txn.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: txn.c,v 11.249 2004/10/15 16:59:44 bostic Exp $
 */

#include "db_config.h"

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

#if TIME_WITH_SYS_TIME
#include <sys/time.h>
#include <time.h>
#else
#if HAVE_SYS_TIME_H
#include <sys/time.h>
#else
#include <time.h>
#endif
#endif

#include <string.h>
#endif

#include "db_int.h"
#include "dbinc/crypto.h"
#include "dbinc/hmac.h"
#include "dbinc/db_page.h"
#include "dbinc/db_shash.h"
#include "dbinc/hash.h"
#include "dbinc/lock.h"
#include "dbinc/log.h"
#include "dbinc/mp.h"
#include "dbinc/txn.h"

#define	SET_LOG_FLAGS(dbenv, txnp, lflags)				\
	do {								\
		lflags = DB_LOG_COMMIT | DB_LOG_PERM;			\
		if (F_ISSET(txnp, TXN_SYNC))				\
			lflags |= DB_FLUSH;				\
		else if (!F_ISSET(txnp, TXN_NOSYNC) &&			\
		    !F_ISSET(dbenv, DB_ENV_TXN_NOSYNC)) {		\
			if (F_ISSET(dbenv, DB_ENV_TXN_WRITE_NOSYNC))	\
				lflags |= DB_LOG_WRNOSYNC;		\
			else						\
				lflags |= DB_FLUSH;			\
		}							\
	} while (0)

/*
 * __txn_isvalid enumerated types.  We cannot simply use the transaction
 * statuses, because different statuses need to be handled differently
 * depending on the caller.
 */
typedef enum {
	TXN_OP_ABORT,
	TXN_OP_COMMIT,
	TXN_OP_DISCARD,
	TXN_OP_PREPARE
} txnop_t;

static int  __txn_abort_pp __P((DB_TXN *));
static int  __txn_begin_int __P((DB_TXN *, int));
static int  __txn_commit_pp __P((DB_TXN *, u_int32_t));
static int  __txn_discard_pp __P((DB_TXN *, u_int32_t));
static int  __txn_end __P((DB_TXN *, int));
static int  __txn_isvalid __P((const DB_TXN *, TXN_DETAIL **, txnop_t));
static int  __txn_undo __P((DB_TXN *));
static int  __txn_dispatch_undo __P((DB_ENV *,
		DB_TXN *, DBT *, DB_LSN *, void *));
static void __txn_set_begin_lsnp __P((DB_TXN *txn, DB_LSN **));

/*
 * __txn_begin_pp --
 *	DB_ENV->txn_begin pre/post processing.
 *
 * PUBLIC: int __txn_begin_pp __P((DB_ENV *, DB_TXN *, DB_TXN **, u_int32_t));
 */
int
__txn_begin_pp(dbenv, parent, txnpp, flags)
	DB_ENV *dbenv;
	DB_TXN *parent, **txnpp;
	u_int32_t flags;
{
	int rep_check, ret;

	PANIC_CHECK(dbenv);
	ENV_REQUIRES_CONFIG(dbenv, dbenv->tx_handle, "txn_begin", DB_INIT_TXN);

	if ((ret = __db_fchk(dbenv,
	    "txn_begin", flags,
	    DB_DEGREE_2 | DB_DIRTY_READ | DB_TXN_NOWAIT |
	    DB_TXN_NOSYNC | DB_TXN_SYNC)) != 0)
		return (ret);
	if ((ret = __db_fcchk(dbenv,
	    "txn_begin", flags, DB_TXN_NOSYNC, DB_TXN_SYNC)) != 0)
		return (ret);

	if (parent == NULL) {
		rep_check = IS_ENV_REPLICATED(dbenv) ? 1 : 0;
		if (rep_check)
			__op_rep_enter(dbenv);
	} else
		rep_check = 0;
	ret = __txn_begin(dbenv, parent, txnpp, flags);
	/*
	 * We only decrement the count if the operation fails.
	 * Otherwise the count will be decremented when the
	 * txn is resolved by txn_commit, txn_abort, etc.
	 */
	if (ret != 0 && rep_check)
		__op_rep_exit(dbenv);

	return (ret);
}

/*
 * __txn_begin --
 *	DB_ENV->txn_begin.
 *
 * This is a wrapper to the actual begin process.  Normal transaction begin
 * allocates a DB_TXN structure for the caller, while XA transaction begin
 * does not.  Other than that, both call into common __txn_begin_int code.
 *
 * Internally, we use TXN_DETAIL structures, but the DB_TXN structure
 * provides access to the transaction ID and the offset in the transaction
 * region of the TXN_DETAIL structure.
 *
 * PUBLIC: int __txn_begin __P((DB_ENV *, DB_TXN *, DB_TXN **, u_int32_t));
 */
int
__txn_begin(dbenv, parent, txnpp, flags)
	DB_ENV *dbenv;
	DB_TXN *parent, **txnpp;
	u_int32_t flags;
{
	DB_LOCKREGION *region;
	DB_TXN *txn;
	int ret;

	*txnpp = NULL;
	if ((ret = __os_calloc(dbenv, 1, sizeof(DB_TXN), &txn)) != 0)
		return (ret);

	txn->mgrp = dbenv->tx_handle;
	txn->parent = parent;
	TAILQ_INIT(&txn->kids);
	TAILQ_INIT(&txn->events);
	STAILQ_INIT(&txn->logs);
	txn->flags = TXN_MALLOC;
	if (LF_ISSET(DB_DEGREE_2))
		F_SET(txn, TXN_DEGREE_2);
	if (LF_ISSET(DB_DIRTY_READ))
		F_SET(txn, TXN_DIRTY_READ);
	if (LF_ISSET(DB_TXN_NOSYNC))
		F_SET(txn, TXN_NOSYNC);
	if (LF_ISSET(DB_TXN_SYNC))
		F_SET(txn, TXN_SYNC);
	if (LF_ISSET(DB_TXN_NOWAIT))
		F_SET(txn, TXN_NOWAIT);

	if ((ret = __txn_begin_int(txn, 0)) != 0)
		goto err;

	if (parent != NULL)
		TAILQ_INSERT_HEAD(&parent->kids, txn, klinks);

	if (LOCKING_ON(dbenv)) {
		region = ((DB_LOCKTAB *)dbenv->lk_handle)->reginfo.primary;
		if (parent != NULL) {
			ret = __lock_inherit_timeout(dbenv,
			    parent->txnid, txn->txnid);
			/* No parent locker set yet. */
			if (ret == EINVAL) {
				parent = NULL;
				ret = 0;
			}
			if (ret != 0)
				goto err;
		}

		/*
		 * Parent is NULL if we have no parent
		 * or it has no timeouts set.
		 */
		if (parent == NULL && region->tx_timeout != 0)
			if ((ret = __lock_set_timeout(dbenv, txn->txnid,
			    region->tx_timeout, DB_SET_TXN_TIMEOUT)) != 0)
				goto err;
	}

	*txnpp = txn;
	return (0);

err:
	__os_free(dbenv, txn);
	return (ret);
}

/*
 * __txn_xa_begin --
 *	XA version of txn_begin.
 *
 * PUBLIC: int __txn_xa_begin __P((DB_ENV *, DB_TXN *));
 */
int
__txn_xa_begin(dbenv, txn)
	DB_ENV *dbenv;
	DB_TXN *txn;
{
	PANIC_CHECK(dbenv);

	/*
	 * We need to initialize the transaction structure, but must be careful
	 * not to smash the links.  We manually initialize the structure.
	 */
	txn->mgrp = dbenv->tx_handle;
	TAILQ_INIT(&txn->kids);
	TAILQ_INIT(&txn->events);
	STAILQ_INIT(&txn->logs);
	txn->parent = NULL;
	ZERO_LSN(txn->last_lsn);
	txn->txnid = TXN_INVALID;
	txn->tid = 0;
	txn->cursors = 0;
	memset(&txn->lock_timeout, 0, sizeof(db_timeout_t));
	memset(&txn->expire, 0, sizeof(db_timeout_t));

	return (__txn_begin_int(txn, 0));
}

/*
 * __txn_compensate_begin
 *	Begin an compensation transaction.  This is a special interface
 * that is used only for transactions that must be started to compensate
 * for actions during an abort.  Currently only used for allocations.
 *
 * PUBLIC: int __txn_compensate_begin __P((DB_ENV *, DB_TXN **txnp));
 */
int
__txn_compensate_begin(dbenv, txnpp)
	DB_ENV *dbenv;
	DB_TXN **txnpp;
{
	DB_TXN *txn;
	int ret;

	PANIC_CHECK(dbenv);

	if ((ret = __os_calloc(dbenv, 1, sizeof(DB_TXN), &txn)) != 0)
		return (ret);

	txn->mgrp = dbenv->tx_handle;
	TAILQ_INIT(&txn->kids);
	TAILQ_INIT(&txn->events);
	STAILQ_INIT(&txn->logs);
	txn->flags = TXN_COMPENSATE | TXN_MALLOC;

	*txnpp = txn;
	return (__txn_begin_int(txn, 1));
}

/*
 * __txn_begin_int --
 *	Normal DB version of txn_begin.
 */
static int
__txn_begin_int(txn, internal)
	DB_TXN *txn;
	int internal;
{
	DB_ENV *dbenv;
	DB_LSN null_lsn;
	DB_TXNMGR *mgr;
	DB_TXNREGION *region;
	TXN_DETAIL *td;
	size_t off;
	u_int32_t id, *ids;
	int nids, ret;

	mgr = txn->mgrp;
	dbenv = mgr->dbenv;
	region = mgr->reginfo.primary;

	R_LOCK(dbenv, &mgr->reginfo);
	if (!F_ISSET(txn, TXN_COMPENSATE) && F_ISSET(region, TXN_IN_RECOVERY)) {
		__db_err(dbenv, "operation not permitted during recovery");
		ret = EINVAL;
		goto err;
	}

	/* Make sure that we aren't still recovering prepared transactions. */
	if (!internal && region->stat.st_nrestores != 0) {
		__db_err(dbenv,
    "recovery of prepared but not yet committed transactions is incomplete");
		ret = EINVAL;
		goto err;
	}

	/*
	 * Allocate a new transaction id. Our current valid range can span
	 * the maximum valid value, so check for it and wrap manually.
	 */
	if (region->last_txnid == TXN_MAXIMUM &&
	    region->cur_maxid != TXN_MAXIMUM)
		region->last_txnid = TXN_MINIMUM - 1;

	if (region->last_txnid == region->cur_maxid) {
		if ((ret = __os_malloc(dbenv,
		    sizeof(u_int32_t) * region->maxtxns, &ids)) != 0)
			goto err;
		nids = 0;
		for (td = SH_TAILQ_FIRST(&region->active_txn, __txn_detail);
		    td != NULL;
		    td = SH_TAILQ_NEXT(td, links, __txn_detail))
			ids[nids++] = td->txnid;
		region->last_txnid = TXN_MINIMUM - 1;
		region->cur_maxid = TXN_MAXIMUM;
		if (nids != 0)
			__db_idspace(ids, nids,
			    &region->last_txnid, &region->cur_maxid);
		__os_free(dbenv, ids);
		if (DBENV_LOGGING(dbenv) &&
		    (ret = __txn_recycle_log(dbenv, NULL, &null_lsn,
			0, region->last_txnid + 1, region->cur_maxid)) != 0)
			goto err;
	}

	/* Allocate a new transaction detail structure. */
	if ((ret =
	    __db_shalloc(&mgr->reginfo, sizeof(TXN_DETAIL), 0, &td)) != 0) {
		__db_err(dbenv,
		    "Unable to allocate memory for transaction detail");
		goto err;
	}

	/* Place transaction on active transaction list. */
	SH_TAILQ_INSERT_HEAD(&region->active_txn, td, links, __txn_detail);

	id = ++region->last_txnid;
	++region->stat.st_nbegins;
	if (++region->stat.st_nactive > region->stat.st_maxnactive)
		region->stat.st_maxnactive = region->stat.st_nactive;

	td->txnid = id;
	ZERO_LSN(td->last_lsn);
	ZERO_LSN(td->begin_lsn);
	if (txn->parent != NULL)
		td->parent = txn->parent->off;
	else
		td->parent = INVALID_ROFF;
	td->status = TXN_RUNNING;
	td->flags = 0;
	td->xa_status = 0;

	off = R_OFFSET(&mgr->reginfo, td);
	R_UNLOCK(dbenv, &mgr->reginfo);

	ZERO_LSN(txn->last_lsn);
	txn->txnid = id;
	txn->off = (u_int32_t)off;

	txn->abort = __txn_abort_pp;
	txn->commit = __txn_commit_pp;
	txn->discard = __txn_discard_pp;
	txn->id = __txn_id;
	txn->prepare = __txn_prepare;
	txn->set_timeout = __txn_set_timeout;
	txn->set_begin_lsnp = __txn_set_begin_lsnp;

	/*
	 * If this is a transaction family, we must link the child to the
	 * maximal grandparent in the lock table for deadlock detection.
	 */
	if (txn->parent != NULL && LOCKING_ON(dbenv))
		if ((ret = __lock_addfamilylocker(dbenv,
		    txn->parent->txnid, txn->txnid)) != 0)
			return (ret);

	if (F_ISSET(txn, TXN_MALLOC)) {
		MUTEX_THREAD_LOCK(dbenv, mgr->mutexp);
		TAILQ_INSERT_TAIL(&mgr->txn_chain, txn, links);
		MUTEX_THREAD_UNLOCK(dbenv, mgr->mutexp);
	}

	return (0);

err:	R_UNLOCK(dbenv, &mgr->reginfo);
	return (ret);
}

/*
 * __txn_commit_pp --
 *	Interface routine to TXN->commit.
 */
static int
__txn_commit_pp(txnp, flags)
	DB_TXN *txnp;
	u_int32_t flags;
{
	DB_ENV *dbenv;
	int not_child, ret;

	dbenv = txnp->mgrp->dbenv;
	not_child = txnp->parent == NULL;
	ret = __txn_commit(txnp, flags);
	if (not_child && IS_ENV_REPLICATED(dbenv))
		__op_rep_exit(dbenv);
	return (ret);
}

/*
 * __txn_commit --
 *	Commit a transaction.
 *
 * PUBLIC: int __txn_commit __P((DB_TXN *, u_int32_t));
 */
int
__txn_commit(txnp, flags)
	DB_TXN *txnp;
	u_int32_t flags;
{
	DBT list_dbt;
	DB_ENV *dbenv;
	DB_LOCKREQ request;
	DB_TXN *kid;
	TXN_DETAIL *td;
	u_int32_t lflags;
	int ret, t_ret;

	dbenv = txnp->mgrp->dbenv;

	PANIC_CHECK(dbenv);

	if ((ret = __txn_isvalid(txnp, &td, TXN_OP_COMMIT)) != 0)
		return (ret);

	/*
	 * We clear flags that are incorrect, ignoring any flag errors, and
	 * default to synchronous operations.  By definition, transaction
	 * handles are dead when we return, and this error should never
	 * happen, but we don't want to fail in the field 'cause the app is
	 * specifying the wrong flag for some reason.
	 */
	if (__db_fchk(dbenv,
	    "DB_TXN->commit", flags, DB_TXN_NOSYNC | DB_TXN_SYNC) != 0)
		flags = DB_TXN_SYNC;
	if (__db_fcchk(dbenv,
	    "DB_TXN->commit", flags, DB_TXN_NOSYNC, DB_TXN_SYNC) != 0)
		flags = DB_TXN_SYNC;
	if (LF_ISSET(DB_TXN_NOSYNC)) {
		F_CLR(txnp, TXN_SYNC);
		F_SET(txnp, TXN_NOSYNC);
	}
	if (LF_ISSET(DB_TXN_SYNC)) {
		F_CLR(txnp, TXN_NOSYNC);
		F_SET(txnp, TXN_SYNC);
	}

	/*
	 * Commit any unresolved children.  If anyone fails to commit,
	 * then try to abort the rest of the kids and then abort the parent.
	 * Abort should never fail; if it does, we bail out immediately.
	 */
	while ((kid = TAILQ_FIRST(&txnp->kids)) != NULL)
		if ((ret = __txn_commit(kid, flags)) != 0)
			while ((kid = TAILQ_FIRST(&txnp->kids)) != NULL)
				if ((t_ret = __txn_abort(kid)) != 0)
					return (__db_panic(dbenv, t_ret));

	/*
	 * If there are any log records, write a log record and sync the log,
	 * else do no log writes.  If the commit is for a child transaction,
	 * we do not need to commit the child synchronously since it may still
	 * abort (if its parent aborts), and otherwise its parent or ultimate
	 * ancestor will write synchronously.
	 */
	if (DBENV_LOGGING(dbenv) && (!IS_ZERO_LSN(txnp->last_lsn) ||
	    STAILQ_FIRST(&txnp->logs) != NULL)) {
		if (txnp->parent == NULL) {
			/*
			 * We are about to free all the read locks for this
			 * transaction below.  Some of those locks might be
			 * handle locks which should not be freed, because
			 * they will be freed when the handle is closed. Check
			 * the events and preprocess any trades now so we don't
			 * release the locks below.
			 */
			if ((ret =
			    __txn_doevents(dbenv, txnp, TXN_PREPARE, 1)) != 0)
				goto err;

			memset(&request, 0, sizeof(request));
			if (LOCKING_ON(dbenv)) {
				request.op = DB_LOCK_PUT_READ;
				if (IS_REP_MASTER(dbenv) &&
				    !IS_ZERO_LSN(txnp->last_lsn)) {
					memset(&list_dbt, 0, sizeof(list_dbt));
					request.obj = &list_dbt;
				}
				ret = __lock_vec(dbenv,
				    txnp->txnid, 0, &request, 1, NULL);
			}

			if (ret == 0 && !IS_ZERO_LSN(txnp->last_lsn)) {
				SET_LOG_FLAGS(dbenv, txnp, lflags);
				ret = __txn_regop_log(dbenv, txnp,
				    &txnp->last_lsn, lflags, TXN_COMMIT,
				    (int32_t)time(NULL), request.obj);
			}

			if (request.obj != NULL && request.obj->data != NULL)
				__os_free(dbenv, request.obj->data);
			if (ret != 0)
				goto err;
		} else {
			/* Log the commit in the parent! */
			if (!IS_ZERO_LSN(txnp->last_lsn) &&
			    (ret = __txn_child_log(dbenv,
			    txnp->parent, &txnp->parent->last_lsn,
			    0, txnp->txnid, &txnp->last_lsn)) != 0) {
				goto err;
			}
			if (STAILQ_FIRST(&txnp->logs) != NULL) {
				/*
				 * Put the child first so we back it out first.