X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Flibmdb%2Fmdb.c;h=b814169b7100b28507ee562573019ac60ab6bb55;hb=1f8c3369b90f54d7fd58c5e327343ca1c708e813;hp=a6852418173c441863759264fd0a51d89eaf54f5;hpb=70a4f6f29d7cc488a3e03598bf62bb93a9332361;p=openldap

diff --git a/libraries/libmdb/mdb.c b/libraries/libmdb/mdb.c
index a685241817..b814169b71 100644
--- a/libraries/libmdb/mdb.c
+++ b/libraries/libmdb/mdb.c
@@ -363,18 +363,19 @@ struct MDB_cursor {
 
 #define METADATA(p)	 ((void *)((char *)(p) + PAGEHDRSZ))
 
+/* We guarantee 2-byte alignment for nodes */
 typedef struct MDB_node {
-#define mn_pgno		 mn_p.np_pgno
-#define mn_dsize	 mn_p.np_dsize
-	union {
-		pgno_t		 np_pgno;	/* child page number */
-		uint32_t	 np_dsize;	/* leaf data size */
-	} mn_p;
-	unsigned int	mn_flags:4;
-	unsigned int	mn_ksize:12;			/* key size */
+	/* lo and hi are used for data size on leaf nodes and for
+	 * child pgno on branch nodes. On 64 bit platforms, flags
+	 * is also used for pgno. (branch nodes ignore flags)
+	 */
+	unsigned short	mn_lo;
+	unsigned short	mn_hi;
+	unsigned short	mn_flags;
 #define F_BIGDATA	 0x01			/* data put on overflow page */
 #define F_SUBDATA	 0x02			/* data is a sub-database */
 #define F_DUPDATA	 0x04			/* data has duplicates */
+	unsigned short	mn_ksize;				/* key size */
 	char		mn_data[1];
 } MDB_node;
 
@@ -440,7 +441,7 @@ struct MDB_env {
 	pthread_key_t	me_txkey;	/* thread-key for readers */
 	MDB_dpage	*me_dpages;
 	pgno_t		me_free_pgs[MDB_IDL_UM_SIZE];
-	ID2			me_dirty_list[MDB_IDL_DB_SIZE];
+	ID2			me_dirty_list[MDB_IDL_UM_SIZE];
 	LAZY_RWLOCK_DEF(me_dblock);
 #ifdef _WIN32
 	HANDLE		me_rmutex;		/* Windows mutexes don't reside in shared mem */
@@ -455,8 +456,19 @@ struct MDB_env {
 #define NODEPTR(p, i)	 ((MDB_node *)((char *)(p) + (p)->mp_ptrs[i]))
 #define NODEKEY(node)	 (void *)((node)->mn_data)
 #define NODEDATA(node)	 (void *)((char *)(node)->mn_data + (node)->mn_ksize)
-#define NODEPGNO(node)	 ((node)->mn_pgno)
-#define NODEDSZ(node)	 ((node)->mn_dsize)
+#if LONG_MAX == 0x7fffffff
+#define NODEPGNO(node)	 ((node)->mn_lo | ((node)->mn_hi << 16))
+#define SETPGNO(node,pgno)	do { \
+	(node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16;} while(0)
+#else
+#define NODEPGNO(node)	 ((node)->mn_lo | ((node)->mn_hi << 16) | ((unsigned long)(node)->mn_flags << 32))
+#define SETPGNO(node,pgno)	do { \
+	(node)->mn_lo = (pgno) & 0xffff; (node)->mn_hi = (pgno) >> 16; \
+	(node)->mn_flags = (pgno) >> 32; } while(0)
+#endif
+#define NODEDSZ(node)	 ((node)->mn_lo | ((unsigned)(node)->mn_hi << 16))
+#define SETDSZ(node,size)	do { \
+	(node)->mn_lo = (size) & 0xffff; (node)->mn_hi = (size) >> 16;} while(0)
 #define NODEKSZ(node)	 ((node)->mn_ksize)
 #define LEAF2KEY(p, i, ks)	((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
 
@@ -486,10 +498,6 @@ static int  mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
 static void mdb_del_node(MDB_page *mp, indx_t indx, int ksize);
 static int mdb_del0(MDB_cursor *mc, unsigned int ki,
     MDB_pageparent *mpp, MDB_node *leaf);
-#if 0
-static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
-    MDB_val *key, MDB_val *data, unsigned int flags);
-#endif
 static int  mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data);
 
 static int	mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
@@ -531,7 +539,7 @@ static size_t	mdb_branch_size(MDB_env *env, MDB_val *key);
 
 static void mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi);
 
-static MDB_cmp_func	memncmp, memnrcmp, intcmp;
+static MDB_cmp_func	memncmp, memnrcmp, intcmp, cintcmp;
 
 #ifdef _WIN32
 static SECURITY_DESCRIPTOR mdb_null_sd;
@@ -578,7 +586,7 @@ mdb_dkey(MDB_val *key, char *buf)
 	unsigned int i;
 	if (key->mv_size > MAXKEYSIZE)
 		return "MAXKEYSIZE";
-#if 0
+#if 1
 	for (i=0; i<key->mv_size; i++)
 		ptr += sprintf(ptr, "%02x", *c++);
 #else
@@ -744,7 +752,7 @@ mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp)
 			return ENOMEM;
 		DPRINTF("touched db %u page %lu -> %lu", dbi, mp->mp_pgno, dp->p.mp_pgno);
 		assert(mp->mp_pgno != dp->p.mp_pgno);
-		mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
+		mdb_midl_append(txn->mt_free_pgs, mp->mp_pgno);
 		pgno = dp->p.mp_pgno;
 		memcpy(&dp->p, mp, txn->mt_env->me_psize);
 		mp = &dp->p;
@@ -753,7 +761,7 @@ mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp)
 
 		/* Update the page number to new touched page. */
 		if (pp->mp_parent != NULL)
-			NODEPGNO(NODEPTR(pp->mp_parent, pp->mp_pi)) = mp->mp_pgno;
+			SETPGNO(NODEPTR(pp->mp_parent, pp->mp_pi), mp->mp_pgno);
 		pp->mp_page = mp;
 	}
 	return 0;
@@ -1017,6 +1025,7 @@ mdb_txn_commit(MDB_txn *txn)
 		mc.mc_snum = 0;
 		mdb_search_page(txn, FREE_DBI, &key, &mc, 1, &mpp);
 
+		mdb_midl_sort(txn->mt_free_pgs);
 #if DEBUG > 1
 		{
 			unsigned int i;
@@ -1928,6 +1937,7 @@ mdb_env_close(MDB_env *env)
 	free(env);
 }
 
+/* only for aligned ints */
 static int
 intcmp(const MDB_val *a, const MDB_val *b)
 {
@@ -1945,6 +1955,26 @@ intcmp(const MDB_val *a, const MDB_val *b)
 	}
 }
 
+/* ints must always be the same size */
+static int
+cintcmp(const MDB_val *a, const MDB_val *b)
+{
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+	unsigned char *u, *c;
+	int x;
+
+	u = a->mv_data + a->mv_size;
+	c = b->mv_data + a->mv_size;
+	while(u > (unsigned char *)a->mv_data) {
+		x = *--u - *--c;
+		if (x) break;
+	}
+	return x;
+#else
+	return memcmp(a->mv_data, b->mv_data, a->mv_size);
+#endif
+}
+
 static int
 memncmp(const MDB_val *a, const MDB_val *b)
 {
@@ -1979,7 +2009,7 @@ memnrcmp(const MDB_val *a, const MDB_val *b)
 	else
 		p1_lim = p1 - b->mv_size;
 
-	while (p1 >= p1_lim) {
+	while (p1 > p1_lim) {
 		diff = *p1 - *p2;
 		if (diff)
 			return diff;
@@ -2056,6 +2086,8 @@ mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, MDB_val *key,
 
 	if (rc > 0) {	/* Found entry is less than the key. */
 		i++;	/* Skip to get the smallest entry larger than key. */
+		if (!IS_LEAF2(mp))
+			node = NODEPTR(mp, i);
 	}
 	if (exactp)
 		*exactp = (rc == 0);
@@ -2266,14 +2298,14 @@ mdb_read_data(MDB_txn *txn, MDB_node *leaf, MDB_val *data)
 	int rc;
 
 	if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
-		data->mv_size = leaf->mn_dsize;
+		data->mv_size = NODEDSZ(leaf);
 		data->mv_data = NODEDATA(leaf);
 		return MDB_SUCCESS;
 	}
 
 	/* Read overflow data.
 	 */
-	data->mv_size = leaf->mn_dsize;
+	data->mv_size = NODEDSZ(leaf);
 	memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
 	if ((rc = mdb_get_page(txn, pgno, &omp))) {
 		DPRINTF("read overflow page %lu failed", pgno);
@@ -2290,7 +2322,7 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi,
 {
 	MDB_cursor	mc;
 	MDB_xcursor	mx;
-	int exact;
+	int exact = 0;
 	DKBUF;
 
 	assert(key);
@@ -2542,11 +2574,17 @@ mdb_cursor_set(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
 		}
 		rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_cmp(key, &nodekey);
 		if (rc == 0) {
+			/* Probably happens rarely, but first node on the page
+			 * was the one we wanted.
+			 */
+			top->mp_ki = 0;
 set1:
-			/* we're already on the right page */
 			mpp.mp_page = top->mp_page;
+			if (exactp)
+				*exactp = 1;
 			rc = 0;
-			goto set2;
+			leaf = NODEPTR(top->mp_page, top->mp_ki);
+			goto set3;
 		}
 		if (rc > 0) {
 			unsigned int i;
@@ -2559,7 +2597,17 @@ set1:
 					MDB_SET_KEY(leaf, &nodekey);
 				}
 				rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_cmp(key, &nodekey);
-				if (rc <= 0) goto set1;
+				if (rc == 0) {
+					/* last node was the one we wanted */
+					top->mp_ki = NUMKEYS(top->mp_page)-1;
+					goto set1;
+				}
+				if (rc < 0) {
+					/* This is definitely the right page, skip search_page */
+					mpp.mp_page = top->mp_page;
+					rc = 0;
+					goto set2;
+				}
 			}
 			/* If any parents have right-sibs, search.
 			 * Otherwise, there's nothing further.
@@ -2570,7 +2618,8 @@ set1:
 					break;
 			if (i == cursor->mc_snum - 1) {
 				/* There are no other pages */
-				goto set1;
+				top->mp_ki = NUMKEYS(top->mp_page);
+				return MDB_NOTFOUND;
 			}
 		}
 	}
@@ -2602,6 +2651,7 @@ set2:
 		leaf = NODEPTR(mpp.mp_page, 0);
 	}
 
+set3:
 	cursor->mc_flags |= C_INITIALIZED;
 	cursor->mc_flags &= ~C_EOF;
 
@@ -2634,7 +2684,7 @@ set2:
 			MDB_val d2;
 			if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
 				return rc;
-			rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_cmp(data, &d2);
+			rc = cursor->mc_txn->mt_dbxs[cursor->mc_dbi].md_dcmp(data, &d2);
 			if (rc) {
 				if (op == MDB_GET_BOTH || rc > 0)
 					return MDB_NOTFOUND;
@@ -3098,7 +3148,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
 		MDB_pageparent mp2;
 
 		if (flags != MDB_NODUPDATA) {
-/*			mdb_xcursor_init2(mc); */
+			mdb_xcursor_init2(mc);
 			rc = mdb_cursor_del(&mc->mc_xcursor->mx_cursor, 0);
 			mdb_xcursor_fini(mc);
 			/* If sub-DB still has entries, we're done */
@@ -3159,7 +3209,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
 #endif
 						{
 							/* free it */
-							mdb_midl_insert(mc->mc_txn->mt_free_pgs, pg);
+							mdb_midl_append(mc->mc_txn->mt_free_pgs, pg);
 						}
 					}
 					rc = mdb_sibling(&mc->mc_xcursor->mx_cursor, 1);
@@ -3178,7 +3228,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
 #endif
 			{
 				/* free it */
-				mdb_midl_insert(mc->mc_txn->mt_free_pgs,
+				mdb_midl_append(mc->mc_txn->mt_free_pgs,
 					mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_root);
 			}
 		}
@@ -3224,6 +3274,7 @@ mdb_leaf_size(MDB_env *env, MDB_val *key, MDB_val *data)
 		/* put on overflow page */
 		sz -= data->mv_size - sizeof(pgno_t);
 	}
+	sz += sz & 1;
 
 	return sz + sizeof(indx_t);
 }
@@ -3299,6 +3350,7 @@ mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
 			node_size += data->mv_size;
 		}
 	}
+	node_size += node_size & 1;
 
 	if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
 		DPRINTF("not enough room in page %lu, got %u ptrs",
@@ -3325,9 +3377,9 @@ mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp, indx_t indx,
 	node->mn_ksize = (key == NULL) ? 0 : key->mv_size;
 	node->mn_flags = flags;
 	if (IS_LEAF(mp))
-		node->mn_dsize = data->mv_size;
+		SETDSZ(node,data->mv_size);
 	else
-		NODEPGNO(node) = pgno;
+		SETPGNO(node,pgno);
 
 	if (key)
 		memcpy(NODEKEY(node), key->mv_data, key->mv_size);
@@ -3381,6 +3433,7 @@ mdb_del_node(MDB_page *mp, indx_t indx, int ksize)
 		else
 			sz += NODEDSZ(node);
 	}
+	sz += sz & 1;
 
 	ptr = mp->mp_ptrs[indx];
 	numkeys = NUMKEYS(mp);
@@ -3422,6 +3475,7 @@ mdb_xcursor_init0(MDB_cursor *mc)
 	mx->mx_dbxs[dbn+1].md_rel = mx->mx_dbxs[dbn].md_rel;
 	mx->mx_dbxs[dbn+1].md_dirty = 0;
 	mx->mx_txn.mt_numdbs = dbn+2;
+	mx->mx_txn.mt_u = mc->mc_txn->mt_u;
 
 	mx->mx_cursor.mc_xcursor = NULL;
 	mx->mx_cursor.mc_txn = &mx->mx_txn;
@@ -3438,6 +3492,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node)
 	mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1];
 	if (mc->mc_dbi > 1) {
 		mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi];
+		mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty;
 		dbn = 3;
 	} else {
 		dbn = 2;
@@ -3449,7 +3504,6 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_page *mp, MDB_node *node)
 	mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
 	mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
 	mx->mx_txn.mt_next_pgno = mc->mc_txn->mt_next_pgno;
-	mx->mx_txn.mt_u = mc->mc_txn->mt_u;
 	mx->mx_cursor.mc_snum = 0;
 	mx->mx_cursor.mc_flags = 0;
 }
@@ -3463,6 +3517,7 @@ mdb_xcursor_init2(MDB_cursor *mc)
 	mx->mx_dbs[1] = mc->mc_txn->mt_dbs[1];
 	if (mc->mc_dbi > 1) {
 		mx->mx_dbs[2] = mc->mc_txn->mt_dbs[mc->mc_dbi];
+		mx->mx_dbxs[2].md_dirty = mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty;
 		dbn = 3;
 	} else {
 		dbn = 2;
@@ -3479,8 +3534,6 @@ mdb_xcursor_fini(MDB_cursor *mc)
 	mc->mc_txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
 	mc->mc_txn->mt_dbs[0] = mx->mx_dbs[0];
 	mc->mc_txn->mt_dbs[1] = mx->mx_dbs[1];
-	mc->mc_txn->mt_dbxs[0].md_dirty = mx->mx_dbxs[0].md_dirty;
-	mc->mc_txn->mt_dbxs[1].md_dirty = mx->mx_dbxs[1].md_dirty;
 	if (mc->mc_dbi > 1) {
 		mc->mc_txn->mt_dbs[mc->mc_dbi] = mx->mx_dbs[2];
 		mc->mc_txn->mt_dbxs[mc->mc_dbi].md_dirty = mx->mx_dbxs[2].md_dirty;
@@ -3644,7 +3697,16 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
 		data.mv_size = 0;
 		data.mv_data = NULL;
 	} else {
-		srcnode = NODEPTR(src->mp_page, srcindx);
+		if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
+			/* must find the lowest key below src */
+			MDB_pageparent mpp;
+			mpp.mp_page = src->mp_page;
+			mpp.mp_pi = 0;
+			mdb_search_page_root(txn, dbi, NULL, NULL, 0, &mpp);
+			srcnode = NODEPTR(mpp.mp_page, 0);
+		} else {
+			srcnode = NODEPTR(src->mp_page, srcindx);
+		}
 		key.mv_size = NODEKSZ(srcnode);
 		key.mv_data = NODEKEY(srcnode);
 		data.mv_size = NODEDSZ(srcnode);
@@ -3668,17 +3730,17 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
 	 */
 	mdb_del_node(src->mp_page, srcindx, key.mv_size);
 
-	/* The key value just changed due to del_node, find it again.
-	 */
-	if (!IS_LEAF2(src->mp_page)) {
-		srcnode = NODEPTR(src->mp_page, srcindx);
-		key.mv_data = NODEKEY(srcnode);
-	}
-
 	/* Update the parent separators.
 	 */
 	if (srcindx == 0) {
 		if (src->mp_pi != 0) {
+			if (IS_LEAF2(src->mp_page)) {
+				key.mv_data = LEAF2KEY(src->mp_page, srcindx, key.mv_size);
+			} else {
+				srcnode = NODEPTR(src->mp_page, srcindx);
+				key.mv_size = NODEKSZ(srcnode);
+				key.mv_data = NODEKEY(srcnode);
+			}
 			DPRINTF("update separator for source page %lu to [%s]",
 				src->mp_page->mp_pgno, DKEY(&key));
 			if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
@@ -3694,6 +3756,13 @@ mdb_move_node(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *src, indx_t srcindx,
 
 	if (dstindx == 0) {
 		if (dst->mp_pi != 0) {
+			if (IS_LEAF2(src->mp_page)) {
+				key.mv_data = LEAF2KEY(dst->mp_page, 0, key.mv_size);
+			} else {
+				srcnode = NODEPTR(dst->mp_page, 0);
+				key.mv_size = NODEKSZ(srcnode);
+				key.mv_data = NODEKEY(srcnode);
+			}
 			DPRINTF("update separator for destination page %lu to [%s]",
 				dst->mp_page->mp_pgno, DKEY(&key));
 			if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
@@ -3814,10 +3883,10 @@ mdb_rebalance(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mpp)
 			txn->mt_dbs[dbi].md_root = P_INVALID;
 			txn->mt_dbs[dbi].md_depth = 0;
 			txn->mt_dbs[dbi].md_leaf_pages = 0;
-			mdb_midl_insert(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
+			mdb_midl_append(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
 		} else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
 			DPUTS("collapsing root page!");
-			mdb_midl_insert(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
+			mdb_midl_append(txn->mt_free_pgs, mpp->mp_page->mp_pgno);
 			txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
 			if ((rc = mdb_get_page(txn, txn->mt_dbs[dbi].md_root, &root)))
 				return rc;
@@ -3895,7 +3964,7 @@ mdb_del0(MDB_cursor *mc, unsigned int ki, MDB_pageparent *mpp, MDB_node *leaf)
 		ovpages = OVPAGES(NODEDSZ(leaf), mc->mc_txn->mt_env->me_psize);
 		for (i=0; i<ovpages; i++) {
 			DPRINTF("freed ov page %lu", pg);
-			mdb_midl_insert(mc->mc_txn->mt_free_pgs, pg);
+			mdb_midl_append(mc->mc_txn->mt_free_pgs, pg);
 			pg++;
 		}
 	}
@@ -4077,6 +4146,7 @@ mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
 					psize += sizeof(pgno_t);
 				else
 					psize += NODEDSZ(node);
+				psize += psize & 1;
 				if (psize > pmax) {
 					split_indx = i;
 					break;
@@ -4091,6 +4161,7 @@ mdb_split(MDB_txn *txn, MDB_dbi dbi, MDB_page **mpp, unsigned int *newindxp,
 					psize += sizeof(pgno_t);
 				else
 					psize += NODEDSZ(node);
+				psize += psize & 1;
 				if (psize > pmax) {
 					split_indx = i+1;
 					break;
@@ -4180,7 +4251,7 @@ newsep:
 			rkey.mv_size = node->mn_ksize;
 			if (IS_LEAF(&mdp->p)) {
 				rdata.mv_data = NODEDATA(node);
-				rdata.mv_size = node->mn_dsize;
+				rdata.mv_size = NODEDSZ(node);
 			} else
 				pgno = NODEPGNO(node);
 			flags = node->mn_flags;
@@ -4207,164 +4278,6 @@ newsep:
 	return rc;
 }
 
-#if 0
-static int
-mdb_put0(MDB_txn *txn, MDB_dbi dbi,
-    MDB_val *key, MDB_val *data, unsigned int flags)
-{
-	int		 rc = MDB_SUCCESS, exact;
-	unsigned int	 ki;
-	MDB_node	*leaf;
-	MDB_pageparent	mpp;
-	MDB_val	xdata, *rdata, dkey;
-	MDB_db dummy;
-	char dbuf[PAGESIZE];
-	int do_sub = 0;
-	size_t nsize;
-	DKBUF;
-
-	DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
-		dbi, DKEY(key), key->mv_size, data->mv_size);
-
-	dkey.mv_size = 0;
-	mpp.mp_parent = NULL;
-	mpp.mp_pi = 0;
-	rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
-	if (rc == MDB_SUCCESS) {
-		leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
-		if (leaf && exact) {
-			if (flags == MDB_NOOVERWRITE) {
-				DPRINTF("duplicate key [%s]", DKEY(key));
-				return MDB_KEYEXIST;
-			}
-			/* there's only a key anyway, so this is a no-op */
-			if (IS_LEAF2(mpp.mp_page))
-				return MDB_SUCCESS;
-
-			if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
-				/* Was a single item before, must convert now */
-				if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
-					dkey.mv_size = NODEDSZ(leaf);
-					dkey.mv_data = dbuf;
-					memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
-					/* data matches, ignore it */
-					if (!mdb_dcmp(txn, dbi, data, &dkey))
-						return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
-					memset(&dummy, 0, sizeof(dummy));
-					if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) {
-						dummy.md_pad = data->mv_size;
-						dummy.md_flags = MDB_DUPFIXED;
-						if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP)
-							dummy.md_flags |= MDB_INTEGERKEY;
-					}
-					dummy.md_root = P_INVALID;
-					if (dkey.mv_size == sizeof(MDB_db)) {
-						memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
-						goto put_sub;
-					}
-					mdb_del_node(mpp.mp_page, ki, 0);
-					do_sub = 1;
-					rdata = &xdata;
-					xdata.mv_size = sizeof(MDB_db);
-					xdata.mv_data = &dummy;
-					goto new_sub;
-				}
-				goto put_sub;
-			}
-			/* same size, just replace it */
-			if (!F_ISSET(leaf->mn_flags, F_BIGDATA) &&
-				NODEDSZ(leaf) == data->mv_size) {
-				memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
-				goto done;
-			}
-			mdb_del_node(mpp.mp_page, ki, 0);
-		}
-		if (leaf == NULL) {		/* append if not found */
-			ki = NUMKEYS(mpp.mp_page);
-			DPRINTF("appending key at index %i", ki);
-		}
-	} else if (rc == MDB_NOTFOUND) {
-		MDB_dpage *dp;
-		/* new file, just write a root leaf page */
-		DPUTS("allocating new root leaf page");
-		if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
-			return ENOMEM;
-		}
-		mpp.mp_page = &dp->p;
-		txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
-		txn->mt_dbs[dbi].md_depth++;
-		txn->mt_dbxs[dbi].md_dirty = 1;
-		if ((txn->mt_dbs[dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED)
-			mpp.mp_page->mp_flags |= P_LEAF2;
-		ki = 0;
-	}
-	else
-		goto done;
-
-	assert(IS_LEAF(mpp.mp_page));
-	DPRINTF("there are %u keys, should insert new key at index %i",
-		NUMKEYS(mpp.mp_page), ki);
-
-	rdata = data;
-
-new_sub:
-	nsize = IS_LEAF2(mpp.mp_page) ? key->mv_size : mdb_leaf_size(txn->mt_env, key, rdata);
-	if (SIZELEFT(mpp.mp_page) < nsize) {
-		rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
-	} else {
-		/* There is room already in this leaf page. */
-		rc = mdb_add_node(txn, dbi, mpp.mp_page, ki, key, rdata, 0, 0);
-	}
-
-	if (rc != MDB_SUCCESS)
-		txn->mt_flags |= MDB_TXN_ERROR;
-	else {
-		/* Remember if we just added a subdatabase */
-		if (flags & F_SUBDATA) {
-			leaf = NODEPTR(mpp.mp_page, ki);
-			leaf->mn_flags |= F_SUBDATA;
-		}
-
-		/* Now store the actual data in the child DB. Note that we're
-		 * storing the user data in the keys field, so there are strict
-		 * size limits on dupdata. The actual data fields of the child
-		 * DB are all zero size.
-		 */
-		if (do_sub) {
-			MDB_cursor mc;
-			MDB_xcursor mx;
-
-			leaf = NODEPTR(mpp.mp_page, ki);
-put_sub:
-			mc.mc_txn = txn;
-			mc.mc_dbi = dbi;
-			mc.mc_flags = 0;
-			mc.mc_xcursor = &mx;
-			mdb_xcursor_init0(&mc);
-			mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
-			xdata.mv_size = 0;
-			xdata.mv_data = "";
-			if (flags == MDB_NODUPDATA)
-				flags = MDB_NOOVERWRITE;
-			/* converted, write the original data first */
-			if (dkey.mv_size) {
-				rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
-				if (rc) return rc;
-				leaf->mn_flags |= F_DUPDATA;
-			}
-			rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
-			mdb_xcursor_fini(&mc);
-			memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
-				sizeof(MDB_db));
-		}
-		txn->mt_dbs[dbi].md_entries++;
-	}
-
-done:
-	return rc;
-}
-#endif
-
 int
 mdb_put(MDB_txn *txn, MDB_dbi dbi,
     MDB_val *key, MDB_val *data, unsigned int flags)
@@ -4463,12 +4376,10 @@ mdb_env_stat(MDB_env *env, MDB_stat *arg)
 static void
 mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi)
 {
-	if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
-#if __BYTE_ORDER == __LITTLE_ENDIAN
-		|MDB_INTEGERKEY
-#endif
-	))
+	if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEKEY)
 		txn->mt_dbxs[dbi].md_cmp = memnrcmp;
+	else if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERKEY)
+		txn->mt_dbxs[dbi].md_cmp = cintcmp;
 	else
 		txn->mt_dbxs[dbi].md_cmp = memncmp;
 
@@ -4477,11 +4388,7 @@ mdb_default_cmp(MDB_txn *txn, MDB_dbi dbi)
 			if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED)
 				txn->mt_dbxs[dbi].md_dcmp = intcmp;
 			else
-#if __BYTE_ORDER == __LITTLE_ENDIAN
-				txn->mt_dbxs[dbi].md_dcmp = memnrcmp;
-#else
-				txn->mt_dbxs[dbi].md_dcmp = memncmp;
-#endif
+				txn->mt_dbxs[dbi].md_dcmp = cintcmp;
 		} else if (txn->mt_dbs[dbi].md_flags & MDB_REVERSEDUP) {
 			txn->mt_dbxs[dbi].md_dcmp = memnrcmp;
 		} else {