I have some code use libev on how to deal with connection timeout as below (please refer to http://lists.schmorp.de/pipermail/libev/2011q2/001365.html):
sd = create_socket()
set_socket_nonblock(sd)
connect("127.0.0.1", port) // connect to an invalid port
ev_io_init(&w_io, connect_cb, sd, EV_WRITE)
ev_io_start(...)
ev_timer_init(&w_timer, timeout_cb, 5.0, 0)
ev_timer_start(...)
and in someplace perform ev_run. The connect_cb is called and in this callback function I checked the revents with EV_ERROR, the result is no error. This is strange because I provide an invalid port number which is not listening in local machine. Anyway, I try to send a message in the connect_cb function, got an error 111, which means that connection refused. I'm confused! How to check if the connection is established correctly when use non-block socket?
getsockopt is possible way to get if the connection has some error happen:
int err;
socklen_t len = sizeof(err);
getsockopt(sd, SOL_SOCKET, SO_ERROR, &err, &len);
if (err) {
// error happen
} else {
connection is OK
}
Related
I'm using the standard way, shown in many examples, to create a UDP socket in C++, keep it alive and keep sending stuff over it.
I only call sendto. I never call recvfrom. In this code fragment I only use sendto once and then sleep for 5 seconds:
C code:
static int do_the_rest( int sock )
{
struct sockaddr_in server_addr;
bzero(&server_addr,sizeof(server_addr));
server_addr.sin_family = AF_INET;
server_addr.sin_addr.s_addr=inet_addr("192.168.32.32");
server_addr.sin_port=htons(32000);
char sendline[100];
bzero(sendline,sizeof(sendline));
const struct sockaddr * addr = (const struct sockaddr*)&server_addr;
sendto(sock,sendline,sizeof(sendline),0,addr,sizeof(server_addr));
sleep( 5 );
return 0;
}
int main(int argc, char**argv)
{
int sock;
sock=socket(AF_INET,SOCK_DGRAM,0);
if( sock < 0 ) {
perror( "socket" );
return 1;
}
int ret = do_the_rest( sock );
close( sock );
return ret;
}
Now, if I run "netstat -na", I can identify that the system seems to listen on the local port of the socket (I remove the lines in my program that print the local port, for clarity):
netstat -na:
Active Internet connections (servers and established)
Proto Recv-Q Send-Q Local Address Foreign Address State
...
udp 0 304 0.0.0.0:53735 0.0.0.0:*
When I try something similar in Java, I also seem to get some listening, although it looks a bit different (perhaps IPv6?):
Java code:
import java.io.*;
import java.net.*;
class Udp {
public static void main(String[] args) throws Throwable {
DatagramSocket sock = new DatagramSocket(null);
try {
InetAddress ipAddress = InetAddress.getByName("192.168.32.32");
byte[] sendData = new byte[50000];
DatagramPacket sendPacket = new DatagramPacket(
sendData, sendData.length, ipAddress, 32000);
sock.send(sendPacket);
Thread.sleep(5000);
} finally {
sock.close();
}
}
}
netstat -na:
Active Internet connections (servers and established)
Proto Recv-Q Send-Q Local Address Foreign Address State
...
udp 0 0 :::37053 :::*
I understand this is done in order to support a possible recvfrom (receive in Java) that may follow. However, is there a way to tell the socket not to listen to incoming packets at all?
Thanks
Now, if I run "netstat -na", I can identify that the system seems to listen on the local port of the socket
UDP sockets have a kernel buffer for incoming messages. This buffer is maintained by the kernel regardless whether you call recv/recvfrom/recvmsg from user-space code.
You can use shutdown() with its how parameter set to SHUT_RD to disable reads on the socket, which may have the intended behavior of stopping the listening, maybe even freeing the kernel's receive buffer. But the receive port has to remain allocated for the socket's lifetime, as it is used as the source port for the outgoing packets you are sending with sendto(). You cannot avoid that.
It is not only 'listening' on that port, it is sending via that port. The port allocation is required. The socket gets bound to a port as soon as you call sendto(), unless it is already bound.
Is there an easy way to check if a given memcached_st* is successfully connected to a memcached server?
I'm connecting via memcached_server_add_with_weight, and it's returning MEMCACHED_SUCCESS when I give it spurious hostnames. Similarly, calling memcached_last_error_errno immediately after the call to memcached_server_add_with_weight gives me MEMCACHED_SUCCESS.
One interesting way to do this is checking the actual socket descriptor.
If libmemcached successfully connected to the server then the socket descriptor is positive, otherwise it is -1.
std::shared_ptr<memcached_st> es (memcached_create (NULL), [](memcached_st* msp) {memcached_free (msp);});
memcached_server_add_with_weight (es.get(), "server1", 9201, 100);
memcached_server_add (es.get(), "server2", 9201);
memcached_server_add (es.get(), "server3", 9201);
memcached_server_fn serverVisitor = [](const memcached_st *ptr, memcached_server_instance_st server, void *context) {
if (server->fd < 0) throw std::runtime_error (std::string ("libmemcached connection to ") + server->hostname + " failed!");
return MEMCACHED_SUCCESS;
};
memcached_server_cursor (es.get(), &serverVisitor, NULL, 1);
I have simple server and client in UDP (WinSocks/C++).
I send datagram client -> server via sendto, and reply from server to client using the ip and port obtained from recvfrom function.
I found out that:
Every sendto from client is being sent from different port
When trying to reply from server Windows returns WSAECONNRESET (which mean that port is closed - http://support.microsoft.com/kb/263823)
How can I properly answer client from server (ie force port binding on client when sending using sendto?)
Edit: Adding some source code:
bool InitClient()
{
internal->sock = socket(PF_INET, SOCK_DGRAM, 0);
char8 yes = 1;
setsockopt(internal->sock, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(int32));
return internal->sock != -1;
}
void Send(const IpAddress & target, const uint16 port, const char8 * data, int32 size )
{
sockaddr_in trgt;
memset(&trgt, 0, sizeof(trgt));
trgt.sin_family = AF_INET;
trgt.sin_port = htons(port);
trgt.sin_addr.s_addr = target.GetRaw();
if(sendto(internal->sock, (const char8 *)data, size, 0, (PSOCKADDR)&trgt, sizeof(trgt)) == SOCKET_ERROR)
{
LOG("Network sending error: %d", WSAGetLastError());
}
}
Call the "bind" function to specify a local port to send from. Example of using port 4567 below. Make sure to check the return value from bind.Call this code after you create the socket.
sockaddr_in local = {};
local.family = AF_INET;
local.port = htons(4567);
local.addr = INADDR_ANY;
bind(internal->sock,(sockaddr*)&local, sizeof(local));
If you bind to port zero instead of 4567 then the os will pick a random port for you and use it for all subsequent send and receives. You can call getsockname to discover which port the os picked for you after calling bind.
I wrote a small C/C++ Windows Mobile 6.5 client-application that is connecting to a server and sends some data to this server. The server is in my internal wireless network and is not reacheable outside.
The weird behaviour I'm having:
1) Even if the WIFI is not started on my mobile device, the connect() from the client-application returns success (!= SOCKET_ERROR), which is not the case b/c the server is reacheable only in the wireless network.
2) If the WIFI is not started on my mobile device, if there is a Sleep(1000) between the connect() and the send(), the send() fails with WSAECONNRESET, BUT if there is no Sleep() between the connect() and send() the send() succeeds! (only when doing the read() I finally get the WSAECONNRESET error).
Can somebody pls point me some tips why do I have this behaviour. It's pretty scary that without actually being able to reach the server I still get success for the connect() and for the send() :(
As requested, here is a sample code:
#include <windows.h>
#include <Winsock2.h>
#include "dbgview.h"
# define FxMemZero(buf,len) RtlZeroMemory ((VOID*)(buf),(SIZE_T)(len))
# define FxMemCopy(dst,src,len) RtlCopyMemory ((VOID*)(dst),(CONST VOID*)(src),(SIZE_T)(len))
int WINAPI WinMain(HINSTANCE hInstance,
HINSTANCE hPrevInstance,
LPTSTR lpCmdLine,
int nCmdShow)
{
SOCKET proxy_connection;
WSADATA wsadata;
if( 0 != WSAStartup (MAKEWORD(1, 1), &wsadata))
return -1;
proxy_connection = WSASocket (AF_INET, SOCK_STREAM, IPPROTO_TCP, NULL, 0, 0);
if(proxy_connection == INVALID_SOCKET) {
// error creating the socket
DbgViewTraceError((L"main", L"error creating socket."));
return -1;
}
// try to connect
UINT proxy_ip_ = 0x00000000;
CHAR* proxy_0_ = "192.168.1.105";
UINT proxy_port = 3100;
// get the proxy ip
{
struct hostent *he_;
if((he_ = gethostbyname(proxy_0_)) == NULL) {
DbgViewTraceWarning((L"main", L"error %d resolving hostname %hs", WSAGetLastError(), proxy_0_));
return -1;
}
FxMemCopy((PBYTE)&proxy_ip_, (PBYTE)he_->h_addr, he_->h_length);
}
// prepare the connection data
sockaddr_in saddr_;
FxMemZero(&saddr_,sizeof(sockaddr_in));
saddr_.sin_family = AF_INET;
saddr_.sin_addr.S_un.S_addr = proxy_ip_;// address
saddr_.sin_port = htons((USHORT)proxy_port);
// do the conection
if(SOCKET_ERROR == connect(proxy_connection, (SOCKADDR*) &saddr_, sizeof(saddr_))) {
// error connecting to the proxy
DbgViewTraceWarning(( L"main", L"error %d connecting to %hs:%d", WSAGetLastError(), proxy_0_, proxy_port));
closesocket(proxy_connection);
proxy_connection = INVALID_SOCKET;
return -1;
}
DbgViewTraceInfo(( L"main", L"SUCCESS. connected to %hs:%d.", proxy_0_, proxy_port));
CHAR* buffer_ = "Momo";
UINT count_ = strlen(buffer_);
DWORD total_ = 0;
DWORD sent_ = 0;
while(total_ < count_) {
// ISSUE: IF the WIFI is not started on the mobile, the connect returns success AND the send() returns success, even though with putty
// on the mobile, a telnet on 192.168.1.105:3100 will fail with: "Network error: Connection reset by peer"
// IF I add a long-enough Sleep() between the connect() and the send(), the send() will fail with: WSAECONNRESET
//Sleep(5000);
if(SOCKET_ERROR == (sent_ = send(proxy_connection, (const char*)buffer_ + total_, count_ - total_, 0))) {
// error sending data to the socket
DbgViewTraceError((L"main", L"error %d sending data to proxy", WSAGetLastError()));
return -1;
}
total_ += sent_;
}
DbgViewTraceInfo((L"main", L"send() SUCCESS"));
return 0;
}
The results are:
1) Without Sleep():
main [INFO ] SUCCESS. connected to 192.168.1.105:3100.
main [INFO ] send() SUCCESS
2) With Sleep():
main [INFO ] SUCCESS. connected to 192.168.1.105:3100.
main [ERROR ] error 10054 sending data to proxy
So the questions are:
1) Why the connect() succeeds? How can I be sure that there is actually a real connection?
2) Why the send() succeeds?
3) Why with a Sleep() in between connect() and send() the behaviour is different?
The problem seems to be ActiveSync. If ActiveSync is running, I get the behavior described above (connect() and send() report success, even though they are not). If ActiveSync is not running, gethostbyname() fails with:
WSAENETDOWN -> if WIFI is disabled
WSAHOST_NOT_FOUND -> if WIFI is enabled
which is correct!
How can this be? What is ActiveSync doing that is ruining everything? How can I avoid this problem? I mean, I can't be sure that the user is running my application when there is no ActiveSync running, so what can I do to avoid this behavior when ActiveSync is running?
Thx,
MeCoco
Looks like you are at least misusing struct sockaddr_in. Try more modern API for address conversion - Windows has InetPton - and see if that fixes the issues.
I'm using SO_REUSEADDR option, but I'm not sure why am getting
Resource temporary unvailable option.
I'm testing client server code on 127.0.0.1
if ( (sockfd = socket(AF_INET, SOCK_DGRAM, 0)) < 0 )
{
perror("socket() error!!\n");
exit(1);
}
if ( setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof(reuse) ) < 0 ) {
perror("SO_REUSEADDR failed::");
exit(1);
}
while(1) {
nbytes_read = recvfrom(sockfd, (void *)&recvd_msg, sizeof(recvd_msg),
flags, &from, &from_len);
printf("nbytes_read = %d\n", nbytes_read);
if(nbytes_read == -1) {
perror("client: recvfrom() failed");
return FAILED;
}
if (nbytes_read > 0) {
if(recvd_msg.hdr.msgtype == DATA)
printf("recvd %d bytes from server\n", recvd_msg.hdr.payload_size);
ftp_show_payload(&recvd_msg);
}
if(recvd_msg.hdr.is_last == TRUE) {
break;
}
}
Error message:
" client: recvfrom() failed: Resource temporarily unavailable"
errno:11
After trying to run client for 3-4 times, I get the data, I'm not sure whats happening.
Also, this problem is on Ubuntu Linux, but when I run the same client server on Solaris,
it works fine!!
SO_REUSEADDR is useful when you use bind(), but here you are not using bind.
I dont see any problem if recvfrom() returns -1
Use bind() and replace your call recvfrom() with recv(). recv() will receive all the packets at the port you used in your bind call.
Are you trimming out any other socket configuration? EAGAIN is typically returned when you read a non-blocking socket and there's no data available. The manpage for recvfrom lists the possible errnos that will be set on failure with an explanation for each one.
Your test is invalid. recvfrom() can return zero, which doesn't indicate an error. It is only valid to call perror() if you get -1. So you may not have a problem at all ..
I don't see why you're using SO_REUSEADDR at all here, as you're not binding to a specific port.