RE: RE : [lwip-users] Receiving RSTs

[aparna.dutta]

I had initially tested my client code using the Windows socket API, but apparently the behaviour of the Lwip socket API is slightly different. I have put in the fix by closing the socket everytime connect() fails and creating a new socket again. Hopefully this would solve the issue.

 

Thanks a lot for all your replies,

 

(By the way, the  “sys_mbox_post: mbox not full" assertion is in sys_mbox_post() function in sys_arch.c of the VDSP port of Lwip for Blackfin)

 

 

Regards,

Aparna

 

 

---

From: address@hidden [mailto:address@hidden On Behalf Of Frédéric BERNON
Sent: Wednesday, December 12, 2007 12:59 PM
To: Mailing list for lwIP users
Subject: Re: RE : [lwip-users] Receiving RSTs

 

I did this test with last CVS HEAD (with a Vista workstation):

 

static void

tcp_connect_thread(void *arg)

{

int sock;

struct sockaddr_in local;

struct sockaddr_in to;

int iResult;

u32_t server_address;

 

/* initialize server address */

server_address = inet_addr("192.168.0.1");/*my pc*/

/* if we got a valid server address... */

if (server_address!=0) {

/* create new socket */

sock = socket( AF_INET, SOCK_STREAM, 0);

if (sock>=0) {

/* prepare local address */

memset(&local, 0, sizeof(local));

local.sin_family = AF_INET;

local.sin_port = htons(INADDR_ANY);

local.sin_addr.s_addr = htonl(INADDR_ANY);

/* bind to local address */

if (bind( sock, (struct sockaddr *)&local, sizeof(local))==0) {

/* prepare server address */

memset(&to, 0, sizeof(to));

to.sin_family = AF_INET;

to.sin_port = htons(62345);/* a port which is not opened*/

to.sin_addr.s_addr = server_address;

 

/* connect to server */

if ((iResult=connect( sock, (struct sockaddr *)&to, sizeof(to)))>=0) {

printf("connected success==%i [%i]\n", iResult, errno);

} else {

printf("connected error==%i [%i]\n", iResult, errno);

}

/* sleep */

sys_msleep(60000);/*let the time to see if any other packet is received*/

printf("close now\n");

 

}

/* close the socket */

closesocket(sock);

}

}

}

 

I got :

 

Starting lwIP, loopback interface IP is 127.0.0.1
Starting lwIP, local interface IP is 192.168.0.2
 0: \Device\NPF_{A6293F5E-B3DF-4E8D-8D01-C65870EEDD1E}
 1: \Device\NPF_{2DA42D7E-7436-40CD-901E-5D73107798B0}
MAC: 00:18:F3:E3:8D:8D
err_tcp==-4
connected error==-1 [0]
close now

-4 = ERR_ABRT

 

What I can said is the PC never send any RST packet (checked with wireshark). Of course, perhaps it's different with another pc OS.

 

But in all cases, in err_tcp, we set to NULL conn->pcb.tcp. So, we can't reuse the pcb. The only solution seems to close the socket and reopen it. If on RST, we receive any error, perhaps we could add a line like:

static void

err_tcp(void *arg, err_t err)

{

struct netconn *conn;

conn = arg;

LWIP_ASSERT("conn != NULL", (conn != NULL));

if (conn->pcb.tcp == NULL) return; /*<<<<<<<<<<<<<<<<<<<<<<<<<<*/

conn->pcb.tcp = NULL;

conn->err = err;

 

----- Original Message -----

From: "Jonathan Larmour" <address@hidden>

To: "Mailing list for lwIP users" <address@hidden>

Sent: Tuesday, December 11, 2007 11:45 PM

Subject: Re: RE : [lwip-users] Receiving RSTs

 

> address@hidden wrote:
>> Jonathan Larmour wrote:
>>
>>> I wonder from what Aparna says:
>>>
>>> 
>>>
>>>>>         When the Windows server is not up, the PC responds with a RST
>>>>>         for a received SYN. The connect() API returns an error, and the
>>>>>         client retries connect() after every few seconds.
>>>>>      
>>>
>>>
>>> whether he/she is not closing the socket before calling connect again?
>>>
>>> Maybe the netconn API needs to explicitly disallow connects if
>>> conn->err is
>>> set?
>>>
>>> Either way, if the socket isn't being closed, Aparna needs to fix
>>> his/her code.
>>>  
>>
>> Normally, the socket API allows calling connect for a second time (when
>> the first call fails and the socket is not already connected).
>
> Oh yes, of course you are right. Aparna's code should be valid for a BSD
> sockets API, sorry.
>
>> LwIP
>> might not allow this, in which case we can argue if there is
>> documentation missing or an assert (or a code change)...
>
> I think in the short term, disallowing based on non-zero conn->err would
> be best. This could be solved, but probably only when we have the richer
> set of synchronisation primitives we've been thinking about elsewhere.
>
> Jifl
> --
> eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
> Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
> Registered in England and Wales: Reg No 4422071.
> ------["The best things in life aren't things."]------      Opinions==mine
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users
>