Bug 644608 - RHDS 8.1->8.2 upgrade fails to properly migrate ACIs

https://bugzilla.redhat.com/show_bug.cgi?id=644608

Description: Upgrade script ##upgradednformat.pl generates an
ancestorid index file in which the entry IDs are not sorted
in the index values.
This was caused because
1. upgradednformat.pl copies db files to a work dir
     /path/to/db/<inst>/dnupgrade/<inst>
   and run upgradednformat against the work dir.
2. Since ancestorid index needs to be created from the scratch,
   import/upgradednformat code deletes the index and recreates
   it.
3. When creating the index file, dblayer_open_file checks if
   the to-be-opened index file is (in the standard location AND
   the file exists) or not.  In this case, the condition is
   satisfied.  Thus, the db is created using an absolute path
   and closed once.  Then, it is reopened using a relative path.
4. Before opening an index file, callback functions for libdb
   are set.  idl_new_compare_dups is one of them which is used
   to sort entry IDs in the secondary index attribute values.
   The setting is discarded by the close described in 3.
   This patch resets the db callbacks and flags after the close.

Note: cherry-picked 058299aeaf48e34f4359f00cc05eb7186a80fc48

1 files changed, 84 insertions, 74 deletions

diff --git a/ldap/servers/slapd/back-ldbm/dblayer.c b/ldap/servers/slapd/back-ldbm/dblayer.c
index 5c1f6e6c..a052d00b 100644
--- a/ldap/servers/slapd/back-ldbm/dblayer.c
+++ b/ldap/servers/slapd/back-ldbm/dblayer.c
@@ -2774,6 +2774,79 @@ dblayer_remove_env(struct ldbminfo *li)
 #define DB_DUPSORT 0
 #endif
 
+static int
+_dblayer_set_db_callbacks(dblayer_private *priv, DB *dbp, struct attrinfo *ai)
+{
+    int rc = 0;
+
+    /* With the new idl design, the large 8Kbyte pages we use are not
+       optimal. The page pool churns very quickly as we add new IDs under a
+       sustained add load. Smaller pages stop this happening so much and
+       consequently make us spend less time flushing dirty pages on checkpoints.
+       But 8K is still a good page size for id2entry. So we now allow different
+       page sizes for the primary and secondary indices. 
+       Filed as bug: 604654
+     */
+    if (idl_get_idl_new()) {
+        rc = dbp->set_pagesize(
+                        dbp,
+                        (priv->dblayer_index_page_size == 0) ?
+                        DBLAYER_INDEX_PAGESIZE : priv->dblayer_index_page_size);
+    } else {
+        rc = dbp->set_pagesize(
+                        dbp,
+                        (priv->dblayer_page_size == 0) ?
+                        DBLAYER_PAGESIZE : priv->dblayer_page_size);
+    }
+    if (rc)
+        return rc;
+
+#if 1000*DB_VERSION_MAJOR + 100*DB_VERSION_MINOR < 3300
+    rc = dbp->set_malloc(dbp, (void *)slapi_ch_malloc);
+    if (rc)
+        return rc;
+#endif
+
+    if (idl_get_idl_new() && !(ai->ai_indexmask & INDEX_VLV)) {
+        rc = dbp->set_flags(dbp, DB_DUP | DB_DUPSORT);
+        if (rc)
+            return rc;
+
+        rc = dbp->set_dup_compare( dbp, idl_new_compare_dups);
+        if (rc)
+            return rc;
+    }
+
+    if (ai->ai_indexmask & INDEX_VLV) {
+        /*
+         * Need index with record numbers for
+         * Virtual List View index
+         */
+        rc = dbp->set_flags(dbp, DB_RECNUM);
+        if (rc)
+            return rc;
+    } else if (ai->ai_key_cmp_fn) { /* set in attr_index_config() */
+        /*
+          This is so that we can have ordered keys in the index, so that
+          greater than/less than searches work on indexed attrs.  We had
+          to introduce this when we changed the integer key format from
+          a 32/64 bit value to a normalized string value.  The default
+          bdb key cmp is based on length and lexicographic order, which
+          does not work with integer strings.
+
+          NOTE: If we ever need to use app_private for something else, we
+          will have to create some sort of data structure with different
+          fields for different uses.  We will also need to have a new()
+          function that creates and allocates that structure, and a
+          destroy() function that destroys the structure, and make sure
+          to call it when the DB* is closed and/or freed.
+        */
+        dbp->app_private = (void *)ai->ai_key_cmp_fn;
+        dbp->set_bt_compare(dbp, dblayer_bt_compare);
+    }
+    return rc;
+}
+
 /* Routines for opening and closing random files in the DB_ENV.
    Used by ldif2db merging code currently.
 
@@ -2781,7 +2854,9 @@ dblayer_remove_env(struct ldbminfo *li)
        Success: 0
     Failure: -1
  */
-int dblayer_open_file(backend *be, char* indexname, int open_flag, struct attrinfo *ai, DB **ppDB)
+int
+dblayer_open_file(backend *be, char* indexname, int open_flag, 
+                  struct attrinfo *ai, DB **ppDB)
 {
     struct ldbminfo *li = (struct ldbminfo *) be->be_database->plg_private;
     ldbm_instance *inst = (ldbm_instance *) be->be_instance_info;
@@ -2840,79 +2915,9 @@ int dblayer_open_file(backend *be, char* indexname, int open_flag, struct attrin
         goto out;
 
     dbp = *ppDB;
-/* With the new idl design, the large 8Kbyte pages we use are not
-   optimal. The page pool churns very quickly as we add new IDs under a
-   sustained add load. Smaller pages stop this happening so much and
-   consequently make us spend less time flushing dirty pages on checkpoints.
-   But 8K is still a good page size for id2entry. So we now allow different
-   page sizes for the primary and secondary indices. 
-   Filed as bug: 604654
- */
-    if (idl_get_idl_new()) {
-        return_value = dbp->set_pagesize(
-                        dbp,
-                        (priv->dblayer_index_page_size == 0) ?
-                        DBLAYER_INDEX_PAGESIZE : priv->dblayer_index_page_size
-                        );
-    } else {
-        return_value = dbp->set_pagesize(
-                        dbp,
-                        (priv->dblayer_page_size == 0) ?
-                        DBLAYER_PAGESIZE : priv->dblayer_page_size
-                        );
-    }
-    if (0 != return_value)
-        goto out;
-
-#if 1000*DB_VERSION_MAJOR + 100*DB_VERSION_MINOR < 3300
-    return_value = dbp->set_malloc(dbp, (void *)slapi_ch_malloc);
-    if (0 != return_value) {
+    return_value = _dblayer_set_db_callbacks(priv, dbp, ai);
+    if (return_value)
         goto out;
-    }
-#endif
-
-    if (idl_get_idl_new() && !(ai->ai_indexmask & INDEX_VLV)) {
-        return_value = dbp->set_flags(dbp, DB_DUP | DB_DUPSORT);
-        if (0 != return_value)
-            goto out;
-
-        if (ai->ai_dup_cmp_fn) {
-            /* If set, use the special dup compare callback */
-            return_value = dbp->set_dup_compare(dbp, ai->ai_dup_cmp_fn);
-        } else {
-            return_value = dbp->set_dup_compare(dbp, idl_new_compare_dups);
-        }
-        if (0 != return_value)
-            goto out;
-    }
-
-    if (ai->ai_indexmask & INDEX_VLV) {
-        /*
-         * Need index with record numbers for
-         * Virtual List View index
-         */
-        return_value = dbp->set_flags(dbp, DB_RECNUM);
-        if (0 != return_value)
-            goto out;
-    } else if (ai->ai_key_cmp_fn) { /* set in attr_index_config() */
-        /*
-          This is so that we can have ordered keys in the index, so that
-          greater than/less than searches work on indexed attrs.  We had
-          to introduce this when we changed the integer key format from
-          a 32/64 bit value to a normalized string value.  The default
-          bdb key cmp is based on length and lexicographic order, which
-          does not work with integer strings.
-
-          NOTE: If we ever need to use app_private for something else, we
-          will have to create some sort of data structure with different
-          fields for different uses.  We will also need to have a new()
-          function that creates and allocates that structure, and a
-          destroy() function that destroys the structure, and make sure
-          to call it when the DB* is closed and/or freed.
-        */
-        dbp->app_private = (void *)ai->ai_key_cmp_fn;
-        dbp->set_bt_compare(dbp, dblayer_bt_compare);
-    }
 
     /* The subname argument allows applications to have
      * subdatabases, i.e., multiple databases inside of a single
@@ -2920,8 +2925,10 @@ int dblayer_open_file(backend *be, char* indexname, int open_flag, struct attrin
      * are both numerous and reasonably small, in order to
      * avoid creating a large number of underlying files.
      */
+    /* If inst_parent_dir_name is not the primary DB dir &&
+     * the index file does not exist */
     if ((charray_get_index(priv->dblayer_data_directories,
-                           inst->inst_parent_dir_name) != 0) &&
+                           inst->inst_parent_dir_name) > 0) &&
         !dblayer_inst_exists(inst, file_name))
     {
         char inst_dir[MAXPATHLEN];
@@ -2947,6 +2954,9 @@ int dblayer_open_file(backend *be, char* indexname, int open_flag, struct attrin
             goto out;
         }
         dbp = *ppDB;
+        return_value = _dblayer_set_db_callbacks(priv, dbp, ai);
+        if (return_value)
+            goto out;
 
         slapi_ch_free_string(&abs_file_name);
         if (inst_dirp != inst_dir)