Bug 2004825

Created attachment 1823531 [details]
The possibly intended patch

Description of problem:

As reported in the corresponding Bodhi updates:

F36: https://bodhi.fedoraproject.org/updates/FEDORA-2021-d6e603fa20
F35: https://bodhi.fedoraproject.org/updates/FEDORA-2021-f0e003a376
F34: https://bodhi.fedoraproject.org/updates/FEDORA-2021-29ab43c382

these updates break at least two other packages (freeipa, nordugrid-arc).

Due to the negative karma, the F35 and F34 updates will (hopefully) not be pushed to stable. But the F36/rawhide update is causing problems.

The libdb update in rawhide breaks the build of nordugrid-arc, because it causes one of its tests to fail on the ppc64le koji builders. This failure has been confirmed on the ppc64le-test.fedorainfracloud.org test machine.

Here is a link to a scratch build:
https://koji.fedoraproject.org/koji/taskinfo?taskID=75744622

The nordugrid-arc package needs to be rebuilt in rawhide due to the OpenSSL 3.0 migration. So this problem is blocking that.

Version-Release number of selected component (if applicable):

libdb-5.3.28-49.fc36
libdb-5.3.28-49.fc35
libdb-5.3.28-48.fc34

How reproducible:

Always

Steps to Reproduce:
1. Build nordugrid-arc on rowhide for ppc64le

Actual results:
Build failure due to libdb related test failure.

Expected results:
Successful build

Additional info:

This is copied from the comment on the F35 bodhi update:

The descrition for this update is "improve previous patch to cover more cases".

The previouss patch was:

diff -ur db-5.3.28/src/os/os_map.c db-patched/src/os/os_map.c
--- db-5.3.28/src/os/os_map.c   2013-09-09 17:35:09.000000000 +0200
+++ db-patched/src/os/os_map.c  2021-09-06 14:32:28.792139908 +0200
@@ -213,6 +213,10 @@
    if (rp->max < rp->size)
        rp->max = rp->size;
    if (ret == 0 && F_ISSET(infop, REGION_CREATE)) {
+#ifdef HAVE_MLOCK
+       if (F_ISSET(env, ENV_LOCKDOWN))
+           rp->size = rp->max;
+#endif
        if (F_ISSET(dbenv, DB_ENV_REGION_INIT))
            ret = __db_file_write(env, infop->fhp,
                rp->size / MEGABYTE, rp->size % MEGABYTE, 0x00);

The new patch is (after cleaning up unintentional indentation mistakes that makes it longer than necessary):

diff -ur db-5.3.28/src/os/os_map.c db_patch/src/os/os_map.c
--- db-5.3.28/src/os/os_map.c   2013-09-09 17:35:09.000000000 +0200
+++ db_patch/src/os/os_map.c    2021-09-09 07:33:12.027328265 +0200
@@ -213,6 +213,9 @@
    if (rp->max < rp->size)
        rp->max = rp->size;
    if (ret == 0 && F_ISSET(infop, REGION_CREATE)) {
+
+       rp->max = rp->size;
+
        if (F_ISSET(dbenv, DB_ENV_REGION_INIT))
            ret = __db_file_write(env, infop->fhp,
                rp->size / MEGABYTE, rp->size % MEGABYTE, 0x00);

This removes the condition, which is consistent with "to cover more cases". However, it also swaps the assignment from rp->size = rp->max; to rp->max = rp->size;, which was probably unintentional.

Was the patch intended to be:

diff -ur db-5.3.28/src/os/os_map.c db_patch/src/os/os_map.c
--- db-5.3.28/src/os/os_map.c   2013-09-09 17:35:09.000000000 +0200
+++ db_patch/src/os/os_map.c    2021-09-09 07:33:12.027328265 +0200
@@ -213,6 +213,9 @@
    if (rp->max < rp->size)
        rp->max = rp->size;
    if (ret == 0 && F_ISSET(infop, REGION_CREATE)) {
+
+       rp->size = rp->max;
+
        if (F_ISSET(dbenv, DB_ENV_REGION_INIT))
            ret = __db_file_write(env, infop->fhp,
                rp->size / MEGABYTE, rp->size % MEGABYTE, 0x00);


This version of the patch is attached to this bugzilla ticket.