+++ /dev/null
-/*
- * Copyright 2000, International Business Machines Corporation and others.
- * All Rights Reserved.
- *
- * This software has been released under the terms of the IBM Public
- * License. For details, see the LICENSE file in the top-level source
- * directory or online at http://www.openafs.org/dl/license10.html
- */
-
-#error I thought venus/lock.h was never included, what's going on?
-
-#if 0
-/* I believe this file is never actually used */
-
-#ifndef __AFSLOCK_INCLUDE__
-#define __AFSLOCK_INCLUDE__ 1
-
-#warning UMR-INCLUDE venus/lock.h
-
-/*******************************************************************\
-* *
-* Information Technology Center *
-* Carnegie-Mellon University *
-* *
-* *
-* *
-\*******************************************************************/
-
-/*
- Include file for using Vice locking routines.
-*/
-
-/* The following macros allow multi statement macros to be defined safely, i.e.
- - the multi statement macro can be the object of an if statement;
- - the call to the multi statement macro may be legally followed by a semi-colon.
- BEGINMAC and ENDMAC have been tested with both the portable C compiler and
- Hi-C. Both compilers were from the Palo Alto 4.2BSD software releases, and
- both optimized out the constant loop code. For an example of the use
- of BEGINMAC and ENDMAC, see the definition for ReleaseWriteLock, below.
- An alternative to this, using "if(1)" for BEGINMAC is not used because it
- may generate worse code with pcc, and may generate warning messages with hi-C.
-*/
-
-#define BEGINMAC do {
-#define ENDMAC } while (0)
-
-struct afs_bozoLock {
- short count; /* count of excl locks */
- char flags; /* bit 1: is anyone waiting? */
- char spare; /* for later */
- char *proc; /* process holding the lock, really a struct proc * */
-};
-
-#define AFS_BOZONWAITING 1 /* someone is waiting for this lock */
-
-/* all locks wait on excl_locked except for READ_LOCK, which waits on readers_reading */
-struct afs_lock {
- unsigned char wait_states; /* type of lockers waiting */
- unsigned char excl_locked; /* anyone have boosted, shared or write lock? */
- unsigned char readers_reading; /* # readers actually with read locks */
- unsigned char num_waiting; /* probably need this soon */
-};
-
-#define READ_LOCK 1
-#define WRITE_LOCK 2
-#define SHARED_LOCK 4
-/* this next is not a flag, but rather a parameter to Afs_Lock_Obtain */
-#define BOOSTED_LOCK 6
-
-/* next defines wait_states for which we wait on excl_locked */
-#define EXCL_LOCKS (WRITE_LOCK|SHARED_LOCK)
-
-#define ObtainReadLock(lock)\
- if (!((lock)->excl_locked & WRITE_LOCK))\
- (lock) -> readers_reading++;\
- else\
- Afs_Lock_Obtain(lock, READ_LOCK)
-
-#define ObtainWriteLock(lock)\
- if (!(lock)->excl_locked && !(lock)->readers_reading)\
- (lock) -> excl_locked = WRITE_LOCK;\
- else\
- Afs_Lock_Obtain(lock, WRITE_LOCK)
-
-#define ObtainSharedLock(lock)\
- if (!(lock)->excl_locked)\
- (lock) -> excl_locked = SHARED_LOCK;\
- else\
- Afs_Lock_Obtain(lock, SHARED_LOCK)
-
-#define UpgradeSToWLock(lock)\
- if (!(lock)->readers_reading)\
- (lock)->excl_locked = WRITE_LOCK;\
- else\
- Afs_Lock_Obtain(lock, BOOSTED_LOCK)
-
-/* this must only be called with a WRITE or boosted SHARED lock! */
-#define ConvertWToSLock(lock)\
- BEGINMAC\
- (lock)->excl_locked = SHARED_LOCK; \
- if((lock)->wait_states) \
- Afs_Lock_ReleaseR(lock); \
- ENDMAC
-
-#define ConvertWToRLock(lock) \
- BEGINMAC\
- (lock)->excl_locked &= ~(SHARED_LOCK | WRITE_LOCK);\
- (lock)->readers_reading++;\
- Afs_Lock_ReleaseR(lock);\
- ENDMAC
-
-#define ConvertSToRLock(lock) \
- BEGINMAC\
- (lock)->excl_locked &= ~(SHARED_LOCK | WRITE_LOCK);\
- (lock)->readers_reading++;\
- Afs_Lock_ReleaseR(lock);\
- ENDMAC
-
-#define ReleaseReadLock(lock)\
- BEGINMAC\
- if (!--(lock)->readers_reading && (lock)->wait_states)\
- Afs_Lock_ReleaseW(lock) ; \
- ENDMAC
-
-#define ReleaseWriteLock(lock)\
- BEGINMAC\
- (lock)->excl_locked &= ~WRITE_LOCK;\
- if ((lock)->wait_states) Afs_Lock_ReleaseR(lock);\
- ENDMAC
-
-/* can be used on shared or boosted (write) locks */
-#define ReleaseSharedLock(lock)\
- BEGINMAC\
- (lock)->excl_locked &= ~(SHARED_LOCK | WRITE_LOCK);\
- if ((lock)->wait_states) Afs_Lock_ReleaseR(lock);\
- ENDMAC
-
-/* I added this next macro to make sure it is safe to nuke a lock -- Mike K. */
-#define LockWaiters(lock)\
- ((int) ((lock)->num_waiting))
-
-#define CheckLock(lock)\
- ((lock)->excl_locked? (int) -1 : (int) (lock)->readers_reading)
-
-#define WriteLocked(lock)\
- ((lock)->excl_locked & WRITE_LOCK)
-
-/*
-
-You can also use the lock package for handling parent locks for independently-lockable sets of
-small objects. The concept here is that the parent lock is at the same level in the
-locking hierarchy as the little locks, but certain restrictions apply.
-
-The general usage pattern is as follows. You have a set of entries to search. When searching it, you
-have a "scan" lock on the table. If you find what you're looking for, you drop the lock down
-to a "hold" lock, lock the entry, and release the parent lock. If you don't find what
-you're looking for, you create the entry, downgrade the "scan" lock to a "hold" lock,
-lock the entry and unlock the parent.
-
-To delete an item from the table, you initially obtain a "purge" lock on the parent. Unlike all
-of the other parent lock modes described herein, in order to obtain a "purge" lock mode, you
-must have released all locks on any items in the table. Once you have obtained the parent
-lock in "purge" mode, you should check to see if the entry is locked. If its not locked, you
-are free to delete the entry, knowing that no one else can attempt to obtain a lock
-on the entry while you have the purge lock held on the parent. Unfortunately, if it *is* locked,
-you can not lock it yourself and wait for the other dude to release it, since the entry's locker
-may need to lock another entry before unlocking the entry you want (which would result in
-deadlock). Instead, then, you must release the parent lock, and try again "later" (see Lock_Wait
-for assistance in waiting until later). Unfortunately, this is the best locking paradigm I've yet
-come up with.
-
-What are the advantages to this scheme? First, the use of the parent lock ensures that
-two people don't try to add the same entry at the same time or delete an entry while someone
-else is adding it. It also ensures that when one process is deleting an entry, no one else is
-preparing to lock the entry. Furthermore, when obtaining a lock on a little entry, you
-are only holding a "hold" lock on the parent lock, so that others may come in and search
-the table during this time. Thus it will not hold up the system if a little entry takes
-a great deal of time to free up.
-
-Here's how to compute the compatibility matrix:
-
-The invariants are:
-
-add no deletions, additions allowed, additions will be performed, will obtain little locks
-hold no deletions, additions allowed, no additions will be performed, will obtain little locks
-purge no deletions or additions allowed, deletions will be performed, don't obtain little locks
-
-When we compute the locking matrix, we note that hold is compatible with hold and add.
-Add is compatible only with hold. purge is not compatible with anything. This is the same
-matrix as obtained by mapping add->S, hold->read and purge->write locks. Thus we
-can use the locks above to solve this problem, and we do.
-
-*/
-#endif /* __AFSLOCK_INCLUDE__ */
-
-
-#endif
git://git.openafs.org / openafs.git / commitdiff