dash/src/httpserver.cpp

680 lines
21 KiB
C++
Raw Normal View History

evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
// Copyright (c) 2015 The Bitcoin Core developers
// Distributed under the MIT software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include "httpserver.h"
#include "chainparamsbase.h"
#include "compat.h"
#include "util.h"
#include "netbase.h"
#include "rpcprotocol.h" // For HTTP status codes
#include "sync.h"
#include "ui_interface.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <signal.h>
#include <event2/event.h>
#include <event2/http.h>
#include <event2/thread.h>
#include <event2/buffer.h>
#include <event2/util.h>
#include <event2/keyvalq_struct.h>
#ifdef EVENT__HAVE_NETINET_IN_H
#include <netinet/in.h>
#ifdef _XOPEN_SOURCE_EXTENDED
#include <arpa/inet.h>
#endif
#endif
#include <boost/algorithm/string/case_conv.hpp> // for to_lower()
#include <boost/foreach.hpp>
#include <boost/scoped_ptr.hpp>
/** Maximum size of http request (request line + headers) */
static const size_t MAX_HEADERS_SIZE = 8192;
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
/** HTTP request work item */
class HTTPWorkItem : public HTTPClosure
{
public:
HTTPWorkItem(HTTPRequest* req, const std::string &path, const HTTPRequestHandler& func):
req(req), path(path), func(func)
{
}
void operator()()
{
func(req.get(), path);
}
boost::scoped_ptr<HTTPRequest> req;
private:
std::string path;
HTTPRequestHandler func;
};
/** Simple work queue for distributing work over multiple threads.
* Work items are simply callable objects.
*/
template <typename WorkItem>
class WorkQueue
{
private:
/** Mutex protects entire object */
CWaitableCriticalSection cs;
CConditionVariable cond;
/* XXX in C++11 we can use std::unique_ptr here and avoid manual cleanup */
std::deque<WorkItem*> queue;
bool running;
size_t maxDepth;
int numThreads;
/** RAII object to keep track of number of running worker threads */
class ThreadCounter
{
public:
WorkQueue &wq;
ThreadCounter(WorkQueue &w): wq(w)
{
boost::lock_guard<boost::mutex> lock(wq.cs);
wq.numThreads += 1;
}
~ThreadCounter()
{
boost::lock_guard<boost::mutex> lock(wq.cs);
wq.numThreads -= 1;
wq.cond.notify_all();
}
};
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
public:
WorkQueue(size_t maxDepth) : running(true),
maxDepth(maxDepth),
numThreads(0)
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
{
}
/*( Precondition: worker threads have all stopped
* (call WaitExit)
*/
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
~WorkQueue()
{
while (!queue.empty()) {
delete queue.front();
queue.pop_front();
}
}
/** Enqueue a work item */
bool Enqueue(WorkItem* item)
{
boost::unique_lock<boost::mutex> lock(cs);
if (queue.size() >= maxDepth) {
return false;
}
queue.push_back(item);
cond.notify_one();
return true;
}
/** Thread function */
void Run()
{
ThreadCounter count(*this);
while (true) {
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
WorkItem* i = 0;
{
boost::unique_lock<boost::mutex> lock(cs);
while (running && queue.empty())
cond.wait(lock);
if (!running)
break;
i = queue.front();
queue.pop_front();
}
(*i)();
delete i;
}
}
/** Interrupt and exit loops */
void Interrupt()
{
boost::unique_lock<boost::mutex> lock(cs);
running = false;
cond.notify_all();
}
/** Wait for worker threads to exit */
void WaitExit()
{
boost::unique_lock<boost::mutex> lock(cs);
while (numThreads > 0){
cond.wait(lock);
}
}
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
/** Return current depth of queue */
size_t Depth()
{
boost::unique_lock<boost::mutex> lock(cs);
return queue.size();
}
};
struct HTTPPathHandler
{
HTTPPathHandler() {}
HTTPPathHandler(std::string prefix, bool exactMatch, HTTPRequestHandler handler):
prefix(prefix), exactMatch(exactMatch), handler(handler)
{
}
std::string prefix;
bool exactMatch;
HTTPRequestHandler handler;
};
/** HTTP module state */
//! libevent event loop
static struct event_base* eventBase = 0;
//! HTTP server
struct evhttp* eventHTTP = 0;
//! List of subnets to allow RPC connections from
static std::vector<CSubNet> rpc_allow_subnets;
//! Work queue for handling longer requests off the event loop thread
static WorkQueue<HTTPClosure>* workQueue = 0;
//! Handlers for (sub)paths
std::vector<HTTPPathHandler> pathHandlers;
//! Bound listening sockets
std::vector<evhttp_bound_socket *> boundSockets;
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
/** Check if a network address is allowed to access the HTTP server */
static bool ClientAllowed(const CNetAddr& netaddr)
{
if (!netaddr.IsValid())
return false;
BOOST_FOREACH (const CSubNet& subnet, rpc_allow_subnets)
if (subnet.Match(netaddr))
return true;
return false;
}
/** Initialize ACL list for HTTP server */
static bool InitHTTPAllowList()
{
rpc_allow_subnets.clear();
rpc_allow_subnets.push_back(CSubNet("127.0.0.0/8")); // always allow IPv4 local subnet
rpc_allow_subnets.push_back(CSubNet("::1")); // always allow IPv6 localhost
if (mapMultiArgs.count("-rpcallowip")) {
const std::vector<std::string>& vAllow = mapMultiArgs["-rpcallowip"];
BOOST_FOREACH (std::string strAllow, vAllow) {
CSubNet subnet(strAllow);
if (!subnet.IsValid()) {
uiInterface.ThreadSafeMessageBox(
strprintf("Invalid -rpcallowip subnet specification: %s. Valid are a single IP (e.g. 1.2.3.4), a network/netmask (e.g. 1.2.3.4/255.255.255.0) or a network/CIDR (e.g. 1.2.3.4/24).", strAllow),
"", CClientUIInterface::MSG_ERROR);
return false;
}
rpc_allow_subnets.push_back(subnet);
}
}
std::string strAllowed;
BOOST_FOREACH (const CSubNet& subnet, rpc_allow_subnets)
strAllowed += subnet.ToString() + " ";
LogPrint("http", "Allowing HTTP connections from: %s\n", strAllowed);
return true;
}
/** HTTP request method as string - use for logging only */
static std::string RequestMethodString(HTTPRequest::RequestMethod m)
{
switch (m) {
case HTTPRequest::GET:
return "GET";
break;
case HTTPRequest::POST:
return "POST";
break;
case HTTPRequest::HEAD:
return "HEAD";
break;
case HTTPRequest::PUT:
return "PUT";
break;
default:
return "unknown";
}
}
/** HTTP request callback */
static void http_request_cb(struct evhttp_request* req, void* arg)
{
std::auto_ptr<HTTPRequest> hreq(new HTTPRequest(req));
LogPrint("http", "Received a %s request for %s from %s\n",
RequestMethodString(hreq->GetRequestMethod()), hreq->GetURI(), hreq->GetPeer().ToString());
// Early address-based allow check
if (!ClientAllowed(hreq->GetPeer())) {
hreq->WriteReply(HTTP_FORBIDDEN);
return;
}
// Early reject unknown HTTP methods
if (hreq->GetRequestMethod() == HTTPRequest::UNKNOWN) {
hreq->WriteReply(HTTP_BADMETHOD);
return;
}
// Find registered handler for prefix
std::string strURI = hreq->GetURI();
std::string path;
std::vector<HTTPPathHandler>::const_iterator i = pathHandlers.begin();
std::vector<HTTPPathHandler>::const_iterator iend = pathHandlers.end();
for (; i != iend; ++i) {
bool match = false;
if (i->exactMatch)
match = (strURI == i->prefix);
else
match = (strURI.substr(0, i->prefix.size()) == i->prefix);
if (match) {
path = strURI.substr(i->prefix.size());
break;
}
}
// Dispatch to worker thread
if (i != iend) {
std::auto_ptr<HTTPWorkItem> item(new HTTPWorkItem(hreq.release(), path, i->handler));
assert(workQueue);
if (workQueue->Enqueue(item.get()))
item.release(); /* if true, queue took ownership */
else
item->req->WriteReply(HTTP_INTERNAL, "Work queue depth exceeded");
} else {
hreq->WriteReply(HTTP_NOTFOUND);
}
}
/** Callback to reject HTTP requests after shutdown. */
static void http_reject_request_cb(struct evhttp_request* req, void*)
{
LogPrint("http", "Rejecting request while shutting down\n");
evhttp_send_error(req, HTTP_SERVUNAVAIL, NULL);
}
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
/** Event dispatcher thread */
static void ThreadHTTP(struct event_base* base, struct evhttp* http)
{
2016-03-04 08:25:16 +01:00
RenameThread("dash-http");
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
LogPrint("http", "Entering http event loop\n");
event_base_dispatch(base);
// Event loop will be interrupted by InterruptHTTPServer()
LogPrint("http", "Exited http event loop\n");
}
/** Bind HTTP server to specified addresses */
static bool HTTPBindAddresses(struct evhttp* http)
{
int defaultPort = GetArg("-rpcport", BaseParams().RPCPort());
std::vector<std::pair<std::string, uint16_t> > endpoints;
// Determine what addresses to bind to
if (!mapArgs.count("-rpcallowip")) { // Default to loopback if not allowing external IPs
endpoints.push_back(std::make_pair("::1", defaultPort));
endpoints.push_back(std::make_pair("127.0.0.1", defaultPort));
if (mapArgs.count("-rpcbind")) {
LogPrintf("WARNING: option -rpcbind was ignored because -rpcallowip was not specified, refusing to allow everyone to connect\n");
}
} else if (mapArgs.count("-rpcbind")) { // Specific bind address
const std::vector<std::string>& vbind = mapMultiArgs["-rpcbind"];
for (std::vector<std::string>::const_iterator i = vbind.begin(); i != vbind.end(); ++i) {
int port = defaultPort;
std::string host;
SplitHostPort(*i, port, host);
endpoints.push_back(std::make_pair(host, port));
}
} else { // No specific bind address specified, bind to any
endpoints.push_back(std::make_pair("::", defaultPort));
endpoints.push_back(std::make_pair("0.0.0.0", defaultPort));
}
// Bind addresses
for (std::vector<std::pair<std::string, uint16_t> >::iterator i = endpoints.begin(); i != endpoints.end(); ++i) {
LogPrint("http", "Binding RPC on address %s port %i\n", i->first, i->second);
evhttp_bound_socket *bind_handle = evhttp_bind_socket_with_handle(http, i->first.empty() ? NULL : i->first.c_str(), i->second);
if (bind_handle) {
boundSockets.push_back(bind_handle);
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
} else {
LogPrintf("Binding RPC on address %s port %i failed.\n", i->first, i->second);
}
}
return !boundSockets.empty();
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
}
/** Simple wrapper to set thread name and run work queue */
static void HTTPWorkQueueRun(WorkQueue<HTTPClosure>* queue)
{
2016-03-02 12:30:36 +01:00
RenameThread("dash-httpworker");
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
queue->Run();
}
/** libevent event log callback */
static void libevent_log_cb(int severity, const char *msg)
{
#ifndef EVENT_LOG_WARN
// EVENT_LOG_WARN was added in 2.0.19; but before then _EVENT_LOG_WARN existed.
# define EVENT_LOG_WARN _EVENT_LOG_WARN
#endif
if (severity >= EVENT_LOG_WARN) // Log warn messages and higher without debug category
LogPrintf("libevent: %s\n", msg);
else
LogPrint("libevent", "libevent: %s\n", msg);
}
bool InitHTTPServer()
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
{
struct evhttp* http = 0;
struct event_base* base = 0;
if (!InitHTTPAllowList())
return false;
if (GetBoolArg("-rpcssl", false)) {
uiInterface.ThreadSafeMessageBox(
"SSL mode for RPC (-rpcssl) is no longer supported.",
"", CClientUIInterface::MSG_ERROR);
return false;
}
// Redirect libevent's logging to our own log
event_set_log_callback(&libevent_log_cb);
#if LIBEVENT_VERSION_NUMBER >= 0x02010100
// If -debug=libevent, set full libevent debugging.
// Otherwise, disable all libevent debugging.
if (LogAcceptCategory("libevent"))
event_enable_debug_logging(EVENT_DBG_ALL);
else
event_enable_debug_logging(EVENT_DBG_NONE);
#endif
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
#ifdef WIN32
evthread_use_windows_threads();
#else
evthread_use_pthreads();
#endif
base = event_base_new(); // XXX RAII
if (!base) {
LogPrintf("Couldn't create an event_base: exiting\n");
return false;
}
/* Create a new evhttp object to handle requests. */
http = evhttp_new(base); // XXX RAII
if (!http) {
LogPrintf("couldn't create evhttp. Exiting.\n");
event_base_free(base);
return false;
}
evhttp_set_timeout(http, GetArg("-rpcservertimeout", DEFAULT_HTTP_SERVER_TIMEOUT));
evhttp_set_max_headers_size(http, MAX_HEADERS_SIZE);
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
evhttp_set_max_body_size(http, MAX_SIZE);
evhttp_set_gencb(http, http_request_cb, NULL);
if (!HTTPBindAddresses(http)) {
LogPrintf("Unable to bind any endpoint for RPC server\n");
evhttp_free(http);
event_base_free(base);
return false;
}
LogPrint("http", "Initialized HTTP server\n");
int workQueueDepth = std::max((long)GetArg("-rpcworkqueue", DEFAULT_HTTP_WORKQUEUE), 1L);
LogPrintf("HTTP: creating work queue of depth %d\n", workQueueDepth);
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
workQueue = new WorkQueue<HTTPClosure>(workQueueDepth);
eventBase = base;
eventHTTP = http;
return true;
}
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
boost::thread threadHTTP;
bool StartHTTPServer()
{
LogPrint("http", "Starting HTTP server\n");
int rpcThreads = std::max((long)GetArg("-rpcthreads", DEFAULT_HTTP_THREADS), 1L);
LogPrintf("HTTP: starting %d worker threads\n", rpcThreads);
threadHTTP = boost::thread(boost::bind(&ThreadHTTP, eventBase, eventHTTP));
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
for (int i = 0; i < rpcThreads; i++)
boost::thread(boost::bind(&HTTPWorkQueueRun, workQueue));
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
return true;
}
void InterruptHTTPServer()
{
LogPrint("http", "Interrupting HTTP server\n");
if (eventHTTP) {
// Unlisten sockets
BOOST_FOREACH (evhttp_bound_socket *socket, boundSockets) {
evhttp_del_accept_socket(eventHTTP, socket);
}
// Reject requests on current connections
evhttp_set_gencb(eventHTTP, http_reject_request_cb, NULL);
}
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
if (workQueue)
workQueue->Interrupt();
}
void StopHTTPServer()
{
LogPrint("http", "Stopping HTTP server\n");
if (workQueue) {
LogPrint("http", "Waiting for HTTP worker threads to exit\n");
#ifndef WIN32
// ToDo: Disabling WaitExit() for Windows platforms is an ugly workaround for the wallet not
// closing during a repair-restart. It doesn't hurt, though, because threadHTTP.timed_join
// below takes care of this and sends a loopbreak.
workQueue->WaitExit();
#endif
delete workQueue;
}
if (eventBase) {
LogPrint("http", "Waiting for HTTP event thread to exit\n");
// Give event loop a few seconds to exit (to send back last RPC responses), then break it
// Before this was solved with event_base_loopexit, but that didn't work as expected in
// at least libevent 2.0.21 and always introduced a delay. In libevent
// master that appears to be solved, so in the future that solution
// could be used again (if desirable).
// (see discussion in https://github.com/bitcoin/bitcoin/pull/6990)
#if BOOST_VERSION >= 105000
if (!threadHTTP.try_join_for(boost::chrono::milliseconds(2000))) {
#else
if (!threadHTTP.timed_join(boost::posix_time::milliseconds(2000))) {
#endif
LogPrintf("HTTP event loop did not exit within allotted time, sending loopbreak\n");
event_base_loopbreak(eventBase);
threadHTTP.join();
}
}
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
if (eventHTTP) {
evhttp_free(eventHTTP);
eventHTTP = 0;
}
if (eventBase) {
event_base_free(eventBase);
eventBase = 0;
}
LogPrint("http", "Stopped HTTP server\n");
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
}
struct event_base* EventBase()
{
return eventBase;
}
static void httpevent_callback_fn(evutil_socket_t, short, void* data)
{
// Static handler: simply call inner handler
HTTPEvent *self = ((HTTPEvent*)data);
self->handler();
if (self->deleteWhenTriggered)
delete self;
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
}
HTTPEvent::HTTPEvent(struct event_base* base, bool deleteWhenTriggered, const boost::function<void(void)>& handler):
deleteWhenTriggered(deleteWhenTriggered), handler(handler)
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
{
ev = event_new(base, -1, 0, httpevent_callback_fn, this);
assert(ev);
}
HTTPEvent::~HTTPEvent()
{
event_free(ev);
}
void HTTPEvent::trigger(struct timeval* tv)
{
if (tv == NULL)
event_active(ev, 0, 0); // immediately trigger event in main thread
else
evtimer_add(ev, tv); // trigger after timeval passed
}
HTTPRequest::HTTPRequest(struct evhttp_request* req) : req(req),
replySent(false)
{
}
HTTPRequest::~HTTPRequest()
{
if (!replySent) {
// Keep track of whether reply was sent to avoid request leaks
LogPrintf("%s: Unhandled request\n", __func__);
WriteReply(HTTP_INTERNAL, "Unhandled request");
}
// evhttpd cleans up the request, as long as a reply was sent.
}
std::pair<bool, std::string> HTTPRequest::GetHeader(const std::string& hdr)
{
const struct evkeyvalq* headers = evhttp_request_get_input_headers(req);
assert(headers);
const char* val = evhttp_find_header(headers, hdr.c_str());
if (val)
return std::make_pair(true, val);
else
return std::make_pair(false, "");
}
std::string HTTPRequest::ReadBody()
{
struct evbuffer* buf = evhttp_request_get_input_buffer(req);
if (!buf)
return "";
size_t size = evbuffer_get_length(buf);
/** Trivial implementation: if this is ever a performance bottleneck,
* internal copying can be avoided in multi-segment buffers by using
* evbuffer_peek and an awkward loop. Though in that case, it'd be even
* better to not copy into an intermediate string but use a stream
* abstraction to consume the evbuffer on the fly in the parsing algorithm.
*/
const char* data = (const char*)evbuffer_pullup(buf, size);
if (!data) // returns NULL in case of empty buffer
return "";
std::string rv(data, size);
evbuffer_drain(buf, size);
return rv;
}
void HTTPRequest::WriteHeader(const std::string& hdr, const std::string& value)
{
struct evkeyvalq* headers = evhttp_request_get_output_headers(req);
assert(headers);
evhttp_add_header(headers, hdr.c_str(), value.c_str());
}
/** Closure sent to main thread to request a reply to be sent to
* a HTTP request.
* Replies must be sent in the main loop in the main http thread,
* this cannot be done from worker threads.
*/
void HTTPRequest::WriteReply(int nStatus, const std::string& strReply)
{
assert(!replySent && req);
// Send event to main http thread to send reply message
struct evbuffer* evb = evhttp_request_get_output_buffer(req);
assert(evb);
evbuffer_add(evb, strReply.data(), strReply.size());
HTTPEvent* ev = new HTTPEvent(eventBase, true,
boost::bind(evhttp_send_reply, req, nStatus, (const char*)NULL, (struct evbuffer *)NULL));
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
ev->trigger(0);
replySent = true;
req = 0; // transferred back to main thread
}
CService HTTPRequest::GetPeer()
{
evhttp_connection* con = evhttp_request_get_connection(req);
CService peer;
if (con) {
// evhttp retains ownership over returned address string
const char* address = "";
uint16_t port = 0;
evhttp_connection_get_peer(con, (char**)&address, &port);
peer = CService(address, port);
}
return peer;
}
std::string HTTPRequest::GetURI()
{
return evhttp_request_get_uri(req);
}
HTTPRequest::RequestMethod HTTPRequest::GetRequestMethod()
{
switch (evhttp_request_get_command(req)) {
case EVHTTP_REQ_GET:
return GET;
break;
case EVHTTP_REQ_POST:
return POST;
break;
case EVHTTP_REQ_HEAD:
return HEAD;
break;
case EVHTTP_REQ_PUT:
return PUT;
break;
default:
return UNKNOWN;
break;
}
}
void RegisterHTTPHandler(const std::string &prefix, bool exactMatch, const HTTPRequestHandler &handler)
{
LogPrint("http", "Registering HTTP handler for %s (exactmatch %d)\n", prefix, exactMatch);
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
pathHandlers.push_back(HTTPPathHandler(prefix, exactMatch, handler));
}
void UnregisterHTTPHandler(const std::string &prefix, bool exactMatch)
{
std::vector<HTTPPathHandler>::iterator i = pathHandlers.begin();
std::vector<HTTPPathHandler>::iterator iend = pathHandlers.end();
for (; i != iend; ++i)
if (i->prefix == prefix && i->exactMatch == exactMatch)
break;
if (i != iend)
{
LogPrint("http", "Unregistering HTTP handler for %s (exactmatch %d)\n", prefix, exactMatch);
evhttpd implementation - *Replace usage of boost::asio with [libevent2](http://libevent.org/)*. boost::asio is not part of C++11, so unlike other boost there is no forwards-compatibility reason to stick with it. Together with #4738 (convert json_spirit to UniValue), this rids Bitcoin Core of the worst offenders with regard to compile-time slowness. - *Replace spit-and-duct-tape http server with evhttp*. Front-end http handling is handled by libevent, a work queue (with configurable depth and parallelism) is used to handle application requests. - *Wrap HTTP request in C++ class*; this makes the application code mostly HTTP-server-neutral - *Refactor RPC to move all http-specific code to a separate file*. Theoreticaly this can allow building without HTTP server but with another RPC backend, e.g. Qt's debug console (currently not implemented) or future RPC mechanisms people may want to use. - *HTTP dispatch mechanism*; services (e.g., RPC, REST) register which URL paths they want to handle. By using a proven, high-performance asynchronous networking library (also used by Tor) and HTTP server, problems such as #5674, #5655, #344 should be avoided. What works? bitcoind, bitcoin-cli, bitcoin-qt. Unit tests and RPC/REST tests pass. The aim for now is everything but SSL support. Configuration options: - `-rpcthreads`: repurposed as "number of work handler threads". Still defaults to 4. - `-rpcworkqueue`: maximum depth of work queue. When this is reached, new requests will return a 500 Internal Error. - `-rpctimeout`: inactivity time, in seconds, after which to disconnect a client. - `-debug=http`: low-level http activity logging
2015-01-23 07:53:17 +01:00
pathHandlers.erase(i);
}
}