samba-TNG smbd crashing

[Tom Crummey]

Hello Jeremy,

Luke asked me to contact you direct about the problem I'm having with
smbd crashing.

As I see it there are two issues to address:

1) Fix the problem causing the crash 
2) Define what should happen to locks when an instance of smbd crashes.

1) Background to the problem:

samba-TNG cvs update 11:00BST 05/04/00 Solaris 2.7 gcc 2.8.1 Sparc 64

I realise this isn't the latest version, but I don't think anyone has
addressed the problem as yet.

I have several Win2k clients using a samba-TNG server for domain logins
and profiles. Sometimes smbd crashes while the client is performing the
login script and leaves an oplock against it: (Extract from smbstatus:)

1163   DENY_WRITE RDONLY     EXCLUSIVE+BATCH  
/opt/samba-TNG/etc/netlogon/UNKNOWN.exe   Mon Apr 10 09:55:47 2000
1163   DENY_NONE  RDONLY     LEVEL_II         
/opt/samba-TNG/etc/netlogon/kixtart.scr   Mon Apr 10 09:55:58 2000

A core is not dumped because on Solaris a process will not produce
a core if its euid is different from its real uid. 

I have logs at level 8:

[2000/04/10 09:55:58, 5] rpc_parse/parse_prs.c:_prs_uint32s(890)
                  0210 sub_auths : 00000015 ccfe2690 ace38d7c c4a0db2f
[2000/04/10 09:55:58, 4] smbd/uid.c:become_userk(84)
  Skipping become_user - already user
[2000/04/10 09:55:58, 3] smbd/oplock.c:oplock_break(1075)
  oplock_break: returning success for dev = 800005, inode = 682
  Current exclusive_oplocks_open = 1
[2000/04/10 09:55:58, 4] smbd/open.c:open_file_shared(1012)
  calling open_file with flags=0x0 flags2=0x0 mode=0744
[2000/04/10 09:55:58, 3] smbd/files.c:fd_get_already_open(132)
  Re-used file_fd_struct dev = 800005, inode = 682, ref_count = 2
[2000/04/10 09:55:58, 0] lib/fault.c:fault_report(40)
  ===============================================================
[2000/04/10 09:55:58, 0] lib/fault.c:fault_report(41)
  INTERNAL ERROR: Signal 11 in pid 1163 (TNG-prealpha)
  Please read the file BUGS.txt in the distribution
[2000/04/10 09:55:58, 0] lib/fault.c:fault_report(43)
  ===============================================================
[2000/04/10 09:55:58, 0] lib/util.c:smb_panic(2627)
  PANIC: internal error
[2000/04/10 09:55:58, 6] param/loadparm.c:lp_file_list_changed(2057)
  lp_file_list_changed()
  file /opt/samba-TNG/lib/smb.conf -> /opt/samba-TNG/lib/smb.conf  last 
mod_time: Sun Apr  9 23:11:53 2000

[2000/04/10 09:55:58, 5] smbd/connection.c:claim_connection(84)
  claiming  100000

I don't know if this is enough to track the problem. If it isn't, 
let me know and I'll try and hack the code to allow me to get a core
dump.

Unfortunately, this problem does not occur on every login, so I can't
reproduce it at will, though it seems to be something to do with
re-using file structs.

2) When smbd crashes what should happen to file locks? Should they
be released by the crashing smbd? Should they be handed to another
smbd? Leaving them open is not satisfactory, especially when there
isn't a way to easily remove them by hand. Also, should smbd be forced
to drop a core when it crashes no matter what euid it has at the time?
i.e. change the euid back to root and then force the core dump.
(I'm not sure if this is possible, but perhaps others can comment.)

If you need any further information, please email me.

Tom.

----------------------------------------------------------------------------
 Tom Crummey, Systems and Network Manager,      EMAIL: tom at ee.ucl.ac.uk
 Department of Electronic and Electrical Engineering,                  
 University College London,                       TEL: +44 (0)20 7679 3898    
 Torrington Place,                                FAX: +44 (0)20 7388 9325
 London, UK, WC1E 7JE.                         
----------------------------------------------------------------------------