struct ctdb_db_context *ctdb_db;
struct ctdb_ltdb_header hdr;
TDB_DATA key;
+ uint8_t keydata[1];
};
struct delete_records_list {
{
struct delete_record_data *dd;
uint32_t hash;
+ size_t len;
- dd = talloc_zero(tree, struct delete_record_data);
+ len = offsetof(struct delete_record_data, keydata) + key.dsize;
+
+ dd = (struct delete_record_data *)talloc_size(tree, len);
if (dd == NULL) {
DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
return -1;
}
+ talloc_set_name_const(dd, "struct delete_record_data");
dd->ctdb = ctdb;
dd->ctdb_db = ctdb_db;
dd->key.dsize = key.dsize;
- dd->key.dptr = talloc_memdup(dd, key.dptr, key.dsize);
- if (dd->key.dptr == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
- }
+ dd->key.dptr = dd->keydata;
+ memcpy(dd->keydata, key.dptr, key.dsize);
dd->hdr = *hdr;
tdb_data = tdb_fetch(ctdb_db->ltdb->tdb, dd->key);
if (tdb_data.dsize < sizeof(struct ctdb_ltdb_header)) {
- /* Does not exist or not a ctdb record. Skip. */
+ DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
+ "on database db[%s] does not exist or is not"
+ " a ctdb-record. skipping.\n",
+ hash, ctdb_db->db_name));
vdata->delete_skipped++;
goto done;
}
if (tdb_data.dsize > sizeof(struct ctdb_ltdb_header)) {
- /* The record has been recycled (filled with data). Skip. */
+ DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
+ "on database db[%s] has been recycled. "
+ "skipping.\n",
+ hash, ctdb_db->db_name));
vdata->delete_skipped++;
goto done;
}
header = (struct ctdb_ltdb_header *)tdb_data.dptr;
if (header->dmaster != ctdb->pnn) {
- /* The record has been migrated off the node. Skip. */
+ DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
+ "on database db[%s] has been migrated away. "
+ "skipping.\n",
+ hash, ctdb_db->db_name));
vdata->delete_skipped++;
goto done;
}
* The record has been migrated off the node and back again.
* But not requeued for deletion. Skip it.
*/
+ DEBUG(DEBUG_INFO, (__location__ ": record with hash [0x%08x] "
+ "on database db[%s] seems to have been "
+ "migrated away and back again (with empty "
+ "data). skipping.\n",
+ hash, ctdb_db->db_name));
vdata->delete_skipped++;
goto done;
}
lmaster = ctdb_lmaster(ctdb_db->ctdb, &dd->key);
if (lmaster != ctdb->pnn) {
- /* we are not lmaster - strange */
+ DEBUG(DEBUG_INFO, (__location__ ": not lmaster for record in "
+ "delete list (key hash [0x%08x], db[%s]). "
+ "Strange! skipping.\n",
+ hash, ctdb_db->db_name));
vdata->delete_skipped++;
goto done;
}
}
/**
- * Proces the delete list:
+ * Process the delete list:
* Send the records to delete to all other nodes with the
* try_delete_records control.
*/
{
int ret, i;
struct ctdb_context *ctdb = ctdb_db->ctdb;
+ struct delete_records_list *recs;
+ TDB_DATA indata, outdata;
+ struct ctdb_node_map *nodemap;
+ uint32_t *active_nodes;
+ int num_active_nodes;
+
+ if (vdata->delete_count == 0) {
+ return 0;
+ }
vdata->delete_left = vdata->delete_count;
- if (vdata->delete_count > 0) {
- struct delete_records_list *recs;
- TDB_DATA indata, outdata;
- int32_t res;
- struct ctdb_node_map *nodemap;
- uint32_t *active_nodes;
- int num_active_nodes;
+ recs = talloc_zero(vdata, struct delete_records_list);
+ if (recs == NULL) {
+ DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
+ return -1;
+ }
+ recs->records = (struct ctdb_marshall_buffer *)
+ talloc_zero_size(vdata,
+ offsetof(struct ctdb_marshall_buffer, data));
+ if (recs->records == NULL) {
+ DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
+ return -1;
+ }
+ recs->records->db_id = ctdb_db->db_id;
- recs = talloc_zero(vdata, struct delete_records_list);
- if (recs == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
- }
- recs->records = (struct ctdb_marshall_buffer *)
- talloc_zero_size(vdata,
- offsetof(struct ctdb_marshall_buffer, data));
- if (recs->records == NULL) {
- DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
- return -1;
- }
- recs->records->db_id = ctdb_db->db_id;
+ /*
+ * traverse the tree of all records we want to delete and
+ * create a blob we can send to the other nodes.
+ */
+ trbt_traversearray32(vdata->delete_list, 1,
+ delete_marshall_traverse, recs);
- /*
- * traverse the tree of all records we want to delete and
- * create a blob we can send to the other nodes.
- */
- trbt_traversearray32(vdata->delete_list, 1,
- delete_marshall_traverse, recs);
+ indata.dsize = talloc_get_size(recs->records);
+ indata.dptr = (void *)recs->records;
- indata.dsize = talloc_get_size(recs->records);
- indata.dptr = (void *)recs->records;
+ /*
+ * now tell all the active nodes to delete all these records
+ * (if possible)
+ */
- /*
- * now tell all the active nodes to delete all these records
- * (if possible)
- */
+ ret = ctdb_ctrl_getnodemap(ctdb, TIMELIMIT(),
+ CTDB_CURRENT_NODE,
+ recs, /* talloc context */
+ &nodemap);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,(__location__ " unable to get node map\n"));
+ return -1;
+ }
+
+ active_nodes = list_of_active_nodes(ctdb, nodemap,
+ nodemap, /* talloc context */
+ false /* include self */);
+ /* yuck! ;-) */
+ num_active_nodes = talloc_get_size(active_nodes)/sizeof(*active_nodes);
+
+ for (i = 0; i < num_active_nodes; i++) {
+ struct ctdb_marshall_buffer *records;
+ struct ctdb_rec_data *rec;
+ int32_t res;
- ret = ctdb_ctrl_getnodemap(ctdb, TIMELIMIT(),
- CTDB_CURRENT_NODE,
- recs, /* talloc context */
- &nodemap);
- if (ret != 0) {
- DEBUG(DEBUG_ERR,(__location__ " unable to get node map\n"));
+ ret = ctdb_control(ctdb, active_nodes[i], 0,
+ CTDB_CONTROL_TRY_DELETE_RECORDS, 0,
+ indata, recs, &outdata, &res,
+ NULL, NULL);
+ if (ret != 0 || res != 0) {
+ DEBUG(DEBUG_ERR, ("Failed to delete records on "
+ "node %u: ret[%d] res[%d]\n",
+ active_nodes[i], ret, res));
return -1;
}
- active_nodes = list_of_active_nodes(ctdb, nodemap,
- nodemap, /* talloc context */
- false /* include self */);
- /* yuck! ;-) */
- num_active_nodes = talloc_get_size(active_nodes)/sizeof(*active_nodes);
-
- for (i = 0; i < num_active_nodes; i++) {
- struct ctdb_marshall_buffer *records;
- struct ctdb_rec_data *rec;
-
- ret = ctdb_control(ctdb, active_nodes[i], 0,
- CTDB_CONTROL_TRY_DELETE_RECORDS, 0,
- indata, recs, &outdata, &res,
- NULL, NULL);
- if (ret != 0 || res != 0) {
- DEBUG(DEBUG_ERR, ("Failed to delete records on "
- "node %u: ret[%d] res[%d]\n",
- active_nodes[i], ret, res));
+ /*
+ * outdata contains the list of records coming back
+ * from the node: These are the records that the
+ * remote node could not delete.
+ *
+ * NOTE: There is a problem here:
+ *
+ * When a node failed to delete the record, but
+ * others succeeded, we may have created gaps in the
+ * history of the record. Hence when a node dies, an
+ * closed file handle might be resurrected or an open
+ * file handle might be lost, leading to blocked access
+ * or data corruption.
+ *
+ * TODO: This needs to be fixed!
+ */
+ records = (struct ctdb_marshall_buffer *)outdata.dptr;
+ rec = (struct ctdb_rec_data *)&records->data[0];
+ while (records->count-- > 1) {
+ TDB_DATA reckey, recdata;
+ struct ctdb_ltdb_header *rechdr;
+ struct delete_record_data *dd;
+
+ reckey.dptr = &rec->data[0];
+ reckey.dsize = rec->keylen;
+ recdata.dptr = &rec->data[reckey.dsize];
+ recdata.dsize = rec->datalen;
+
+ if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
+ DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
return -1;
}
-
- /*
- * outdata contains the list of records coming back
- * from the node: These are the records that the
- * remote node could not delete.
- *
- * NOTE: There is a problem here:
- *
- * When a node failed to delete the record, but
- * others succeeded, we may have created gaps in the
- * history of the record. Hence when a node dies, an
- * closed file handle might be resurrected or an open
- * file handle might be lost, leading to blocked access
- * or data corruption.
- *
- * TODO: This needs to be fixed!
- */
- records = (struct ctdb_marshall_buffer *)outdata.dptr;
- rec = (struct ctdb_rec_data *)&records->data[0];
- while (records->count-- > 1) {
- TDB_DATA reckey, recdata;
- struct ctdb_ltdb_header *rechdr;
- struct delete_record_data *dd;
-
- reckey.dptr = &rec->data[0];
- reckey.dsize = rec->keylen;
- recdata.dptr = &rec->data[reckey.dsize];
- recdata.dsize = rec->datalen;
-
- if (recdata.dsize < sizeof(struct ctdb_ltdb_header)) {
- DEBUG(DEBUG_CRIT,(__location__ " bad ltdb record\n"));
- return -1;
- }
- rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
- recdata.dptr += sizeof(*rechdr);
- recdata.dsize -= sizeof(*rechdr);
-
- dd = (struct delete_record_data *)trbt_lookup32(
- vdata->delete_list,
- (uint32_t)tdb_jenkins_hash(&reckey));
- if (dd != NULL) {
- /*
- * The other node could not delete the
- * record and it is the first node that
- * failed. So we should remove it from
- * the tree and update statistics.
- */
- talloc_free(dd);
- vdata->delete_remote_error++;
- vdata->delete_left--;
- }
-
- rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec);
+ rechdr = (struct ctdb_ltdb_header *)recdata.dptr;
+ recdata.dptr += sizeof(*rechdr);
+ recdata.dsize -= sizeof(*rechdr);
+
+ dd = (struct delete_record_data *)trbt_lookup32(
+ vdata->delete_list,
+ (uint32_t)tdb_jenkins_hash(&reckey));
+ if (dd != NULL) {
+ /*
+ * The other node could not delete the
+ * record and it is the first node that
+ * failed. So we should remove it from
+ * the tree and update statistics.
+ */
+ talloc_free(dd);
+ vdata->delete_remote_error++;
+ vdata->delete_left--;
}
- }
- /* free nodemap and active_nodes */
- talloc_free(nodemap);
+ rec = (struct ctdb_rec_data *)(rec->length + (uint8_t *)rec);
+ }
}
+ /* free nodemap and active_nodes */
+ talloc_free(nodemap);
+
if (vdata->delete_left > 0) {
/*
* The only records remaining in the tree are those
git.samba.org / ctdb.git / blobdiff