]> granicus.if.org Git - postgresql/blobdiff - src/backend/executor/execMain.c
Ensure tableoid reads correctly in EvalPlanQual-manufactured tuples.
[postgresql] / src / backend / executor / execMain.c
index 3b664d09265e1565e7270254cafdf7affde1edf1..13ceffae5c46390e1bb912dea8c7294ac7e17ccc 100644 (file)
  *     ExecutorRun accepts direction and count arguments that specify whether
  *     the plan is to be executed forwards, backwards, and for how many tuples.
  *     In some cases ExecutorRun may be called multiple times to process all
- *     the tuples for a plan.  It is also acceptable to stop short of executing
+ *     the tuples for a plan.  It is also acceptable to stop short of executing
  *     the whole plan (but only if it is a SELECT).
  *
  *     ExecutorFinish must be called after the final ExecutorRun call and
  *     before ExecutorEnd.  This can be omitted only in case of EXPLAIN,
  *     which should also omit ExecutorRun.
  *
- * Portions Copyright (c) 1996-2013, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1996-2015, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  *
@@ -42,6 +42,7 @@
 #include "access/transam.h"
 #include "access/xact.h"
 #include "catalog/namespace.h"
+#include "commands/matview.h"
 #include "commands/trigger.h"
 #include "executor/execdebug.h"
 #include "foreign/fdwapi.h"
@@ -55,6 +56,7 @@
 #include "utils/acl.h"
 #include "utils/lsyscache.h"
 #include "utils/memutils.h"
+#include "utils/rls.h"
 #include "utils/snapmgr.h"
 #include "utils/tqual.h"
 
@@ -81,11 +83,23 @@ static void ExecutePlan(EState *estate, PlanState *planstate,
                        DestReceiver *dest);
 static bool ExecCheckRTEPerms(RangeTblEntry *rte);
 static void ExecCheckXactReadOnly(PlannedStmt *plannedstmt);
-static char *ExecBuildSlotValueDescription(TupleTableSlot *slot,
+static char *ExecBuildSlotValueDescription(Oid reloid,
+                                                         TupleTableSlot *slot,
+                                                         TupleDesc tupdesc,
+                                                         Bitmapset *modifiedCols,
                                                          int maxfieldlen);
 static void EvalPlanQualStart(EPQState *epqstate, EState *parentestate,
                                  Plan *planTree);
 
+/*
+ * Note that this macro also exists in commands/trigger.c.  There does not
+ * appear to be any good header to put it into, given the structures that
+ * it uses, so we let them be duplicated.  Be sure to update both if one needs
+ * to be changed, however.
+ */
+#define GetModifiedColumns(relinfo, estate) \
+       (rt_fetch((relinfo)->ri_RangeTableIndex, (estate)->es_range_table)->modifiedCols)
+
 /* end of local decls */
 
 
@@ -327,12 +341,12 @@ standard_ExecutorRun(QueryDesc *queryDesc,
  *             ExecutorFinish
  *
  *             This routine must be called after the last ExecutorRun call.
- *             It performs cleanup such as firing AFTER triggers.      It is
+ *             It performs cleanup such as firing AFTER triggers.  It is
  *             separate from ExecutorEnd because EXPLAIN ANALYZE needs to
  *             include these actions in the total runtime.
  *
  *             We provide a function hook variable that lets loadable plugins
- *             get control when ExecutorFinish is called.      Such a plugin would
+ *             get control when ExecutorFinish is called.  Such a plugin would
  *             normally call standard_ExecutorFinish().
  *
  * ----------------------------------------------------------------
@@ -499,6 +513,12 @@ ExecutorRewind(QueryDesc *queryDesc)
  *
  * Returns true if permissions are adequate.  Otherwise, throws an appropriate
  * error if ereport_on_violation is true, or simply returns false otherwise.
+ *
+ * Note that this does NOT address row level security policies (aka: RLS).  If
+ * rows will be returned to the user as a result of this permission check
+ * passing, then RLS also needs to be consulted (and check_enable_rls()).
+ *
+ * See rewrite/rowsecurity.c.
  */
 bool
 ExecCheckRTPerms(List *rangeTable, bool ereport_on_violation)
@@ -539,7 +559,6 @@ ExecCheckRTEPerms(RangeTblEntry *rte)
        AclMode         remainingPerms;
        Oid                     relOid;
        Oid                     userid;
-       Bitmapset  *tmpset;
        int                     col;
 
        /*
@@ -563,7 +582,7 @@ ExecCheckRTEPerms(RangeTblEntry *rte)
         * userid to check as: current user unless we have a setuid indication.
         *
         * Note: GetUserId() is presently fast enough that there's no harm in
-        * calling it separately for each RTE.  If that stops being true, we could
+        * calling it separately for each RTE.  If that stops being true, we could
         * call it once in ExecCheckRTPerms and pass the userid down from there.
         * But for now, no need for the extra clutter.
         */
@@ -606,12 +625,13 @@ ExecCheckRTEPerms(RangeTblEntry *rte)
                                        return false;
                        }
 
-                       tmpset = bms_copy(rte->selectedCols);
-                       while ((col = bms_first_member(tmpset)) >= 0)
+                       col = -1;
+                       while ((col = bms_next_member(rte->selectedCols, col)) >= 0)
                        {
-                               /* remove the column number offset */
-                               col += FirstLowInvalidHeapAttributeNumber;
-                               if (col == InvalidAttrNumber)
+                               /* bit #s are offset by FirstLowInvalidHeapAttributeNumber */
+                               AttrNumber      attno = col + FirstLowInvalidHeapAttributeNumber;
+
+                               if (attno == InvalidAttrNumber)
                                {
                                        /* Whole-row reference, must have priv on all cols */
                                        if (pg_attribute_aclcheck_all(relOid, userid, ACL_SELECT,
@@ -620,12 +640,11 @@ ExecCheckRTEPerms(RangeTblEntry *rte)
                                }
                                else
                                {
-                                       if (pg_attribute_aclcheck(relOid, col, userid,
+                                       if (pg_attribute_aclcheck(relOid, attno, userid,
                                                                                          ACL_SELECT) != ACLCHECK_OK)
                                                return false;
                                }
                        }
-                       bms_free(tmpset);
                }
 
                /*
@@ -648,24 +667,24 @@ ExecCheckRTEPerms(RangeTblEntry *rte)
                                        return false;
                        }
 
-                       tmpset = bms_copy(rte->modifiedCols);
-                       while ((col = bms_first_member(tmpset)) >= 0)
+                       col = -1;
+                       while ((col = bms_next_member(rte->modifiedCols, col)) >= 0)
                        {
-                               /* remove the column number offset */
-                               col += FirstLowInvalidHeapAttributeNumber;
-                               if (col == InvalidAttrNumber)
+                               /* bit #s are offset by FirstLowInvalidHeapAttributeNumber */
+                               AttrNumber      attno = col + FirstLowInvalidHeapAttributeNumber;
+
+                               if (attno == InvalidAttrNumber)
                                {
                                        /* whole-row reference can't happen here */
                                        elog(ERROR, "whole-row update is not implemented");
                                }
                                else
                                {
-                                       if (pg_attribute_aclcheck(relOid, col, userid,
+                                       if (pg_attribute_aclcheck(relOid, attno, userid,
                                                                                          remainingPerms) != ACLCHECK_OK)
                                                return false;
                                }
                        }
-                       bms_free(tmpset);
                }
        }
        return true;
@@ -828,7 +847,7 @@ InitPlan(QueryDesc *queryDesc, int eflags)
                erm->prti = rc->prti;
                erm->rowmarkId = rc->rowmarkId;
                erm->markType = rc->markType;
-               erm->noWait = rc->noWait;
+               erm->waitPolicy = rc->waitPolicy;
                ItemPointerSetInvalid(&(erm->curCtid));
                estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
        }
@@ -864,7 +883,8 @@ InitPlan(QueryDesc *queryDesc, int eflags)
                 * it is a parameterless subplan (not initplan), we suggest that it be
                 * prepared to handle REWIND efficiently; otherwise there is no need.
                 */
-               sp_eflags = eflags & EXEC_FLAG_EXPLAIN_ONLY;
+               sp_eflags = eflags
+                       & (EXEC_FLAG_EXPLAIN_ONLY | EXEC_FLAG_WITH_NO_DATA);
                if (bms_is_member(i, plannedstmt->rewindPlanIDs))
                        sp_eflags |= EXEC_FLAG_REWIND;
 
@@ -975,7 +995,7 @@ CheckValidResultRel(Relation resultRel, CmdType operation)
                                                  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
                                                   errmsg("cannot insert into view \"%s\"",
                                                                  RelationGetRelationName(resultRel)),
-                                                  errhint("To make the view insertable, provide an unconditional ON INSERT DO INSTEAD rule or an INSTEAD OF INSERT trigger.")));
+                                                  errhint("To enable inserting into the view, provide an INSTEAD OF INSERT trigger or an unconditional ON INSERT DO INSTEAD rule.")));
                                        break;
                                case CMD_UPDATE:
                                        if (!trigDesc || !trigDesc->trig_update_instead_row)
@@ -983,7 +1003,7 @@ CheckValidResultRel(Relation resultRel, CmdType operation)
                                                  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
                                                   errmsg("cannot update view \"%s\"",
                                                                  RelationGetRelationName(resultRel)),
-                                                  errhint("To make the view updatable, provide an unconditional ON UPDATE DO INSTEAD rule or an INSTEAD OF UPDATE trigger.")));
+                                                  errhint("To enable updating the view, provide an INSTEAD OF UPDATE trigger or an unconditional ON UPDATE DO INSTEAD rule.")));
                                        break;
                                case CMD_DELETE:
                                        if (!trigDesc || !trigDesc->trig_delete_instead_row)
@@ -991,7 +1011,7 @@ CheckValidResultRel(Relation resultRel, CmdType operation)
                                                  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
                                                   errmsg("cannot delete from view \"%s\"",
                                                                  RelationGetRelationName(resultRel)),
-                                                  errhint("To make the view updatable, provide an unconditional ON DELETE DO INSTEAD rule or an INSTEAD OF DELETE trigger.")));
+                                                  errhint("To enable deleting from the view, provide an INSTEAD OF DELETE trigger or an unconditional ON DELETE DO INSTEAD rule.")));
                                        break;
                                default:
                                        elog(ERROR, "unrecognized CmdType: %d", (int) operation);
@@ -999,10 +1019,11 @@ CheckValidResultRel(Relation resultRel, CmdType operation)
                        }
                        break;
                case RELKIND_MATVIEW:
-                       ereport(ERROR,
-                                       (errcode(ERRCODE_WRONG_OBJECT_TYPE),
-                                        errmsg("cannot change materialized view \"%s\"",
-                                                       RelationGetRelationName(resultRel))));
+                       if (!MatViewIncrementalMaintenanceIsEnabled())
+                               ereport(ERROR,
+                                               (errcode(ERRCODE_WRONG_OBJECT_TYPE),
+                                                errmsg("cannot change materialized view \"%s\"",
+                                                               RelationGetRelationName(resultRel))));
                        break;
                case RELKIND_FOREIGN_TABLE:
                        /* Okay only if the FDW supports it */
@@ -1098,14 +1119,15 @@ CheckValidRowMarkRel(Relation rel, RowMarkType markType)
                                                        RelationGetRelationName(rel))));
                        break;
                case RELKIND_MATVIEW:
-                       /* Should not get here */
-                       ereport(ERROR,
-                                       (errcode(ERRCODE_WRONG_OBJECT_TYPE),
-                                        errmsg("cannot lock rows in materialized view \"%s\"",
-                                                       RelationGetRelationName(rel))));
+                       /* Allow referencing a matview, but not actual locking clauses */
+                       if (markType != ROW_MARK_REFERENCE)
+                               ereport(ERROR,
+                                               (errcode(ERRCODE_WRONG_OBJECT_TYPE),
+                                          errmsg("cannot lock rows in materialized view \"%s\"",
+                                                         RelationGetRelationName(rel))));
                        break;
                case RELKIND_FOREIGN_TABLE:
-                       /* Should not get here */
+                       /* Should not get here; planner should have used ROW_MARK_COPY */
                        ereport(ERROR,
                                        (errcode(ERRCODE_WRONG_OBJECT_TYPE),
                                         errmsg("cannot lock rows in foreign table \"%s\"",
@@ -1179,7 +1201,7 @@ InitResultRelInfo(ResultRelInfo *resultRelInfo,
  * if so it doesn't matter which one we pick.)  However, it is sometimes
  * necessary to fire triggers on other relations; this happens mainly when an
  * RI update trigger queues additional triggers on other relations, which will
- * be processed in the context of the outer query.     For efficiency's sake,
+ * be processed in the context of the outer query.  For efficiency's sake,
  * we want to have a ResultRelInfo for those triggers too; that can avoid
  * repeated re-opening of the relation.  (It also provides a way for EXPLAIN
  * ANALYZE to report the runtimes of such triggers.)  So we make additional
@@ -1216,7 +1238,7 @@ ExecGetTriggerResultRel(EState *estate, Oid relid)
        /*
         * Open the target relation's relcache entry.  We assume that an
         * appropriate lock is still held by the backend from whenever the trigger
-        * event got queued, so we need take no new lock here.  Also, we need not
+        * event got queued, so we need take no new lock here.  Also, we need not
         * recheck the relkind, so no need for CheckValidResultRel.
         */
        rel = heap_open(relid, NoLock);
@@ -1322,7 +1344,7 @@ ExecPostprocessPlan(EState *estate)
 
        /*
         * Run any secondary ModifyTable nodes to completion, in case the main
-        * query did not fetch all rows from them.      (We do this to ensure that
+        * query did not fetch all rows from them.  (We do this to ensure that
         * such nodes have predictable results.)
         */
        foreach(lc, estate->es_auxmodifytables)
@@ -1583,26 +1605,38 @@ ExecConstraints(ResultRelInfo *resultRelInfo,
                                TupleTableSlot *slot, EState *estate)
 {
        Relation        rel = resultRelInfo->ri_RelationDesc;
-       TupleConstr *constr = rel->rd_att->constr;
+       TupleDesc       tupdesc = RelationGetDescr(rel);
+       TupleConstr *constr = tupdesc->constr;
 
        Assert(constr);
 
        if (constr->has_not_null)
        {
-               int                     natts = rel->rd_att->natts;
+               int                     natts = tupdesc->natts;
                int                     attrChk;
 
                for (attrChk = 1; attrChk <= natts; attrChk++)
                {
-                       if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
+                       if (tupdesc->attrs[attrChk - 1]->attnotnull &&
                                slot_attisnull(slot, attrChk))
+                       {
+                               char       *val_desc;
+                               Bitmapset  *modifiedCols;
+
+                               modifiedCols = GetModifiedColumns(resultRelInfo, estate);
+                               val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
+                                                                                                                slot,
+                                                                                                                tupdesc,
+                                                                                                                modifiedCols,
+                                                                                                                64);
+
                                ereport(ERROR,
                                                (errcode(ERRCODE_NOT_NULL_VIOLATION),
                                                 errmsg("null value in column \"%s\" violates not-null constraint",
-                                                 NameStr(rel->rd_att->attrs[attrChk - 1]->attname)),
-                                                errdetail("Failing row contains %s.",
-                                                                  ExecBuildSlotValueDescription(slot, 64)),
+                                                         NameStr(tupdesc->attrs[attrChk - 1]->attname)),
+                                                val_desc ? errdetail("Failing row contains %s.", val_desc) : 0,
                                                 errtablecol(rel, attrChk)));
+                       }
                }
        }
 
@@ -1611,13 +1645,83 @@ ExecConstraints(ResultRelInfo *resultRelInfo,
                const char *failed;
 
                if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
+               {
+                       char       *val_desc;
+                       Bitmapset  *modifiedCols;
+
+                       modifiedCols = GetModifiedColumns(resultRelInfo, estate);
+                       val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
+                                                                                                        slot,
+                                                                                                        tupdesc,
+                                                                                                        modifiedCols,
+                                                                                                        64);
                        ereport(ERROR,
                                        (errcode(ERRCODE_CHECK_VIOLATION),
                                         errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
                                                        RelationGetRelationName(rel), failed),
-                                        errdetail("Failing row contains %s.",
-                                                          ExecBuildSlotValueDescription(slot, 64)),
+                                        val_desc ? errdetail("Failing row contains %s.", val_desc) : 0,
                                         errtableconstraint(rel, failed)));
+               }
+       }
+}
+
+/*
+ * ExecWithCheckOptions -- check that tuple satisfies any WITH CHECK OPTIONs
+ */
+void
+ExecWithCheckOptions(ResultRelInfo *resultRelInfo,
+                                        TupleTableSlot *slot, EState *estate)
+{
+       Relation        rel = resultRelInfo->ri_RelationDesc;
+       TupleDesc       tupdesc = RelationGetDescr(rel);
+       ExprContext *econtext;
+       ListCell   *l1,
+                          *l2;
+
+       /*
+        * We will use the EState's per-tuple context for evaluating constraint
+        * expressions (creating it if it's not already there).
+        */
+       econtext = GetPerTupleExprContext(estate);
+
+       /* Arrange for econtext's scan tuple to be the tuple under test */
+       econtext->ecxt_scantuple = slot;
+
+       /* Check each of the constraints */
+       forboth(l1, resultRelInfo->ri_WithCheckOptions,
+                       l2, resultRelInfo->ri_WithCheckOptionExprs)
+       {
+               WithCheckOption *wco = (WithCheckOption *) lfirst(l1);
+               ExprState  *wcoExpr = (ExprState *) lfirst(l2);
+
+               /*
+                * WITH CHECK OPTION checks are intended to ensure that the new tuple
+                * is visible (in the case of a view) or that it passes the
+                * 'with-check' policy (in the case of row security).
+                * If the qual evaluates to NULL or FALSE, then the new tuple won't be
+                * included in the view or doesn't pass the 'with-check' policy for the
+                * table.  We need ExecQual to return FALSE for NULL to handle the view
+                * case (the opposite of what we do above for CHECK constraints).
+                */
+               if (!ExecQual((List *) wcoExpr, econtext, false))
+               {
+                       char       *val_desc;
+                       Bitmapset  *modifiedCols;
+
+                       modifiedCols = GetModifiedColumns(resultRelInfo, estate);
+                       val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
+                                                                                                        slot,
+                                                                                                        tupdesc,
+                                                                                                        modifiedCols,
+                                                                                                        64);
+
+                       ereport(ERROR,
+                                       (errcode(ERRCODE_WITH_CHECK_OPTION_VIOLATION),
+                                errmsg("new row violates WITH CHECK OPTION for \"%s\"",
+                                               wco->viewname),
+                                       val_desc ? errdetail("Failing row contains %s.", val_desc) :
+                                                          0));
+               }
        }
 }
 
@@ -1625,58 +1729,149 @@ ExecConstraints(ResultRelInfo *resultRelInfo,
  * ExecBuildSlotValueDescription -- construct a string representing a tuple
  *
  * This is intentionally very similar to BuildIndexValueDescription, but
- * unlike that function, we truncate long field values.  That seems necessary
- * here since heap field values could be very long, whereas index entries
- * typically aren't so wide.
+ * unlike that function, we truncate long field values (to at most maxfieldlen
+ * bytes).  That seems necessary here since heap field values could be very
+ * long, whereas index entries typically aren't so wide.
+ *
+ * Also, unlike the case with index entries, we need to be prepared to ignore
+ * dropped columns.  We used to use the slot's tuple descriptor to decode the
+ * data, but the slot's descriptor doesn't identify dropped columns, so we
+ * now need to be passed the relation's descriptor.
+ *
+ * Note that, like BuildIndexValueDescription, if the user does not have
+ * permission to view any of the columns involved, a NULL is returned.  Unlike
+ * BuildIndexValueDescription, if the user has access to view a subset of the
+ * column involved, that subset will be returned with a key identifying which
+ * columns they are.
  */
 static char *
-ExecBuildSlotValueDescription(TupleTableSlot *slot, int maxfieldlen)
+ExecBuildSlotValueDescription(Oid reloid,
+                                                         TupleTableSlot *slot,
+                                                         TupleDesc tupdesc,
+                                                         Bitmapset *modifiedCols,
+                                                         int maxfieldlen)
 {
        StringInfoData buf;
-       TupleDesc       tupdesc = slot->tts_tupleDescriptor;
+       StringInfoData collist;
+       bool            write_comma = false;
+       bool            write_comma_collist = false;
        int                     i;
+       AclResult       aclresult;
+       bool            table_perm = false;
+       bool            any_perm = false;
 
-       /* Make sure the tuple is fully deconstructed */
-       slot_getallattrs(slot);
+       /*
+        * Check if RLS is enabled and should be active for the relation; if so,
+        * then don't return anything.  Otherwise, go through normal permission
+        * checks.
+        */
+       if (check_enable_rls(reloid, GetUserId(), true) == RLS_ENABLED)
+               return NULL;
 
        initStringInfo(&buf);
 
        appendStringInfoChar(&buf, '(');
 
+       /*
+        * Check if the user has permissions to see the row.  Table-level SELECT
+        * allows access to all columns.  If the user does not have table-level
+        * SELECT then we check each column and include those the user has SELECT
+        * rights on.  Additionally, we always include columns the user provided
+        * data for.
+        */
+       aclresult = pg_class_aclcheck(reloid, GetUserId(), ACL_SELECT);
+       if (aclresult != ACLCHECK_OK)
+       {
+               /* Set up the buffer for the column list */
+               initStringInfo(&collist);
+               appendStringInfoChar(&collist, '(');
+       }
+       else
+               table_perm = any_perm = true;
+
+       /* Make sure the tuple is fully deconstructed */
+       slot_getallattrs(slot);
+
        for (i = 0; i < tupdesc->natts; i++)
        {
+               bool            column_perm = false;
                char       *val;
                int                     vallen;
 
-               if (slot->tts_isnull[i])
-                       val = "null";
-               else
+               /* ignore dropped columns */
+               if (tupdesc->attrs[i]->attisdropped)
+                       continue;
+
+               if (!table_perm)
                {
-                       Oid                     foutoid;
-                       bool            typisvarlena;
+                       /*
+                        * No table-level SELECT, so need to make sure they either have
+                        * SELECT rights on the column or that they have provided the
+                        * data for the column.  If not, omit this column from the error
+                        * message.
+                        */
+                       aclresult = pg_attribute_aclcheck(reloid, tupdesc->attrs[i]->attnum,
+                                                                                         GetUserId(), ACL_SELECT);
+                       if (bms_is_member(tupdesc->attrs[i]->attnum - FirstLowInvalidHeapAttributeNumber,
+                                                         modifiedCols) || aclresult == ACLCHECK_OK)
+                       {
+                               column_perm = any_perm = true;
 
-                       getTypeOutputInfo(tupdesc->attrs[i]->atttypid,
-                                                         &foutoid, &typisvarlena);
-                       val = OidOutputFunctionCall(foutoid, slot->tts_values[i]);
-               }
+                               if (write_comma_collist)
+                                       appendStringInfoString(&collist, ", ");
+                               else
+                                       write_comma_collist = true;
 
-               if (i > 0)
-                       appendStringInfoString(&buf, ", ");
+                               appendStringInfoString(&collist, NameStr(tupdesc->attrs[i]->attname));
+                       }
+               }
 
-               /* truncate if needed */
-               vallen = strlen(val);
-               if (vallen <= maxfieldlen)
-                       appendStringInfoString(&buf, val);
-               else
+               if (table_perm || column_perm)
                {
-                       vallen = pg_mbcliplen(val, vallen, maxfieldlen);
-                       appendBinaryStringInfo(&buf, val, vallen);
-                       appendStringInfoString(&buf, "...");
+                       if (slot->tts_isnull[i])
+                               val = "null";
+                       else
+                       {
+                               Oid                     foutoid;
+                               bool            typisvarlena;
+
+                               getTypeOutputInfo(tupdesc->attrs[i]->atttypid,
+                                                                 &foutoid, &typisvarlena);
+                               val = OidOutputFunctionCall(foutoid, slot->tts_values[i]);
+                       }
+
+                       if (write_comma)
+                               appendStringInfoString(&buf, ", ");
+                       else
+                               write_comma = true;
+
+                       /* truncate if needed */
+                       vallen = strlen(val);
+                       if (vallen <= maxfieldlen)
+                               appendStringInfoString(&buf, val);
+                       else
+                       {
+                               vallen = pg_mbcliplen(val, vallen, maxfieldlen);
+                               appendBinaryStringInfo(&buf, val, vallen);
+                               appendStringInfoString(&buf, "...");
+                       }
                }
        }
 
+       /* If we end up with zero columns being returned, then return NULL. */
+       if (!any_perm)
+               return NULL;
+
        appendStringInfoChar(&buf, ')');
 
+       if (!table_perm)
+       {
+               appendStringInfoString(&collist, ") = ");
+               appendStringInfoString(&collist, buf.data);
+
+               return collist.data;
+       }
+
        return buf.data;
 }
 
@@ -1794,7 +1989,7 @@ EvalPlanQual(EState *estate, EPQState *epqstate,
        /*
         * Get and lock the updated version of the row; if fail, return NULL.
         */
-       copyTuple = EvalPlanQualFetch(estate, relation, lockmode,
+       copyTuple = EvalPlanQualFetch(estate, relation, lockmode, LockWaitBlock,
                                                                  tid, priorXmax);
 
        if (copyTuple == NULL)
@@ -1807,7 +2002,7 @@ EvalPlanQual(EState *estate, EPQState *epqstate,
        *tid = copyTuple->t_self;
 
        /*
-        * Need to run a recheck subquery.      Initialize or reinitialize EPQ state.
+        * Need to run a recheck subquery.  Initialize or reinitialize EPQ state.
         */
        EvalPlanQualBegin(epqstate, estate);
 
@@ -1853,11 +2048,15 @@ EvalPlanQual(EState *estate, EPQState *epqstate,
  *     estate - executor state data
  *     relation - table containing tuple
  *     lockmode - requested tuple lock mode
+ *     wait_policy - requested lock wait policy
  *     *tid - t_ctid from the outdated tuple (ie, next updated version)
  *     priorXmax - t_xmax from the outdated tuple
  *
  * Returns a palloc'd copy of the newest tuple version, or NULL if we find
  * that there is no newest version (ie, the row was deleted not updated).
+ * We also return NULL if the tuple is locked and the wait policy is to skip
+ * such tuples.
+ *
  * If successful, we have locked the newest tuple version, so caller does not
  * need to worry about it changing anymore.
  *
@@ -1866,6 +2065,7 @@ EvalPlanQual(EState *estate, EPQState *epqstate,
  */
 HeapTuple
 EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
+                                 LockWaitPolicy wait_policy,
                                  ItemPointer tid, TransactionId priorXmax)
 {
        HeapTuple       copyTuple = NULL;
@@ -1890,7 +2090,7 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
 
                        /*
                         * If xmin isn't what we're expecting, the slot must have been
-                        * recycled and reused for an unrelated tuple.  This implies that
+                        * recycled and reused for an unrelated tuple.  This implies that
                         * the latest version of the row was deleted, so we need do
                         * nothing.  (Should be safe to examine xmin without getting
                         * buffer's content lock, since xmin never changes in an existing
@@ -1909,12 +2109,30 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
 
                        /*
                         * If tuple is being updated by other transaction then we have to
-                        * wait for its commit/abort.
+                        * wait for its commit/abort, or die trying.
                         */
                        if (TransactionIdIsValid(SnapshotDirty.xmax))
                        {
                                ReleaseBuffer(buffer);
-                               XactLockTableWait(SnapshotDirty.xmax);
+                               switch (wait_policy)
+                               {
+                                       case LockWaitBlock:
+                                               XactLockTableWait(SnapshotDirty.xmax,
+                                                                                 relation, &tuple.t_self,
+                                                                                 XLTW_FetchUpdated);
+                                               break;
+                                       case LockWaitSkip:
+                                               if (!ConditionalXactLockTableWait(SnapshotDirty.xmax))
+                                                       return NULL; /* skip instead of waiting */
+                                               break;
+                                       case LockWaitError:
+                                               if (!ConditionalXactLockTableWait(SnapshotDirty.xmax))
+                                                       ereport(ERROR,
+                                                                       (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
+                                                                        errmsg("could not obtain lock on row in relation \"%s\"",
+                                                                                       RelationGetRelationName(relation))));
+                                               break;
+                               }
                                continue;               /* loop back to repeat heap_fetch */
                        }
 
@@ -1925,7 +2143,7 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
                         * heap_lock_tuple() will throw an error, and so would any later
                         * attempt to update or delete the tuple.  (We need not check cmax
                         * because HeapTupleSatisfiesDirty will consider a tuple deleted
-                        * by our transaction dead, regardless of cmax.) Wee just checked
+                        * by our transaction dead, regardless of cmax.) We just checked
                         * that priorXmax == xmin, so we can test that variable instead of
                         * doing HeapTupleHeaderGetXmin again.
                         */
@@ -1941,7 +2159,7 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
                         */
                        test = heap_lock_tuple(relation, &tuple,
                                                                   estate->es_output_cid,
-                                                                  lockmode, false /* wait */ ,
+                                                                  lockmode, wait_policy,
                                                                   false, &buffer, &hufd);
                        /* We now have two pins on the buffer, get rid of one */
                        ReleaseBuffer(buffer);
@@ -1987,6 +2205,10 @@ EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
                                        /* tuple was deleted, so give up */
                                        return NULL;
 
+                               case HeapTupleWouldBlock:
+                                       ReleaseBuffer(buffer);
+                                       return NULL;
+
                                default:
                                        ReleaseBuffer(buffer);
                                        elog(ERROR, "unrecognized heap_lock_tuple status: %u",
@@ -2079,7 +2301,7 @@ EvalPlanQualInit(EPQState *epqstate, EState *estate,
 /*
  * EvalPlanQualSetPlan -- set or change subplan of an EPQState.
  *
- * We need this so that ModifyTuple can deal with multiple subplans.
+ * We need this so that ModifyTable can deal with multiple subplans.
  */
 void
 EvalPlanQualSetPlan(EPQState *epqstate, Plan *subplan, List *auxrowmarks)
@@ -2129,7 +2351,7 @@ EvalPlanQualGetTuple(EPQState *epqstate, Index rti)
 
 /*
  * Fetch the current row values for any non-locked relations that need
- * to be scanned by an EvalPlanQual operation. origslot must have been set
+ * to be scanned by an EvalPlanQual operation.  origslot must have been set
  * to contain the current result row (top-level row) that we need to recheck.
  */
 void
@@ -2216,7 +2438,9 @@ EvalPlanQualFetchRowMarks(EPQState *epqstate)
                        /* build a temporary HeapTuple control structure */
                        tuple.t_len = HeapTupleHeaderGetDatumLength(td);
                        ItemPointerSetInvalid(&(tuple.t_self));
-                       tuple.t_tableOid = InvalidOid;
+                       /* relation might be a foreign table, if so provide tableoid */
+                       tuple.t_tableOid = getrelid(erm->rti,
+                                                                               epqstate->estate->es_range_table);
                        tuple.t_data = td;
 
                        /* copy and store tuple */
@@ -2316,6 +2540,14 @@ EvalPlanQualStart(EPQState *epqstate, EState *parentestate, Plan *planTree)
         * the snapshot, rangetable, result-rel info, and external Param info.
         * They need their own copies of local state, including a tuple table,
         * es_param_exec_vals, etc.
+        *
+        * The ResultRelInfo array management is trickier than it looks.  We
+        * create a fresh array for the child but copy all the content from the
+        * parent.  This is because it's okay for the child to share any
+        * per-relation state the parent has already created --- but if the child
+        * sets up any ResultRelInfo fields, such as its own junkfilter, that
+        * state must *not* propagate back to the parent.  (For one thing, the
+        * pointed-to data is in a memory context that won't last long enough.)
         */
        estate->es_direction = ForwardScanDirection;
        estate->es_snapshot = parentestate->es_snapshot;
@@ -2324,9 +2556,19 @@ EvalPlanQualStart(EPQState *epqstate, EState *parentestate, Plan *planTree)
        estate->es_plannedstmt = parentestate->es_plannedstmt;
        estate->es_junkFilter = parentestate->es_junkFilter;
        estate->es_output_cid = parentestate->es_output_cid;
-       estate->es_result_relations = parentestate->es_result_relations;
-       estate->es_num_result_relations = parentestate->es_num_result_relations;
-       estate->es_result_relation_info = parentestate->es_result_relation_info;
+       if (parentestate->es_num_result_relations > 0)
+       {
+               int                     numResultRelations = parentestate->es_num_result_relations;
+               ResultRelInfo *resultRelInfos;
+
+               resultRelInfos = (ResultRelInfo *)
+                       palloc(numResultRelations * sizeof(ResultRelInfo));
+               memcpy(resultRelInfos, parentestate->es_result_relations,
+                          numResultRelations * sizeof(ResultRelInfo));
+               estate->es_result_relations = resultRelInfos;
+               estate->es_num_result_relations = numResultRelations;
+       }
+       /* es_result_relation_info must NOT be copied */
        /* es_trig_target_relations must NOT be copied */
        estate->es_rowMarks = parentestate->es_rowMarks;
        estate->es_top_eflags = parentestate->es_top_eflags;
@@ -2358,7 +2600,7 @@ EvalPlanQualStart(EPQState *epqstate, EState *parentestate, Plan *planTree)
 
        /*
         * Each EState must have its own es_epqScanDone state, but if we have
-        * nested EPQ checks they should share es_epqTuple arrays.      This allows
+        * nested EPQ checks they should share es_epqTuple arrays.  This allows
         * sub-rechecks to inherit the values being examined by an outer recheck.
         */
        estate->es_epqScanDone = (bool *) palloc0(rtsize * sizeof(bool));
@@ -2415,7 +2657,7 @@ EvalPlanQualStart(EPQState *epqstate, EState *parentestate, Plan *planTree)
  *
  * This is a cut-down version of ExecutorEnd(); basically we want to do most
  * of the normal cleanup, but *not* close result relations (which we are
- * just sharing from the outer query). We do, however, have to close any
+ * just sharing from the outer query).  We do, however, have to close any
  * trigger target relations that got opened, since those are not shared.
  * (There probably shouldn't be any of the latter, but just in case...)
  */