我正在尝试创建一个HTTP代理，根据HTTP请求中的GET / CONNET主机名，某些连接的优先级高于其他连接。
这个想法是根据给定的主机名列表来完成具有更高优先级的请求，每个主机名都有一定的优先级。
待连接将由 accepter 线程存储在四个不同的队列中（每个队列对应一个优先级：最大，中等，最小和未分类）;然后， accepter 将 fork()一个子进程，它将按优先级顺序出列并处理挂起的连接。通过这样做， accepter 线程将始终接受新连接和每个排队连接
简而言之，这是我的代理：
socket()循环返回客户端套接字，accept()来自客户端，解析请求并根据HTTP请求中的主机名，我的自定义结构将排入正确的队列;然后它将recv()，因此一个进程将出列并处理连接; 新代理：不再fork()，我创建了一个包含四个线程的线程池（一个“接受者”和三个“连接器”：因为我打算将此代理放在上面我的RPi 2，有一个四核处理器，我以为至少有四个线程是好的）。我现在有一个fork()和两个mutex。除线程，互斥锁和条件变量外，代码几乎相同。这些是线程调用的新函数：
入队：此线程包含condition_variables循环，它从客户端接收，解析HTTP请求，查找主机名并根据其优先级排队{{ 1}} struct（代码开头的typedefed）;
dequeue ：这个线程包含出队和管理连接循环，它从队列中获取accept()结构，检索客户端套接字（我从{{1 }）}，解析主机名并管理GET或CONNECT请求。
问题：总是一样的，当谈到管理CONNECT请求时，来自客户端的info_conn总是返回0：我知道当连接的另一端有recv（）时返回0断开连接，但这不是我想要的！
基于线程方法，这是一个微不足道的生产者/消费者问题（弹出并推送到队列）所以我认为排队和出队的线程交替是正确的。
我的（新）代码：
info_conn 环境：我的笔记本电脑运行Ubuntu 14.04，x86_64;使用SwitchyOmega插件在Chrome上测试代理，该插件允许重定向特定端口（我将传递给我的代理的同一端口）上的流量，使用accept()编译。
输出（针对Netflix和YouTube试用，它们都有同样的问题，即recv()，#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/time.h>
#include <arpa/inet.h>
#include <unistd.h>
#include <thread>
#include <iostream>
#include <netdb.h>
#include <queue>
#include <list>
#include <vector>
#include <condition_variable>
#include <cstdlib>
using namespace std;
#define GET 0
#define CONNECT 1
#define DEFAULTCOLOR "\033[0m"
#define RED "\033[22;31m"
#define YELLOW "\033[1;33m"
#define GREEN "\033[0;0;32m"
#define MAX_SIZE 1000
#define CONNECT_200_OK "HTTP/1.1 200 Connection established\r\nProxy-agent: myproxy\r\n\r\n"
// my custom struct stored in queues
typedef struct info_connection {
int client_fd;
string host;
string payload;
int request;
} info_conn;
queue<info_conn>q1;
queue<info_conn>q2;
queue<info_conn>q3;
queue<info_conn>q4;
vector<thread> workers;
condition_variable cond_read, cond_write;
mutex mtx;
void enqueue(int sock_client);
void dequeue(void);
int main(int argc, char *argv[]) {
int socket_desc;
struct sockaddr_in server;
socket_desc = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
if (socket_desc == -1) {
perror("socket()");
exit(-1);
}
server.sin_family = AF_INET;
server.sin_addr.s_addr = INADDR_ANY;
if (argc == 2)
server.sin_port = htons(atoi(argv[1]));
printf("listening to port %d\n", atoi(argv[1]));
if (bind(socket_desc,(struct sockaddr *)&server, sizeof(server)) < 0) {
perror("bind failed. Error");
exit(-1);
}
printf("binded\n");
listen(socket_desc, 10);
printf("listen\n");
// thread pool, because I suck at forking
workers.push_back(thread(enqueue, socket_desc));
workers.push_back(thread(dequeue));
workers.push_back(thread(dequeue));
workers.push_back(thread(dequeue));
for (thread& t : workers) {
t.join();
}
return 0;
}
void enqueue(int sock_client) {
printf("enqueue()\n");
int client_sock;
struct sockaddr_in *client_struct;
unsigned int clilen;
bzero((char*)&client_struct, sizeof(client_struct));
clilen = sizeof(client_struct);
char host_name[128];
char buff[4096];
int n_recv, n_send;
char *start_row, *end_row, *tmp_ptr, *tmp_start;
int req;
while( (client_sock = accept(sock_client, (struct sockaddr *)&client_struct, &clilen)) ) {
memset(host_name, 0, sizeof(host_name));
n_recv = recv(client_sock, buff, sizeof(buff), 0);
if (n_recv < 0) {
perror("recv()");
break;
}
start_row = end_row = buff;
while ((end_row = strstr(start_row, "\r\n")) != NULL) {
int row_len = end_row - start_row;
if (row_len == 0)
break;
if (strncmp(buff, "GET ", 4) == 0) {
req = GET;
tmp_start = start_row + 4;
tmp_ptr = strstr(tmp_start, "//");
int len = tmp_ptr - tmp_start;
tmp_start = tmp_start + len + 2;
tmp_ptr = strchr(tmp_start, '/');
len = tmp_ptr - tmp_start;
strncpy(host_name, tmp_start, len);
break;
}
else if (strncmp(buff, "CONNECT ", 8) == 0) {
req = CONNECT;
tmp_start = start_row + 8;
tmp_ptr = strchr(tmp_start, ':');
int host_len = tmp_ptr - tmp_start;
strncpy(host_name, tmp_start, host_len);
break;
}
start_row = end_row + 2;
/* if ((start_row - buff) >= strlen(buff))
break;*/
}
unique_lock<mutex> locker(mtx, defer_lock);
locker.lock();
cond_write.wait(locker, [](){
return (q1.size() < MAX_SIZE || q2.size() < MAX_SIZE || q3.size() < MAX_SIZE || q4.size() < MAX_SIZE);
});
cout << "(DEBUG) thread " << this_thread::get_id() << " wants to insert, queues not full " <<
q1.size() << ' ' << q2.size() << ' ' << q3.size() << ' ' << q4.size() << '\n';
int priority = 0;
info_conn info_c;
info_c.client_fd = client_sock;
info_c.host = host_name;
info_c.request = req;
info_c.payload = string(buff);
cout << "(DEBUG) thread " << this_thread::get_id() << " looking for " << host_name <<
" queues" << '\n';
if (strcmp(host_name, "www.netflix.com") == 0) {
priority = 1;
printf("hostname = www.netflix.com, priority %d\n", priority);
q1.push(info_c);
}
else if (strcmp(host_name, "www.youtube.com") == 0) {
priority = 2;
printf("hostname = www.youtube.com, priority %d\n", priority);
q2.push(info_c);
}
else if (strcmp(host_name, "www.facebook.com") == 0) {
priority = 3;
printf("hostname = www.facebook.com, priority %d\n", priority);
q3.push(info_c);
}
else {
priority = 4;
printf("hostname %s not found in queues\n", host_name);
q4.push(info_c);
}
cout << GREEN << "(DEBUG) thread " << this_thread::get_id() << " inserted " <<
q1.size() << ' ' << q2.size() << ' ' << q3.size() << ' ' << q4.size() << DEFAULTCOLOR<< '\n';
locker.unlock();
cond_read.notify_all();
}
if (client_sock < 0) {
perror("accept failed");
exit(-1);
}
}
void dequeue(void) {
int fd_client = -1;
int fd_server = -1;
struct sockaddr_in server;
int what_request;
char host_name[128];
char buffer[1500];
int n_send, n_recv;
size_t length;
info_conn req;
// CONNECT
int r, max;
int send_200_OK;
int read_from_client = 0;
int read_from_server = 0;
int send_to_client = 0;
int send_to_server = 0;
struct timeval timeout;
char buff[8192];
fd_set fdset;
printf("dequeue()\n");
while (true) {
unique_lock<mutex> locker(mtx, defer_lock);
locker.lock();
cond_read.wait(locker, [](){
return (q1.size() > 0 || q2.size() > 0 || q3.size() > 0 || q4.size() > 0);
});
cout << "(DEBUG) thread " << this_thread::get_id() << " wants to remove, queues not empty " <<
q1.size() << ' ' << q2.size() << ' ' << q3.size() << ' ' << q4.size() << '\n';
if (q1.size() > 0) {
req = q1.front();
q1.pop();
}
else if (q2.size() > 0) {
req = q2.front();
q2.pop();
}
else if (q3.size() > 0) {
req = q3.front();
q3.pop();
}
else if (q4.size() > 0) {
req = q4.front();
q4.pop();
}
cout << YELLOW <<"(DEBUG) thread " << this_thread::get_id() << " removed, " <<
q1.size() << ' ' << q2.size() << ' ' << q3.size() << ' ' << q4.size() << DEFAULTCOLOR<<'\n';
locker.unlock();
// notify one, because I have only one "producer" thread
cond_write.notify_one();
fd_client = req.client_fd;
//memcpy(host_name, req.host.c_str(), strlen(req.host));
length = req.host.copy(host_name, req.host.size(), 0);
host_name[length] = '\0';
what_request = req.request;
//memcpy(buffer, req.payload, req.payload.size());
length = req.payload.copy(buffer, req.payload.size(), 0);
buffer[length] = '\0';
what_request = req.request;
//cout << RED <<"(DEBUG) thread " << this_thread::get_id() << " copied packet payload " <<
// buffer << DEFAULTCOLOR<<'\n';
struct addrinfo* result;
struct addrinfo* res;
int error;
struct sockaddr_in *resolve;
fd_server = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
if (fd_server < 0) {
perror("socket()");
exit(-1);
}
cout << "(DEBUG) thread " << this_thread::get_id() << " fd_server " << fd_server << '\n';
error = getaddrinfo(host_name, NULL, NULL, &result);
if (error != 0) {
if (error == EAI_SYSTEM) {
perror("getaddrinfo");
} else {
fprintf(stderr, "error in getaddrinfo for (%s): %s\n", host_name, gai_strerror(error));
}
exit(EXIT_FAILURE);
}
if (what_request == GET) {
server.sin_port = htons(80);
}
else if (what_request == CONNECT) {
server.sin_port = htons(443);
}
server.sin_family = AF_INET;
cout << "(DEBUG) thread " << this_thread::get_id() << " getaddrinfo()" << '\n';
for (res = result; res != NULL; res = res->ai_next) {
if (res->ai_family == AF_INET) {
resolve = (struct sockaddr_in *)res->ai_addr;
//char *ip = inet_ntoa(resolve->sin_addr);
//printf("%s\n", ip);
server.sin_addr.s_addr = resolve->sin_addr.s_addr;
if (connect(fd_server, (struct sockaddr *)&server, sizeof (struct sockaddr_in)) < 0) {
fflush(stdout);
perror("connect()");
}
else {
cout << "(DEBUG) thread " << this_thread::get_id() << " connected to " << inet_ntoa(server.sin_addr) << '\n';
}
break;
}
}
// dealing with GET
if (what_request == GET) {
cout << "thread " << this_thread::get_id() << " dealing GET " << host_name <<
" sending to server " << buffer << '\n';
n_send = send(fd_server, buffer, strlen(buffer)+1, 0);
if (n_send < 0) {
cout << "thread " << this_thread::get_id() << " error sending GET request to server" << '\n';
perror("send()");
break;
}
do {
memset(buffer, 0, sizeof(buffer));
n_recv = recv(fd_server, buffer, sizeof(buffer), 0);
cout << "thread " << this_thread::get_id() << " GET: " << host_name << " read from recv() " << n_recv << " bytes, " <<
fd_client << "<->" << fd_server << '\n';
n_send = send(fd_client, buffer, n_recv, 0);
} while (n_recv > 0);
if (n_recv < 0) {
cout << RED << "thread " << this_thread::get_id() << " error sending GET response from server to client" << DEFAULTCOLOR<<'\n';
perror("send()");
break;
}
close(fd_client);
close(fd_server);
cout << "thread " << this_thread::get_id() <<
" done with GET request, quitting\n";
}
// dealing with CONNECT
else if (what_request == CONNECT) {
cout << "thread " << this_thread::get_id() << " dealing CONNECT " << host_name << '\n';
max = fd_server >= fd_client ? fd_server+1 : fd_client+1;
send_200_OK = send(fd_client, CONNECT_200_OK, sizeof(CONNECT_200_OK), 0);
if (send_200_OK < 0) {
perror("send() 200 OK to client");
break;
}
cout << "thread " << this_thread::get_id() << " SENT 200 OK to client " << '\n';
int tot_recvd;
int tot_sent;
// TCP tunnel
while(true) {
memset(buff, 0, sizeof(buff));
FD_ZERO(&fdset);
FD_SET(fd_client, &fdset);
FD_SET(fd_server, &fdset);
timeout.tv_sec = 15;
timeout.tv_usec = 0;
r = select(max, &fdset, NULL, NULL, &timeout);
if (r < 0) {
perror("select()");
close(fd_client);
close(fd_server);
break;
}
if (r == 0) { // select timed out
printf("tunnel(): select() request timeout 408\n");
close(fd_client);
close(fd_server);
break;
}
if (FD_ISSET(fd_client, &fdset)) {
tot_recvd = 0;
tot_sent = 0;
do {
read_from_client = recv(fd_client, &(buff[tot_recvd]), sizeof(buff), 0);
tot_recvd += read_from_client;
cout << "thread " << this_thread::get_id() <<
" select(), reading from client " << fd_client <<
" " << read_from_client << " bytes, " << fd_client<< " <-> " << fd_server<<'\n';
if (buff[tot_recvd-1] == '\0') {
break;
}
} while (read_from_client > 0);
if (read_from_client < 0) {
perror("recv()");
close(fd_client);
close(fd_server);
break;
}
if (read_from_client == 0) {
// this always happens!!!
}
send_to_server = send(fd_server, buff, read_from_client, 0);
if (send_to_server < 0) {
perror("send() to client");
close(fd_client);
close(fd_server);
break;
}
}
if (FD_ISSET(fd_server, &fdset)) {
tot_recvd = 0;
tot_sent = 0;
do {
read_from_server = recv(fd_server, &(buff[tot_recvd]), sizeof(buff), 0);
tot_recvd += read_from_server;
cout << "thread " << this_thread::get_id() <<
" select(), reading from server " << fd_client <<
" " << read_from_server << " bytes, " << fd_client<< " <-> " << fd_server<<'\n';
if (buff[tot_recvd-1] == '\0') {
break;
}
} while (read_from_server > 0);
if (read_from_server < 0) {
perror("read()");
close(fd_client);
close(fd_server);
break;
}
if (read_from_server == 0) {
cout << "thread " << this_thread::get_id() << " select(), server closed conn" << '\n';
close(fd_client);
close(fd_server);
break;
}
send_to_client = send(fd_client, buff, read_from_server, 0);
if (send_to_client < 0) {
perror("send() to client");
close(fd_client);
close(fd_server);
break;
}
}
}
cout << "thread " << this_thread::get_id() << " done with CONNECT request\n";
}
}
}
返回0）：
g++ -std=c++11 -pedantic -Wall -o funwithproxyfork funwithproxyfork.cpp -lpthread然后它什么也没说。
答案 0 :(得分：-1)
通过检查代码，这似乎是正常的，并且由于HTTP / 1.1的工作方式。
您可能正在使用一些支持HTTP / 1.1流水线操作的客户端。当HTTP / 1.1流水线操作生效时，服务器将保持连接打开，以防客户端想要发送另一个请求。如果客户端没有，则客户端关闭连接。
您的代码似乎希望服务器在响应HTTP请求后关闭连接，并且您不希望客户端首先关闭其连接端。 HTTP / 1.1不是这样，您可以让客户端或服务器首先关闭连接。关闭连接的任何一个都是正常的。
所以，这里没有问题，除了我在评论中单独提到的几个问题，与手头的recv()问题无关。此外，许多地方的代码都没有充分检查send()的返回值，并假设已发送所有请求的字节。这是错误的。您不能保证send()将发送所请求的确切字节数。它实际上可以发送更少，这在返回值中表示，返回值是实际发送的字节数。
此代理将在高流量负载下开始失败，因为将发送的字节数比请求的少，但代码无法检测到这种情况，并正确处理它。例如，它将从服务器读取2000个字节，尝试将它们发送到客户端send()报告发送了1000个字节，代码继续其快乐方式，并且客户端最终没有收到整个响应从服务器。接连发生了。
此外，这里还有一些其他竞争条件可能导致代理与完全支持流水线操作的HTTP / 1.1客户端“楔入”或锁定。但如果你开始遇到这些问题，这将是另一个要问的问题......