I'm running Samba 2.2.3a on NetBSD. I've been having an annoying problem with winbindd dying inexplicably. I finally found a test that seems to (eventually) kill it: while wbinfo -t > /dev/null; do /usr/bin/true; done The problem is a memory leak in winbindd. Actually, wbinfo -t will still work, but wbinfo -u doesn't (response.extra_data is always null). Is this a known problem? Any fixes? Ken
Ken, We have the same problem with our setup on RedHat 6.2 (2.2.20 ACL patched). We have two servers running identical hardware and software and winbind dies every couple of hours, sometimes more often, sometimes less often. It dies during the night and at weekends so it does not seem to be load dependent. You say it is a memory leak ? There was a nasty one in 2.2.2 but that does not seem to be the case with 2.2.3a - just the dying is a problem. I have had a RedHat 7.1 machine running continuous tests on winbind for the last five days and it has not died once. I am going to add the ACL patches next and see if that affects winbind. Are you using ACLs ? Noel -----Original Message----- From: Ken Cross [mailto:kcross@nssolutions.com] Sent: 12 March 2002 14:04 To: samba@lists.samba.org Subject: [Samba] Memory leak in winbindd I'm running Samba 2.2.3a on NetBSD. I've been having an annoying problem with winbindd dying inexplicably. I finally found a test that seems to (eventually) kill it: while wbinfo -t > /dev/null; do /usr/bin/true; done The problem is a memory leak in winbindd. Actually, wbinfo -t will still work, but wbinfo -u doesn't (response.extra_data is always null). Is this a known problem? Any fixes? Ken -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Hi there, I am having the same problem with the winbindd daemon. I am NOT using ACL... It simply dies on me:-) I am running Red Hat 7.2 with the latest samba. Martijn Tigchelaar. -----Original Message----- From: Noel Kelly [mailto:nkelly@tarsus.co.uk] Sent: Wednesday, March 13, 2002 09:24 To: 'Ken Cross'; samba@lists.samba.org Subject: RE: [Samba] Memory leak in winbindd Ken, We have the same problem with our setup on RedHat 6.2 (2.2.20 ACL patched). We have two servers running identical hardware and software and winbind dies every couple of hours, sometimes more often, sometimes less often. It dies during the night and at weekends so it does not seem to be load dependent. You say it is a memory leak ? There was a nasty one in 2.2.2 but that does not seem to be the case with 2.2.3a - just the dying is a problem. I have had a RedHat 7.1 machine running continuous tests on winbind for the last five days and it has not died once. I am going to add the ACL patches next and see if that affects winbind. Are you using ACLs ? Noel -----Original Message----- From: Ken Cross [mailto:kcross@nssolutions.com] Sent: 12 March 2002 14:04 To: samba@lists.samba.org Subject: [Samba] Memory leak in winbindd I'm running Samba 2.2.3a on NetBSD. I've been having an annoying problem with winbindd dying inexplicably. I finally found a test that seems to (eventually) kill it: while wbinfo -t > /dev/null; do /usr/bin/true; done The problem is a memory leak in winbindd. Actually, wbinfo -t will still work, but wbinfo -u doesn't (response.extra_data is always null). Is this a known problem? Any fixes? Ken -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Hi Jeremy, I reproduced the winbindd core on HP-UX 11.0 with the little script that Noel sent; There was NO activity on this winbindd server except for the script doing the wbinfo -t's; (only nmbd and winbindd were running). I started the script at about 17:21, and it dumped core at 22:00. The log.winbindd showed: [2002/03/13 17:21:38, 1] lib/debug.c:(248) INFO: Debug class all level = 1 (pid 14703 from pid 14703) [2002/03/13 17:21:38, 0] lib/util_unistr.c:(582) load_unicode_map: filename /usr/local/samba/lib/codepages/unicode_map.850 does not exist. [2002/03/13 17:21:38, 0] lib/util_unistr.c:(582) load_unicode_map: filename /usr/local/samba/lib/codepages/unicode_map.ISO8859- 1 does not exist. [2002/03/13 17:21:38, 1] nsswitch/winbindd_util.c:(137) getting trusted domain list [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) adding domain WT1 [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) adding domain ATL-WTEC [2002/03/13 21:59:32, 0] lib/util.c:(1064) PANIC: init_unistr2: malloc fail And the backtrace showed: (gdb) bt #0 0xc01082b8 in kill () from /usr/lib/libc.2 #1 0xc00a52e8 in raise () from /usr/lib/libc.2 #2 0xc00e5c8c in abort_C () from /usr/lib/libc.2 #3 0xc00e5ce4 in abort () from /usr/lib/libc.2 #4 0x3e1e0 in smb_panic () #5 0x65624 in init_unistr2 () #6 0x75410 in init_q_req_chal () #7 0xaddc0 in cli_net_req_chal () #8 0xaea64 in cli_nt_setup_creds () #9 0x10814 in cm_get_netlogon_cli () #10 0xf0a0 in winbindd_check_machine_acct () #11 0x71e4 in process_request () #12 0x73f8 in process_packet () #13 0x79f8 in process_loop () #14 0x7f18 in main () Does this match the issue you just fixed, or is this different? Hope this helps, don -----Original Message----- From: jra@samba.org [mailto:jra@samba.org] Sent: Wednesday, March 13, 2002 11:13 PM To: Ken Cross Cc: 'Noel Kelly'; samba@lists.samba.org Subject: Re: [Samba] Memory leak in winbindd On Wed, Mar 13, 2002 at 08:12:26AM -0500, Ken Cross wrote:> Noel: > > There's definitely a memory leak (try the little script at the bottom > and it's obvious -- the wbinfo -t does it). > > I'm not certain that the memory leak causes it to die or not. Actually, > on my system it isn't dead -- it just doesn't return user/group data any > more. wbinfo -t works, but not wbinfo -u. I think the memory leak > fixed in 2.2.3 was in smbd not winbind, wasn't it?I've just fixed a bug in SAMBA_2_2 CVS that could cause winbindd to core dump on a lookup name/sid returning SID_NAME_UNKNOWN. If you could cvs checkout this branch and see if it fixes the winbindd dying problem I'd appreciate it. Thanks, Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
Sorry Jeremy - hold off and I'll try it again (reproduction) tonight with -g... don -----Original Message----- From: jra@samba.org [mailto:jra@samba.org] Sent: Thursday, March 14, 2002 12:47 PM To: MCCALL,DON (HP-USA,ex1) Cc: 'jra@samba.org'; Ken Cross; 'Noel Kelly'; samba@lists.samba.org Subject: Re: [Samba] Memory leak in winbindd On Thu, Mar 14, 2002 at 10:45:41AM -0500, MCCALL,DON (HP-USA,ex1) wrote:> Hi Jeremy, > I reproduced the winbindd core on HP-UX 11.0 > with the little script that Noel sent; There was > NO activity on this winbindd server except for > the script doing the wbinfo -t's; (only nmbd > and winbindd were running). I started the script > at about 17:21, and it dumped core at 22:00. > The log.winbindd showed: > > [2002/03/13 17:21:38, 1] lib/debug.c:(248) > INFO: Debug class all level = 1 (pid 14703 from pid 14703) > [2002/03/13 17:21:38, 0] lib/util_unistr.c:(582) > load_unicode_map: filename/usr/local/samba/lib/codepages/unicode_map.850> does > not exist. > [2002/03/13 17:21:38, 0] lib/util_unistr.c:(582) > load_unicode_map: filename > /usr/local/samba/lib/codepages/unicode_map.ISO8859- > 1 does not exist. > [2002/03/13 17:21:38, 1] nsswitch/winbindd_util.c:(137) > getting trusted domain list > [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) > adding domain WT1 > [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) > adding domain ATL-WTEC > [2002/03/13 21:59:32, 0] lib/util.c:(1064) > PANIC: init_unistr2: malloc fail > > And the backtrace showed: > > > (gdb) bt > #0 0xc01082b8 in kill () from /usr/lib/libc.2 > #1 0xc00a52e8 in raise () from /usr/lib/libc.2 > #2 0xc00e5c8c in abort_C () from /usr/lib/libc.2 > #3 0xc00e5ce4 in abort () from /usr/lib/libc.2 > #4 0x3e1e0 in smb_panic () > #5 0x65624 in init_unistr2 () > #6 0x75410 in init_q_req_chal () > #7 0xaddc0 in cli_net_req_chal () > #8 0xaea64 in cli_nt_setup_creds () > #9 0x10814 in cm_get_netlogon_cli () > #10 0xf0a0 in winbindd_check_machine_acct () > #11 0x71e4 in process_request () > #12 0x73f8 in process_packet () > #13 0x79f8 in process_loop () > #14 0x7f18 in main () > > Does this match the issue you just fixed, or is this different? > > Hope this helps,No it's different.... I don't suppose you could do the same with *symbols* enabled :-) :-). That might actually help in getting the line numbers at fault.... Just a thought ! (Yes I know I'm being facecious here, but for heavens sake - you know how to compile with -g I'm sure :-) :-) :-). Now I've got to grub through to try and find a call sequence that matches, whereas with one -g you could have pointed me right at it ! Oh well, shouldn't complain about free help I suppose :-) :-). Cheers, Jeremy. (Who is grouchy 'cos he's still in bloody Roseville :-) :-). -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
No Problem, J! I don't generally compile with -g because of (would you believe it) I'm THAT tight on disk space on the 11.x system I have available to test on... But I SHOULD have gone back and recompiled winbindd with -g; will do so next time... ;-> Thanks for the quick action! don -----Original Message----- From: jra@samba.org [mailto:jra@samba.org] Sent: Thursday, March 14, 2002 13:18 To: MCCALL,DON (HP-USA,ex1) Cc: 'jra@samba.org'; Ken Cross; 'Noel Kelly'; samba@lists.samba.org Subject: Re: [Samba] Memory leak in winbindd On Thu, Mar 14, 2002 at 10:45:41AM -0500, MCCALL,DON (HP-USA,ex1) wrote:> [2002/03/13 17:21:38, 0] lib/util_unistr.c:(582) > load_unicode_map: filename > /usr/local/samba/lib/codepages/unicode_map.ISO8859- > 1 does not exist. > [2002/03/13 17:21:38, 1] nsswitch/winbindd_util.c:(137) > getting trusted domain list > [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) > adding domain WT1 > [2002/03/13 17:21:39, 1] nsswitch/winbindd_util.c:(103) > adding domain ATL-WTEC > [2002/03/13 21:59:32, 0] lib/util.c:(1064) > PANIC: init_unistr2: malloc failOk, sorry for being so grumpy - it was fairly obvious where the problem lies - this looks like a memory leak due to the main_loop_talloc not being freed in winbindd. I've just committed a fix for this - this should fix the winbindd memory leak problem people have reported. Sorry for complaining about the bug report - it did allow me to find it :-). Thanks (sheepishly), Jeremy. -- To unsubscribe from this list go to the following URL and read the instructions: http://lists.samba.org/mailman/listinfo/samba
I have been running the CVS code for SAMBA_2_2 (2.2.4 pre) from AM Tuesday 19th since last night. We have not had one crash but unfortunately the winbindd process over the fifteen hour period managed to consume 290Mb of RAM! I have restarted now and will keep a closer eye on memory. Noel -----Original Message----- From: jra@samba.org [mailto:jra@samba.org] Sent: 14 March 2002 04:13 To: Ken Cross Cc: 'Noel Kelly'; samba@lists.samba.org Subject: Re: [Samba] Memory leak in winbindd On Wed, Mar 13, 2002 at 08:12:26AM -0500, Ken Cross wrote:> Noel: > > There's definitely a memory leak (try the little script at the bottom > and it's obvious -- the wbinfo -t does it). > > I'm not certain that the memory leak causes it to die or not. Actually, > on my system it isn't dead -- it just doesn't return user/group data any > more. wbinfo -t works, but not wbinfo -u. I think the memory leak > fixed in 2.2.3 was in smbd not winbind, wasn't it?I've just fixed a bug in SAMBA_2_2 CVS that could cause winbindd to core dump on a lookup name/sid returning SID_NAME_UNKNOWN. If you could cvs checkout this branch and see if it fixes the winbindd dying problem I'd appreciate it. Thanks, Jeremy.