From: Kahrl <kahrl@gmx.net>
Date: Thu, 29 Aug 2013 03:04:56 +0000 (+0200)
Subject: Implement httpfetch module and initialize it from main()
X-Git-Url: http://81.2.79.47:8989/gitweb/?a=commitdiff_plain;h=0ea3e6dbe2288854d9d4a971fc6539c2e740a95a;p=zefram%2Fminetest%2Fminetest_engine.git

Implement httpfetch module and initialize it from main()

Add curl_parallel_limit setting that will replace media_fetch_threads in
a later commit.

Fix a typo in MutexedQueue::pop_back() that made it impossible to compile
code that used this function. (Noticed this while implementing httpfetch.)
---

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index a30dc985..104e5640 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -188,7 +188,7 @@ message (STATUS "LuaJIT library: ${LUA_LIBRARY}")
 message (STATUS "LuaJIT headers: ${LUA_INCLUDE_DIR}")
 
 set(USE_LUAJIT 0)
-if(LUA_LIBRARY AND LUA_INCLUDE_DIR) 
+if(LUA_LIBRARY AND LUA_INCLUDE_DIR)
 	message (STATUS "LuaJIT found.")
 	set(USE_LUAJIT 1)
 else(LUA_LIBRARY AND LUA_INCLUDE_DIR)
@@ -307,6 +307,7 @@ set(common_SRCS
 	pathfinder.cpp
 	convert_json.cpp
 	gettext.cpp
+	httpfetch.cpp
 	${JTHREAD_SRCS}
 	${common_SCRIPT_SRCS}
 	${UTIL_SRCS}
@@ -500,7 +501,7 @@ if(MSVC)
 	# Flags for C files (sqlite)
 	# /MT = Link statically with standard library stuff
 	set(CMAKE_C_FLAGS_RELEASE "/O2 /Ob2 /MT")
-	
+
 	if(BUILD_SERVER)
 		set_target_properties(${PROJECT_NAME}server PROPERTIES
 				COMPILE_DEFINITIONS "SERVER")
@@ -508,13 +509,13 @@ if(MSVC)
 
 else()
 	# Probably GCC
-	
+
 	if(WARN_ALL)
 		set(RELEASE_WARNING_FLAGS "-Wall")
 	else()
 		set(RELEASE_WARNING_FLAGS "")
 	endif()
-	
+
 	if(NOT APPLE AND NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
 		CHECK_CXX_COMPILER_FLAG("-Wno-unused-but-set-variable" HAS_UNUSED_BUT_SET_VARIABLE_WARNING)
 		if(HAS_UNUSED_BUT_SET_VARIABLE_WARNING)
@@ -537,7 +538,7 @@ else()
 	if(USE_GPROF)
 		set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -pg")
 	endif()
-	
+
 	if(BUILD_SERVER)
 		set_target_properties(${PROJECT_NAME}server PROPERTIES
 				COMPILE_DEFINITIONS "SERVER")
diff --git a/src/defaultsettings.cpp b/src/defaultsettings.cpp
index 9b407b1c..662717c8 100644
--- a/src/defaultsettings.cpp
+++ b/src/defaultsettings.cpp
@@ -140,6 +140,7 @@ void set_default_settings(Settings *settings)
 	settings->setDefault("enable_particles", "true");
 
 	settings->setDefault("media_fetch_threads", "8");
+	settings->setDefault("curl_parallel_limit", "8");
 
 	settings->setDefault("serverlist_url", "servers.minetest.net");
 	settings->setDefault("serverlist_file", "favoriteservers.txt");
diff --git a/src/httpfetch.cpp b/src/httpfetch.cpp
new file mode 100644
index 00000000..4342a8b2
--- /dev/null
+++ b/src/httpfetch.cpp
@@ -0,0 +1,718 @@
+/*
+Minetest
+Copyright (C) 2013 celeron55, Perttu Ahola <celeron55@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU Lesser General Public License as published by
+the Free Software Foundation; either version 2.1 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU Lesser General Public License for more details.
+
+You should have received a copy of the GNU Lesser General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+*/
+
+#include "httpfetch.h"
+#include <iostream>
+#include <sstream>
+#include <list>
+#include <map>
+#include <errno.h>
+#include "jthread/jevent.h"
+#include "config.h"
+#include "exceptions.h"
+#include "debug.h"
+#include "log.h"
+#include "util/container.h"
+#include "util/thread.h"
+#include "socket.h" // for select()
+
+JMutex g_httpfetch_mutex;
+std::map<unsigned long, std::list<HTTPFetchResult> > g_httpfetch_results;
+
+static void httpfetch_deliver_result(const HTTPFetchResult &fetchresult)
+{
+	unsigned long caller = fetchresult.caller;
+	if (caller != HTTPFETCH_DISCARD) {
+		JMutexAutoLock lock(g_httpfetch_mutex);
+		g_httpfetch_results[caller].push_back(fetchresult);
+	}
+}
+
+static void httpfetch_request_clear(unsigned long caller);
+
+unsigned long httpfetch_caller_alloc()
+{
+	JMutexAutoLock lock(g_httpfetch_mutex);
+
+	// Check each caller ID except HTTPFETCH_DISCARD
+	const unsigned long discard = HTTPFETCH_DISCARD;
+	for (unsigned long caller = discard + 1; caller != discard; ++caller) {
+		std::map<unsigned long, std::list<HTTPFetchResult> >::iterator
+			it = g_httpfetch_results.find(caller);
+		if (it == g_httpfetch_results.end()) {
+			verbosestream<<"httpfetch_caller_alloc: allocating "
+					<<caller<<std::endl;
+			// Access element to create it
+			g_httpfetch_results[caller];
+			return caller;
+		}
+	}
+
+	assert("httpfetch_caller_alloc: ran out of caller IDs" == 0);
+	return discard;
+}
+
+void httpfetch_caller_free(unsigned long caller)
+{
+	verbosestream<<"httpfetch_caller_free: freeing "
+			<<caller<<std::endl;
+
+	httpfetch_request_clear(caller);
+	if (caller != HTTPFETCH_DISCARD) {
+		JMutexAutoLock lock(g_httpfetch_mutex);
+		g_httpfetch_results.erase(caller);
+	}
+}
+
+bool httpfetch_async_get(unsigned long caller, HTTPFetchResult &fetchresult)
+{
+	JMutexAutoLock lock(g_httpfetch_mutex);
+
+	// Check that caller exists
+	std::map<unsigned long, std::list<HTTPFetchResult> >::iterator
+		it = g_httpfetch_results.find(caller);
+	if (it == g_httpfetch_results.end())
+		return false;
+
+	// Check that result queue is nonempty
+	std::list<HTTPFetchResult> &callerresults = it->second;
+	if (callerresults.empty())
+		return false;
+
+	// Pop first result
+	fetchresult = callerresults.front();
+	callerresults.pop_front();
+	return true;
+}
+
+#if USE_CURL
+#include <curl/curl.h>
+
+/*
+	USE_CURL is on: use cURL based httpfetch implementation
+*/
+
+static size_t httpfetch_writefunction(
+		char *ptr, size_t size, size_t nmemb, void *userdata)
+{
+	std::ostringstream *stream = (std::ostringstream*)userdata;
+	size_t count = size * nmemb;
+	stream->write(ptr, count);
+	return count;
+}
+
+static size_t httpfetch_discardfunction(
+		char *ptr, size_t size, size_t nmemb, void *userdata)
+{
+	return size * nmemb;
+}
+
+class CurlHandlePool
+{
+	std::list<CURL*> handles;
+
+public:
+	CurlHandlePool() {}
+	~CurlHandlePool()
+	{
+		for (std::list<CURL*>::iterator it = handles.begin();
+				it != handles.end(); ++it) {
+			curl_easy_cleanup(*it);
+		}
+	}
+	CURL * alloc()
+	{
+		CURL *curl;
+		if (handles.empty()) {
+			curl = curl_easy_init();
+			if (curl == NULL) {
+				errorstream<<"curl_easy_init returned NULL"<<std::endl;
+			}
+		}
+		else {
+			curl = handles.front();
+			handles.pop_front();
+		}
+		return curl;
+	}
+	void free(CURL *handle)
+	{
+		if (handle)
+			handles.push_back(handle);
+	}
+};
+
+struct HTTPFetchOngoing
+{
+	CurlHandlePool *pool;
+	CURL *curl;
+	CURLM *multi;
+	HTTPFetchRequest request;
+	HTTPFetchResult result;
+	std::ostringstream oss;
+	char *post_fields;
+	struct curl_slist *httpheader;
+
+	HTTPFetchOngoing(HTTPFetchRequest request_, CurlHandlePool *pool_):
+		pool(pool_),
+		curl(NULL),
+		multi(NULL),
+		request(request_),
+		result(request_),
+		oss(std::ios::binary),
+		httpheader(NULL)
+	{
+		curl = pool->alloc();
+		if (curl != NULL) {
+			// Set static cURL options
+			curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1);
+			curl_easy_setopt(curl, CURLOPT_FAILONERROR, 1);
+			curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1);
+			curl_easy_setopt(curl, CURLOPT_MAXREDIRS, 1);
+
+#if LIBCURL_VERSION_NUM >= 0x071304
+			// Restrict protocols so that curl vulnerabilities in
+			// other protocols don't affect us.
+			// These settings were introduced in curl 7.19.4.
+			long protocols =
+				CURLPROTO_HTTP |
+				CURLPROTO_HTTPS |
+				CURLPROTO_FTP |
+				CURLPROTO_FTPS;
+			curl_easy_setopt(curl, CURLOPT_PROTOCOLS, protocols);
+			curl_easy_setopt(curl, CURLOPT_REDIR_PROTOCOLS, protocols);
+#endif
+
+			// Set cURL options based on HTTPFetchRequest
+			curl_easy_setopt(curl, CURLOPT_URL,
+					request.url.c_str());
+			curl_easy_setopt(curl, CURLOPT_TIMEOUT_MS,
+					request.timeout);
+			curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT_MS,
+					request.connect_timeout);
+			// Set up a write callback that writes to the
+			// ostringstream ongoing->oss, unless the data
+			// is to be discarded
+			if (request.caller == HTTPFETCH_DISCARD) {
+				curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION,
+						httpfetch_discardfunction);
+				curl_easy_setopt(curl, CURLOPT_WRITEDATA, NULL);
+			}
+			else {
+				curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION,
+						httpfetch_writefunction);
+				curl_easy_setopt(curl, CURLOPT_WRITEDATA, &oss);
+			}
+			// Set POST (or GET) data
+			if (request.post_fields.empty()) {
+				curl_easy_setopt(curl, CURLOPT_HTTPGET, 1);
+			}
+			else {
+				curl_easy_setopt(curl, CURLOPT_POST, 1);
+				curl_easy_setopt(curl, CURLOPT_POSTFIELDSIZE,
+						request.post_fields.size());
+				curl_easy_setopt(curl, CURLOPT_POSTFIELDS,
+						request.post_fields.c_str());
+				// request.post_fields must now *never* be
+				// modified until CURLOPT_POSTFIELDS is cleared
+			}
+			// Set additional HTTP headers
+			for (size_t i = 0; i < request.extra_headers.size(); ++i) {
+				httpheader = curl_slist_append(
+					httpheader,
+					request.extra_headers[i].c_str());
+			}
+			curl_easy_setopt(curl, CURLOPT_HTTPHEADER, httpheader);
+		}
+	}
+
+	CURLcode start(CURLM *multi_)
+	{
+		if (curl == NULL)
+			return CURLE_FAILED_INIT;
+
+		if (multi_) {
+			// Multi interface (async)
+			CURLMcode mres = curl_multi_add_handle(multi_, curl);
+			if (mres != CURLM_OK) {
+				errorstream<<"curl_multi_add_handle"
+					<<" returned error code "<<mres
+					<<std::endl;
+				return CURLE_FAILED_INIT;
+			}
+			multi = multi_; // store for curl_multi_remove_handle
+			return CURLE_OK;
+		}
+		else {
+			// Easy interface (sync)
+			return curl_easy_perform(curl);
+		}
+	}
+
+	void complete(CURLcode res)
+	{
+		result.succeeded = (res == CURLE_OK);
+		result.timeout = (res == CURLE_OPERATION_TIMEDOUT);
+		result.data = oss.str();
+
+		// Get HTTP/FTP response code
+		result.response_code = 0;
+		if (curl != NULL) {
+			if (curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE,
+					&result.response_code) != CURLE_OK) {
+				result.response_code = 0;
+			}
+		}
+
+		if (res != CURLE_OK) {
+			infostream<<request.url<<" not found ("
+				<<curl_easy_strerror(res)<<")"
+				<<" (response code "<<result.response_code<<")"
+				<<std::endl;
+		}
+	}
+
+	~HTTPFetchOngoing()
+	{
+		if (multi != NULL) {
+			CURLMcode mres = curl_multi_remove_handle(multi, curl);
+			if (mres != CURLM_OK) {
+				errorstream<<"curl_multi_remove_handle"
+					<<" returned error code "<<mres
+					<<std::endl;
+			}
+		}
+
+		// Set safe options for the reusable cURL handle
+		curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION,
+				httpfetch_discardfunction);
+		curl_easy_setopt(curl, CURLOPT_WRITEDATA, NULL);
+		curl_easy_setopt(curl, CURLOPT_POSTFIELDS, NULL);
+		if (httpheader != NULL) {
+			curl_easy_setopt(curl, CURLOPT_HTTPHEADER, NULL);
+			curl_slist_free_all(httpheader);
+		}
+
+		// Store the cURL handle for reuse
+		pool->free(curl);
+	}
+};
+
+class CurlFetchThread : public SimpleThread
+{
+protected:
+	enum RequestType {
+		RT_FETCH,
+		RT_CLEAR,
+		RT_WAKEUP,
+	};
+
+	struct Request {
+		RequestType type;
+		HTTPFetchRequest fetchrequest;
+		Event *event;
+	};
+
+	CURLM *m_multi;
+	MutexedQueue<Request> m_requests;
+	size_t m_parallel_limit;
+
+	// Variables exclusively used within thread
+	std::vector<HTTPFetchOngoing*> m_all_ongoing;
+	std::list<HTTPFetchRequest> m_queued_fetches;
+
+public:
+	CurlFetchThread(int parallel_limit)
+	{
+		if (parallel_limit >= 1)
+			m_parallel_limit = parallel_limit;
+		else
+			m_parallel_limit = 1;
+	}
+
+	void requestFetch(const HTTPFetchRequest &fetchrequest)
+	{
+		Request req;
+		req.type = RT_FETCH;
+		req.fetchrequest = fetchrequest;
+		req.event = NULL;
+		m_requests.push_back(req);
+	}
+
+	void requestClear(unsigned long caller, Event *event)
+	{
+		Request req;
+		req.type = RT_CLEAR;
+		req.fetchrequest.caller = caller;
+		req.event = event;
+		m_requests.push_back(req);
+	}
+
+	void requestWakeUp()
+	{
+		Request req;
+		req.type = RT_WAKEUP;
+		req.event = NULL;
+		m_requests.push_back(req);
+	}
+
+protected:
+	// Handle a request from some other thread
+	// E.g. new fetch; clear fetches for one caller; wake up
+	void processRequest(const Request &req)
+	{
+		if (req.type == RT_FETCH) {
+			// New fetch, queue until there are less
+			// than m_parallel_limit ongoing fetches
+			m_queued_fetches.push_back(req.fetchrequest);
+
+			// see processQueued() for what happens next
+
+		}
+		else if (req.type == RT_CLEAR) {
+			unsigned long caller = req.fetchrequest.caller;
+
+			// Abort all ongoing fetches for the caller
+			for (std::vector<HTTPFetchOngoing*>::iterator
+					it = m_all_ongoing.begin();
+					it != m_all_ongoing.end();) {
+				if ((*it)->request.caller == caller) {
+					delete (*it);
+					it = m_all_ongoing.erase(it);
+				}
+				else
+					++it;
+			}
+
+			// Also abort all queued fetches for the caller
+			for (std::list<HTTPFetchRequest>::iterator
+					it = m_queued_fetches.begin();
+					it != m_queued_fetches.end();) {
+				if ((*it).caller == caller)
+					it = m_queued_fetches.erase(it);
+				else
+					++it;
+			}
+		}
+		else if (req.type == RT_WAKEUP) {
+			// Wakeup: Nothing to do, thread is awake at this point
+		}
+
+		if (req.event != NULL)
+			req.event->signal();
+	}
+
+	// Start new ongoing fetches if m_parallel_limit allows
+	void processQueued(CurlHandlePool *pool)
+	{
+		while (m_all_ongoing.size() < m_parallel_limit &&
+				!m_queued_fetches.empty()) {
+			HTTPFetchRequest request = m_queued_fetches.front();
+			m_queued_fetches.pop_front();
+
+			// Create ongoing fetch data and make a cURL handle
+			// Set cURL options based on HTTPFetchRequest
+			HTTPFetchOngoing *ongoing =
+				new HTTPFetchOngoing(request, pool);
+
+			// Initiate the connection (curl_multi_add_handle)
+			CURLcode res = ongoing->start(m_multi);
+			if (res == CURLE_OK) {
+				m_all_ongoing.push_back(ongoing);
+			}
+			else {
+				ongoing->complete(res);
+				httpfetch_deliver_result(ongoing->result);
+				delete ongoing;
+			}
+		}
+	}
+
+	// Process CURLMsg (indicates completion of a fetch)
+	void processCurlMessage(CURLMsg *msg)
+	{
+		// Determine which ongoing fetch the message pertains to
+		size_t i = 0;
+		bool found = false;
+		for (i = 0; i < m_all_ongoing.size(); ++i) {
+			if (m_all_ongoing[i]->curl == msg->easy_handle) {
+				found = true;
+				break;
+			}
+		}
+		if (msg->msg == CURLMSG_DONE && found) {
+			// m_all_ongoing[i] succeeded or failed.
+			HTTPFetchOngoing *ongoing = m_all_ongoing[i];
+			ongoing->complete(msg->data.result);
+			httpfetch_deliver_result(ongoing->result);
+			delete ongoing;
+			m_all_ongoing.erase(m_all_ongoing.begin() + i);
+		}
+	}
+
+	// Wait for a request from another thread, or timeout elapses
+	void waitForRequest(long timeout)
+	{
+		if (m_queued_fetches.empty()) {
+			try {
+				Request req = m_requests.pop_front(timeout);
+				processRequest(req);
+			}
+			catch (ItemNotFoundException &e) {}
+		}
+	}
+
+	// Wait until some IO happens, or timeout elapses
+	void waitForIO(long timeout)
+	{
+		fd_set read_fd_set;
+		fd_set write_fd_set;
+		fd_set exc_fd_set;
+		int max_fd;
+		long select_timeout = -1;
+		struct timeval select_tv;
+		CURLMcode mres;
+
+		FD_ZERO(&read_fd_set);
+		FD_ZERO(&write_fd_set);
+		FD_ZERO(&exc_fd_set);
+
+		mres = curl_multi_fdset(m_multi, &read_fd_set,
+				&write_fd_set, &exc_fd_set, &max_fd);
+		if (mres != CURLM_OK) {
+			errorstream<<"curl_multi_fdset"
+				<<" returned error code "<<mres
+				<<std::endl;
+			select_timeout = 0;
+		}
+
+		mres = curl_multi_timeout(m_multi, &select_timeout);
+		if (mres != CURLM_OK) {
+			errorstream<<"curl_multi_timeout"
+				<<" returned error code "<<mres
+				<<std::endl;
+			select_timeout = 0;
+		}
+
+		// Limit timeout so new requests get through
+		if (select_timeout < 0 || select_timeout > timeout)
+			select_timeout = timeout;
+
+		if (select_timeout > 0) {
+			select_tv.tv_sec = select_timeout / 1000;
+			select_tv.tv_usec = (select_timeout % 1000) * 1000;
+			int retval = select(max_fd + 1, &read_fd_set,
+					&write_fd_set, &exc_fd_set,
+					&select_tv);
+			if (retval == -1) {
+				#ifdef _WIN32
+				errorstream<<"select returned error code "
+					<<WSAGetLastError()<<std::endl;
+				#else
+				errorstream<<"select returned error code "
+					<<errno<<std::endl;
+				#endif
+			}
+		}
+	}
+
+	void * Thread()
+	{
+		ThreadStarted();
+		log_register_thread("CurlFetchThread");
+		DSTACK(__FUNCTION_NAME);
+
+		CurlHandlePool pool;
+
+		m_multi = curl_multi_init();
+		if (m_multi == NULL) {
+			errorstream<<"curl_multi_init returned NULL\n";
+			return NULL;
+		}
+
+		assert(m_all_ongoing.empty());
+
+		while (getRun()) {
+			BEGIN_DEBUG_EXCEPTION_HANDLER
+
+			/*
+				Handle new async requests
+			*/
+
+			while (!m_requests.empty()) {
+				Request req = m_requests.pop_front();
+				processRequest(req);
+			}
+			processQueued(&pool);
+
+			/*
+				Handle ongoing async requests
+			*/
+
+			int still_ongoing = 0;
+			while (curl_multi_perform(m_multi, &still_ongoing) ==
+					CURLM_CALL_MULTI_PERFORM)
+				/* noop */;
+
+			/*
+				Handle completed async requests
+			*/
+			if (still_ongoing < (int) m_all_ongoing.size()) {
+				CURLMsg *msg;
+				int msgs_in_queue;
+				msg = curl_multi_info_read(m_multi, &msgs_in_queue);
+				while (msg != NULL) {
+					processCurlMessage(msg);
+					msg = curl_multi_info_read(m_multi, &msgs_in_queue);
+				}
+			}
+
+			/*
+				If there are ongoing requests, wait for data
+				(with a timeout of 100ms so that new requests
+				can be processed).
+
+				If no ongoing requests, wait for a new request.
+				(Possibly an empty request that signals
+				that the thread should be stopped.)
+			*/
+			if (m_all_ongoing.empty())
+				waitForRequest(100000000);
+			else
+				waitForIO(100);
+
+			END_DEBUG_EXCEPTION_HANDLER(errorstream)
+		}
+
+		// Call curl_multi_remove_handle and cleanup easy handles
+		for (size_t i = 0; i < m_all_ongoing.size(); ++i) {
+			delete m_all_ongoing[i];
+		}
+		m_all_ongoing.clear();
+
+		m_queued_fetches.clear();
+
+		CURLMcode mres = curl_multi_cleanup(m_multi);
+		if (mres != CURLM_OK) {
+			errorstream<<"curl_multi_cleanup"
+				<<" returned error code "<<mres
+				<<std::endl;
+		}
+
+		return NULL;
+	}
+};
+
+CurlFetchThread *g_httpfetch_thread = NULL;
+
+void httpfetch_init(int parallel_limit)
+{
+	verbosestream<<"httpfetch_init: parallel_limit="<<parallel_limit
+			<<std::endl;
+
+	CURLcode res = curl_global_init(CURL_GLOBAL_DEFAULT);
+	assert(res == CURLE_OK);
+
+	g_httpfetch_thread = new CurlFetchThread(parallel_limit);
+}
+
+void httpfetch_cleanup()
+{
+	verbosestream<<"httpfetch_cleanup: cleaning up"<<std::endl;
+
+	g_httpfetch_thread->setRun(false);
+	g_httpfetch_thread->requestWakeUp();
+	g_httpfetch_thread->stop();
+	delete g_httpfetch_thread;
+
+	curl_global_cleanup();
+}
+
+void httpfetch_async(const HTTPFetchRequest &fetchrequest)
+{
+	g_httpfetch_thread->requestFetch(fetchrequest);
+	if (!g_httpfetch_thread->IsRunning())
+		g_httpfetch_thread->Start();
+}
+
+static void httpfetch_request_clear(unsigned long caller)
+{
+	if (g_httpfetch_thread->IsRunning()) {
+		Event event;
+		g_httpfetch_thread->requestClear(caller, &event);
+		event.wait();
+	}
+	else {
+		g_httpfetch_thread->requestClear(caller, NULL);
+	}
+}
+
+void httpfetch_sync(const HTTPFetchRequest &fetchrequest,
+		HTTPFetchResult &fetchresult)
+{
+	// Create ongoing fetch data and make a cURL handle
+	// Set cURL options based on HTTPFetchRequest
+	CurlHandlePool pool;
+	HTTPFetchOngoing ongoing(fetchrequest, &pool);
+	// Do the fetch (curl_easy_perform)
+	CURLcode res = ongoing.start(NULL);
+	// Update fetchresult
+	ongoing.complete(res);
+	fetchresult = ongoing.result;
+}
+
+#else  // USE_CURL
+
+/*
+	USE_CURL is off:
+
+	Dummy httpfetch implementation that always returns an error.
+*/
+
+void httpfetch_init(int parallel_limit)
+{
+}
+
+void httpfetch_cleanup()
+{
+}
+
+void httpfetch_async(const HTTPFetchRequest &fetchrequest)
+{
+	errorstream<<"httpfetch_async: unable to fetch "<<fetchrequest.url
+			<<" because USE_CURL=0"<<std::endl;
+
+	HTTPFetchResult fetchresult(fetchrequest); // sets succeeded = false etc.
+	httpfetch_deliver_result(fetchresult);
+}
+
+static void httpfetch_request_clear(unsigned long caller)
+{
+}
+
+void httpfetch_sync(const HTTPFetchRequest &fetchrequest,
+		HTTPFetchResult &fetchresult)
+{
+	errorstream<<"httpfetch_sync: unable to fetch "<<fetchrequest.url
+			<<" because USE_CURL=0"<<std::endl;
+
+	fetchresult = HTTPFetchResult(fetchrequest); // sets succeeded = false etc.
+}
+
+#endif  // USE_CURL
diff --git a/src/httpfetch.h b/src/httpfetch.h
new file mode 100644
index 00000000..56a198ba
--- /dev/null
+++ b/src/httpfetch.h
@@ -0,0 +1,126 @@
+/*
+Minetest
+Copyright (C) 2013 celeron55, Perttu Ahola <celeron55@gmail.com>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU Lesser General Public License as published by
+the Free Software Foundation; either version 2.1 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU Lesser General Public License for more details.
+
+You should have received a copy of the GNU Lesser General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+*/
+
+#ifndef HTTPFETCH_HEADER
+#define HTTPFETCH_HEADER
+
+#include <string>
+#include <vector>
+#include "config.h"
+
+// Can be used in place of "caller" in asynchronous transfers to discard result
+// (used as default value of "caller")
+#define HTTPFETCH_DISCARD 0
+
+struct HTTPFetchRequest
+{
+	std::string url;
+
+	// Identifies the caller (for asynchronous requests)
+	// Ignored by httpfetch_sync
+	unsigned long caller;
+
+	// Some number that identifies the request
+	// (when the same caller issues multiple httpfetch_async calls)
+	unsigned long request_id;
+
+	// Timeout for the whole transfer, in milliseconds
+	long timeout;
+
+	// Timeout for the connection phase, in milliseconds
+	long connect_timeout;
+
+	// POST data (should be application/x-www-form-urlencoded
+	// unless a Content-Type header is specified in extra_headers)
+	// If this is empty a GET request is done instead.
+	std::string post_fields;
+
+	// If not empty, should contain entries such as "Accept: text/html"
+	std::vector<std::string> extra_headers;
+
+	HTTPFetchRequest()
+	{
+		url = "";
+		caller = HTTPFETCH_DISCARD;
+		request_id = 0;
+		timeout = 0;
+		connect_timeout = 0;
+	}
+};
+
+struct HTTPFetchResult
+{
+	bool succeeded;
+	bool timeout;
+	long response_code;
+	std::string data;
+	// The caller and request_id from the corresponding HTTPFetchRequest.
+	unsigned long caller;
+	unsigned long request_id;
+
+	HTTPFetchResult()
+	{
+		succeeded = false;
+		timeout = false;
+		response_code = 0;
+		data = "";
+		caller = HTTPFETCH_DISCARD;
+		request_id = 0;
+	}
+
+	HTTPFetchResult(const HTTPFetchRequest &fetchrequest)
+	{
+		succeeded = false;
+		timeout = false;
+		response_code = 0;
+		data = "";
+		caller = fetchrequest.caller;
+		request_id = fetchrequest.request_id;
+	}
+};
+
+// Initializes the httpfetch module
+void httpfetch_init(int parallel_limit);
+
+// Stops the httpfetch thread and cleans up resources
+void httpfetch_cleanup();
+
+// Starts an asynchronous HTTP fetch request
+void httpfetch_async(const HTTPFetchRequest &fetchrequest);
+
+// If any fetch for the given caller ID is complete, removes it from the
+// result queue, sets fetchresult and returns true. Otherwise returns false.
+bool httpfetch_async_get(unsigned long caller, HTTPFetchResult &fetchresult);
+
+// Allocates a caller ID for httpfetch_async
+// Not required if you want to set caller = HTTPFETCH_DISCARD
+unsigned long httpfetch_caller_alloc();
+
+// Frees a caller ID allocated with httpfetch_caller_alloc
+// Note: This can be expensive, because the httpfetch thread is told
+// to stop any ongoing fetches for the given caller.
+void httpfetch_caller_free(unsigned long caller);
+
+// Performs a synchronous HTTP request. This blocks and therefore should
+// only be used from background threads.
+void httpfetch_sync(const HTTPFetchRequest &fetchrequest,
+		HTTPFetchResult &fetchresult);
+
+
+#endif // !HTTPFETCH_HEADER
diff --git a/src/main.cpp b/src/main.cpp
index 57c2f06a..239d6824 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -77,6 +77,7 @@ with this program; if not, write to the Free Software Foundation, Inc.,
 #include "subgame.h"
 #include "quicktune.h"
 #include "serverlist.h"
+#include "httpfetch.h"
 #include "guiEngine.h"
 #include "mapsector.h"
 
@@ -1001,6 +1002,9 @@ int main(int argc, char *argv[])
 	assert(res == CURLE_OK);
 #endif
 
+	// Initialize HTTP fetcher
+	httpfetch_init(g_settings->getS32("curl_parallel_limit"));
+
 	/*
 		Run unit tests
 	*/
@@ -1858,6 +1862,9 @@ int main(int argc, char *argv[])
 		}
 	}
 
+	// Stop httpfetch thread (if started)
+	httpfetch_cleanup();
+
 	END_DEBUG_EXCEPTION_HANDLER(errorstream)
 
 	debugstreams_deinit();
diff --git a/src/util/container.h b/src/util/container.h
index fdd76cc5..e83c3cd3 100644
--- a/src/util/container.h
+++ b/src/util/container.h
@@ -297,7 +297,8 @@ public:
 
 				if(!m_list.empty())
 				{
-					typename std::list<T>::iterator last = m_list.back();
+					typename std::list<T>::iterator last = m_list.end();
+					last--;
 					T t = *last;
 					m_list.erase(last);
 					return t;