I use Samba 2 and then 3.0.1 ... 3.0.6 on a server for years. Everything works fine until I upgrade from 3.0.5 to 3.0.6. I've compiled Samba 3.0.6 as I do usually : ./configure --with-nis --with-smbmount && make && make install Nothing else has changed on the server (neither smb.conf). When I transfer files from a WinXP computer to the Linux samba server the transfert is very very slow (at least ten times slower than with Samba 3.0.5 with the same WinXP PC). I downgrade Samba back to 3.0.5 and everything works fine again. The server is a Dell PowerEdge 2300 running Linux Mandrake 9.1 with nss_ldap, NIS ypbind and LVM on ext2fs filesystems. I have the same problem on another Dell PowerEdge 2400 server with roughtly the same Linux and Samba config. Both servers use Linux Trustees ACL (http://trustees.sf.net) without change for years.
I install new samba 3.0.6, and cannot read (resp reading is time-outed) from WinXP (win98 works fine) from samba shared disks. I cannot downgrade back to 3.0.5 because samba generate error: [2004/08/25 08:57:02, 0] passdb/pdb_tdb.c:tdbsam_tdbopen(201) TDBSAM version unknown: 2 [2004/08/25 08:57:02, 0] passdb/pdb_tdb.c:tdbsam_getsampwnam(427) pdb_getsampwnam: Unable to open TDB passwd (/var/lib/samba/passdb.tdb)! samba is unusable for normal work
same problem occurred, when accessing samba 3.0.6 from linux box via smbfs and smbclient, transfer stop after receiving 16k of data, sending working fine directory listing working form small directories (I think it is related with 16k limit)
next testing (with samba log level=9): when copying short file (shorter then 16k), file is transfered, but connection is not closed - client is still waiting, connection must be closed (interrupted) manually (debug-short_file.log) when copying long file (>16k), transfer stopped after 16k, must be interrupted (debug-long_file.log)
Created attachment 618 [details] lgo from samba server for short file
Created attachment 619 [details] debug from samba server for long file (>16k)
I can confirm the same basic problem. samba-3.0.6 is terrbly broken when talking to XP SP1. I use debian and had to downgrade to samba-3.0.5. I'm using debian unstable, and I was able to downgrade because I only have a minimal setup, so I blew away all my .tdb files and remade them as necessary. The symptoms are obvious: XP just hangs in blocking network calls for minutes; it seems to be waiting for some samba response that never comes. The same happens in XP command consoles. Again, no configuration changes at all. Simply switching to the 3.0.5 packages fixed all the problems I was experiencing. The following debian bug id seems related: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=267766
Could you add sniffs of 3.0.5 compared to 3.0.6 for the same operations? The log file does not really show any reason why the client disconnects. Volker
Created attachment 623 [details] log from samba server v3.0.5 - long file (over 16k) ok, here is log from 3.0.5 version samba server with same log level (level 5) same operation (pruduction system, same configuration like in buggy 3.0.6 server)
Created attachment 624 [details] log from samba server v3.0.5 - short file
Sorry, but I did not mean 'log file' but 'sniff'. To take a sniff you need to install the tcpdump program and run: tcpdump -i <eth0> -n -s 1500 -w /tmp/sniff.cap host <client-ip> Replace <eth0> with the interface name your client comes via, and replace <client-ip> with the client's IP address. Then restart Samba, retry the operation that fails. Then stop the tcpdump program. /tmp/sniff.cap should contain the data we need. Volker
aah ok, but I must install samba 3.0.6 to other server (i can't break work in office), so it take some time...
Created attachment 631 [details] tcpdump capture of 3.0.5, "ls; get mirror"
Created attachment 632 [details] tcpdump capture of 3.0.6, "ls; get mirror", hang, ctrl-c
from debian bugs database:http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=267766 all reported problems was spoted on servers running kernel 2.4.20 (vanilla or patched) - me too. after upgrading to 2.4.27, resp. 2.6.8 kernel, all working fine, without any problem. I think, this problem is kernel related (resp. networking in kernel 2.4.20)
Your last comment triggers something: Could you do us a favor and try 2.4.20 with 'use sendfile = no'? Volker
ok, but i have 2.4.20 only on main server, so i can restart it after workhours (approx. 19:00 CEST - 8 hours from now) (I can't install it on other machine, it is compiled directly for server hw, but i try to found 2.4.20 kernel from archives)
last news: I was installed 2.4.20 kernel on my laptop with samba 3.0.6 - problem appeared as was expected with parameter use sendfile = no , all working fine do you need tcpdump captures ?
Jeremy, please add the necessary tests to configure.in to black list sendfile on 2.4.20. Thanks.
*** Bug 1942 has been marked as a duplicate of this bug. ***
fixed. please test 3.0.11
sorry for the same, cleaning up the database to prevent unecessary reopens of bugs.