Samba 1.9.16pl11: dropping connections between Solaris 2.5.1 and WinNT4SP3

[Dutchyn, Chris]

Hi all:

I am experiencing a problem where WinNT and Samba are dropping
connections during file transfers.  I have the following smb.conf
settings:

	keep alive = 30
	dead time = 1440
	socket options = IPTOS_THROUGHPUT SO_KEEPALIVE

(although I tried IPTOS_NODELAY with no apparent benefit).  The client
is copying a large number of files up to the server, then there is a
pause at a random file, followed by a "connection closed" message on the
client.  It appears as if the smbd server process restarts.  Sure, NT
automatically reconnects and starts copying from the next file, but the
user has to dismiss the error dialog and copy the failed file
separately.  This drastically diminishes the utility of samba.  The
relevant samba.log file fragment is below.

Christopher Dutchyn			cdutchyn at aec.ca
AECWest SA


<last successful file open for write>
	switch message SMBopenX (pid 2206)
	Skipping become_user - already user
	unix_clean_name [./w2/Install/Fixes/ASP.DLL]
	reduce_name [w2/Install/Fixes/ASP.DLL] [/export/LogArch]
	chdir to w2/Install/Fixes
	chdir to /export/LogArch
	reduced to w2/Install/Fixes/ASP.DLL
	calling open_file with flags=0x1 flags2=0x300 mode=0664
	Allocated new file_fd_struct 0, dev = ffffffff, inode = ffffffff
	07/15/97 16:43:18 cdutchyn opened file w2/Install/Fixes/ASP.DLL
read=No write=Yes (numopen=1 fnum=1)
	chdir to /
	07/15/97 16:43:18 Transaction 110 of length 52
	switch message SMBwrite (pid 2206)
	chdir to /export/LogArch
	07/15/97 16:43:18 write fnum=1 cnum=85 num=0 wrote=0
	07/15/97 16:43:18 Transaction 111 of length 54
	switch message SMBsetattrE (pid 2206)
	Skipping become_user - already user
	07/15/97 16:43:18 reply_setattrE fnum=1 cnum=85
	07/15/97 16:43:18 Transaction 112 of length 61440
<let's write a block>
	switch message SMBwrite (pid 2206)
	Skipping become_user - already user
	07/15/97 16:43:18 write fnum=1 cnum=85 num=61388 wrote=61388
	07/15/97 16:43:18 Transaction 113 of length 104
<lots more writing ... deleted because it's pretty much the same as the
previous block>
	...
<whew! last block>
	switch message SMBwrite (pid 2206)
	Skipping become_user - already user
	07/15/97 16:43:18 write fnum=1 cnum=85 num=784 wrote=784
	07/15/97 16:43:18 Transaction 132 of length 46
<close the file>
	switch message SMBclose (pid 2206)
	Skipping become_user - already user
	fd_attempt_close on file_fd_struct 0, fd = 8, dev = 2840005,
inode = b, open_flags = 2, ref_count = 1.
	07/15/97 16:43:18 cdutchyn closed file w2/Install/Fixes/ASP.DLL
(numopen=0)
	07/15/97 16:43:18 close fd=-1 fnum=1 cnum=85 (numopen=0)
	07/15/97 16:43:18 Transaction 133 of length 95
<open file which disconnects>
	switch message SMBopenX (pid 2206)
	Skipping become_user - already user
	unix_clean_name [./w2/Install/Fixes/ASP.PDB]
	reduce_name [w2/Install/Fixes/ASP.PDB] [/export/LogArch]
	chdir to w2/Install/Fixes
	chdir to /export/LogArch
	reduced to w2/Install/Fixes/ASP.PDB
	calling open_file with flags=0x1 flags2=0x300 mode=0664
	Allocated new file_fd_struct 0, dev = ffffffff, inode = ffffffff
	07/15/97 16:43:18 cdutchyn opened file w2/Install/Fixes/ASP.PDB
read=No write=Yes (numopen=1 fnum=1)
	chdir to /
	07/15/97 16:43:18 Transaction 134 of length 52
	switch message SMBwrite (pid 2206)
	chdir to /export/LogArch
	07/15/97 16:43:18 write fnum=1 cnum=85 num=0 wrote=0
	07/15/97 16:43:18 Transaction 135 of length 54
	switch message SMBsetattrE (pid 2206)
	Skipping become_user - already user
		07/15/97 16:43:18 reply_setattrE fnum=1 cnum=85
<looks good to here, comparing to other logs and the previous success>
		doing parameter default case = lower
<what's this?  it's rereading the smb.conf file!  what happened to my
connection>
	doing parameter guest account = nobody
<skip the rest of the read parameters>
	doing parameter write list = cdutchyn, exp010
	pm_process() returned Yes
	Scanning username map /opt/samba/lib/users.map
	adding IPC service
	07/15/97 16:43:44 changed root to /
	priming nmbd
	sending a packet of len 1 to (127.0.0.1) on port 137 of type
DGRAM
	07/15/97 16:43:44 Transaction 0 of length 72
	netbios connect: name1=BABS             name2=CDUTCHYN       
	Scanning username map /opt/samba/lib/users.map
	Mapped user cdutchyn to cdutchyn
	07/15/97 16:43:44 Transaction 1 of length 174
<well ... a new pid! why?>
	switch message SMBnegprot (pid 2211)
	Requested protocol [PC NETWORK PROGRAM 1.0]