Patch-ID# 100075-08
Keywords: lockd, rpc.lockd, file locking
Synopsis: SunOS 4.1;4.1.1;4.1.2: lockd jumbo patch
Date: 25/Mar/92

SunOS release: 4.1, 4.1.1, 4.1.2
 
Unbundled Product:
 
Topic: rpc.lockd jumbo patch
 
BugId's fixed with this patch: 1077402 1080359 1040406 1046373 1063692 1064183 1071435 1074757

Architectures for which this patch is available: sun4 sun4c sun4m sun3 and sun3x

Problem Description:

	PROBLEMS FIXED BY 100075-08 PATCH

	i) klm protocol was changed but the rpc number was not.

	ii) Shared or exclusive lock relative to EOF does not work
	correctly.

	iii) Deadlock is not detected when locking over NFS.

	iv) Downgrade of a lock does not grant other read locks.

	v) Panic "corrupt file lock list" might occur.

	vi) rpc.lockd does not release memory allocated and grows in size.

	vii) When doing "region locking" some regions do not get properly
	unlocked.

	viii) Granting of locks not working correctly when local and remote
	locks on same file are used at the same time.

	ix) Unlock that is interrupted by signal could leave the file as
	being locked by a process that does not exist.

	x) Core dump of rpc.lockd when unlocking in lm_unlock_region.

	xi) When server is rebooted locks are not reclaimed correctly.

	PROBLEMS FIXED BY 100075-07 PATCH

	WARNING:
	-------
	Due to fixes for iv) v) and vii) listed below, there is a potential 
	interoperabilty problem when 07 patch is the client and 06 patch is the 
	server; ie. lock requests that are being queued in the 06 server may 
	not get granted.
	
	There is no side-effect expected if the 07 patch is the server.

	i) Check for sanity of lock requests and return EINVAL if lock ranges
	is abnormal (ie. start < end).

	ii) Fix problem where locking to EOF request results in the last byte 
	of the file not being locked.

	iii) Fix problems where lock requests tend to get lost when 
	interruptions to the daemon hosed its internal data structures;
	this may also cause the daemon to core dump in certain cases.

	iv) Fix problems where the server lock daemon not sending back 
	NLM_GRANTED_MSG to remote blocked clients.

	v) Fix protocol implementation problem with NLM_GRANTED_MSG.

	vi) Fixed problems where NLM_CANCEL_MSGS and KLM_CANCEL_MSG result
	in applications waiting forever.

	vii) Fix error recovery mechanism to allow clients to reclaim locks 
	after servers crash & recover.

	viii) Fix the krpc call timeout to 5 sec instead of 45 sec for better
	response.  This also fixed the potential problem where results 
	for lock request is missed by the kernel.

	ix) Fix "panic iinactive" on the server.

	x) Fix problem where the temporary file /tmp/foo is not being removed
	after the client lock mgr crashes & recovers.
	
	xi) Fix panic trap data fault when more than 135 processes
	concurrently try an do file locking.

	PROBLEMS FIXED BY 100075-06 PATCH

	i) Fixed problems where locks were getting lost on a heavily
	loaded system, particularly when using shared locks or test
	lock calls.

	ii) Fixed problem where pc-nfs applications were failing with
	"rpc.lockd: unable to unlock a lock" and "rpc.lockd: unable to 
	set a lock."

	iii) Fixed problem with automatic upgrade and downgrade of 
	locks.

	iv) Fixed problem with client reboot (L1 A) and locks not being
	recovered after that.

	v) Fixed problem with signal interrupting lock calls and wrong
	error code returned.

	vi) Fixed problem with infinite retry of lock on unlinked files.

	vii) Fixed problem with local blocking shared locks not being
	granted a lock when one is available.

	viii) Fixed problem with u-area overwrite when doing test lock.

	ix) Fixed problem where messages like "klm_lockmgr: unlock denied?!" 
	and "lock-manager: RPC error: .." no longer appear under normal 
	operations.

	x) Additionally this patch allows the kernel lockf debugging code
	to be turned on and off dynamically by setting/unsetting the
        variable lock_debug_on using the following command:-

                #adb -w -k /vmunix /dev/mem
                physmem XXX
                lock_debug_on/W 1
                ^D


        PROBLEMS FIXED BY 100075-05 PATCH

	i) Fixed problem with running out of file descriptors and
	getting RPC TIMEOUT errors, seen when running large number
	of diskless clients.

	ii) Fixed problem where fd and fd structure was not released
	when doing test lock, eventually running out of file 
	descriptors. This problem would occur when running WP and
	quitting out of the window.

	iii) Fixed problem where rpc.lockd core dumps after a large
	amount of time when running "fame" application.

	iv) Fixed problem where restarting lockd on a client results
	in the server not being able to communicate with the new lockd
	as it has an old client handle that is associated with a now
	invalid port number.

	v) Fixed problem where upgrade from a read lock to a write lock 
	is allowed when remote read locks are outstanding.


INSTALL:
 
As root backup the files to be patched and then install the appropriate patches
as follows:
 
	For 4.1 and 4.1.1:
        mv /usr/sys/`arch -k`/OBJ/kern_descrip.o /usr/sys/`arch -k`/OBJ/kern_descrip.o.FCS

	For all OS levels:
        mv /usr/sys/`arch -k`/OBJ/klm_lockmgr.o /usr/sys/`arch -k`/OBJ/klm_lockmgr.o.FCS
        mv /usr/sys/`arch -k`/OBJ/ufs_lockf.o /usr/sys/`arch -k`/OBJ/ufs_lockf.o.FCS
	mv /usr/etc/rpc.lockd /usr/etc/rpc.lockd.FCS
	
	For 4.1 and 4.1.1:
        cp {4.1.1}/`arch -k`/kern_descrip.o /usr/sys/`arch -k`/OBJ

	For all OS levels:
        cp {4.1,4.1.1,4.1.2}/`arch -k`/klm_lockmgr.o /usr/sys/`arch -k`/OBJ
        cp {4.1,4.1.1,4.1.2}/`arch -k`/ufs_lockf.o /usr/sys/`arch -k`/OBJ
        cp {4.1,4.1.1,4.1.2}/`arch`/rpc.lockd /usr/etc
        chmod 755 /usr/etc/rpc.lockd
 
Then a new kernel will need to be made.
 
Please refer to the System and Network Administration manual
on building and installing a custom kernel.