andykimpe / rpms / 389-ds-base

Forked from rpms/389-ds-base 5 months ago
Clone
Blob Blame History Raw
From f9c0ae9e0c143359ef12c8f5ae3070e34afd5495 Mon Sep 17 00:00:00 2001
From: Ludwig Krispenz <lkrispen@redhat.com>
Date: Wed, 15 Jan 2020 13:40:36 +0100
Subject: [PATCH] Ticket 49624 cont - DB Deadlock on modrdn appears to corrupt
 database and entry cache

Bug: If there are deadlocks a transaction will be retried. In the case
	of modrdn operation there is an error in handling the newsuperior
	dn, which has to be reset when the txn is repeated.
     There is also an error in freeing the entry stored in the pblock which can
	lead to a double free
     There is also a memory leak for ec entries

Fix: check if the newsuperior in the pblock was changed before the retry and
	only then free and reset it.
     check and protect pblock entry from double free
     remove ec entry from cache
     fix the txn_test_thread to run

     There is also a message at shutdown that entries remain in the entry cache
	although no leaks are reported and a hash dump didn't show entries.
	Change log level to avoid confusion

Reviewed by: Thierry, William, Viktor - Thanks
---
 ldap/servers/slapd/back-ldbm/cache.c          |  2 +-
 .../slapd/back-ldbm/db-bdb/bdb_layer.c        |  2 +-
 ldap/servers/slapd/back-ldbm/ldbm_modrdn.c    | 60 +++++++++++++------
 3 files changed, 45 insertions(+), 19 deletions(-)

diff --git a/ldap/servers/slapd/back-ldbm/cache.c b/ldap/servers/slapd/back-ldbm/cache.c
index a03cdaa83..89f958a35 100644
--- a/ldap/servers/slapd/back-ldbm/cache.c
+++ b/ldap/servers/slapd/back-ldbm/cache.c
@@ -723,7 +723,7 @@ entrycache_clear_int(struct cache *cache)
     }
     cache->c_maxsize = size;
     if (cache->c_curentries > 0) {
-        slapi_log_err(SLAPI_LOG_WARNING,
+        slapi_log_err(SLAPI_LOG_CACHE,
                       "entrycache_clear_int", "There are still %" PRIu64 " entries "
                                               "in the entry cache.\n",
                       cache->c_curentries);
diff --git a/ldap/servers/slapd/back-ldbm/db-bdb/bdb_layer.c b/ldap/servers/slapd/back-ldbm/db-bdb/bdb_layer.c
index 5a6a2a2e5..36bf42dab 100644
--- a/ldap/servers/slapd/back-ldbm/db-bdb/bdb_layer.c
+++ b/ldap/servers/slapd/back-ldbm/db-bdb/bdb_layer.c
@@ -3064,7 +3064,7 @@ txn_test_threadmain(void *param)
 
     txn_test_init_cfg(&cfg);
 
-    if(BDB_CONFIG(li)->bdb_enable_transactions) {
+    if(!BDB_CONFIG(li)->bdb_enable_transactions) {
         goto end;
     }
 
diff --git a/ldap/servers/slapd/back-ldbm/ldbm_modrdn.c b/ldap/servers/slapd/back-ldbm/ldbm_modrdn.c
index 433ed88fb..26698012a 100644
--- a/ldap/servers/slapd/back-ldbm/ldbm_modrdn.c
+++ b/ldap/servers/slapd/back-ldbm/ldbm_modrdn.c
@@ -67,6 +67,7 @@ ldbm_back_modrdn(Slapi_PBlock *pb)
     Slapi_DN *dn_newsuperiordn = NULL;
     Slapi_DN dn_parentdn;
     Slapi_DN *orig_dn_newsuperiordn = NULL;
+    Slapi_DN *pb_dn_newsuperiordn = NULL; /* used to check what is currently in the pblock */
     Slapi_Entry *target_entry = NULL;
     Slapi_Entry *original_targetentry = NULL;
     int rc;
@@ -248,30 +249,45 @@ ldbm_back_modrdn(Slapi_PBlock *pb)
             slapi_sdn_set_dn_byref(&dn_newrdn, original_newrdn);
             original_newrdn = slapi_ch_strdup(original_newrdn);
 
-            slapi_pblock_get(pb, SLAPI_MODRDN_NEWSUPERIOR_SDN, &dn_newsuperiordn);
-            slapi_sdn_free(&dn_newsuperiordn);
-            slapi_pblock_set(pb, SLAPI_MODRDN_NEWSUPERIOR_SDN, orig_dn_newsuperiordn);
-            dn_newsuperiordn = slapi_sdn_dup(orig_dn_newsuperiordn);
+            /* we need to restart with the original newsuperiordn which could have
+             * been modified. So check what is in the pblock, if it was changed
+             * free it, reset orig dn in th epblock and recreate a working superior
+             */
+            slapi_pblock_get(pb, SLAPI_MODRDN_NEWSUPERIOR_SDN, &pb_dn_newsuperiordn);
+            if (pb_dn_newsuperiordn != orig_dn_newsuperiordn) {
+                slapi_sdn_free(&pb_dn_newsuperiordn);
+                slapi_pblock_set(pb, SLAPI_MODRDN_NEWSUPERIOR_SDN, orig_dn_newsuperiordn);
+                dn_newsuperiordn = slapi_sdn_dup(orig_dn_newsuperiordn);
+            }
             /* must duplicate ec before returning it to cache,
              * which could free the entry. */
-            if ((tmpentry = backentry_dup(original_entry ? original_entry : ec)) == NULL) {
+            if (!original_entry) {
+                slapi_log_err(SLAPI_LOG_ERR, "ldbm_back_modrdn",
+                    "retrying transaction, but no original entry found\n");
+                ldap_result_code = LDAP_OPERATIONS_ERROR;
+                goto error_return;
+            }
+            if ((tmpentry = backentry_dup(original_entry)) == NULL) {
                 ldap_result_code = LDAP_OPERATIONS_ERROR;
                 goto error_return;
             }
             slapi_pblock_get(pb, SLAPI_MODRDN_EXISTING_ENTRY, &ent);
             if (cache_is_in_cache(&inst->inst_cache, ec)) {
                 CACHE_REMOVE(&inst->inst_cache, ec);
-                if (ent && (ent == ec->ep_entry)) {
-                    /*
-                     * On a retry, it's possible that ec is now stored in the
-                     * pblock as SLAPI_MODRDN_EXISTING_ENTRY.  "ec" will be freed
-                     * by CACHE_RETURN below, so set ent to NULL so don't free
-                     * it again.
-                     */
-                    ent = NULL;
-                }
+            }
+            if (ent && (ent == ec->ep_entry)) {
+                /*
+                 * On a retry, it's possible that ec is now stored in the
+                 * pblock as SLAPI_MODRDN_EXISTING_ENTRY.  "ec" will be freed
+                 * by CACHE_RETURN below, so set ent to NULL so don't free
+                 * it again.
+                 * And it needs to be checked always.
+                 */
+                ent = NULL;
             }
             CACHE_RETURN(&inst->inst_cache, &ec);
+
+            /* LK why do we need this ????? */
             if (!cache_is_in_cache(&inst->inst_cache, e)) {
                 if (CACHE_ADD(&inst->inst_cache, e, NULL) < 0) {
                     slapi_log_err(SLAPI_LOG_CACHE,
@@ -1087,8 +1103,9 @@ ldbm_back_modrdn(Slapi_PBlock *pb)
         if (slapi_sdn_get_dn(dn_newsuperiordn) != NULL) {
             retval = ldbm_ancestorid_move_subtree(be, sdn, &dn_newdn, e->ep_id, children, &txn);
             if (retval != 0) {
-                if (retval == DB_LOCK_DEADLOCK)
+                if (retval == DB_LOCK_DEADLOCK) {
                     continue;
+                }
                 if (retval == DB_RUNRECOVERY || LDBM_OS_ERR_IS_DISKFULL(retval))
                     disk_full = 1;
                 MOD_SET_ERROR(ldap_result_code,
@@ -1108,8 +1125,9 @@ ldbm_back_modrdn(Slapi_PBlock *pb)
                                              e->ep_id, &txn, is_tombstone);
             slapi_rdn_done(&newsrdn);
             if (retval != 0) {
-                if (retval == DB_LOCK_DEADLOCK)
+                if (retval == DB_LOCK_DEADLOCK) {
                     continue;
+                }
                 if (retval == DB_RUNRECOVERY || LDBM_OS_ERR_IS_DISKFULL(retval))
                     disk_full = 1;
                 MOD_SET_ERROR(ldap_result_code, LDAP_OPERATIONS_ERROR, retry_count);
@@ -1500,7 +1518,12 @@ common_return:
     done_with_pblock_entry(pb, SLAPI_MODRDN_NEWPARENT_ENTRY);
     done_with_pblock_entry(pb, SLAPI_MODRDN_TARGET_ENTRY);
     slapi_ch_free_string(&original_newrdn);
-    slapi_sdn_free(&orig_dn_newsuperiordn);
+    slapi_pblock_get(pb, SLAPI_MODRDN_NEWSUPERIOR_SDN, &pb_dn_newsuperiordn);
+    if (pb_dn_newsuperiordn != orig_dn_newsuperiordn) {
+        slapi_sdn_free(&orig_dn_newsuperiordn);
+    } else {
+        slapi_sdn_free(&dn_newsuperiordn);
+    }
     backentry_free(&original_entry);
     backentry_free(&tmpentry);
     slapi_entry_free(original_targetentry);
@@ -1561,6 +1584,9 @@ moddn_unlock_and_return_entry(
     /* Something bad happened so we should give back all the entries */
     if (*targetentry != NULL) {
         cache_unlock_entry(&inst->inst_cache, *targetentry);
+        if (cache_is_in_cache(&inst->inst_cache, *targetentry)) {
+            CACHE_REMOVE(&inst->inst_cache, *targetentry);
+        }
         CACHE_RETURN(&inst->inst_cache, targetentry);
         *targetentry = NULL;
     }
-- 
2.21.1