Bug 7909 - malfunction of udp_get_port() in wildcard ports assigning
Summary: malfunction of udp_get_port() in wildcard ports assigning
Status: RESOLVED CODE_FIX
Alias: None
Product: Networking
Classification: Unclassified
Component: IPV4 (show other bugs)
Hardware: i386 Linux
: P2 normal
Assignee: Stephen Hemminger
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2007-01-31 03:42 UTC by Konstantin Petrov
Modified: 2007-01-31 12:56 UTC (History)
1 user (show)

See Also:
Kernel Version: 2.6.19
Subsystem:
Regression: ---
Bisected commit-id:


Attachments

Description Konstantin Petrov 2007-01-31 03:42:00 UTC
Most recent kernel where this bug did *NOT* occur: 2.6.18.4
Distribution: Debian
Hardware Environment: PCs x86-64 and i686
Software Environment: 
Problem Description:

repeated binding of datagram sockets to some IP with wildcard
port fails after ~three hundred iterations with errno EADDRINUSE.
List of really assigned ports looks strangely:
...PORT[128-port-numbers-gap]PORT[128-port-numbers-gap]...

Steps to reproduce:

See code:

#include <stdio.h>
#include <netinet/in.h>
#include <sys/socket.h>
#include <sys/errno.h>

int
main(int argc, char **argv)
{
    int                                 iut_s;
    int                                 retval;
    struct sockaddr_in                  servaddr,
                                        cliaddr;
    socklen_t                           namelen;
    char                                addr_buf[16];
    int                                 i;

    /* Read command line arguments */
    if (argc != 2)
    {
        printf("\nuse pco_iut <local IP>\n");
        return -1;
    }

    memset(&servaddr, 0, sizeof(servaddr));
    servaddr.sin_family = AF_INET;
    inet_pton(AF_INET, argv[1], &servaddr.sin_addr);
    servaddr.sin_port = 0;

    for (i = 0; i < 1024; i++)
    {
        if ( (iut_s = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
        {
            printf("\nFailed to create socket No %d error %d\n", i + 1,
                    errno);
            return -1;
        }

        if (bind(iut_s,(struct sockaddr *) &servaddr, sizeof(servaddr)) != 0)
        {
            printf("\nFailed to bind socket No %d error %d\n", i + 1,
                    errno);
            return -1;
        }

        namelen = sizeof(cliaddr);
        if (getsockname(iut_s, (struct sockaddr *)&cliaddr, &namelen) != 0)
        {
            printf("\ngetsockname() error errno %d\n", errno);
            return -1;
        }
        else
        {
            inet_ntop(AF_INET, &cliaddr.sin_addr,
                        addr_buf, 16);
            printf("\nBound to %s:%d\n",
                    addr_buf,
                    ntohs(cliaddr.sin_port));
        }
    }

    return 0;
}


The reason of this issue is IMHO in udp_get_port() function
which was added in 2.6.19 version to replace udp_v4_get_port()
in older versions.
 
In ugp_get_port() we see the strange piece of code

.................................................................
                        size = 0;
                        sk_for_each(sk2, node, head)
                                if (++size < best_size_so_far) {
                                        best_size_so_far = size;
                                        best = result;
                                }
                }
                result = best;
                for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result +=
UDP_HTABLE_SIZE) {
..........................................................................

which means that sk_for_each(...) iteration is repeated only once.

No such problem in older kernels where the similar
piece of code in udp_v4_get_port() looks like

..........................................................................
                        size = 0;
                        sk_for_each(sk2, node, list)
                                if (++size >= best_size_so_far)
                                        goto next;
                        best_size_so_far = size;
                        best = result;
                next:;
                }
                result = best;
                for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result +=
UDP_HTABLE_SIZE) {

..........................................................................
Comment 1 Andrew Morton 2007-01-31 10:23:27 UTC

Begin forwarded message:

Date: Wed, 31 Jan 2007 03:50:56 -0800
From: bugme-daemon@bugzilla.kernel.org
To: bugme-new@lists.osdl.org
Subject: [Bugme-new] [Bug 7909] New: malfunction of udp_get_port() in wildcard ports assigning


http://bugzilla.kernel.org/show_bug.cgi?id=7909

           Summary: malfunction of udp_get_port() in wildcard ports
                    assigning
    Kernel Version: 2.6.19
            Status: NEW
          Severity: normal
             Owner: shemminger@osdl.org
         Submitter: Konstantin.Petrov@oktetlabs.ru
                CC: Alexandra.Kossovsky@oktetlabs.ru


Most recent kernel where this bug did *NOT* occur: 2.6.18.4
Distribution: Debian
Hardware Environment: PCs x86-64 and i686
Software Environment: 
Problem Description:

repeated binding of datagram sockets to some IP with wildcard
port fails after ~three hundred iterations with errno EADDRINUSE.
List of really assigned ports looks strangely:
...PORT[128-port-numbers-gap]PORT[128-port-numbers-gap]...

Steps to reproduce:

See code:

#include <stdio.h>
#include <netinet/in.h>
#include <sys/socket.h>
#include <sys/errno.h>

int
main(int argc, char **argv)
{
    int                                 iut_s;
    int                                 retval;
    struct sockaddr_in                  servaddr,
                                        cliaddr;
    socklen_t                           namelen;
    char                                addr_buf[16];
    int                                 i;

    /* Read command line arguments */
    if (argc != 2)
    {
        printf("\nuse pco_iut <local IP>\n");
        return -1;
    }

    memset(&servaddr, 0, sizeof(servaddr));
    servaddr.sin_family = AF_INET;
    inet_pton(AF_INET, argv[1], &servaddr.sin_addr);
    servaddr.sin_port = 0;

    for (i = 0; i < 1024; i++)
    {
        if ( (iut_s = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
        {
            printf("\nFailed to create socket No %d error %d\n", i + 1,
                    errno);
            return -1;
        }

        if (bind(iut_s,(struct sockaddr *) &servaddr, sizeof(servaddr)) != 0)
        {
            printf("\nFailed to bind socket No %d error %d\n", i + 1,
                    errno);
            return -1;
        }

        namelen = sizeof(cliaddr);
        if (getsockname(iut_s, (struct sockaddr *)&cliaddr, &namelen) != 0)
        {
            printf("\ngetsockname() error errno %d\n", errno);
            return -1;
        }
        else
        {
            inet_ntop(AF_INET, &cliaddr.sin_addr,
                        addr_buf, 16);
            printf("\nBound to %s:%d\n",
                    addr_buf,
                    ntohs(cliaddr.sin_port));
        }
    }

    return 0;
}


The reason of this issue is IMHO in udp_get_port() function
which was added in 2.6.19 version to replace udp_v4_get_port()
in older versions.
 
In ugp_get_port() we see the strange piece of code

.................................................................
                        size = 0;
                        sk_for_each(sk2, node, head)
                                if (++size < best_size_so_far) {
                                        best_size_so_far = size;
                                        best = result;
                                }
                }
                result = best;
                for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result +=
UDP_HTABLE_SIZE) {
..........................................................................

which means that sk_for_each(...) iteration is repeated only once.

No such problem in older kernels where the similar
piece of code in udp_v4_get_port() looks like

..........................................................................
                        size = 0;
                        sk_for_each(sk2, node, list)
                                if (++size >= best_size_so_far)
                                        goto next;
                        best_size_so_far = size;
                        best = result;
                next:;
                }
                result = best;
                for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result +=
UDP_HTABLE_SIZE) {

..........................................................................

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

Comment 2 Anonymous Emailer 2007-01-31 11:44:12 UTC
Reply-To: davem@davemloft.net


This was fixed several weeks ago.


Note You need to log in before you can comment on or make changes to this bug.