]> git.sur5r.net Git - openldap/blobdiff - servers/slapd/back-bdb/idl.c
import fix to ITS#5172
[openldap] / servers / slapd / back-bdb / idl.c
index 5c5e0ef304eb3524127699796d603ddf3883f2d5..e557c701bfe0423941434fe3aaa3eec13734444d 100644 (file)
@@ -2,7 +2,7 @@
 /* $OpenLDAP$ */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2000-2005 The OpenLDAP Foundation.
+ * Copyright 2000-2007 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -224,10 +224,9 @@ int bdb_idl_insert( ID *ids, ID id )
        return 0;
 }
 
-#if 0  /* unused */
-static int idl_delete( ID *ids, ID id )
+static int bdb_idl_delete( ID *ids, ID id )
 {
-       unsigned x = bdb_idl_search( ids, id );
+       unsigned x;
 
 #if IDL_DEBUG > 1
        Debug( LDAP_DEBUG_ANY, "delete: %04lx at %d\n", (long) id, x, 0 );
@@ -236,6 +235,23 @@ static int idl_delete( ID *ids, ID id )
        idl_check( ids );
 #endif
 
+       if (BDB_IDL_IS_RANGE( ids )) {
+               /* If deleting a range boundary, adjust */
+               if ( ids[1] == id )
+                       ids[1]++;
+               else if ( ids[2] == id )
+                       ids[2]--;
+               /* deleting from inside a range is a no-op */
+
+               /* If the range has collapsed, re-adjust */
+               if ( ids[1] > ids[2] )
+                       ids[0] = 0;
+               else if ( ids[1] == ids[2] )
+                       ids[1] = 1;
+               return 0;
+       }
+
+       x = bdb_idl_search( ids, id );
        assert( x > 0 );
 
        if( x <= 0 ) {
@@ -264,7 +280,6 @@ static int idl_delete( ID *ids, ID id )
 
        return 0;
 }
-#endif /* unused */
 
 static char *
 bdb_show_key(
@@ -327,17 +342,17 @@ bdb_idl_cache_put(
        bdb_idl_cache_entry_t idl_tmp;
        bdb_idl_cache_entry_t *ee;
 
+       if ( rc == DB_NOTFOUND || BDB_IDL_IS_ZERO( ids ))
+               return;
+
        DBT2bv( key, &idl_tmp.kstr );
 
        ee = (bdb_idl_cache_entry_t *) ch_malloc(
                sizeof( bdb_idl_cache_entry_t ) );
        ee->db = db;
-       if ( rc == DB_NOTFOUND || BDB_IDL_IS_ZERO( ids )) {
-               ee->idl = NULL;
-       } else {
-               ee->idl = (ID*) ch_malloc( BDB_IDL_SIZEOF ( ids ) );
-               BDB_IDL_CPY( ee->idl, ids );
-       }
+       ee->idl = (ID*) ch_malloc( BDB_IDL_SIZEOF ( ids ) );
+       BDB_IDL_CPY( ee->idl, ids );
+
        ee->idl_lru_prev = NULL;
        ee->idl_lru_next = NULL;
        ber_dupbv( &ee->kstr, &idl_tmp.kstr );
@@ -407,6 +422,64 @@ bdb_idl_cache_del(
        ldap_pvt_thread_rdwr_wunlock( &bdb->bi_idl_tree_rwlock );
 }
 
+void
+bdb_idl_cache_add_id(
+       struct bdb_info *bdb,
+       DB                      *db,
+       DBT                     *key,
+       ID                      id )
+{
+       bdb_idl_cache_entry_t *cache_entry, idl_tmp;
+       DBT2bv( key, &idl_tmp.kstr );
+       idl_tmp.db = db;
+       ldap_pvt_thread_rdwr_wlock( &bdb->bi_idl_tree_rwlock );
+       cache_entry = avl_find( bdb->bi_idl_tree, &idl_tmp,
+                                     bdb_idl_entry_cmp );
+       if ( cache_entry != NULL ) {
+               if ( !BDB_IDL_IS_RANGE( cache_entry->idl ) &&
+                       cache_entry->idl[0] < BDB_IDL_DB_MAX ) {
+                       size_t s = BDB_IDL_SIZEOF( cache_entry->idl ) + sizeof(ID);
+                       cache_entry->idl = ch_realloc( cache_entry->idl, s );
+               }
+               bdb_idl_insert( cache_entry->idl, id );
+       }
+       ldap_pvt_thread_rdwr_wunlock( &bdb->bi_idl_tree_rwlock );
+}
+
+void
+bdb_idl_cache_del_id(
+       struct bdb_info *bdb,
+       DB                      *db,
+       DBT                     *key,
+       ID                      id )
+{
+       bdb_idl_cache_entry_t *cache_entry, idl_tmp;
+       DBT2bv( key, &idl_tmp.kstr );
+       idl_tmp.db = db;
+       ldap_pvt_thread_rdwr_wlock( &bdb->bi_idl_tree_rwlock );
+       cache_entry = avl_find( bdb->bi_idl_tree, &idl_tmp,
+                                     bdb_idl_entry_cmp );
+       if ( cache_entry != NULL ) {
+               bdb_idl_delete( cache_entry->idl, id );
+               if ( cache_entry->idl[0] == 0 ) {
+                       if ( avl_delete( &bdb->bi_idl_tree, (caddr_t) cache_entry,
+                                               bdb_idl_entry_cmp ) == NULL ) {
+                               Debug( LDAP_DEBUG_ANY, "=> bdb_idl_cache_del: "
+                                       "AVL delete failed\n",
+                                       0, 0, 0 );
+                       }
+                       --bdb->bi_idl_cache_size;
+                       ldap_pvt_thread_mutex_lock( &bdb->bi_idl_tree_lrulock );
+                       IDL_LRU_DELETE( bdb, cache_entry );
+                       ldap_pvt_thread_mutex_unlock( &bdb->bi_idl_tree_lrulock );
+                       free( cache_entry->kstr.bv_val );
+                       free( cache_entry->idl );
+                       free( cache_entry );
+               }
+       }
+       ldap_pvt_thread_rdwr_wunlock( &bdb->bi_idl_tree_rwlock );
+}
+
 int
 bdb_idl_fetch_key(
        BackendDB       *be,
@@ -1207,8 +1280,11 @@ ID bdb_idl_next( ID *ids, ID *cursor )
        return NOID;
 }
 
-/* Add one ID to an unsorted list. We still maintain a lo/hi reference
- * for fast range compaction.
+#ifdef BDB_HIER
+
+/* Add one ID to an unsorted list. We ensure that the first element is the
+ * minimum and the last element is the maximum, for fast range compaction.
+ *   this means IDLs up to length 3 are always sorted...
  */
 int bdb_idl_append_one( ID *ids, ID id )
 {
@@ -1229,27 +1305,29 @@ int bdb_idl_append_one( ID *ids, ID id )
                        tmp = ids[1];
                        ids[1] = id;
                        id = tmp;
-               } else if ( ids[0] > 1 && id > ids[2] ) {
-                       tmp = ids[2];
-                       ids[2] = id;
+               }
+               if ( ids[0] > 1 && id < ids[ids[0]] ) {
+                       tmp = ids[ids[0]];
+                       ids[ids[0]] = id;
                        id = tmp;
                }
        }
        ids[0]++;
        if ( ids[0] >= BDB_IDL_UM_MAX ) {
                ids[0] = NOID;
+               ids[2] = id;
        } else {
                ids[ids[0]] = id;
        }
        return 0;
 }
 
-/* Append unsorted list b to unsorted list a. Both lists must have their
- * lowest value in slot 1 and highest value in slot 2.
+/* Append sorted list b to sorted list a. The result is unsorted but
+ * a[1] is the min of the result and a[a[0]] is the max.
  */
 int bdb_idl_append( ID *a, ID *b )
 {
-       ID ida, idb;
+       ID ida, idb, tmp, swap = 0;
 
        if ( BDB_IDL_IS_ZERO( b ) ) {
                return 0;
@@ -1260,80 +1338,217 @@ int bdb_idl_append( ID *a, ID *b )
                return 0;
        }
 
+       ida = BDB_IDL_LAST( a );
+       idb = BDB_IDL_LAST( b );
        if ( BDB_IDL_IS_RANGE( a ) || BDB_IDL_IS_RANGE(b) ||
                a[0] + b[0] >= BDB_IDL_UM_MAX ) {
-               ida = IDL_MIN( a[1], b[1] );
-               idb = IDL_MAX( a[2], b[2] );
+               a[2] = IDL_MAX( ida, idb );
+               a[1] = IDL_MIN( a[1], b[1] );
                a[0] = NOID;
-               a[1] = ida;
-               a[2] = idb;
                return 0;
        }
 
-       if ( b[1] < a[1] ) {
-               ida = a[1];
-               a[1] = b[1];
-       } else {
-               ida = b[1];
+       if ( b[0] > 1 && ida > idb ) {
+               swap = idb;
+               a[a[0]] = idb;
+               b[b[0]] = ida;
        }
-       a[0]++;
-       a[a[0]] = ida;
 
-       if ( b[0] > 1 && b[2] > a[2] ) {
-               ida = a[2];
-               a[2] = b[2];
+       if ( b[1] < a[1] ) {
+               tmp = a[1];
+               a[1] = b[1];
        } else {
-               ida = b[2];
+               tmp = b[1];
        }
        a[0]++;
-       a[a[0]] = ida;
+       a[a[0]] = tmp;
 
-       if ( b[0] > 2 ) {
-               int i = b[0] - 2;
-               AC_MEMCPY(a+a[0]+1, b+3, i * sizeof(ID));
+       if ( b[0] > 1 ) {
+               int i = b[0] - 1;
+               AC_MEMCPY(a+a[0]+1, b+2, i * sizeof(ID));
                a[0] += i;
        }
+       if ( swap ) {
+               b[b[0]] = swap;
+       }
        return 0;
-       
 }
 
-/* Sort an IDL using HeapSort */
-static void
-siftDown(ID *ids, int root, int bottom)
+#if 1
+
+/* Quicksort + Insertion sort for small arrays */
+
+#define SMALL  8
+#define        SWAP(a,b)       itmp=(a);(a)=(b);(b)=itmp
+
+void
+bdb_idl_sort( ID *ids, ID *tmp )
 {
-       int child;
-       ID temp;
+       int *istack = (int *)tmp;
+       int i,j,k,l,ir,jstack;
+       ID a, itmp;
 
-       temp = ids[root];
-       while ((child=root*2) <= bottom) {
-               if (child < bottom && ids[child] < ids[child + 1])
-                       child++;
+       if ( BDB_IDL_IS_RANGE( ids ))
+               return;
 
-               if (temp >= ids[child])
-                       break;
-               ids[root] = ids[child];
-               root = child;
+       ir = ids[0];
+       l = 1;
+       jstack = 0;
+       for(;;) {
+               if (ir - l < SMALL) {   /* Insertion sort */
+                       for (j=l+1;j<=ir;j++) {
+                               a = ids[j];
+                               for (i=j-1;i>=1;i--) {
+                                       if (ids[i] <= a) break;
+                                       ids[i+1] = ids[i];
+                               }
+                               ids[i+1] = a;
+                       }
+                       if (jstack == 0) break;
+                       ir = istack[jstack--];
+                       l = istack[jstack--];
+               } else {
+                       k = (l + ir) >> 1;      /* Choose median of left, center, right */
+                       SWAP(ids[k], ids[l+1]);
+                       if (ids[l] > ids[ir]) {
+                               SWAP(ids[l], ids[ir]);
+                       }
+                       if (ids[l+1] > ids[ir]) {
+                               SWAP(ids[l+1], ids[ir]);
+                       }
+                       if (ids[l] > ids[l+1]) {
+                               SWAP(ids[l], ids[l+1]);
+                       }
+                       i = l+1;
+                       j = ir;
+                       a = ids[l+1];
+                       for(;;) {
+                               do i++; while(ids[i] < a);
+                               do j--; while(ids[j] > a);
+                               if (j < i) break;
+                               SWAP(ids[i],ids[j]);
+                       }
+                       ids[l+1] = ids[j];
+                       ids[j] = a;
+                       jstack += 2;
+                       if (ir-i+1 >= j-1) {
+                               istack[jstack] = ir;
+                               istack[jstack-1] = i;
+                               ir = j-1;
+                       } else {
+                               istack[jstack] = j-1;
+                               istack[jstack-1] = l;
+                               l = i;
+                       }
+               }
        }
-       ids[root] = temp;
 }
 
+#else
+
+/* 8 bit Radix sort + insertion sort
+ * 
+ * based on code from http://www.cubic.org/docs/radix.htm
+ * with improvements by mbackes@symas.com and hyc@symas.com
+ *
+ * This code is O(n) but has a relatively high constant factor. For lists
+ * up to ~50 Quicksort is slightly faster; up to ~100 they are even.
+ * Much faster than quicksort for lists longer than ~100. Insertion
+ * sort is actually superior for lists <50.
+ */
+
+#define BUCKETS        (1<<8)
+#define SMALL  50
+
 void
-bdb_idl_sort( ID *ids )
+bdb_idl_sort( ID *ids, ID *tmp )
 {
-       int i;
-       ID temp;
-
-       if ( BDB_IDL_IS_RANGE( ids ))
+       int count, soft_limit, phase = 0, size = ids[0];
+       ID *idls[2];
+       unsigned char *maxv = (unsigned char *)&ids[size];
+
+       if ( BDB_IDL_IS_RANGE( ids ))
+               return;
+
+       /* Use insertion sort for small lists */
+       if ( size <= SMALL ) {
+               int i,j;
+               ID a;
+
+               for (j=1;j<=size;j++) {
+                       a = ids[j];
+                       for (i=j-1;i>=1;i--) {
+                               if (ids[i] <= a) break;
+                               ids[i+1] = ids[i];
+                       }
+                       ids[i+1] = a;
+               }
                return;
+       }
 
-       for (i = ids[0] / 2; i >= 1; i--)
-               siftDown(ids, i, ids[0]);
+       tmp[0] = size;
+       idls[0] = ids;
+       idls[1] = tmp;
 
-       for (i = ids[0]; i > 1; i--)
-       {
-               temp = ids[i];
-               ids[i] = ids[1];
-               ids[1] = temp;
-               siftDown(ids, 1, i-1);
+#if BYTE_ORDER == BIG_ENDIAN
+    for (soft_limit = 0; !maxv[soft_limit]; soft_limit++);
+#else
+    for (soft_limit = sizeof(ID)-1; !maxv[soft_limit]; soft_limit--);
+#endif
+
+       for (
+#if BYTE_ORDER == BIG_ENDIAN
+       count = sizeof(ID)-1; count >= soft_limit; --count
+#else
+       count = 0; count <= soft_limit; ++count
+#endif
+       ) {
+               unsigned int num[BUCKETS], * np, n, sum;
+               int i;
+        ID *sp, *source, *dest;
+        unsigned char *bp, *source_start;
+
+               source = idls[phase]+1;
+               dest = idls[phase^1]+1;
+               source_start =  ((unsigned char *) source) + count;
+
+        np = num;
+        for ( i = BUCKETS; i > 0; --i ) *np++ = 0;
+
+               /* count occurences of every byte value */
+               bp = source_start;
+        for ( i = size; i > 0; --i, bp += sizeof(ID) )
+                               num[*bp]++;
+
+               /* transform count into index by summing elements and storing
+                * into same array
+                */
+        sum = 0;
+        np = num;
+        for ( i = BUCKETS; i > 0; --i ) {
+                n = *np;
+                *np++ = sum;
+                sum += n;
+        }
+
+               /* fill dest with the right values in the right place */
+               bp = source_start;
+        sp = source;
+        for ( i = size; i > 0; --i, bp += sizeof(ID) ) {
+                np = num + *bp;
+                dest[*np] = *sp++;
+                ++(*np);
+        }
+               phase ^= 1;
+       }
+
+       /* copy back from temp if needed */
+       if ( phase ) {
+               ids++; tmp++;
+               for ( count = 0; count < size; ++count ) 
+                       *ids++ = *tmp++;
        }
 }
+#endif /* Quick vs Radix */
+
+#endif /* BDB_HIER */