NDK problem when multiple accept()

user726090

Hi.

I Think I found a bug that I want to report in sockpcb.c, SockCleanPcb().
We have a system that has multiple accept() waiting for Connection. When we change IP-address,
only one of them is signaled and aborted (ECONNABORTED).
What I found out is that the pProtNext is set to NULL in SockClose() and therefor only the first one found is signaled.
I made the following change, marked below.
We are using the latest NDK (2_25_00_09) and recomended xdctools/bios. The problem can be found in eariler NDK:s also.

-----------------------

    /* Cycle through all the entries in the socket table. */
    ps = pSockList [SockProt];
    while (ps != NULL)
    {
        // Save pProtNext because SockClose() sets it to NULL in SockPcbDetach()
        pNextTmp = ps->pProtNext;

        /* Clean the entry only if we have a match */
        if (ps->LIP == IPAddress) {
            /*
             * Look for any sockets that are blocked on accept() call
             * (fix for SDOCM00115513)
             */
            if (ps->OptionFlags & SO_ACCEPTCONN) {
                /* Set an error for this socket to force accept() to return */
                error = ECONNABORTED;
                SockSet((HANDLE)ps, SOL_SOCKET, SO_ERROR, &error,
                        sizeof(error));
                /*
                 * Wake the socket blocked on accept(), allowing accept() to
                 * handle the error and return
                 */
                FdSignalEvent(ps, FD_EVENT_READ);
            }

            SockClose (ps); // pProtNext = NULL;
        }

        /* Get the next entry. */
        // ps = ps->pProtNext; Doesn't work
        ps = pNextTmp;
   }

Regards

over 9 years ago

0 Biser Gatchev-XID over 9 years ago

TI__Guru**** 393215 points

Your post is invisible. Please resend.

0 user726090 over 9 years ago in reply to Biser Gatchev-XID

Intellectual 530 points

Hi.

I Think I found a bug that I want to report in sockpcb.c, SockCleanPcb(). 

We have a system that has multiple accept() waiting for Connection. When we change IP-address, 
only one of them is signaled and aborted (ECONNABORTED).

What I found out is that the pProtNext is set to NULL in SockClose() and therefor only the first one found is signaled.

I made the following change, marked below.

We are using the latest NDK (2_25_00_09) and recomended xdctools/bios. The problem can be found in eariler NDK:s also.


-----------------------

    /* Cycle through all the entries in the socket table. */
    ps = pSockList [SockProt];
    while (ps != NULL)
    {
        // Save pProtNext because SockClose() sets it to NULL in SockPcbDetach() 
        pNextTmp = ps->pProtNext; 

        /* Clean the entry only if we have a match */
        if (ps->LIP == IPAddress) {
            /*
             *  Look for any sockets that are blocked on accept() call
             *  (fix for SDOCM00115513)
             */
            if (ps->OptionFlags & SO_ACCEPTCONN) {
                /* Set an error for this socket to force accept() to return */
                error = ECONNABORTED;
                SockSet((HANDLE)ps, SOL_SOCKET, SO_ERROR, &error,
                        sizeof(error));
                /*
                 *  Wake the socket blocked on accept(), allowing accept() to
                 *  handle the error and return
                 */
                FdSignalEvent(ps, FD_EVENT_READ);
            }
            
            SockClose (ps);  // pProtNext = NULL;
        }

        /* Get the next entry. */
        // ps = ps->pProtNext; Doesn't work
        ps = pNextTmp;
   }

Regards

Per-Anders

0 Biser Gatchev-XID over 9 years ago in reply to user726090

TI__Guru**** 393215 points

I don't see anything again.

0 user726090 over 9 years ago in reply to Biser Gatchev-XID

Intellectual 530 points

Hi.I Think I found a bug that I want to report in sockpcb.c, SockCleanPcb().

We have a system that has multiple accept() waiting for Connection.

When we change IP-address, only one of them is signaled and aborted (ECONNABORTED).
What I found out is that the pProtNext is set to NULL in SockClose() and therefor only the
first one found is signaled. I made the following change, marked below.

We are using the latest NDK (2_25_00_09) and recomended xdctools/bios.
The problem can be found in eariler NDK:s also.
-----------------------

/* Cycle through all the entries in the socket table. */

ps = pSockList [SockProt];
while (ps != NULL)
{
// Save pProtNext because SockClose() sets it to NULL in SockPcbDetach()
pNextTmp = ps->pProtNext;

/* Clean the entry only if we have a match */
if (ps->LIP == IPAddress) {
/*
* Look for any sockets that are blocked on accept() call
* (fix for SDOCM00115513)
*/
if (ps->OptionFlags & SO_ACCEPTCONN) {
/* Set an error for this socket to force accept() to return */
error = ECONNABORTED;
SockSet((HANDLE)ps, SOL_SOCKET, SO_ERROR, &error,
sizeof(error));
/*
* Wake the socket blocked on accept(), allowing accept() to
* handle the error and return
*/
FdSignalEvent(ps, FD_EVENT_READ);
}
SockClose (ps); // pProtNext = NULL;
}

/* Get the next entry. */
// ps = ps->pProtNext; Doesn't work
ps = pNextTmp;
}

Regards

Per-Anders

PS: Hope this works

0 dzhou over 9 years ago in reply to user726090

TI__Genius 9065 points

Hello,

Which device and software are you using?

best regards,

David Zhou

0 user726090 over 9 years ago in reply to dzhou

Intellectual 530 points

Hi!

i have tested this on both IDK437x and EVM57XX board with the latest NDK.

I have an application that has 4 different accept() calls (ftp/http/telnet etc.) waiting for connections.

When I change IP-address on the system via the CfgXXX API:s I see that only one of the accept() calls is aborted.

I tracked this down to the netsrv.c, SPIpNet() function. In that function you call:

SockCleanPcb (SOCKPROT_TCP, pi->IPAddr);
SockCleanPcb (SOCKPROT_UDP, pi->IPAddr);
SockCleanPcb (SOCKPROT_RAW, pi->IPAddr);

In SockCleanPcb() there is a "loop" that goes through all open sockets and determind if they shall be aborted. This works fine for the first found, but because the next pointer is overwritten, only the first is aborted.

I added some code myself in the SockCleanPcb() and now all the accept() calls are aborted as they shall. See my change below.

void SockCleanPcb (uint SockProt, IPN IPAddress)
{
    SOCK    *ps;
    struct _sock *Pnext;
    int error;

    /* Validate the socket protocol family. */
    if( SockProt<SOCKPROT_TCP || SockProt>SOCKPROT_RAW )
        return;

    /* Cycle through all the entries in the socket table. */
    ps = pSockList [SockProt];
    while (ps != NULL)
    {
        Pnext = ps->pProtNext;

        /* Clean the entry only if we have a match */
        if (ps->LIP == IPAddress) {
            /*
             * Look for any sockets that are blocked on accept() call
             * (fix for SDOCM00115513)
             */
            if (ps->OptionFlags & SO_ACCEPTCONN) {
                /* Set an error for this socket to force accept() to return */
                error = ECONNABORTED;
                SockSet((HANDLE)ps, SOL_SOCKET, SO_ERROR, &error,
                        sizeof(error));
                /*
                 * Wake the socket blocked on accept(), allowing accept() to
                 * handle the error and return
                 */
                FdSignalEvent(ps, FD_EVENT_READ);
            }

SockClose (ps); // Destroys the pProtNext
}

        /* Get the next entry. */
        ps = Pnext;
    }
    return;
}

Regards:

Per-Anders

0 user726090 over 9 years ago in reply to dzhou

Intellectual 530 points

Hi! Any feedback on this issue?

Regards

Per-Anders

0 user726090 over 9 years ago in reply to user726090

Intellectual 530 points

Hi!
Any news about this issue?

BR:

0 Garrett Ding over 9 years ago in reply to user726090

TI__Mastermind 43296 points

Hi Per-Anders,

The code change appears to make sense. We will have NDK team reviewed for completeness.

Regards,
Garrett

Processors

Processors forum

NDK problem when multiple accept()