Description of problem: I am experiencing a situation where attempting to browse/connect to a z390 RHEL3 samba server causes the process to peg out at 100% cpu and begin to consume every increasing amounts of memory. I have tested basically the same configuration on a x86 RHEL3 server without problem. Once memory is fully consumed the vmachine hangs, which suggests the virtual memory killer is not working or is disabled. Here is a snapshot from strace of the smb process: munmap(0x100023fa000, 4096) = 0 mremap(0x1000241c000, 718000128, 718004224, MREMAP_MAYMOVE) = 0x1000241c000 open("/etc/passwd", O_RDONLY) = 25 fcntl(25, F_GETFD) = 0 fcntl(25, F_SETFD, FD_CLOEXEC) = 0 fstat(25, {st_mode=S_IFREG|0644, st_size=1492, ...}) = 0 mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x100023fa000 read(25, "root:x:0:0:root:/root:/bin/bash\n"..., 4096) = 1492 read(25, "", 4096) = 0 close(25) = 0 munmap(0x100023fa000, 4096) = 0 mremap(0x1000241c000, 718004224, 718004224, MREMAP_MAYMOVE) = 0x1000241c000 open("/etc/passwd", O_RDONLY) = 25 fcntl(25, F_GETFD) = 0 fcntl(25, F_SETFD, FD_CLOEXEC) = 0 fstat(25, {st_mode=S_IFREG|0644, st_size=1492, ...}) = 0 mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x100023fa000 read(25, "root:x:0:0:root:/root:/bin/bash\n"..., 4096) = 1492 read(25, "", 4096) = 0 close(25) The above repeats endlessly until all memory is consumed. I have added the following to /etc/pam.d/system-auth: auth sufficient /lib64/security/pam_winbind.so account required /lib64/security/pam_winbind.so session required /lib64/security/pam_mkhomedir.so skel=/etc/skel I found it odd that /lib/ did not contain these libs and that I had to use /lib64/. I am able to join this server to active directory and rpc without issue, and kinit _SEEMS_ to connect properly. I have verified that the same /etc/samba/smb.conf and /etc/krb5.conf works on my x86 test server without problem but I can provide those if needed. wbinfo -g and wbinfo -u reports fine, however "getent passwd" and "getent groups" reports the local machine accounts, then hangs when it gets to where it should report the active directory users and groups. Version-Release number of selected component (if applicable): samba-3.0.0-14.3E How reproducible: 1.Configure samba as a windows 2000 active directory member server (with winbind and kerberos) 2.Attempt to browse or connect to a samba share 3.Check for smb processes that are consuming all cpu and growing in memory usage. Steps to Reproduce: 1.Configure samba as a windows 2000 active directory member server (with winbind and kerberos) 2.Attempt to browse or connect to a samba share 3.Check for smb processes that are consuming all cpu and growing in memory usage. Actual results: Hung smb process for for each attempt at a connection. Expected results: Browse the server / mount the share Additional info:
Created attachment 97293 [details] My /etc/krb5.conf
Created attachment 97294 [details] My smb.conf file
I've built samba-3.0.2-2rc2.3E from the latest Samba release candidate. I've put i386 and s390x rpms at http://people.redhat.com/fenlason/.samba Please try them out and see if the problem persists. When the samba team releases 3.0.2 final, I intend to build an errata candidate from it.
It seems z390 (64bit) RHELv3 installs the s390x as well as the s390 (no x) for some rpms. samba-3.0.0-14.3E is one of those packages. Will this be an issue? Thanks.
Update, I went ahead and tested with these rpms. The results are the same getent passwd/group still does not report NT 4.0 domain users/groups and/or active directory users/groups. Trying to mount a share still hangs samba at 100% cpu per connection attempt until samba consumes all memory on the server at which point the server hangs. One thing I find curious is that all the documentation I read suggests that when getent fails but wbinfo succeeds that the problem is most likely that I have not linked libwinbind.so to libnss_winbind.so in /lib. Well libnnss_windbind.so exists but it is in /lib64 and not /lib could this be related? I have confirmed that this problem only appears to affect our s390 servers (tested 3) and not our x86 servers (tested 2).
The problem still exists in samba 3.0.2-EL errata.
the problem is verified/reproduced in internal environment ...
I've built test rpms from samba-3.0.2a with a patch to libnss_winbind.so (and smbd) that fixes the problem in my test environment. I've put them up on http://people.redhat.com/fenlason/.samba/ Can you please try them out, and if they fix the problem, I'll start the process of building "official" rpms.
The new packages fix my issue, I look forward to seeing the offical packages released. Thanks for the support.
An errata has been issued which should help the problem described in this bug report. This report is therefore being closed with a resolution of ERRATA. For more information on the solution and/or where to find the updated files, please follow the link below. You may reopen this bug report if the solution does not work for you. http://rhn.redhat.com/errata/RHSA-2004-259.html