(SAMBA) issue with filehandles being released under 6.1-RELEASE?

[Garrett Cooper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Lowell Gilbert wrote:
> Garrett Cooper <youshi10 at u.washington.edu> writes:
> 
>> Just wondering if anyone else was experiencing the same issues that I
>> am.. it's related to Samba-3.0.22c and sockets / filehandles.
>>
>> For some odd reason every couple days (~2 days) I have to restart the
>> smbd daemon because it eats up 6000+ filehandles, just for sockets I
>> assume (based on netstat output). At the point where it reaches 8000
>> some filehandles open, the system refuses to fork, forcing me to login
>> as root on the console directly instead of via SSH.
>>
>> My machine is a local / preferred master (smbd fights with XP Home
>> clients because they want to be master browsers), with limited access
>> to a few XP clients (<4 clients at any given point in time), plus an
>> XBox using smbclient with XBox Media Center.
>>
>> Helpful info:
>> [root at hoover /home/gcooper]# uname -a
>> FreeBSD hoover.localdomain 6.1-RELEASE-p10 FreeBSD 6.1-RELEASE-p10 #9:
>> Mon Oct 16 02:14:29 PDT 2006
>> gcooper at hoover.localdomain:/usr/obj/usr/src/sys/HOOVER  i386
>>
>>
>> /usr/local/etc/smb.conf:
>> [global]
>>   workgroup = WORKGROUP
>>   encrypt passwords = yes
>>   log file = /var/log/samba/log.%m
>>   log level = 3 passdb:4 auth:4
>> #   log level = 5
>>   socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192
>>   local master =  yes
>>   preferred master = yes
>>   dns proxy = no
>>   guest ok = no
>>
>> [shared]
>>   path = /shared
>>   writeable = yes
>>   public = yes
>>   hosts deny = shiina pinocchio
>>   guest ok = no
>>   create mask = 0775
>>
>> The only changes that I've made to smb.conf between now and when I
>> last accessed samba is that I've removed an unneeded share and removed
>> guest advertisement for my shares (need password / username anyhow to
>> login, so I figured I might as well..).
> 
> This is a Samba issue, not an OS issue, as demonstrated by the fact
> that the system is able to close all of the handles when the daemon 
> is shut down.  
> 
> Have you looked closely at the netstat output to which you referred?
> It would be interesting if all of the stuck connections were coming
> from the same host, or were in a particular state, etc.

	I actually went in depth looking at the code for a long period of time
and it appears that it's expected for the smbd daemon to keep on
malloc'ing file descriptors (in this case sockets) until it can no
longer allocate file descriptors. Then it cleans house on all of the
file descriptors to reclaim them for itself.

	That doesn't seem to free up resources for the OS though--or at least
it wasn't obvious when looking through the code, but maybe the sockets
reclaim themselves due to the implementation of sockets... Needless to
say this problem is new (since 3.0.22x at least), but then again I've
never really had XP clients connecting to my FreeBSD box en-masse like
this before. Usually I shared via SMB with my strictly my XP dualboot
client and the xbox I have in my possession.

	My bug report that I made for this (closed it because I thought the
problem was resolved--process allocation) is:
<https://bugzilla.samba.org/show_bug.cgi?id=4207>.
- -Garrett
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFFWMpl6CkrZkzMC68RAkbPAJ44GsTfcyOj9zEFV9oCXFGzq7tHKwCeKqO0
HoLa/My3E5FVgzqnmF40CU8=
=ZA7e
-----END PGP SIGNATURE-----