amcheck: Fix verify_heapam for tuples where xmin or xmax is 0.
authorRobert Haas <rhaas@postgresql.org>
Thu, 23 Mar 2023 19:29:28 +0000 (15:29 -0400)
committerRobert Haas <rhaas@postgresql.org>
Fri, 24 Mar 2023 14:56:10 +0000 (10:56 -0400)
In such cases, get_xid_status() doesn't set its output parameter (the
third argument), so we shouldn't fall through to code which will test
the value of that parameter. There are five existing calls to
get_xid_status(), three of which seem to already handle this case
properly.  This commit tries to fix the other two.

If we're checking xmin and find that it is invalid (i.e. 0) just
report that as corruption, similar to what's already done in the
three cases that seem correct. If we're checking xmax and find
that's invalid, that's fine: it just means that the tuple hasn't
been updated or deleted.

Thanks to Andres Freund and valgrind for finding this problem, and
also to Andres for having a look at the patch.  This bug seems to go
all the way back to where verify_heapam was first introduced, but
wasn't detected until recently, possibly because of the new test cases
added for update chain verification.  Back-patch to v14, where this
code showed up.

Discussion: http://postgr.es/m/CA+TgmoZAYzQZqyUparXy_ks3OEOfLD9-bEXt8N-2tS1qghX9gQ@mail.gmail.com

contrib/amcheck/verify_heapam.c

index 1b8607c6cc8925ea65fdc1c2bd975bc7c74c7933..10dd31477b0afd5b21874b7b891f79422bea3334 100644 (file)
@@ -1012,7 +1012,9 @@ check_tuple_visibility(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
    switch (get_xid_status(xmin, ctx, &xmin_status))
    {
        case XID_INVALID:
-           break;
+           report_corruption(ctx,
+                             pstrdup("xmin is invalid"));
+           return false;
        case XID_BOUNDS_OK:
            *xmin_commit_status_ok = true;
            *xmin_commit_status = xmin_status;
@@ -1350,6 +1352,9 @@ check_tuple_visibility(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
    xmax = HeapTupleHeaderGetRawXmax(tuphdr);
    switch (get_xid_status(xmax, ctx, &xmax_status))
    {
+       case XID_INVALID:
+           ctx->tuple_could_be_pruned = false;
+           return true;
        case XID_IN_FUTURE:
            report_corruption(ctx,
                              psprintf("xmax %u equals or exceeds next valid transaction ID %u:%u",
@@ -1372,7 +1377,6 @@ check_tuple_visibility(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
                                       XidFromFullTransactionId(ctx->oldest_fxid)));
            return false;       /* corrupt */
        case XID_BOUNDS_OK:
-       case XID_INVALID:
            break;
    }