got an entire nlmsghdr in the buffer, and payload, too. |
| // If not, discard already-processed messages in buffer and read more data. |
| if (((char*) &pNLMsg[1] > (buff + readCount)) || // i.e. *pNLMsg extends off end of buffer |
| ((char*) pNLMsg + pNLMsg->nlmsg_len > (buff + readCount))) |
| { |
| if (buff < (char*) pNLMsg) // we have space to shuffle |
| { |
| // discard processed data |
| readCount -= ((char*) pNLMsg - buff); |
| memmove(buff, pNLMsg, readCount); |
| pNLMsg = (struct nlmsghdr*) buff; |
| |
| // read more data |
| readCount += read(sd, buff + readCount, sizeof buff - readCount); |
| continue; // spin around and revalidate with new readCount |
| } |
| else |
| break; // Otherwise message does not fit in buffer |
| } |
| |
| #if MDNS_DEBUGMSGS |
| PrintNetLinkMsg(pNLMsg); |
| #endif |
| |
| // Process the NetLink message |
| if (pNLMsg->nlmsg_type == RTM_GETLINK || pNLMsg->nlmsg_type == RTM_NEWLINK) |
| result |= 1 << ((struct ifinfomsg*) NLMSG_DATA(pNLMsg))->ifi_index; |
| else if (pNLMsg->nlmsg_type == RTM_DELADDR || pNLMsg->nlmsg_type == RTM_NEWADDR) |
| result |= 1 << ((struct ifaddrmsg*) NLMSG_DATA(pNLMsg))->ifa_index; |
| |
| // Advance pNLMsg to the next message in the buffer |
| if ((pNLMsg->nlmsg_flags & NLM_F_MULTI) != 0 && pNLMsg->nlmsg_type != NLMSG_DONE) |
| { |
| ssize_t len = readCount - ((char*)pNLMsg - buff); |
| pNLMsg = NLMSG_NEXT(pNLMsg, len); |
| } |
| else |
| break; // all done! |
| } |
| |
| return result; |
| } |
| |
| #else // USES_NETLINK |
| |
| // Open a socket that will receive interface change notifications |
| mDNSlocal mStatus OpenIfNotifySocket(int *pFD) |
| { |
| *pFD = socket(AF_ROUTE, SOCK_RAW, 0); |
| |
| if (*pFD < 0) |
| return mStatus_UnknownErr; |
| |
| // Configure read to be non-blocking because inbound msg size is not known in advance |
| (void) fcntl(*pFD, F_SETFL, O_NONBLOCK); |
| |
| return mStatus_NoError; |
| } |
| |
| #if MDNS_DEBUGMSGS |
| mDNSlocal void PrintRoutingSocketMsg(const struct ifa_msghdr *pRSMsg) |
| { |
| const char *kRSMsgTypes[] = { "", "RTM_ADD", "RTM_DELETE", "RTM_CHANGE", "RTM_GET", "RTM_LOSING", |
| "RTM_REDIRECT", "RTM_MISS", "RTM_LOCK", "RTM_OLDADD", "RTM_OLDDEL", "RTM_RESOLVE", |
| "RTM_NEWADDR", "RTM_DELADDR", "RTM_IFINFO", "RTM_NEWMADDR", "RTM_DELMADDR" }; |
| |
| int index = pRSMsg->ifam_type == RTM_IFINFO ? ((struct if_msghdr*) pRSMsg)->ifm_index : pRSMsg->ifam_index; |
| |
| printf("ifa_msghdr len=%d, type=%s, index=%d\n", pRSMsg->ifam_msglen, kRSMsgTypes[pRSMsg->ifam_type], index); |
| } |
| #endif |
| |
| mDNSlocal mDNSu32 ProcessRoutingNotification(int sd) |
| // Read through the messages on sd and if any indicate that any interface records should |
| // be torn down and rebuilt, return affected indices as a bitmask. Otherwise return 0. |
| { |
| ssize_t readCount; |
| char buff[4096]; |
| struct ifa_msghdr *pRSMsg = (struct ifa_msghdr*) buff; |
| mDNSu32 result = 0; |
| |
| readCount = read(sd, buff, sizeof buff); |
| if (readCount < (ssize_t) sizeof(struct ifa_msghdr)) |
| return mStatus_UnsupportedErr; // cannot decipher message |
| |
| #if MDNS_DEBUGMSGS |
| PrintRoutingSocketMsg(pRSMsg); |
| #endif |
| |
| // Process the message |
| if (pRSMsg->ifam_type == RTM_NEWADDR || pRSMsg->ifam_type == RTM_DELADDR || |
| pRSMsg->ifam_type == RTM_IFINFO) |
| { |
| if (pRSMsg->ifam_type == RTM_IFINFO) |
| result |= 1 << ((struct if_msghdr*) pRSMsg)->ifm_index; |
| else |
| result |= 1 << pRSMsg->ifam_index; |
| } |
| |
| return result; |
| } |
| |
| #endif // USES_NETLINK |
| |
| // Called when data appears on interface change notification socket |
| mDNSlocal void InterfaceChangeCallback(int fd, void *context) |
| { |
| IfChangeRec *pChgRec = (IfChangeRec*) context; |
| fd_set readFDs; |
| mDNSu32 changedInterfaces = 0; |
| struct timeval zeroTimeout = { 0, 0 }; |
| |
| (void)fd; // Unused |
| |
| FD_ZERO(&readFDs); |
| FD_SET(pChgRec->NotifySD, &readFDs); |
| |
| do |
| { |
| changedInterfaces |= ProcessRoutingNotification(pChgRec->NotifySD); |
| } |
| while (0 < select(pChgRec->NotifySD + 1, &readFDs, (fd_set*) NULL, (fd_set*) NULL, &zeroTimeout)); |
| |
| // Currently we rebuild the entire interface list whenever any interface change is |
| // detected. If this ever proves to be a performance issue in a multi-homed |
| // configuration, more care should be paid to changedInterfaces. |
| if (changedInterfaces) |
| mDNSPlatformPosixRefreshInterfaceList(pChgRec->mDNS); |
| } |
| |
| // Register with either a Routing Socket or RtNetLink to listen for interface changes. |
| mDNSlocal mStatus WatchForInterfaceChange(mDNS *const m) |
| { |
| mStatus err; |
| IfChangeRec *pChgRec; |
| |
| pChgRec = (IfChangeRec*) mDNSPlatformMemAllocateClear(sizeof *pChgRec); |
| if (pChgRec == NULL) |
| return mStatus_NoMemoryErr; |
| |
| pChgRec->mDNS = m; |
| err = OpenIfNotifySocket(&pChgRec->NotifySD); |
| if (err == 0) |
| err = mDNSPosixAddFDToEventLoop(pChgRec->NotifySD, InterfaceChangeCallback, pChgRec); |
| if (err) |
| mDNSPlatformMemFree(pChgRec); |
| |
| return err; |
| } |
| |
| // Test to see if we're the first client running on UDP port 5353, by trying to bind to 5353 without using SO_REUSEPORT. |
| // If we fail, someone else got here first. That's not a big problem; we can share the port for multicast responses -- |
| // we just need to be aware that we shouldn't expect to successfully receive unicast UDP responses. |
| mDNSlocal mDNSBool mDNSPlatformInit_CanReceiveUnicast(void) |
| { |
| int err; |
| int s = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); |
| struct sockaddr_in s5353; |
| s5353.sin_family = AF_INET; |
| s5353.sin_port = MulticastDNSPort.NotAnInteger; |
| s5353.sin_addr.s_addr = 0; |
| err = bind(s, (struct sockaddr *)&s5353, sizeof(s5353)); |
| close(s); |
| if (err) debugf("No unicast UDP responses"); |
| else debugf("Unicast UDP responses okay"); |
| return(err == 0); |
| } |
| |
| // mDNS core calls this routine to initialise the platform-specific data. |
| mDNSexport mStatus mDNSPlatformInit(mDNS *const m) |
| { |
| int err = 0; |
| struct sockaddr sa; |
| assert(m != NULL); |
| |
| if (mDNSPlatformInit_CanReceiveUnicast()) m->CanReceiveUnicastOn5353 = mDNStrue; |
| |
| // Tell mDNS core the names of this machine. |
| |
| // Set up the nice label |
| m->nicelabel.c[0] = 0; |
| GetUserSpecifiedFriendlyComputerName(&m->nicelabel); |
| if (m->nicelabel.c[0] == 0) MakeDomainLabelFromLiteralString(&m->nicelabel, "Computer"); |
| |
| // Set up the RFC 1034-compliant label |
| m->hostlabel.c[0] = 0; |
| GetUserSpecifiedRFC1034ComputerName(&m->hostlabel); |
| if (m->hostlabel.c[0] == 0) MakeDomainLabelFromLiteralString(&m->hostlabel, "Computer"); |
| |
| mDNS_SetFQDN(m); |
| |
| sa.sa_family = AF_INET; |
| m->p->unicastSocket4 = -1; |
| if (err == mStatus_NoError) err = SetupSocket(&sa, zeroIPPort, 0, &m->p->unicastSocket4); |
| #if HAVE_IPV6 |
| sa.sa_family = AF_INET6; |
| m->p->unicastSocket6 = -1; |
| if (err == mStatus_NoError) err = SetupSocket(&sa, zeroIPPort, 0, &m->p->unicastSocket6); |
| #endif |
| |
| // Tell mDNS core about the network interfaces on this machine. |
| if (err == mStatus_NoError) err = SetupInterfaceList(m); |
| |
| // Tell mDNS core about DNS Servers |
| mDNS_Lock(m); |
| if (err == mStatus_NoError) ParseDNSServers(m, uDNS_SERVERS_FILE); |
| mDNS_Unlock(m); |
| |
| if (err == mStatus_NoError) |
| { |
| err = WatchForInterfaceChange(m); |
| // Failure to observe interface changes is non-fatal. |
| if (err != mStatus_NoError) |
| { |
| fprintf(stderr, "mDNS(%d) WARNING: Unable to detect interface changes (%d).\n", getpid(), err); |
| err = mStatus_NoError; |
| } |
| } |
| |
| #if POSIX_HAS_TLS |
| // Use the SRP TLS shim. |
| mDNSPosixTLSInit(); |
| #endif |
| |
| // We don't do asynchronous initialization on the Posix platform, so by the time |
| // we get here the setup will already have succeeded or failed. If it succeeded, |
| // we should just call mDNSCoreInitComplete() immediately. |
| if (err == mStatus_NoError) |
| mDNSCoreInitComplete(m, mStatus_NoError); |
| |
| return PosixErrorToStatus(err); |
| } |
| |
| // mDNS core calls this routine to clean up the platform-specific data. |
| // In our case all we need to do is to tear down every network interface. |
| mDNSexport void mDNSPlatformClose(mDNS *const m) |
| { |
| int rv; |
| assert(m != NULL); |
| ClearInterfaceList(m); |
| if (m->p->unicastSocket4 != -1) |
| { |
| rv = close(m->p->unicastSocket4); |
| assert(rv == 0); |
| } |
| #if HAVE_IPV6 |
| if (m->p->unicastSocket6 != -1) |
| { |
| rv = close(m->p->unicastSocket6); |
| assert(rv == 0); |
| } |
| #endif |
| } |
| |
| // This is used internally by InterfaceChangeCallback. |
| // It's also exported so that the Standalone Responder (mDNSResponderPosix) |
| // can call it in response to a SIGHUP (mainly for debugging purposes). |
| mDNSexport mStatus mDNSPlatformPosixRefreshInterfaceList(mDNS *const m) |
| { |
| int err; |
| // This is a pretty heavyweight way to process interface changes -- |
| // destroying the entire interface list and then making fresh one from scratch. |
| // We should make it like the OS X version, which leaves unchanged interfaces alone. |
| ClearInterfaceList(m); |
| err = SetupInterfaceList(m); |
| return PosixErrorToStatus(err); |
| } |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark ***** Locking |
| #endif |
| |
| // On the Posix platform, locking is a no-op because we only ever enter |
| // mDNS core on the main thread. |
| |
| // mDNS core calls this routine when it wants to prevent |
| // the platform from reentering mDNS core code. |
| mDNSexport void mDNSPlatformLock (const mDNS *const m) |
| { |
| (void) m; // Unused |
| } |
| |
| // mDNS core calls this routine when it release the lock taken by |
| // mDNSPlatformLock and allow the platform to reenter mDNS core code. |
| mDNSexport void mDNSPlatformUnlock (const mDNS *const m) |
| { |
| (void) m; // Unused |
| } |
| |
| #if COMPILER_LIKES_PRAGMA_MARK |
| #pragma mark ***** Strings |
| #endif |
| |
| mDNSexport void mDNSPlatformStrLCopy(void *dst, const void *src, mDNSu32 len) |
| { |
| mdns_strlcpy((char *)dst, (const char *)src, len); |
| } |
| |
| // mDNS core calls this routine to get the length of a C string. |
| // On the Posix platform this maps directly to the ANSI C strlen. |
| mDNSexport mDNSu32 mDNSPlatformStrLen (const void *src) |
| { |
| return strlen((const char*)src); |
| } |
| |
| // mDNS core calls this routine to copy memory. |
| // On the Posix platform this maps directly to the ANSI C memcpy. |
| mDNSexport void mDNSPlatformMemCopy(void *dst, const void *src, mDNSu32 len) |
| { |
| memcpy(dst, src, len); |
| } |
| |
| // mDNS core calls this routine to test whether blocks of memory are byte-for-byte |
| // identical. On the Posix platform this is a simple wrapper around ANSI C memcmp. |
| mDNSexport mDNSBool mDNSPlatformMemSame(const void *dst, const void *src, mDNSu32 len) |
| { |
| return memcmp(dst, src, len) == 0; |
| } |
| |
| // If the caller wants to know the exact return of memcmp, then use this instead |
| // of mDNSPlatformMemSame |
| mDNSexport int mDNSPlatformMemCmp(const void *dst, const void *src, mDNSu32 len) |
| { |
| return (memcmp(dst, src, len)); |
| } |
| |
| mDNSexport void mDNSPlatformQsort(void *base, int nel, int width, int (*compar)(const void *, const void *)) |
| { |
| return (qsort(base, nel, width, compar)); |
| } |
| |
| // Proxy stub functions |
| mDNSexport mDNSu8 *DNSProxySetAttributes(DNSQuestion *q, DNSMessageHeader *h, DNSMessage *msg, mDNSu8 *ptr, mDNSu8 *limit) |
| { |
| (void) q; |
| (void) h; |
| (void) msg; |
| (void) ptr; |
| (void) limit; |
| |
| return ptr; |
| } |
| |
| // mDNS core calls this routine to clear blocks of memory. |
| // On the Posix platform this is a simple wrapper around ANSI C memset. |
| mDNSexport void mDNSPlatformMemZero(void *dst, mDNSu32 len) |
| { |
| memset(dst, 0, len); |
| } |
| |
| #if !MDNS_MALLOC_DEBUGGING |
| mDNSexport void *mDNSPlatformMemAllocate(mDNSu32 len) { return(mallocL("mDNSPlatformMemAllocate", len)); } |
| mDNSexport void *mDNSPlatformMemAllocateClear(mDNSu32 len) { return(callocL("mDNSPlatformMemAllocateClear", len)); } |
| mDNSexport void mDNSPlatformMemFree (void *mem) { freeL("mDNSPlatformMemFree", mem); } |
| #endif |
| |
| #if _PLATFORM_HAS_STRONG_PRNG_ |
| mDNSexport mDNSu32 mDNSPlatformRandomNumber(void) |
| { |
| return(arc4random()); |
| } |
| #else |
| mDNSexport mDNSu32 mDNSPlatformRandomSeed(void) |
| { |
| struct timeval tv; |
| gettimeofday(&tv, NULL); |
| return(tv.tv_usec); |
| } |
| #endif |
| |
| mDNSexport mDNSs32 mDNSPlatformOneSecond = 1024; |
| |
| mDNSexport mStatus mDNSPlatformTimeInit(void) |
| { |
| // No special setup is required on Posix -- we just use gettimeofday(); |
| // This is not really safe, because gettimeofday can go backwards if the user manually changes the date or time |
| // We should find a better way to do this |
| return(mStatus_NoError); |
| } |
| |
| mDNSexport mDNSs32 mDNSPlatformRawTime(void) |
| { |
| struct timespec tm; |
| int ret = clock_gettime(CLOCK_MONOTONIC, &tm); |
| assert(ret == 0); // This call will only fail if the number of seconds does not fit in an object of type time_t. |
| |
| // tm.tv_sec is seconds since some unspecified starting point (it is usually the system start up time) |
| // tm.tv_nsec is nanoseconds since the start of this second (i.e. values 0 to 999999999) |
| // We use the lower 22 bits of tm.tv_sec for the top 22 bits of our result |
| // and we multiply tm.tv_nsec by 2 / 1953125 to get a value in the range 0-1023 to go in the bottom 10 bits. |
| // This gives us a proper modular (cyclic) counter that has a resolution of roughly 1ms (actually 1/1024 second) |
| // and correctly cycles every 2^22 seconds (4194304 seconds = approx 48 days). |
| |
| return (mDNSs32)(((tm.tv_sec << 10) | (tm.tv_nsec * 2 / 1953125))); |
| } |
| |
| mDNSexport mDNSs32 mDNSPlatformUTC(void) |
| { |
| return time(NULL); |
| } |
| |
| // This should return elapsed time in seconds since boot. Posix doesn't have an API for this, so we currently assume |
| // that time() doesn't get adjusted, which isn't the case. |
| mDNSexport mDNSs32 mDNSPlatformContinuousTimeSeconds(void) |
| { |
| #ifdef CLOCK_BOOTTIME |
| // CLOCK_BOOTTIME is a Linux-specific constant that indicates a monotonic time that includes time asleep |
| const int clockid = CLOCK_BOOTTIME; |
| #else |
| // On MacOS, CLOCK_MONOTONIC is a monotonic time that includes time asleep. However, this may not be the case |
| // on other Posix systems, since the POSIX specification doesn't say one way or the other. E.g. on Linux |
| // time asleep is not accounted for, which is why we prefer CLOCK_BOOTTIME on Linux. |
| const int clockid = CLOCK_MONOTONIC; |
| #endif |
| struct timespec tm; |
| int ret = clock_gettime(clockid, &tm); |
| assert(ret == 0); // This call will only fail if the number of seconds does not fit in an object of type time_t. |
| |
| // We are only accurate to the second. |
| return (mDNSs32)tm.tv_sec; |
| } |
| |
| mDNSexport void mDNSPlatformSendWakeupPacket(mDNSInterfaceID InterfaceID, char *EthAddr, char *IPAddr, int iteration) |
| { |
| (void) InterfaceID; |
| (void) EthAddr; |
| (void) IPAddr; |
| (void) iteration; |
| } |
| |
| mDNSexport mDNSBool mDNSPlatformValidRecordForInterface(const AuthRecord *rr, mDNSInterfaceID InterfaceID) |
| { |
| (void) rr; |
| (void) InterfaceID; |
| |
| return 1; |
| } |
| |
| mDNSexport mDNSBool mDNSPlatformValidQuestionForInterface(DNSQuestion *q, const NetworkInterfaceInfo *intf) |
| { |
| (void) q; |
| (void) intf; |
| |
| return 1; |
| } |
| |
| // Used for debugging purposes. For now, just set the buffer to zero |
| mDNSexport void mDNSPlatformFormatTime(unsigned long te, mDNSu8 *buf, int bufsize) |
| { |
| (void) te; |
| if (bufsize) buf[0] = 0; |
| } |
| |
| mDNSexport void mDNSPlatformSendKeepalive(mDNSAddr *sadd, mDNSAddr *dadd, mDNSIPPort *lport, mDNSIPPort *rport, mDNSu32 seq, mDNSu32 ack, mDNSu16 win) |
| { |
| (void) sadd; // Unused |
| (void) dadd; // Unused |
| (void) lport; // Unused |
| (void) rport; // Unused |
| (void) seq; // Unused |
| (void) ack; // Unused |
| (void) win; // Unused |
| } |
| |
| mDNSexport mStatus mDNSPlatformRetrieveTCPInfo(mDNSAddr *laddr, mDNSIPPort *lport, mDNSAddr *raddr, mDNSIPPort *rport, mDNSTCPInfo *mti) |
| { |
| (void) laddr; // Unused |
| (void) raddr; // Unused |
| (void) lport; // Unused |
| (void) rport; // Unused |
| (void) mti; // Unused |
| |
| return mStatus_NoError; |
| } |
| |
| mDNSexport mStatus mDNSPlatformGetRemoteMacAddr(mDNSAddr *raddr) |
| { |
| (void) raddr; // Unused |
| |
| return mStatus_NoError; |
| } |
| |
| mDNSexport mStatus mDNSPlatformStoreSPSMACAddr(mDNSAddr *spsaddr, char *ifname) |
| { |
| (void) spsaddr; // Unused |
| (void) ifname; // Unused |
| |
| return mStatus_NoError; |
| } |
| |
| mDNSexport mStatus mDNSPlatformClearSPSData(void) |
| { |
| return mStatus_NoError; |
| } |
| |
| mDNSexport mStatus mDNSPlatformStoreOwnerOptRecord(char *ifname, DNSMessage *msg, int length) |
| { |
| (void) ifname; // Unused |
| (void) msg; // Unused |
| (void) length; // Unused |
| return mStatus_UnsupportedErr; |
| } |
| |
| mDNSexport mDNSu16 mDNSPlatformGetUDPPort(UDPSocket *sock) |
| { |
| (void) sock; // unused |
| |
| return (mDNSu16)-1; |
| } |
| |
| mDNSexport mDNSBool mDNSPlatformInterfaceIsD2D(mDNSInterfaceID InterfaceID) |
| { |
| (void) InterfaceID; // unused |
| |
| return mDNSfalse; |
| } |
| |
| mDNSexport void mDNSPlatformSetSocktOpt(void *sock, mDNSTransport_Type transType, mDNSAddr_Type addrType, const DNSQuestion *q) |
| { |
| (void) sock; |
| (void) transType; |
| (void) addrType; |
| (void) q; |
| } |
| |
| mDNSexport mDNSs32 mDNSPlatformGetPID() |
| { |
| return 0; |
| } |
| |
| mDNSlocal void mDNSPosixAddToFDSet(int *nfds, fd_set *readfds, int s) |
| { |
| if (*nfds < s + 1) *nfds = s + 1; |
| FD_SET(s, readfds); |
| } |
| |
| mDNSlocal void mDNSPosixGetFDSetForSelect(mDNS *m, int *nfds, fd_set *readfds, fd_set *writefds) |
| { |
| int numFDs = *nfds; |
| PosixEventSource *iSource; |
| |
| // 2. Build our list of active file descriptors |
| PosixNetworkInterface *info = (PosixNetworkInterface *)(m->HostInterfaces); |
| if (m->p->unicastSocket4 != -1) mDNSPosixAddToFDSet(&numFDs, readfds, m->p->unicastSocket4); |
| #if HAVE_IPV6 |
| if (m->p->unicastSocket6 != -1) mDNSPosixAddToFDSet(&numFDs, readfds, m->p->unicastSocket6); |
| #endif |
| while (info) |
| { |
| if (info->multicastSocket4 != -1) mDNSPosixAddToFDSet(&numFDs, readfds, info->multicastSocket4); |
| #if HAVE_IPV6 |
| if (info->multicastSocket6 != -1) mDNSPosixAddToFDSet(&numFDs, readfds, info->multicastSocket6); |
| #endif |
| info = (PosixNetworkInterface *)(info->coreIntf.next); |
| } |
| |
| // Copy over the event fds. We have to do it this way because client-provided event loops expect |
| // to initialize their FD sets first and then call mDNSPosixGetFDSet() |
| for (iSource = gEventSources; iSource; iSource = iSource->next) |
| { |
| if (iSource->readCallback != NULL) |
| FD_SET(iSource->fd, readfds); |
| if (iSource->writeCallback != NULL) |
| FD_SET(iSource->fd, writefds); |
| if (numFDs <= iSource->fd) |
| numFDs = iSource->fd + 1; |
| } |
| *nfds = numFDs; |
| } |
| |
| mDNSexport void mDNSPosixGetFDSet(mDNS *m, int *nfds, fd_set *readfds, fd_set *writefds, struct timeval *timeout) |
| { |
| mDNSs32 ticks; |
| struct timeval interval; |
| |
| // 1. Call mDNS_Execute() to let mDNSCore do what it needs to do |
| mDNSs32 nextevent = mDNS_Execute(m); |
| |
| // 3. Calculate the time remaining to the next scheduled event (in struct timeval format) |
| ticks = nextevent - mDNS_TimeNow(m); |
| if (ticks < 1) ticks = 1; |
| interval.tv_sec = ticks >> 10; // The high 22 bits are seconds |
| interval.tv_usec = ((ticks & 0x3FF) * 15625) / 16; // The low 10 bits are 1024ths |
| |
| // 4. If client's proposed timeout is more than what we want, then reduce it |
| if (timeout->tv_sec > interval.tv_sec || |
| (timeout->tv_sec == interval.tv_sec && timeout->tv_usec > interval.tv_usec)) |
| *timeout = interval; |
| |
| mDNSPosixGetFDSetForSelect(m, nfds, readfds, writefds); |
| } |
| |
| mDNSexport void mDNSPosixProcessFDSet(mDNS *const m, fd_set *readfds, fd_set *writefds) |
| { |
| PosixNetworkInterface *info; |
| PosixEventSource *iSource; |
| assert(m != NULL); |
| assert(readfds != NULL); |
| info = (PosixNetworkInterface *)(m->HostInterfaces); |
| |
| if (m->p->unicastSocket4 != -1 && FD_ISSET(m->p->unicastSocket4, readfds)) |
| { |
| FD_CLR(m->p->unicastSocket4, readfds); |
| SocketDataReady(m, NULL, m->p->unicastSocket4, NULL); |
| } |
| #if HAVE_IPV6 |
| if (m->p->unicastSocket6 != -1 && FD_ISSET(m->p->unicastSocket6, readfds)) |
| { |
| FD_CLR(m->p->unicastSocket6, readfds); |
| SocketDataReady(m, NULL, m->p->unicastSocket6, NULL); |
| } |
| #endif |
| |
| while (info) |
| { |
| if (info->multicastSocket4 != -1 && FD_ISSET(info->multicastSocket4, readfds)) |
| { |
| FD_CLR(info->multicastSocket4, readfds); |
| SocketDataReady(m, info, info->multicastSocket4, NULL); |
| } |
| #if HAVE_IPV6 |
| if (info->multicastSocket6 != -1 && FD_ISSET(info->multicastSocket6, readfds)) |
| { |
| FD_CLR(info->multicastSocket6, readfds); |
| SocketDataReady(m, info, info->multicastSocket6, NULL); |
| } |
| #endif |
| info = (PosixNetworkInterface *)(info->coreIntf.next); |
| } |
| |
| // Now process routing socket events, discovery relay events and anything else of that ilk. |
| for (iSource = gEventSources; iSource; iSource = iSource->next) |
| { |
| if (iSource->readCallback != NULL && FD_ISSET(iSource->fd, readfds)) |
| { |
| iSource->readCallback(iSource->fd, iSource->readContext); |
| break; // in case callback removed elements from gEventSources |
| } |
| else if (iSource->writeCallback != NULL && FD_ISSET(iSource->fd, writefds)) |
| { |
| mDNSPosixEventCallback writeCallback = iSource->writeCallback; |
| // Write events are one-shot: to get another event, the consumer has to put in a new request. |
| // We reset this before calling the callback just in case the callback requests another write |
| // callback, or deletes the event context from the list. |
| iSource->writeCallback = NULL; |
| writeCallback(iSource->fd, iSource->writeContext); |
| break; // in case callback removed elements from gEventSources |
| } |
| } |
| } |
| |
| mDNSu32 mDNSPlatformEventContextSize = sizeof (PosixEventSource); |
| |
| mDNSlocal void requestIOEvents(PosixEventSource *newSource, const char *taskName, |
| mDNSPosixEventCallback callback, void *context, int flag) |
| { |
| PosixEventSource **epp = &gEventSources; |
| |
| if (newSource->fd >= (int) FD_SETSIZE || newSource->fd < 0) |
| { |
| LogMsg("requestIOEvents called with fd %d > FD_SETSIZE %d.", newSource->fd, FD_SETSIZE); |
| assert(0); |
| } |
| if (callback == NULL) |
| { |
| LogMsg("requestIOEvents called no callback.", newSource->fd, FD_SETSIZE); |
| assert(0); |
| } |
| |
| // See if this event context is already on the list; if it is, no need to scan the list. |
| if (!(newSource->flags & PosixEventFlag_OnList)) |
| { |
| while (*epp) |
| { |
| // This should never happen. |
| if (newSource == *epp) |
| { |
| LogMsg("Event context marked not on list but is on list."); |
| assert(0); |
| } |
| epp = &(*epp)->next; |
| } |
| if (*epp == NULL) |
| { |
| *epp = newSource; |
| newSource->next = NULL; |
| newSource->flags = PosixEventFlag_OnList; |
| } |
| } |
| |
| if (flag & PosixEventFlag_Read) |
| { |
| newSource->readCallback = callback; |
| newSource->readContext = context; |
| newSource->flags |= PosixEventFlag_Read; |
| newSource->readTaskName = taskName; |
| } |
| if (flag & PosixEventFlag_Write) |
| { |
| newSource->writeCallback = callback; |
| newSource->writeContext = context; |
| newSource->flags |= PosixEventFlag_Write; |
| newSource->writeTaskName = taskName; |
| } |
| } |
| |
| mDNSlocal void requestReadEvents(PosixEventSource *eventSource, |
| const char *taskName, mDNSPosixEventCallback callback, void *context) |
| { |
| requestIOEvents(eventSource, taskName, callback, context, PosixEventFlag_Read); |
| } |
| |
| mDNSlocal void requestWriteEvents(PosixEventSource *eventSource, |
| const char *taskName, mDNSPosixEventCallback callback, void *context) |
| { |
| requestIOEvents(eventSource, taskName, callback, context, PosixEventFlag_Write); |
| } |
| |
| // Remove a file descriptor from the set that mDNSPosixRunEventLoopOnce() listens to. |
| mDNSlocal mStatus stopReadOrWriteEvents(int fd, mDNSBool freeContext, mDNSBool removeContext, int flags) |
| { |
| PosixEventSource *iSource, **epp = &gEventSources; |
| |
| while (*epp) |
| { |
| iSource = *epp; |
| if (fd == iSource->fd) |
| { |
| if (flags & PosixEventFlag_Read) |
| { |
| iSource->readCallback = NULL; |
| iSource->readContext = NULL; |
| } |
| if (flags & PosixEventFlag_Write) |
| { |
| iSource->writeCallback = NULL; |
| iSource->writeContext = NULL; |
| } |
| if (iSource->writeCallback == NULL && iSource->readCallback == NULL) |
| { |
| if (removeContext || freeContext) |
| *epp = iSource->next; |
| if (freeContext) |
| mdns_free(iSource); |
| } |
| return mStatus_NoError; |
| } |
| epp = &(*epp)->next; |
| } |
| return mStatus_NoSuchNameErr; |
| } |
| |
| // Some of the mDNSPosix client code relies on being able to add FDs to the event loop without |
| // providing storage for the event-related info. mDNSPosixAddFDToEventLoop and |
| // mDNSPosixRemoveFDFromEventLoop handle the event structure storage automatically. |
| mStatus mDNSPosixAddFDToEventLoop(int fd, mDNSPosixEventCallback callback, void *context) |
| { |
| PosixEventSource *newSource; |
| |
| newSource = (PosixEventSource*) mdns_malloc(sizeof *newSource); |
| if (NULL == newSource) |
| return mStatus_NoMemoryErr; |
| memset(newSource, 0, sizeof *newSource); |
| newSource->fd = fd; |
| |
| requestReadEvents(newSource, "mDNSPosixAddFDToEventLoop", callback, context); |
| return mStatus_NoError; |
| } |
| |
| mStatus mDNSPosixRemoveFDFromEventLoop(int fd) |
| { |
| return stopReadOrWriteEvents(fd, mDNStrue, mDNStrue, PosixEventFlag_Read | PosixEventFlag_Write); |
| } |
| |
| // Simply note the received signal in gEventSignals. |
| mDNSlocal void NoteSignal(int signum) |
| { |
| sigaddset(&gEventSignals, signum); |
| } |
| |
| // Tell the event package to listen for signal and report it in mDNSPosixRunEventLoopOnce(). |
| mStatus mDNSPosixListenForSignalInEventLoop(int signum) |
| { |
| struct sigaction action; |
| mStatus err; |
| |
| mDNSPlatformMemZero(&action, sizeof action); // more portable than member-wise assignment |
| action.sa_handler = NoteSignal; |
| err = sigaction(signum, &action, (struct sigaction*) NULL); |
| |
| sigaddset(&gEventSignalSet, signum); |
| |
| return err; |
| } |
| |
| // Tell the event package to stop listening for signal in mDNSPosixRunEventLoopOnce(). |
| mStatus mDNSPosixIgnoreSignalInEventLoop(int signum) |
| { |
| struct sigaction action; |
| mStatus err; |
| |
| mDNSPlatformMemZero(&action, sizeof action); // more portable than member-wise assignment |
| action.sa_handler = SIG_DFL; |
| err = sigaction(signum, &action, (struct sigaction*) NULL); |
| |
| sigdelset(&gEventSignalSet, signum); |
| |
| return err; |
| } |
| |
| // Do a single pass through the attendent event sources and dispatch any found to their callbacks. |
| // Return as soon as internal timeout expires, or a signal we're listening for is received. |
| mStatus mDNSPosixRunEventLoopOnce(mDNS *m, const struct timeval *pTimeout, |
| sigset_t *pSignalsReceived, mDNSBool *pDataDispatched) |
| { |
| fd_set listenFDs; |
| fd_set writeFDs; |
| int numFDs = 0, numReady; |
| struct timeval timeout = *pTimeout; |
| |
| // 1. Set up the fd_set as usual here. |
| // This example client has no file descriptors of its own, |
| // but a real application would call FD_SET to add them to the set here |
| FD_ZERO(&listenFDs); |
| FD_ZERO(&writeFDs); |
| |
| // 2. Set up the timeout. |
| // MainLoop has already called mDNS_Execute and udsserver_idle, so the timeout we |
| // were passed is already set up. |
| |
| // Include the sockets that are listening to the wire in our select() set |
| mDNSPosixGetFDSetForSelect(m, &numFDs, &listenFDs, &writeFDs); |
| numReady = select(numFDs, &listenFDs, &writeFDs, (fd_set*) NULL, &timeout); |
| |
| if (numReady > 0) |
| { |
| mDNSPosixProcessFDSet(m, &listenFDs, &writeFDs); |
| *pDataDispatched = mDNStrue; |
| } |
| else if (numReady < 0) |
| { |
| if (errno != EINTR) { |
| // This should never happen, represents a coding error, and is not recoverable, since |
| // we'll just sit here spinning and never receive another event. The usual reason for |
| // it to happen is that an FD was closed but not removed from the event list. |
| LogMsg("select failed: %s", strerror(errno)); |
| abort(); |
| } |
| } |
| else |
| *pDataDispatched = mDNSfalse; |
| |
| (void) sigprocmask(SIG_BLOCK, &gEventSignalSet, (sigset_t*) NULL); |
| *pSignalsReceived = gEventSignals; |
| sigemptyset(&gEventSignals); |
| (void) sigprocmask(SIG_UNBLOCK, &gEventSignalSet, (sigset_t*) NULL); |
| |
| return mStatus_NoError; |
| } |