diff options
Diffstat (limited to 'netwerk/base/nsServerSocket.cpp')
-rw-r--r-- | netwerk/base/nsServerSocket.cpp | 550 |
1 files changed, 550 insertions, 0 deletions
diff --git a/netwerk/base/nsServerSocket.cpp b/netwerk/base/nsServerSocket.cpp new file mode 100644 index 0000000000..e0ac43ebd9 --- /dev/null +++ b/netwerk/base/nsServerSocket.cpp @@ -0,0 +1,550 @@ +/* vim:set ts=2 sw=2 et cindent: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "nsSocketTransport2.h" +#include "nsServerSocket.h" +#include "nsProxyRelease.h" +#include "nsError.h" +#include "nsNetCID.h" +#include "prnetdb.h" +#include "prio.h" +#include "nsThreadUtils.h" +#include "mozilla/Attributes.h" +#include "mozilla/EndianUtils.h" +#include "mozilla/net/DNS.h" +#include "nsServiceManagerUtils.h" +#include "nsIFile.h" + +namespace mozilla { +namespace net { + +//----------------------------------------------------------------------------- + +typedef void (nsServerSocket::*nsServerSocketFunc)(void); + +static nsresult PostEvent(nsServerSocket* s, nsServerSocketFunc func) { + nsCOMPtr<nsIRunnable> ev = NewRunnableMethod("net::PostEvent", s, func); + if (!gSocketTransportService) return NS_ERROR_FAILURE; + + return gSocketTransportService->Dispatch(ev, NS_DISPATCH_NORMAL); +} + +//----------------------------------------------------------------------------- +// nsServerSocket +//----------------------------------------------------------------------------- + +nsServerSocket::nsServerSocket() + : mFD(nullptr), + mLock("nsServerSocket.mLock"), + mAttached(false), + mKeepWhenOffline(false) { + this->mAddr.raw.family = 0; + this->mAddr.inet.family = 0; + this->mAddr.inet.port = 0; + this->mAddr.inet.ip = 0; + this->mAddr.ipv6.family = 0; + this->mAddr.ipv6.port = 0; + this->mAddr.ipv6.flowinfo = 0; + this->mAddr.ipv6.scope_id = 0; + this->mAddr.local.family = 0; + // we want to be able to access the STS directly, and it may not have been + // constructed yet. the STS constructor sets gSocketTransportService. + if (!gSocketTransportService) { + // This call can fail if we're offline, for example. + nsCOMPtr<nsISocketTransportService> sts = + do_GetService(NS_SOCKETTRANSPORTSERVICE_CONTRACTID); + } + // make sure the STS sticks around as long as we do + NS_IF_ADDREF(gSocketTransportService); +} + +nsServerSocket::~nsServerSocket() { + Close(); // just in case :) + + // release our reference to the STS + nsSocketTransportService* serv = gSocketTransportService; + NS_IF_RELEASE(serv); +} + +void nsServerSocket::OnMsgClose() { + SOCKET_LOG(("nsServerSocket::OnMsgClose [this=%p]\n", this)); + + if (NS_FAILED(mCondition)) return; + + // tear down socket. this signals the STS to detach our socket handler. + mCondition = NS_BINDING_ABORTED; + + // if we are attached, then we'll close the socket in our OnSocketDetached. + // otherwise, call OnSocketDetached from here. + if (!mAttached) OnSocketDetached(mFD); +} + +void nsServerSocket::OnMsgAttach() { + SOCKET_LOG(("nsServerSocket::OnMsgAttach [this=%p]\n", this)); + + if (NS_FAILED(mCondition)) return; + + mCondition = TryAttach(); + + // if we hit an error while trying to attach then bail... + if (NS_FAILED(mCondition)) { + NS_ASSERTION(!mAttached, "should not be attached already"); + OnSocketDetached(mFD); + } +} + +nsresult nsServerSocket::TryAttach() { + nsresult rv; + + if (!gSocketTransportService) return NS_ERROR_FAILURE; + + // + // find out if it is going to be ok to attach another socket to the STS. + // if not then we have to wait for the STS to tell us that it is ok. + // the notification is asynchronous, which means that when we could be + // in a race to call AttachSocket once notified. for this reason, when + // we get notified, we just re-enter this function. as a result, we are + // sure to ask again before calling AttachSocket. in this way we deal + // with the race condition. though it isn't the most elegant solution, + // it is far simpler than trying to build a system that would guarantee + // FIFO ordering (which wouldn't even be that valuable IMO). see bug + // 194402 for more info. + // + if (!gSocketTransportService->CanAttachSocket()) { + nsCOMPtr<nsIRunnable> event = NewRunnableMethod( + "net::nsServerSocket::OnMsgAttach", this, &nsServerSocket::OnMsgAttach); + if (!event) return NS_ERROR_OUT_OF_MEMORY; + + nsresult rv = gSocketTransportService->NotifyWhenCanAttachSocket(event); + if (NS_FAILED(rv)) return rv; + } + + // + // ok, we can now attach our socket to the STS for polling + // + rv = gSocketTransportService->AttachSocket(mFD, this); + if (NS_FAILED(rv)) return rv; + + mAttached = true; + + // + // now, configure our poll flags for listening... + // + mPollFlags = (PR_POLL_READ | PR_POLL_EXCEPT); + return NS_OK; +} + +void nsServerSocket::CreateClientTransport(PRFileDesc* aClientFD, + const NetAddr& aClientAddr) { + RefPtr<nsSocketTransport> trans = new nsSocketTransport; + if (NS_WARN_IF(!trans)) { + mCondition = NS_ERROR_OUT_OF_MEMORY; + return; + } + + nsresult rv = trans->InitWithConnectedSocket(aClientFD, &aClientAddr); + if (NS_WARN_IF(NS_FAILED(rv))) { + mCondition = rv; + return; + } + + mListener->OnSocketAccepted(this, trans); +} + +//----------------------------------------------------------------------------- +// nsServerSocket::nsASocketHandler +//----------------------------------------------------------------------------- + +void nsServerSocket::OnSocketReady(PRFileDesc* fd, int16_t outFlags) { + NS_ASSERTION(NS_SUCCEEDED(mCondition), "oops"); + NS_ASSERTION(mFD == fd, "wrong file descriptor"); + NS_ASSERTION(outFlags != -1, "unexpected timeout condition reached"); + + if (outFlags & (PR_POLL_ERR | PR_POLL_HUP | PR_POLL_NVAL)) { + NS_WARNING("error polling on listening socket"); + mCondition = NS_ERROR_UNEXPECTED; + return; + } + + PRFileDesc* clientFD; + PRNetAddr prClientAddr; + + // NSPR doesn't tell us the peer address's length (as provided by the + // 'accept' system call), so we can't distinguish between named, + // unnamed, and abstract peer addresses. Clear prClientAddr first, so + // that the path will at least be reliably empty for unnamed and + // abstract addresses, and not garbage when the peer is unnamed. + memset(&prClientAddr, 0, sizeof(prClientAddr)); + + clientFD = PR_Accept(mFD, &prClientAddr, PR_INTERVAL_NO_WAIT); + if (!clientFD) { + NS_WARNING("PR_Accept failed"); + mCondition = NS_ERROR_UNEXPECTED; + return; + } + PR_SetFDInheritable(clientFD, false); + + NetAddr clientAddr(&prClientAddr); + // Accept succeeded, create socket transport and notify consumer + CreateClientTransport(clientFD, clientAddr); +} + +void nsServerSocket::OnSocketDetached(PRFileDesc* fd) { + // force a failure condition if none set; maybe the STS is shutting down :-/ + if (NS_SUCCEEDED(mCondition)) mCondition = NS_ERROR_ABORT; + + if (mFD) { + NS_ASSERTION(mFD == fd, "wrong file descriptor"); + PR_Close(mFD); + mFD = nullptr; + } + + if (mListener) { + mListener->OnStopListening(this, mCondition); + + // need to atomically clear mListener. see our Close() method. + RefPtr<nsIServerSocketListener> listener = nullptr; + { + MutexAutoLock lock(mLock); + listener = ToRefPtr(std::move(mListener)); + } + + // XXX we need to proxy the release to the listener's target thread to work + // around bug 337492. + if (listener) { + NS_ProxyRelease("nsServerSocket::mListener", mListenerTarget, + listener.forget()); + } + } +} + +void nsServerSocket::IsLocal(bool* aIsLocal) { +#if defined(XP_UNIX) + // Unix-domain sockets are always local. + if (mAddr.raw.family == PR_AF_LOCAL) { + *aIsLocal = true; + return; + } +#endif + + // If bound to loopback, this server socket only accepts local connections. + *aIsLocal = PR_IsNetAddrType(&mAddr, PR_IpAddrLoopback); +} + +void nsServerSocket::KeepWhenOffline(bool* aKeepWhenOffline) { + *aKeepWhenOffline = mKeepWhenOffline; +} + +//----------------------------------------------------------------------------- +// nsServerSocket::nsISupports +//----------------------------------------------------------------------------- + +NS_IMPL_ISUPPORTS(nsServerSocket, nsIServerSocket) + +//----------------------------------------------------------------------------- +// nsServerSocket::nsIServerSocket +//----------------------------------------------------------------------------- + +NS_IMETHODIMP +nsServerSocket::Init(int32_t aPort, bool aLoopbackOnly, int32_t aBackLog) { + return InitSpecialConnection(aPort, aLoopbackOnly ? LoopbackOnly : 0, + aBackLog); +} + +NS_IMETHODIMP +nsServerSocket::InitIPv6(int32_t aPort, bool aLoopbackOnly, int32_t aBackLog) { + PRNetAddrValue val; + PRNetAddr addr; + + if (aPort < 0) { + aPort = 0; + } + if (aLoopbackOnly) { + val = PR_IpAddrLoopback; + } else { + val = PR_IpAddrAny; + } + PR_SetNetAddr(val, PR_AF_INET6, aPort, &addr); + + mKeepWhenOffline = false; + return InitWithAddress(&addr, aBackLog); +} + +NS_IMETHODIMP +nsServerSocket::InitWithFilename(nsIFile* aPath, uint32_t aPermissions, + int32_t aBacklog) { +#if defined(XP_UNIX) + nsresult rv; + + nsAutoCString path; + rv = aPath->GetNativePath(path); + if (NS_FAILED(rv)) return rv; + + // Create a Unix domain PRNetAddr referring to the given path. + PRNetAddr addr; + if (path.Length() > sizeof(addr.local.path) - 1) + return NS_ERROR_FILE_NAME_TOO_LONG; + addr.local.family = PR_AF_LOCAL; + memcpy(addr.local.path, path.get(), path.Length()); + addr.local.path[path.Length()] = '\0'; + + rv = InitWithAddress(&addr, aBacklog); + if (NS_FAILED(rv)) return rv; + + return aPath->SetPermissions(aPermissions); +#else + return NS_ERROR_SOCKET_ADDRESS_NOT_SUPPORTED; +#endif +} + +NS_IMETHODIMP +nsServerSocket::InitWithAbstractAddress(const nsACString& aName, + int32_t aBacklog) { + // Abstract socket address is supported on Linux and Android only. + // If not Linux, we should return error. +#if defined(XP_LINUX) + // Create an abstract socket address PRNetAddr referring to the name + PRNetAddr addr; + if (aName.Length() > sizeof(addr.local.path) - 2) { + return NS_ERROR_FILE_NAME_TOO_LONG; + } + addr.local.family = PR_AF_LOCAL; + addr.local.path[0] = 0; + memcpy(addr.local.path + 1, aName.BeginReading(), aName.Length()); + addr.local.path[aName.Length() + 1] = 0; + + return InitWithAddress(&addr, aBacklog); +#else + return NS_ERROR_SOCKET_ADDRESS_NOT_SUPPORTED; +#endif +} + +NS_IMETHODIMP +nsServerSocket::InitSpecialConnection(int32_t aPort, nsServerSocketFlag aFlags, + int32_t aBackLog) { + PRNetAddrValue val; + PRNetAddr addr; + + if (aPort < 0) aPort = 0; + if (aFlags & nsIServerSocket::LoopbackOnly) + val = PR_IpAddrLoopback; + else + val = PR_IpAddrAny; + PR_SetNetAddr(val, PR_AF_INET, aPort, &addr); + + mKeepWhenOffline = ((aFlags & nsIServerSocket::KeepWhenOffline) != 0); + return InitWithAddress(&addr, aBackLog); +} + +NS_IMETHODIMP +nsServerSocket::InitWithAddress(const PRNetAddr* aAddr, int32_t aBackLog) { + NS_ENSURE_TRUE(mFD == nullptr, NS_ERROR_ALREADY_INITIALIZED); + nsresult rv; + + // + // configure listening socket... + // + + mFD = PR_OpenTCPSocket(aAddr->raw.family); + if (!mFD) { + NS_WARNING("unable to create server socket"); + return ErrorAccordingToNSPR(PR_GetError()); + } + + PR_SetFDInheritable(mFD, false); + + PRSocketOptionData opt; + + opt.option = PR_SockOpt_Reuseaddr; + opt.value.reuse_addr = true; + PR_SetSocketOption(mFD, &opt); + + opt.option = PR_SockOpt_Nonblocking; + opt.value.non_blocking = true; + PR_SetSocketOption(mFD, &opt); + + if (PR_Bind(mFD, aAddr) != PR_SUCCESS) { + NS_WARNING("failed to bind socket"); + goto fail; + } + + if (aBackLog < 0) aBackLog = 5; // seems like a reasonable default + + if (PR_Listen(mFD, aBackLog) != PR_SUCCESS) { + NS_WARNING("cannot listen on socket"); + goto fail; + } + + // get the resulting socket address, which may be different than what + // we passed to bind. + if (PR_GetSockName(mFD, &mAddr) != PR_SUCCESS) { + NS_WARNING("cannot get socket name"); + goto fail; + } + + // Set any additional socket defaults needed by child classes + rv = SetSocketDefaults(); + if (NS_WARN_IF(NS_FAILED(rv))) { + goto fail; + } + + // wait until AsyncListen is called before polling the socket for + // client connections. + return NS_OK; + +fail: + rv = ErrorAccordingToNSPR(PR_GetError()); + Close(); + return rv; +} + +NS_IMETHODIMP +nsServerSocket::Close() { + { + MutexAutoLock lock(mLock); + // we want to proxy the close operation to the socket thread if a listener + // has been set. otherwise, we should just close the socket here... + if (!mListener) { + if (mFD) { + PR_Close(mFD); + mFD = nullptr; + } + return NS_OK; + } + } + return PostEvent(this, &nsServerSocket::OnMsgClose); +} + +namespace { + +class ServerSocketListenerProxy final : public nsIServerSocketListener { + ~ServerSocketListenerProxy() = default; + + public: + explicit ServerSocketListenerProxy(nsIServerSocketListener* aListener) + : mListener(new nsMainThreadPtrHolder<nsIServerSocketListener>( + "ServerSocketListenerProxy::mListener", aListener)), + mTarget(GetCurrentEventTarget()) {} + + NS_DECL_THREADSAFE_ISUPPORTS + NS_DECL_NSISERVERSOCKETLISTENER + + class OnSocketAcceptedRunnable : public Runnable { + public: + OnSocketAcceptedRunnable( + const nsMainThreadPtrHandle<nsIServerSocketListener>& aListener, + nsIServerSocket* aServ, nsISocketTransport* aTransport) + : Runnable("net::ServerSocketListenerProxy::OnSocketAcceptedRunnable"), + mListener(aListener), + mServ(aServ), + mTransport(aTransport) {} + + NS_DECL_NSIRUNNABLE + + private: + nsMainThreadPtrHandle<nsIServerSocketListener> mListener; + nsCOMPtr<nsIServerSocket> mServ; + nsCOMPtr<nsISocketTransport> mTransport; + }; + + class OnStopListeningRunnable : public Runnable { + public: + OnStopListeningRunnable( + const nsMainThreadPtrHandle<nsIServerSocketListener>& aListener, + nsIServerSocket* aServ, nsresult aStatus) + : Runnable("net::ServerSocketListenerProxy::OnStopListeningRunnable"), + mListener(aListener), + mServ(aServ), + mStatus(aStatus) {} + + NS_DECL_NSIRUNNABLE + + private: + nsMainThreadPtrHandle<nsIServerSocketListener> mListener; + nsCOMPtr<nsIServerSocket> mServ; + nsresult mStatus; + }; + + private: + nsMainThreadPtrHandle<nsIServerSocketListener> mListener; + nsCOMPtr<nsIEventTarget> mTarget; +}; + +NS_IMPL_ISUPPORTS(ServerSocketListenerProxy, nsIServerSocketListener) + +NS_IMETHODIMP +ServerSocketListenerProxy::OnSocketAccepted(nsIServerSocket* aServ, + nsISocketTransport* aTransport) { + RefPtr<OnSocketAcceptedRunnable> r = + new OnSocketAcceptedRunnable(mListener, aServ, aTransport); + return mTarget->Dispatch(r, NS_DISPATCH_NORMAL); +} + +NS_IMETHODIMP +ServerSocketListenerProxy::OnStopListening(nsIServerSocket* aServ, + nsresult aStatus) { + RefPtr<OnStopListeningRunnable> r = + new OnStopListeningRunnable(mListener, aServ, aStatus); + return mTarget->Dispatch(r, NS_DISPATCH_NORMAL); +} + +NS_IMETHODIMP +ServerSocketListenerProxy::OnSocketAcceptedRunnable::Run() { + mListener->OnSocketAccepted(mServ, mTransport); + return NS_OK; +} + +NS_IMETHODIMP +ServerSocketListenerProxy::OnStopListeningRunnable::Run() { + mListener->OnStopListening(mServ, mStatus); + return NS_OK; +} + +} // namespace + +NS_IMETHODIMP +nsServerSocket::AsyncListen(nsIServerSocketListener* aListener) { + // ensuring mFD implies ensuring mLock + NS_ENSURE_TRUE(mFD, NS_ERROR_NOT_INITIALIZED); + NS_ENSURE_TRUE(mListener == nullptr, NS_ERROR_IN_PROGRESS); + { + MutexAutoLock lock(mLock); + mListener = new ServerSocketListenerProxy(aListener); + mListenerTarget = GetCurrentEventTarget(); + } + + // Child classes may need to do additional setup just before listening begins + nsresult rv = OnSocketListen(); + if (NS_WARN_IF(NS_FAILED(rv))) { + return rv; + } + + return PostEvent(this, &nsServerSocket::OnMsgAttach); +} + +NS_IMETHODIMP +nsServerSocket::GetPort(int32_t* aResult) { + // no need to enter the lock here + uint16_t port; + if (mAddr.raw.family == PR_AF_INET) + port = mAddr.inet.port; + else if (mAddr.raw.family == PR_AF_INET6) + port = mAddr.ipv6.port; + else + return NS_ERROR_FAILURE; + + *aResult = static_cast<int32_t>(NetworkEndian::readUint16(&port)); + return NS_OK; +} + +NS_IMETHODIMP +nsServerSocket::GetAddress(PRNetAddr* aResult) { + // no need to enter the lock here + memcpy(aResult, &mAddr, sizeof(mAddr)); + return NS_OK; +} + +} // namespace net +} // namespace mozilla |