Samba-2.07pre2 still crashes under Linux and HP-UX

[David Collier-Brown - Sun Canada]

Jason Haar <Jason.Haar at trimble.co.nz> wrote:
| Well I've just had a crash last night, and I've copies the log files off.
| But I see nothing to suggest what happened. Also I've enabled core-dumps,
| and no core dump exits, even though smbd died...
| 
| If you look at the smb.conf file, you'll see I've put in a HUGE sleep
| statement to be called via "panic action". Well I can tell you there was no
| left-over smbd process!!

	Cool!  That implies it tore down the connection "normally"...
	
	[For the readers of samba at samba.org: clients are failing
	mysteriously on an otherwise stable server. This is repeatable
	on Linux and HP/UX both]

	The log starts at [2000/03/23 13:54:11], and
	at 15:02:38 we get a Gethostbyaddr failure. The sequence was:
---
[2000/03/23 15:02:38, 0] lib/util_sock.c:client_name(999)
  getpeername failed. Error was Bad file descriptor
[2000/03/23 15:02:38, 0] lib/util_sock.c:client_name(999)
  getpeername failed. Error was Bad file descriptor
[2000/03/23 15:02:38, 1] lib/util_sock.c:client_name(1007)
  Gethostbyaddr failed for 155.63.51.194
---
then, a little later,
---
[2000/03/23 15:03:44, 1] lib/util_sock.c:client_name(1007)
  Gethostbyaddr failed for 155.63.51.52
[2000/03/23 15:03:44, 1] lib/util_sock.c:client_name(1007)
  Gethostbyaddr failed for 155.63.34.14	
---
The getpeername failures continue: it repeats many times
apparently trying to look up new clients as they connect.
I see 110 failures, apparently two per client looked up
Do you have about 55 clients?
---
Around 19:13:00, 5 hours later, we see the first actual login,
starting with an SMBnegprot (negotiate protocol) packet.
We see a bunch of connections, each getting an smb process:
---  
  switch message SMBnegprot (pid 21613)
  switch message SMBnegprot (pid 21837)
  switch message SMBnegprot (pid 22131)
  switch message SMBnegprot (pid 22408)
  switch message SMBnegprot (pid 22567)
  switch message SMBnegprot (pid 23350)
  switch message SMBnegprot (pid 23648)
  switch message SMBnegprot (pid 24090)
---

around 8 PM, a "TRANSFER" account logs on, looks at stuff
and logs off.
---
[2000/03/23 20:39:01, 3] smbd/password.c:register_vuid(277)
  User name: transfer   Real name: Transfer Mechanism Account
[2000/03/23 20:39:01, 3] smbd/ipc.c:api_RNetShareEnum(1608)
  RNetShareEnum gave 8 entries of 8 (1 65504 245 65504)
[2000/03/23 20:39:01, 4] rpc_client/cli_netlogon.c:cli_net_sam_logon(361)
  cli_net_sam_logon: srv:\\NZ01ADC01 mc:CROM clnt 78244259E48CFB34 38d9d825 

[2000/03/23 20:39:01, 3 smbd/server.c:exit_server(435)
  Server exit (normal exit)
transfer at crom-crom.trimble.co.nz-Samba.log.old100644      0      0      154037!
  smb_vwv[7]=0 (0x0)
---

	This implies that the process seen from Samba is normal:
	something else is going on, and besides that, we have a problem
	with getpeername.
	
	Anyone here recognize the getpeername problem?
	
	
--dave
[Jason: cna you give us a detailed description of what the person at
the PC client sees when the world goes bad?  We're going to look at
clients and packet logs next...]
--
David Collier-Brown in Boston
Phone: (781) 442-0734, Room BUR03-3632