db_dispatch.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.
 */
/*
 * Copyright (c) 1995, 1996
 *	The President and Fellows of Harvard University.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Margo Seltzer.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * $Id: db_dispatch.c,v 11.169 2004/10/27 16:44:26 ubell Exp $
 */

#include "db_config.h"

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

#include <stdlib.h>
#include <string.h>
#endif

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

#ifndef HAVE_FTRUNCATE
static int __db_limbo_fix __P((DB *,
    DB_TXN *, DB_TXNLIST *, db_pgno_t *, DBMETA *, db_limbo_state));
static int __db_limbo_bucket __P((DB_ENV *,
	     DB_TXN *, DB_TXNLIST *, db_limbo_state));
static int __db_limbo_move __P((DB_ENV *, DB_TXN *, DB_TXN *, DB_TXNLIST *));
static int __db_limbo_prepare __P(( DB *, DB_TXN *, DB_TXNLIST *));
static int __db_lock_move __P((DB_ENV *,
		u_int8_t *, db_pgno_t, db_lockmode_t, DB_TXN *, DB_TXN *));
static int __db_txnlist_pgnoadd __P((DB_ENV *, DB_TXNHEAD *,
		int32_t, u_int8_t [DB_FILE_ID_LEN], char *, db_pgno_t));
#endif
static int __db_txnlist_find_internal __P((DB_ENV *,
    void *, db_txnlist_type, u_int32_t, u_int8_t[DB_FILE_ID_LEN],
    DB_TXNLIST **, int, u_int32_t *));

/*
 * __db_dispatch --
 *
 * This is the transaction dispatch function used by the db access methods.
 * It is designed to handle the record format used by all the access
 * methods (the one automatically generated by the db_{h,log,read}.sh
 * scripts in the tools directory).  An application using a different
 * recovery paradigm will supply a different dispatch function to txn_open.
 *
 * PUBLIC: int __db_dispatch __P((DB_ENV *,
 * PUBLIC:     int (**)__P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *)),
 * PUBLIC:     size_t, DBT *, DB_LSN *, db_recops, void *));
 */
int
__db_dispatch(dbenv, dtab, dtabsize, db, lsnp, redo, info)
	DB_ENV *dbenv;		/* The environment. */
	int (**dtab)__P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *));
	size_t dtabsize;	/* Size of the dtab. */
	DBT *db;		/* The log record upon which to dispatch. */
	DB_LSN *lsnp;		/* The lsn of the record being dispatched. */
	db_recops redo;		/* Redo this op (or undo it). */
	void *info;
{
	DB_LSN prev_lsn;
	u_int32_t rectype, status, txnid;
	int make_call, ret;

	memcpy(&rectype, db->data, sizeof(rectype));
	memcpy(&txnid, (u_int8_t *)db->data + sizeof(rectype), sizeof(txnid));
	make_call = ret = 0;

	/* If we don't have a dispatch table, it's hard to dispatch. */
	DB_ASSERT(dtab != NULL);

	/*
	 * If we find a record that is in the user's number space and they
	 * have specified a recovery routine, let them handle it.  If they
	 * didn't specify a recovery routine, then we expect that they've
	 * followed all our rules and registered new recovery functions.
	 */
	switch (redo) {
	case DB_TXN_ABORT:
	case DB_TXN_APPLY:
	case DB_TXN_PRINT:
		make_call = 1;
		break;
	case DB_TXN_OPENFILES:
		/*
		 * We collect all the transactions that have
		 * "begin" records, those with no previous LSN,
		 * so that we do not abort partial transactions.
		 * These are known to be undone, otherwise the
		 * log would not have been freeable.
		 */
		memcpy(&prev_lsn, (u_int8_t *)db->data +
		    sizeof(rectype) + sizeof(txnid), sizeof(prev_lsn));
		if (txnid != 0 && prev_lsn.file == 0 && (ret =
		    __db_txnlist_add(dbenv, info, txnid, TXN_OK, NULL)) != 0)
			return (ret);

		/* FALLTHROUGH */
	case DB_TXN_POPENFILES:
		if (rectype == DB___dbreg_register ||
		    rectype == DB___txn_child ||
		    rectype == DB___txn_ckp || rectype == DB___txn_recycle)
			return (dtab[rectype](dbenv, db, lsnp, redo, info));
		break;
	case DB_TXN_BACKWARD_ROLL:
		/*
		 * Running full recovery in the backward pass. In general,
		 * we only process records during this pass that belong
		 * to aborted transactions.  Unfortunately, there are several
		 * exceptions:
		 * 1. If this is a meta-record, one not associated with
		 *    a transaction, then we must always process it.
		 * 2. If this is a transaction commit/abort, we must
		 *    always process it, so that we know the status of
		 *    every transaction.
		 * 3. If this is a child commit, we need to process it
		 *    because the outcome of the child transaction depends
		 *    on the outcome of the parent.
		 * 4. If this is a dbreg_register record, we must always
		 *    process is because they contain non-transactional
		 *    closes that must be properly handled.
		 * 5. If this is a noop, we must always undo it so that we
		 *    properly handle any aborts before a file was closed.
		 * 6. If this a file remove, we need to process it to
		 *    determine if the on-disk file is the same as the
		 *    one being described.
		 */
		switch (rectype) {
		/*
		 * These either do not belong to a transaction or (regop)
		 * must be processed regardless of the status of the
		 * transaction.
		 */
		case DB___txn_regop:
		case DB___txn_recycle:
		case DB___txn_ckp:
			make_call = 1;
			break;
		/*
		 * These belong to a transaction whose status must be
		 * checked.
		 */
		case DB___txn_child:
		case DB___db_noop:
		case DB___fop_file_remove:
		case DB___dbreg_register:
			make_call = 1;

			/* FALLTHROUGH */
		default:
			if (txnid == 0)
				break;

			ret = __db_txnlist_find(dbenv, info, txnid, &status);

			/* If not found, this is an incomplete abort.  */
			if (ret == DB_NOTFOUND)
				return (__db_txnlist_add(dbenv,
				    info, txnid, TXN_IGNORE, lsnp));
			if (ret != 0)
				return (ret);

			/*
			 * If we ignore the transaction, ignore the operation
			 * UNLESS this is a child commit in which case we need
			 * to make sure that the child also gets marked as
			 * ignore.
			 */
			if (status == TXN_IGNORE && rectype != DB___txn_child) {
				make_call = 0;
				break;
			}
			if (status == TXN_COMMIT)
				break;

			/* Set make_call in case we came through default */
			make_call = 1;
			if (status == TXN_OK &&
			    (ret = __db_txnlist_update(dbenv,
			    info, txnid, rectype == DB___txn_xa_regop ?
			    TXN_PREPARE : TXN_ABORT, NULL, &status, 0)) != 0)
				return (ret);
		}
		break;
	case DB_TXN_FORWARD_ROLL:
		/*
		 * In the forward pass, if we haven't seen the transaction,
		 * do nothing, else recover it.
		 *
		 * We need to always redo DB___db_noop records, so that we
		 * properly handle any commits after the file was closed.
		 */
		switch (rectype) {
		case DB___txn_recycle:
		case DB___txn_ckp:
		case DB___db_noop:
			make_call = 1;
			break;

		default:
			if (txnid != 0) {
				ret = __db_txnlist_find(dbenv,
				    info, txnid, &status);

				if (ret == DB_NOTFOUND)
					/* Break out out of if clause. */
					;
				else if (ret != 0)
					return (ret);
				else if (status == TXN_COMMIT) {
					make_call = 1;
					break;
				}
			}

#ifndef HAVE_FTRUNCATE
			if (status != TXN_IGNORE &&
			    (rectype == DB___ham_metagroup ||
			    rectype == DB___ham_groupalloc ||
			    rectype == DB___db_pg_alloc)) {
				/*
				 * Because we do not have truncate
				 * all allocation records must be reprocessed
				 * during rollforward in case the file was
				 * just created.  It may not have been
				 * present during the backward pass.
				 */
				make_call = 1;
				redo = DB_TXN_BACKWARD_ALLOC;
			} else
#endif
			if (rectype == DB___dbreg_register) {
				/*
				 * This may be a transaction dbreg_register.
				 * If it is, we only make the call on a COMMIT,
				 * which we checked above. If it's not, then we
				 * should always make the call, because we need
				 * the file open information.
				 */
				if (txnid == 0)
					make_call = 1;
			}
		}
		break;
	case DB_TXN_BACKWARD_ALLOC:
	default:
		return (__db_unknown_flag(
		    dbenv, "__db_dispatch", (u_int32_t)redo));
	}

	if (make_call) {
		/*
		 * If the debug flag is set then we are logging
		 * records for a non-durable update so that they
		 * may be examined for diagnostic purposes.
		 * So only make the call if we are printing,
		 * otherwise we need to extract the previous
		 * lsn so undo will work properly.
		 */
		if (rectype & DB_debug_FLAG) {
			if (redo == DB_TXN_PRINT)
				rectype &= ~DB_debug_FLAG;
			else {
				memcpy(lsnp,
				    (u_int8_t *)db->data +
				    sizeof(rectype) +
				    sizeof(txnid), sizeof(*lsnp));
				return (0);
			}
		}
		if (rectype >= DB_user_BEGIN && dbenv->app_dispatch != NULL)
			return (dbenv->app_dispatch(dbenv, db, lsnp, redo));
		else {
			/*
			 * The size of the dtab table argument is the same as
			 * the standard table, use the standard table's size
			 * as our sanity check.
			 */
			if (rectype > dtabsize || dtab[rectype] == NULL) {
				__db_err(dbenv,
				    "Illegal record type %lu in log",
				    (u_long)rectype);
				return (EINVAL);
			}
			return (dtab[rectype](dbenv, db, lsnp, redo, info));
		}
	}

	return (0);
}

/*
 * __db_add_recovery --
 *
 * PUBLIC: int __db_add_recovery __P((DB_ENV *,
 * PUBLIC:   int (***)(DB_ENV *, DBT *, DB_LSN *, db_recops, void *), size_t *,
 * PUBLIC:   int (*)(DB_ENV *, DBT *, DB_LSN *, db_recops, void *), u_int32_t));
 */
int
__db_add_recovery(dbenv, dtab, dtabsize, func, ndx)
	DB_ENV *dbenv;
	int (***dtab) __P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *));
	size_t *dtabsize;
	int (*func) __P((DB_ENV *, DBT *, DB_LSN *, db_recops, void *));
	u_int32_t ndx;
{
	size_t i, nsize;
	int ret;

	/* Check if we have to grow the table. */
	if (ndx >= *dtabsize) {
		nsize = ndx + 40;
		if ((ret =
		    __os_realloc(dbenv, nsize * sizeof((*dtab)[0]), dtab)) != 0)
			return (ret);
		for (i = *dtabsize; i < nsize; ++i)
			(*dtab)[i] = NULL;
		*dtabsize = nsize;
	}

	(*dtab)[ndx] = func;
	return (0);
}

/*
 * __db_txnlist_init --
 *	Initialize transaction linked list.
 *
 * PUBLIC: int __db_txnlist_init __P((DB_ENV *,
 * PUBLIC:     u_int32_t, u_int32_t, DB_LSN *, void *));
 */
int
__db_txnlist_init(dbenv, low_txn, hi_txn, trunc_lsn, retp)
	DB_ENV *dbenv;
	u_int32_t low_txn, hi_txn;
	DB_LSN *trunc_lsn;
	void *retp;
{
	DB_TXNHEAD *headp;
	u_int32_t size, tmp;
	int ret;

	/*
	 * Size a hash table.
	 *	If low is zero then we are being called during rollback
	 * and we need only one slot.
	 *	Hi maybe lower than low if we have recycled txnid's.
	 *	The numbers here are guesses about txn density, we can afford
	 * to look at a few entries in each slot.
	 */
	if (low_txn == 0)
		size = 1;
	else {
		if (hi_txn < low_txn) {
			tmp = hi_txn;
			hi_txn = low_txn;
			low_txn = tmp;
		}
		tmp = hi_txn - low_txn;
		/* See if we wrapped around. */
		if (tmp > (TXN_MAXIMUM - TXN_MINIMUM) / 2)
			tmp = (low_txn - TXN_MINIMUM) + (TXN_MAXIMUM - hi_txn);
		size = tmp / 5;
		if (size < 100)
			size = 100;
	}
	if ((ret = __os_malloc(dbenv,
	    sizeof(DB_TXNHEAD) + size * sizeof(headp->head), &headp)) != 0)
		return (ret);

	memset(headp, 0, sizeof(DB_TXNHEAD) + size * sizeof(headp->head));
	headp->maxid = hi_txn;
	headp->generation = 0;
	headp->nslots = size;
	headp->gen_alloc = 8;
	if ((ret = __os_malloc(dbenv, headp->gen_alloc *
	    sizeof(headp->gen_array[0]), &headp->gen_array)) != 0) {
		__os_free(dbenv, headp);
		return (ret);
	}
	headp->gen_array[0].generation = 0;
	headp->gen_array[0].txn_min = TXN_MINIMUM;
	headp->gen_array[0].txn_max = TXN_MAXIMUM;
	if (trunc_lsn != NULL) {
		headp->trunc_lsn = *trunc_lsn;
		headp->maxlsn = *trunc_lsn;
	} else {
		ZERO_LSN(headp->trunc_lsn);
		ZERO_LSN(headp->maxlsn);
	}
	ZERO_LSN(headp->ckplsn);

	*(void **)retp = headp;
	return (0);
}

/*
 * __db_txnlist_add --
 *	Add an element to our transaction linked list.
 *
 * PUBLIC: int __db_txnlist_add __P((DB_ENV *,
 * PUBLIC:     void *, u_int32_t, u_int32_t, DB_LSN *));
 */
int
__db_txnlist_add(dbenv, listp, txnid, status, lsn)
	DB_ENV *dbenv;
	void *listp;
	u_int32_t txnid, status;
	DB_LSN *lsn;
{
	DB_TXNHEAD *hp;
	DB_TXNLIST *elp;
	int ret;

	if ((ret = __os_malloc(dbenv, sizeof(DB_TXNLIST), &elp)) != 0)
		return (ret);

	hp = (DB_TXNHEAD *)listp;
	LIST_INSERT_HEAD(&hp->head[DB_TXNLIST_MASK(hp, txnid)], elp, links);

	elp->type = TXNLIST_TXNID;
	elp->u.t.txnid = txnid;
	elp->u.t.status = status;
	elp->u.t.generation = hp->generation;
	if (txnid > hp->maxid)
		hp->maxid = txnid;
	if (lsn != NULL && IS_ZERO_LSN(hp->maxlsn) && status == TXN_COMMIT)
		hp->maxlsn = *lsn;

	DB_ASSERT(lsn == NULL ||
	    status != TXN_COMMIT || log_compare(&hp->maxlsn, lsn) >= 0);

	return (0);
}

/*
 * __db_txnlist_remove --
 *	Remove an element from our transaction linked list.
 *
 * PUBLIC: int __db_txnlist_remove __P((DB_ENV *, void *, u_int32_t));
 */
int
__db_txnlist_remove(dbenv, listp, txnid)
	DB_ENV *dbenv;
	void *listp;
	u_int32_t txnid;
{
	DB_TXNLIST *entry;
	u_int32_t status;

	return (__db_txnlist_find_internal(dbenv,
	    listp, TXNLIST_TXNID, txnid, NULL, &entry, 1, &status));
}

/*
 * __db_txnlist_ckp --
 *	Used to record the maximum checkpoint that will be retained
 * after recovery.  Typically this is simply the max checkpoint, but
 * if we are doing client replication recovery or timestamp-based
 * recovery, we are going to virtually truncate the log and we need
 * to retain the last checkpoint before the truncation point.
 *
 * PUBLIC: void __db_txnlist_ckp __P((DB_ENV *, void *, DB_LSN *));
 */
void
__db_txnlist_ckp(dbenv, listp, ckp_lsn)
	DB_ENV *dbenv;
	void *listp;
	DB_LSN *ckp_lsn;
{
	DB_TXNHEAD *hp;

	COMPQUIET(dbenv, NULL);

	hp = (DB_TXNHEAD *)listp;

	if (IS_ZERO_LSN(hp->ckplsn) && !IS_ZERO_LSN(hp->maxlsn) &&
	    log_compare(&hp->maxlsn, ckp_lsn) >= 0)
		hp->ckplsn = *ckp_lsn;
}

/*
 * __db_txnlist_end --
 *	Discard transaction linked list.
 *
 * PUBLIC: void __db_txnlist_end __P((DB_ENV *, void *));
 */
void
__db_txnlist_end(dbenv, listp)
	DB_ENV *dbenv;
	void *listp;
{
	DB_TXNHEAD *hp;
	DB_TXNLIST *p;
	u_int32_t i;