smbd processes "stuck" for several days

[mogambo]

Hello,

I am observing a strange phenomenon with our deployment of Samba
3.4.7-58.  Samba server is configured to authenticate against ADS.
There are no issues related to authentication.  There is apparently
only one client connected to the Samba server.  The machine has been
up for 22 days, and I see 4k+ smbd processes running at this time.

# smbstatus

Samba version 3.4.7-58.om2
PID     Username      Group         Machine
-------------------------------------------------------------------
2426      ...

# ps auxww | grep -c smbd
4082

# netstat -nap | grep -c smbd
4037
# netstat -nap | grep smbd | grep -c unix
4034
# netstat -nap | grep smbd | grep -v unix
tcp        0      0 0.0.0.0:139                 0.0.0.0:*
     LISTEN      28440/smbd
tcp        0      0 0.0.0.0:445                 0.0.0.0:*
     LISTEN      28440/smbd
tcp        0      0 192.168.161.254:445         192.168.160.190:50807
     ESTABLISHED 2426/smbd

# netstat -nap | grep smbd
tcp        0      0 0.0.0.0:139                 0.0.0.0:*
     LISTEN      28440/smbd
tcp        0      0 0.0.0.0:445                 0.0.0.0:*
     LISTEN      28440/smbd
tcp        0      0 192.168.161.254:445         192.168.160.190:50807
     ESTABLISHED 2426/smbd
unix  2      [ ]         DGRAM                    97725306 8589/smbd
unix  2      [ ]         DGRAM                    97725064 8563/smbd
unix  2      [ ]         DGRAM                    97699449 1187/smbd
unix  2      [ ]         DGRAM                    97699095 1086/smbd
...

I cannot find anything useful under /var/log/samba.

# grep -nr 8589 /var/log/samba
# grep -nr 8563 /var/log/samba
/var/log/samba/ing-ct-02.log.old:868:  ing-ct-02 (192.168.160.192)
connect to service smbshare initially as user ing-ct-02 domain FORK
(uid=100001, gid=100001) (pid 8563)
#

However, /var/log/messages has these messages:

Jul 24 00:55:38 smbd[8563]: [2012/07/24 00:55:38,  0]
lib/util_sock.c:539(read_fd_with_timeout)
Jul 24 00:55:38 smbd[8589]: [2012/07/24 00:55:38,  0]
lib/util_sock.c:539(read_fd_with_timeout)
Jul 24 00:55:38 smbd[8589]: [2012/07/24 00:55:38,  0]
lib/util_sock.c:1491(get_peer_addr_internal)
Jul 24 00:55:38 smbd[8563]: [2012/07/24 00:55:38,  0]
lib/util_sock.c:1491(get_peer_addr_internal)
Jul 24 00:55:38 smbd[8589]:   getpeername failed. Error was Transport
endpoint is not connected
Jul 24 00:55:38 smbd[8563]:   getpeername failed. Error was Transport
endpoint is not connected
Jul 24 00:55:38 smbd[8589]:   read_fd_with_timeout: client 0.0.0.0
read error = Connection reset by peer.
Jul 24 00:55:38 smbd[8563]:   read_fd_with_timeout: client 0.0.0.0
read error = Connection reset by peer.

The errors do not look good, but I have observed them on devices that
do not have lingering smbd processes.  There are coredump messages in
the log, which may be fixed in later version:

======= Backtrace: =========
/lib/libc.so.6[0xbbe424]
/lib/libc.so.6(__libc_free+0x77)[0xbbe95f]
*** glibc detected *** smbd: free(): invalid pointer: 0x009417f0 ***
/lib/libcom_err.so.2(remove_error_table+0x4b)[0x11eabb]

Any ideas where I can stat looking?

Thanks,
-m.