hash_page.c

这是linux下运行的mysql软件包,可用于linux 下安装 php + mysql + apach 的网络配置

	} else
		LSN_NOT_LOGGED(new_lsn);

	/* Move lsn onto page. */
	LSN(hcp->page) = new_lsn;	/* Structure assignment. */

	__ham_putitem(dbp, hcp->page, pkey, key_type);
	__ham_putitem(dbp, hcp->page, pdata, data_type);

	/*
	 * For splits, we are going to update item_info's page number
	 * field, so that we can easily return to the same page the
	 * next time we come in here.  For other operations, this shouldn't
	 * matter, since odds are this is the last thing that happens before
	 * we return to the user program.
	 */
	hcp->pgno = PGNO(hcp->page);

	/*
	 * XXX
	 * Maybe keep incremental numbers here.
	 */
	if (!STD_LOCKING(dbc)) {
		hcp->hdr->nelem++;
		if ((ret = __ham_dirty_meta(dbc)) != 0)
			return (ret);
	}

	if (do_expand || (hcp->hdr->ffactor != 0 &&
	    (u_int32_t)H_NUMPAIRS(hcp->page) > hcp->hdr->ffactor))
		F_SET(hcp, H_EXPAND);
	return (0);
}

/*
 * Special __putitem call used in splitting -- copies one entry to
 * another.  Works for all types of hash entries (H_OFFPAGE, H_KEYDATA,
 * H_DUPLICATE, H_OFFDUP).  Since we log splits at a high level, we
 * do not need to do any logging here.
 *
 * PUBLIC: void __ham_copy_item __P((DB *, PAGE *, u_int32_t, PAGE *));
 */
void
__ham_copy_item(dbp, src_page, src_ndx, dest_page)
	DB *dbp;
	PAGE *src_page;
	u_int32_t src_ndx;
	PAGE *dest_page;
{
	u_int32_t len;
	size_t pgsize;
	void *src, *dest;
	db_indx_t *inp;

	pgsize = dbp->pgsize;
	inp = P_INP(dbp, dest_page);
	/*
	 * Copy the key and data entries onto this new page.
	 */
	src = P_ENTRY(dbp, src_page, src_ndx);

	/* Set up space on dest. */
	len = (u_int32_t)LEN_HITEM(dbp, src_page, pgsize, src_ndx);
	HOFFSET(dest_page) -= len;
	inp[NUM_ENT(dest_page)] = HOFFSET(dest_page);
	dest = P_ENTRY(dbp, dest_page, NUM_ENT(dest_page));
	NUM_ENT(dest_page)++;

	memcpy(dest, src, len);
}

/*
 *
 * Returns:
 *	pointer on success
 *	NULL on error
 *
 * PUBLIC: int __ham_add_ovflpage __P((DBC *, PAGE *, int, PAGE **));
 */
int
__ham_add_ovflpage(dbc, pagep, release, pp)
	DBC *dbc;
	PAGE *pagep;
	int release;
	PAGE **pp;
{
	DB *dbp;
	DB_LSN new_lsn;
	DB_MPOOLFILE *mpf;
	PAGE *new_pagep;
	int ret;

	dbp = dbc->dbp;
	mpf = dbp->mpf;

	if ((ret = __db_new(dbc, P_HASH, &new_pagep)) != 0)
		return (ret);

	if (DBC_LOGGING(dbc)) {
		if ((ret = __ham_newpage_log(dbp, dbc->txn, &new_lsn, 0,
		    PUTOVFL, PGNO(pagep), &LSN(pagep),
		    PGNO(new_pagep), &LSN(new_pagep), PGNO_INVALID, NULL)) != 0)
			return (ret);
	} else
		LSN_NOT_LOGGED(new_lsn);

	/* Move lsn onto page. */
	LSN(pagep) = LSN(new_pagep) = new_lsn;
	NEXT_PGNO(pagep) = PGNO(new_pagep);

	PREV_PGNO(new_pagep) = PGNO(pagep);

	if (release)
		ret = mpf->put(mpf, pagep, DB_MPOOL_DIRTY);

	*pp = new_pagep;
	return (ret);
}

/*
 * PUBLIC: int __ham_get_cpage __P((DBC *, db_lockmode_t));
 */
int
__ham_get_cpage(dbc, mode)
	DBC *dbc;
	db_lockmode_t mode;
{
	DB *dbp;
	DB_LOCK tmp_lock;
	DB_MPOOLFILE *mpf;
	HASH_CURSOR *hcp;
	int ret;

	dbp = dbc->dbp;
	mpf = dbp->mpf;
	hcp = (HASH_CURSOR *)dbc->internal;
	ret = 0;

	/*
	 * There are four cases with respect to buckets and locks.
	 * 1. If there is no lock held, then if we are locking, we should
	 *    get the lock.
	 * 2. If there is a lock held, it's for the current bucket, and it's
	 *    for the right mode, we don't need to do anything.
	 * 3. If there is a lock held for the current bucket but it's not
	 *    strong enough, we need to upgrade.
	 * 4. If there is a lock, but it's for a different bucket, then we need
	 *    to release the existing lock and get a new lock.
	 */
	LOCK_INIT(tmp_lock);
	if (STD_LOCKING(dbc)) {
		if (hcp->lbucket != hcp->bucket &&	/* Case 4 */
		    (ret = __TLPUT(dbc, hcp->lock)) != 0)
			return (ret);

		if ((LOCK_ISSET(hcp->lock) &&
		    (hcp->lock_mode == DB_LOCK_READ &&
		    mode == DB_LOCK_WRITE))) {
			/* Case 3. */
			tmp_lock = hcp->lock;
			LOCK_INIT(hcp->lock);
		}

		/* Acquire the lock. */
		if (!LOCK_ISSET(hcp->lock))
			/* Cases 1, 3, and 4. */
			if ((ret = __ham_lock_bucket(dbc, mode)) != 0)
				return (ret);

		if (ret == 0) {
			hcp->lock_mode = mode;
			hcp->lbucket = hcp->bucket;
			if (LOCK_ISSET(tmp_lock))
				/* Case 3: release the original lock. */
				ret =
				    dbp->dbenv->lock_put(dbp->dbenv, &tmp_lock);
		} else if (LOCK_ISSET(tmp_lock))
			hcp->lock = tmp_lock;
	}

	if (ret == 0 && hcp->page == NULL) {
		if (hcp->pgno == PGNO_INVALID)
			hcp->pgno = BUCKET_TO_PAGE(hcp, hcp->bucket);
		if ((ret = mpf->get(mpf,
		    &hcp->pgno, DB_MPOOL_CREATE, &hcp->page)) != 0)
			return (ret);
	}

	return (0);
}

/*
 * Get a new page at the cursor, putting the last page if necessary.
 * If the flag is set to H_ISDUP, then we are talking about the
 * duplicate page, not the main page.
 *
 * PUBLIC: int __ham_next_cpage __P((DBC *, db_pgno_t, int));
 */
int
__ham_next_cpage(dbc, pgno, dirty)
	DBC *dbc;
	db_pgno_t pgno;
	int dirty;
{
	DB *dbp;
	DB_MPOOLFILE *mpf;
	HASH_CURSOR *hcp;
	PAGE *p;
	int ret;

	dbp = dbc->dbp;
	mpf = dbp->mpf;
	hcp = (HASH_CURSOR *)dbc->internal;

	if (hcp->page != NULL &&
	    (ret = mpf->put(mpf, hcp->page, dirty ? DB_MPOOL_DIRTY : 0)) != 0)
		return (ret);
	hcp->page = NULL;

	if ((ret = mpf->get(mpf, &pgno, DB_MPOOL_CREATE, &p)) != 0)
		return (ret);

	hcp->page = p;
	hcp->pgno = pgno;
	hcp->indx = 0;

	return (0);
}

/*
 * __ham_lock_bucket --
 *	Get the lock on a particular bucket.
 *
 * PUBLIC: int __ham_lock_bucket __P((DBC *, db_lockmode_t));
 */
int
__ham_lock_bucket(dbc, mode)
	DBC *dbc;
	db_lockmode_t mode;
{
	HASH_CURSOR *hcp;
	db_pgno_t pgno;
	int gotmeta, ret;

	hcp = (HASH_CURSOR *)dbc->internal;
	gotmeta = hcp->hdr == NULL ? 1 : 0;
	if (gotmeta)
		if ((ret = __ham_get_meta(dbc)) != 0)
			return (ret);
	pgno = BUCKET_TO_PAGE(hcp, hcp->bucket);
	if (gotmeta)
		if ((ret = __ham_release_meta(dbc)) != 0)
			return (ret);

	ret = __db_lget(dbc, 0, pgno, mode, 0, &hcp->lock);

	hcp->lock_mode = mode;
	return (ret);
}

/*
 * __ham_dpair --
 *	Delete a pair on a page, paying no attention to what the pair
 *	represents.  The caller is responsible for freeing up duplicates
 *	or offpage entries that might be referenced by this pair.
 *
 *	Recovery assumes that this may be called without the metadata
 *	page pinned.
 *
 * PUBLIC: void __ham_dpair __P((DB *, PAGE *, u_int32_t));
 */
void
__ham_dpair(dbp, p, indx)
	DB *dbp;
	PAGE *p;
	u_int32_t indx;
{
	db_indx_t delta, n, *inp;
	u_int8_t *dest, *src;

	inp = P_INP(dbp, p);
	/*
	 * Compute "delta", the amount we have to shift all of the
	 * offsets.  To find the delta, we just need to calculate
	 * the size of the pair of elements we are removing.
	 */
	delta = H_PAIRSIZE(dbp, p, dbp->pgsize, indx);

	/*
	 * The hard case: we want to remove something other than
	 * the last item on the page.  We need to shift data and
	 * offsets down.
	 */
	if ((db_indx_t)indx != NUM_ENT(p) - 2) {
		/*
		 * Move the data: src is the first occupied byte on
		 * the page. (Length is delta.)
		 */
		src = (u_int8_t *)p + HOFFSET(p);

		/*
		 * Destination is delta bytes beyond src.  This might
		 * be an overlapping copy, so we have to use memmove.
		 */
		dest = src + delta;
		memmove(dest, src, inp[H_DATAINDEX(indx)] - HOFFSET(p));
	}

	/* Adjust page metadata. */
	HOFFSET(p) = HOFFSET(p) + delta;
	NUM_ENT(p) = NUM_ENT(p) - 2;

	/* Adjust the offsets. */
	for (n = (db_indx_t)indx; n < (db_indx_t)(NUM_ENT(p)); n++)
		inp[n] = inp[n + 2] + delta;

}

/*
 * __ham_c_delpg --
 *
 * Adjust the cursors after we've emptied a page in a bucket, taking
 * care that when we move cursors pointing to deleted items, their
 * orders don't collide with the orders of cursors on the page we move
 * them to (since after this function is called, cursors with the same
 * index on the two pages will be otherwise indistinguishable--they'll
 * all have pgno new_pgno).  There are three cases:
 *
 *	1) The emptied page is the first page in the bucket.  In this
 *	case, we've copied all the items from the second page into the
 *	first page, so the first page is new_pgno and the second page is
 *	old_pgno.  new_pgno is empty, but can have deleted cursors
 *	pointing at indx 0, so we need to be careful of the orders
 *	there.  This is DB_HAM_DELFIRSTPG.
 *
 *	2) The page is somewhere in the middle of a bucket.  Our caller
 *	can just delete such a page, so it's old_pgno.  old_pgno is
 *	empty, but may have deleted cursors pointing at indx 0, so we
 *	need to be careful of indx 0 when we move those cursors to
 *	new_pgno.  This is DB_HAM_DELMIDPG.
 *
 *	3) The page is the last in a bucket.  Again the empty page is
 *	old_pgno, and again it should only have cursors that are deleted
 *	and at indx == 0.  This time, though, there's no next page to
 *	move them to, so we set them to indx == num_ent on the previous
 *	page--and indx == num_ent is the index whose cursors we need to
 *	be careful of.  This is DB_HAM_DELLASTPG.
 */
static int
__ham_c_delpg(dbc, old_pgno, new_pgno, num_ent, op, orderp)
	DBC *dbc;
	db_pgno_t old_pgno, new_pgno;
	u_int32_t num_ent;
	db_ham_mode op;
	u_int32_t *orderp;
{
	DB *dbp, *ldbp;
	DB_ENV *dbenv;
	DB_LSN lsn;
	DB_TXN *my_txn;
	DBC *cp;
	HASH_CURSOR *hcp;
	int found, ret;
	db_indx_t indx;
	u_int32_t order;

	/* Which is the worrisome index? */
	indx = (op == DB_HAM_DELLASTPG) ? num_ent : 0;

	dbp = dbc->dbp;
	dbenv = dbp->dbenv;

	my_txn = IS_SUBTRANSACTION(dbc->txn) ? dbc->txn : NULL;
	found = 0;

	MUTEX_THREAD_LOCK(dbenv, dbenv->dblist_mutexp);
	/*
	 * Find the highest order of any cursor our movement
	 * may collide with.
	 */
	order = 1;
	for (ldbp = __dblist_get(dbenv, dbp->adj_fileid);
	    ldbp != NULL && ldbp->adj_fileid == dbp->adj_fileid;
	    ldbp = LIST_NEXT(ldbp, dblistlinks)) {
		MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
		for (cp = TAILQ_FIRST(&ldbp->active_queue); cp != NULL;
		    cp = TAILQ_NEXT(cp, links)) {
			if (cp == dbc || cp->dbtype != DB_HASH)
				continue;
			hcp = (HASH_CURSOR *)cp->internal;
			if (hcp->pgno == new_pgno) {
				if (hcp->indx == indx &&
				    F_ISSET(hcp, H_DELETED) &&
				    hcp->order >= order)
					order = hcp->order + 1;
				DB_ASSERT(op != DB_HAM_DELFIRSTPG ||
				    hcp->indx == NDX_INVALID ||
				    (hcp->indx == 0 &&
				    F_ISSET(hcp, H_DELETED)));
			}
		}
		MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
	}

	for (ldbp = __dblist_get(dbenv, dbp->adj_fileid);
	    ldbp != NULL && ldbp->adj_fileid == dbp->adj_fileid;
	    ldbp = LIST_NEXT(ldbp, dblistlinks)) {
		MUTEX_THREAD_LOCK(dbenv, dbp->mutexp);
		for (cp = TAILQ_FIRST(&ldbp->active_queue); cp != NULL;
		    cp = TAILQ_NEXT(cp, links)) {
			if (cp == dbc || cp->dbtype != DB_HASH)
				continue;

			hcp = (HASH_CURSOR *)cp->internal;

			if (hcp->pgno == old_pgno) {
				switch (op) {
				case DB_HAM_DELFIRSTPG:
					/*
					 * We're moving all items,
					 * regardless of index.
					 */
					hcp->pgno = new_pgno;

					/*
					 * But we have to be careful of
					 * the order values.
					 */
					if (hcp->indx == indx)
						hcp->order += order;
					break;
				case DB_HAM_DELMIDPG:
					hcp->pgno = new_pgno;
					DB_ASSERT(hcp->indx == 0 &&
					    F_ISSET(hcp, H_DELETED));
					hcp->order += order;
					break;
				case DB_HAM_DELLASTPG:
					hcp->pgno = new_pgno;
					DB_ASSERT(hcp->indx == 0 &&
					    F_ISSET(hcp, H_DELETED));
					hcp->indx = indx;
					hcp->order += order;
					break;
				default:
					DB_ASSERT(0);
					return (__db_panic(dbenv, EINVAL));
				}
				if (my_txn != NULL && cp->txn != my_txn)
					found = 1;
			}
		}
		MUTEX_THREAD_UNLOCK(dbenv, dbp->mutexp);
	}
	MUTEX_THREAD_UNLOCK(dbenv, dbenv->dblist_mutexp);

	if (found != 0 && DBC_LOGGING(dbc)) {
		if ((ret = __ham_chgpg_log(dbp, my_txn, &lsn, 0, op,
		    old_pgno, new_pgno, indx, order)) != 0)
			return (ret);
	}
	*orderp = order;
	return (0);
}