/cygdrive/c/src/openafs/openafs.git/repo/src/ubik/lock.c File Reference

Locks hang off of each transaction, with all the transaction hanging off of the appropriate dbase. More...

#include <afsconfig.h>
#include <afs/param.h>
#include <roken.h>
#include <afs/opr.h>
#include <lock.h>
#include "ubik.h"
#include "ubik_int.h"

Defines
#define	UBIK_INTERNALS   1
#define	WouldReadBlock(lock)   ((((lock)->excl_locked & WRITE_LOCK) || (lock)->wait_states) ? 0 : 1)
#define	WouldWriteBlock(lock)   ((((lock)->excl_locked & WRITE_LOCK) || (lock)->readers_reading) ? 0 : 1)
Functions
void	ulock_Init (void)
	locks
int	ulock_getLock (struct ubik_trans *atrans, int atype, int await)
	Set a transaction lock.
void	ulock_relLock (struct ubik_trans *atrans)
	Release the transaction lock.
void	ulock_Debug (struct ubik_debug *aparm)
	debugging hooks
Variables
struct Lock	rwlock

---

Detailed Description

Locks hang off of each transaction, with all the transaction hanging off of the appropriate dbase.

This package expects to be used in a two-phase locking protocol, so it doesn't provide a way to release anything but all of the locks in the transaction.

At present, it doesn't support the setting of more than one byte-position lock at a time, that is the length field must be 1. This doesn't mean that a single transaction can't set more than one lock, however.

It is the responsibility of the user to avoid deadlock by setting locks in a partial order.

#EWOULDBLOCK has been replaced in this file by #EAGAIN. Many Unix's but not all (eg. HP) do not replace #EWOULDBLOCK with #EAGAIN. The bad news is this goes over the wire. The good news is that the code path is never triggered as it requires ulock_getLock to be called with await = 0. And ulock_SetLock isn't even used in this code base. Since NT doesn't have a native #EAGAIN, we are replacing all instances of #EWOULDBLOCK with #EAGAIN.

---

Function Documentation

int ulock_getLock	(	struct ubik_trans *	atrans,
		int	atype,
		int	await
	)

Set a transaction lock.

Parameters:

atype	is #LOCKREAD or #LOCKWRITE.
await	is TRUE if you want to wait for the lock instead of returning #EWOULDBLOCK.

Note:

The #DBHOLD lock must be held.