mp_bh.c

这是国外的resip协议栈

/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 1996-2004
 *	Sleepycat Software.  All rights reserved.
 *
 * $Id: mp_bh.c,v 11.99 2004/10/15 16:59:42 bostic Exp $
 */

#include "db_config.h"

#ifndef NO_SYSTEM_INCLUDES
#include <sys/types.h>

#include <string.h>
#endif

#include "db_int.h"
#include "dbinc/db_shash.h"
#include "dbinc/mp.h"
#include "dbinc/log.h"
#include "dbinc/db_page.h"

static int __memp_pgwrite
	       __P((DB_ENV *, DB_MPOOLFILE *, DB_MPOOL_HASH *, BH *));

/*
 * __memp_bhwrite --
 *	Write the page associated with a given buffer header.
 *
 * PUBLIC: int __memp_bhwrite __P((DB_MPOOL *,
 * PUBLIC:      DB_MPOOL_HASH *, MPOOLFILE *, BH *, int));
 */
int
__memp_bhwrite(dbmp, hp, mfp, bhp, open_extents)
	DB_MPOOL *dbmp;
	DB_MPOOL_HASH *hp;
	MPOOLFILE *mfp;
	BH *bhp;
	int open_extents;
{
	DB_ENV *dbenv;
	DB_MPOOLFILE *dbmfp;
	DB_MPREG *mpreg;
	int ret;

	dbenv = dbmp->dbenv;

	/*
	 * If the file has been removed or is a closed temporary file, we're
	 * done -- the page-write function knows how to handle the fact that
	 * we don't have (or need!) any real file descriptor information.
	 */
	if (mfp->deadfile)
		return (__memp_pgwrite(dbenv, NULL, hp, bhp));

	/*
	 * Walk the process' DB_MPOOLFILE list and find a file descriptor for
	 * the file.  We also check that the descriptor is open for writing.
	 */
	MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);
	for (dbmfp = TAILQ_FIRST(&dbmp->dbmfq);
	    dbmfp != NULL; dbmfp = TAILQ_NEXT(dbmfp, q))
		if (dbmfp->mfp == mfp && !F_ISSET(dbmfp, MP_READONLY)) {
			++dbmfp->ref;
			break;
		}
	MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);

	if (dbmfp != NULL) {
		/*
		 * Temporary files may not have been created.  We only handle
		 * temporary files in this path, because only the process that
		 * created a temporary file will ever flush buffers to it.
		 */
		if (dbmfp->fhp == NULL) {
			/* We may not be allowed to create backing files. */
			if (mfp->no_backing_file)
				return (EPERM);

			MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);
			if (dbmfp->fhp == NULL)
				ret = __db_appname(dbenv, DB_APP_TMP, NULL,
				    F_ISSET(dbenv, DB_ENV_DIRECT_DB) ?
				    DB_OSO_DIRECT : 0, &dbmfp->fhp, NULL);
			else
				ret = 0;
			MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);
			if (ret != 0) {
				__db_err(dbenv,
				    "unable to create temporary backing file");
				return (ret);
			}
		}

		goto pgwrite;
	}

	/*
	 * There's no file handle for this file in our process.
	 *
	 * !!!
	 * It's the caller's choice if we're going to open extent files.
	 */
	if (!open_extents && F_ISSET(mfp, MP_EXTENT))
		return (EPERM);

	/*
	 * !!!
	 * Don't try to attach to temporary files.  There are two problems in
	 * trying to do that.  First, if we have different privileges than the
	 * process that "owns" the temporary file, we might create the backing
	 * disk file such that the owning process couldn't read/write its own
	 * buffers, e.g., memp_trickle running as root creating a file owned
	 * as root, mode 600.  Second, if the temporary file has already been
	 * created, we don't have any way of finding out what its real name is,
	 * and, even if we did, it was already unlinked (so that it won't be
	 * left if the process dies horribly).  This decision causes a problem,
	 * however: if the temporary file consumes the entire buffer cache,
	 * and the owner doesn't flush the buffers to disk, we could end up
	 * with resource starvation, and the memp_trickle thread couldn't do
	 * anything about it.  That's a pretty unlikely scenario, though.
	 *
	 * Note we should never get here when the temporary file in question
	 * has already been closed in another process, in which case it should
	 * be marked dead.
	 */
	if (F_ISSET(mfp, MP_TEMP))
		return (EPERM);

	/*
	 * It's not a page from a file we've opened.  If the file requires
	 * input/output processing, see if this process has ever registered
	 * information as to how to write this type of file.  If not, there's
	 * nothing we can do.
	 */
	if (mfp->ftype != 0) {
		MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);
		for (mpreg = LIST_FIRST(&dbmp->dbregq);
		    mpreg != NULL; mpreg = LIST_NEXT(mpreg, q))
			if (mpreg->ftype == mfp->ftype)
				break;
		MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);
		if (mpreg == NULL)
			return (EPERM);
	}

	/*
	 * Try and open the file, specifying the known underlying shared area.
	 *
	 * !!!
	 * There's no negative cache, so we may repeatedly try and open files
	 * that we have previously tried (and failed) to open.
	 */
	if ((ret = __memp_fcreate(dbenv, &dbmfp)) != 0)
		return (ret);
	if ((ret = __memp_fopen(dbmfp,
	    mfp, NULL, DB_DURABLE_UNKNOWN, 0, mfp->stat.st_pagesize)) != 0) {
		(void)__memp_fclose(dbmfp, 0);

		/*
		 * Ignore any error if the file is marked dead, assume the file
		 * was removed from under us.
		 */
		if (!mfp->deadfile)
			return (ret);

		dbmfp = NULL;
	}

pgwrite:
	ret = __memp_pgwrite(dbenv, dbmfp, hp, bhp);
	if (dbmfp == NULL)
		return (ret);

	/*
	 * Discard our reference, and, if we're the last reference, make sure
	 * the file eventually gets closed.
	 */
	MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);
	if (dbmfp->ref == 1)
		F_SET(dbmfp, MP_FLUSH);
	else
		--dbmfp->ref;
	MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);

	return (ret);
}

/*
 * __memp_pgread --
 *	Read a page from a file.
 *
 * PUBLIC: int __memp_pgread __P((DB_MPOOLFILE *, DB_MUTEX *, BH *, int));
 */
int
__memp_pgread(dbmfp, mutexp, bhp, can_create)
	DB_MPOOLFILE *dbmfp;
	DB_MUTEX *mutexp;
	BH *bhp;
	int can_create;
{
	DB_ENV *dbenv;
	MPOOLFILE *mfp;
	size_t len, nr;
	u_int32_t pagesize;
	int ret;

	dbenv = dbmfp->dbenv;
	mfp = dbmfp->mfp;
	pagesize = mfp->stat.st_pagesize;

	/* We should never be called with a dirty or a locked buffer. */
	DB_ASSERT(!F_ISSET(bhp, BH_DIRTY | BH_DIRTY_CREATE | BH_LOCKED));

	/* Lock the buffer and swap the hash bucket lock for the buffer lock. */
	F_SET(bhp, BH_LOCKED | BH_TRASH);
	MUTEX_LOCK(dbenv, &bhp->mutex);
	MUTEX_UNLOCK(dbenv, mutexp);

	/*
	 * Temporary files may not yet have been created.  We don't create
	 * them now, we create them when the pages have to be flushed.
	 */
	nr = 0;
	if (dbmfp->fhp != NULL)
		if ((ret = __os_io(dbenv, DB_IO_READ,
		    dbmfp->fhp, bhp->pgno, pagesize, bhp->buf, &nr)) != 0)
			goto err;

	/*
	 * The page may not exist; if it doesn't, nr may well be 0, but we
	 * expect the underlying OS calls not to return an error code in
	 * this case.
	 */
	if (nr < pagesize) {
		/*
		 * Don't output error messages for short reads.  In particular,
		 * DB recovery processing may request pages never written to
		 * disk or for which only some part have been written to disk,
		 * in which case we won't find the page.  The caller must know
		 * how to handle the error.
		 */
		if (can_create == 0) {
			ret = DB_PAGE_NOTFOUND;
			goto err;
		}

		/* Clear any bytes that need to be cleared. */
		len = mfp->clear_len == 0 ? pagesize : mfp->clear_len;
		memset(bhp->buf, 0, len);

#if defined(DIAGNOSTIC) || defined(UMRW)
		/*
		 * If we're running in diagnostic mode, corrupt any bytes on
		 * the page that are unknown quantities for the caller.
		 */
		if (len < pagesize)
			memset(bhp->buf + len, CLEAR_BYTE, pagesize - len);
#endif
		++mfp->stat.st_page_create;
	} else
		++mfp->stat.st_page_in;

	/* Call any pgin function. */
	ret = mfp->ftype == 0 ? 0 : __memp_pg(dbmfp, bhp, 1);

	/* Unlock the buffer and reacquire the hash bucket lock. */
err:	MUTEX_UNLOCK(dbenv, &bhp->mutex);
	MUTEX_LOCK(dbenv, mutexp);

	/*
	 * If no errors occurred, the data is now valid, clear the BH_TRASH
	 * flag; regardless, clear the lock bit and let other threads proceed.
	 */
	F_CLR(bhp, BH_LOCKED);
	if (ret == 0)
		F_CLR(bhp, BH_TRASH);

	return (ret);
}

/*
 * __memp_pgwrite --
 *	Write a page to a file.
 */
static int
__memp_pgwrite(dbenv, dbmfp, hp, bhp)
	DB_ENV *dbenv;
	DB_MPOOLFILE *dbmfp;