Strange connect behavior

Strange connect behavior

[Videau Brice]

[-- Attachment #1: Type: text/plain, Size: 1034 bytes --]

Hello,

While writing some client server application in c, we noticed a strange 
behavior : if we try to connect endlessly to a given local port where 
nobody is listening, and if the port is >= to 32768, after several 
thousands tries ( Connection refused ) connect will return 0.
This behavior is not exhibited when port is < 32768.

We confirmed this behavior in kernel 2.6.17-10, 2.6.18-1, 2.6.8, on x86 
and 2.4.21-32 on ia64, on several hardware configurations.
Distribution is debian or ubuntu.

Attached is a source file that demonstrate this behavior.
./a.out port_number

Sample execution :

./a.out 35489
Out port : 35489
connect try 1 failed : Connection refused
connect try 2 failed : Connection refused
connect try 3 failed : Connection refused
.....
connect try 6089 failed : Connection refused
connect try 6090 failed : Connection refused
Connection success : 6091 try
Connection closed
last error : Connection refused


Is this behavior to be expected?
Can it be disabled?

Thanks in advance.

Regards,

Brice Videau

[-- Attachment #2: client.c --]
[-- Type: text/x-csrc, Size: 1536 bytes --]

#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <string.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <errno.h>

int main(int argc, char **argv) {
   unsigned short out_port = 35000;
   char * endptr;
   struct sockaddr_in server_address;
   int out_socket;
   int i=1;

   if( argc > 1 ) {
      out_port = (unsigned short) strtol(argv[1], &endptr, 10);
      errno = 0;
      if(errno != 0 || out_port == 0 || endptr == argv[1] ) {
         perror("Invalid port number");
         exit(1);
      }
   }
   printf( "Out port : %hu\n", out_port );
   fflush(NULL);

   memset( &server_address, 0, sizeof(struct sockaddr_in) );
   server_address.sin_family = AF_INET;
   server_address.sin_port = htons(out_port);
   if ( inet_pton( AF_INET, "127.0.0.1", &(server_address.sin_addr) ) <= 0 ) {
      perror("inet_pton error");
      exit(1);
   }

   if ( (out_socket = socket(AF_INET, SOCK_STREAM, 0)) < 0) {
      perror("socket error");
      exit(1);
   }

   while (connect(out_socket, (struct sockaddr *) &server_address, sizeof(struct sockaddr_in)) < 0)
   {
      fprintf(stderr,"connect try %i failed : ",i);
      perror("");
      fflush(NULL);
      i++;
   }

   printf("Connection success : %i try\n",i);
   fflush(NULL);

   if( close(out_socket) < 0 ) {
      perror("close error");
      exit(1);
   }

   printf("Connection closed\n");
   fflush(NULL);

   fprintf(stderr,"last error : ");
   perror("");

   return 0;
}

Re: Strange connect behavior

[Eric Dumazet]

On Monday 30 October 2006 11:24, Videau Brice wrote:
> Hello,
>
> While writing some client server application in c, we noticed a strange
> behavior : if we try to connect endlessly to a given local port where
> nobody is listening, and if the port is >= to 32768, after several
> thousands tries ( Connection refused ) connect will return 0.
> This behavior is not exhibited when port is < 32768.
>

Hello Brice

Yes, it's quite possible your attempts are hitting themselves.

Hint :

cat /proc/sys/net/ipv4/ip_local_port_range

When you connect(), TCP stack automatically chose a source port and bind your 
outgoing socket. If the chosen port happens to be 35489 (your 'destination' 
port), then the connect succeeds (you're connected to yourself), as specified 
by TCP specs.


> We confirmed this behavior in kernel 2.6.17-10, 2.6.18-1, 2.6.8, on x86
> and 2.4.21-32 on ia64, on several hardware configurations.
> Distribution is debian or ubuntu.
>
> Attached is a source file that demonstrate this behavior.
> ./a.out port_number
>
> Sample execution :
>
> ./a.out 35489
> Out port : 35489
> connect try 1 failed : Connection refused
> connect try 2 failed : Connection refused
> connect try 3 failed : Connection refused
> .....
> connect try 6089 failed : Connection refused
> connect try 6090 failed : Connection refused
> Connection success : 6091 try
> Connection closed
> last error : Connection refused
>
>
> Is this behavior to be expected?
> Can it be disabled?
>
> Thanks in advance.
>
> Regards,
>
> Brice Videau

Re: Strange connect behavior

[Videau Brice]

It all makes sense now!

Thank you very much.

Regards,

Brice



Eric Dumazet wrote:
> On Monday 30 October 2006 11:24, Videau Brice wrote:
>   
>> Hello,
>>
>> While writing some client server application in c, we noticed a strange
>> behavior : if we try to connect endlessly to a given local port where
>> nobody is listening, and if the port is >= to 32768, after several
>> thousands tries ( Connection refused ) connect will return 0.
>> This behavior is not exhibited when port is < 32768.
>>
>>     
>
> Hello Brice
>
> Yes, it's quite possible your attempts are hitting themselves.
>
> Hint :
>
> cat /proc/sys/net/ipv4/ip_local_port_range
>
> When you connect(), TCP stack automatically chose a source port and bind your 
> outgoing socket. If the chosen port happens to be 35489 (your 'destination' 
> port), then the connect succeeds (you're connected to yourself), as specified 
> by TCP specs.
>
>
>   
>> We confirmed this behavior in kernel 2.6.17-10, 2.6.18-1, 2.6.8, on x86
>> and 2.4.21-32 on ia64, on several hardware configurations.
>> Distribution is debian or ubuntu.
>>
>> Attached is a source file that demonstrate this behavior.
>> ./a.out port_number
>>
>> Sample execution :
>>
>> ./a.out 35489
>> Out port : 35489
>> connect try 1 failed : Connection refused
>> connect try 2 failed : Connection refused
>> connect try 3 failed : Connection refused
>> .....
>> connect try 6089 failed : Connection refused
>> connect try 6090 failed : Connection refused
>> Connection success : 6091 try
>> Connection closed
>> last error : Connection refused
>>
>>
>> Is this behavior to be expected?
>> Can it be disabled?
>>
>> Thanks in advance.
>>
>> Regards,
>>
>> Brice Videau
>>     
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
>