systemd/0736-test-temporarily-disable-coredumps-in-testsuite-17.0.patch

From 54c44b19c1018400c38da8f8be597536d14e7afa Mon Sep 17 00:00:00 2001
From: Frantisek Sumsal <frantisek@sumsal.cz>
Date: Thu, 18 Apr 2024 22:39:31 +0200
Subject: [PATCH] test: temporarily disable coredumps in testsuite-17.03.sh

Since f387005b54 we started generating coredumps by default (up to
certain size). This change has one unintentional effect on our test
suite - if a sanitized binary (udevd worker here) crashes and ASan is
instructed to allow core dumping (via disable_coredump=0 and
use_madv_dontdump=1), we try to dump a relatively big core file (~80
MiB), and since the test suite configures systemd-coredumpd to dump the
cores into the journal, we try to append it to the journal message about
the crash. However, journal complains that the message with the coredump
is too big so the crash report is not written, and we end up with
coredumpctl not showing the crash, which the test in this case uses to
monitor if the udevd worker's job timed out:

[   17.873463] systemd-udevd[1617]: null: Worker [1625] processing SEQNUM=3588 is taking a long time
[   17.876823] systemd-udevd[1625]: null: Spawned process '/bin/sleep 60' [1626] is taking longer than 3s to complete
...
[   24.223459] systemd-udevd[1617]: null: Worker [1625] processing SEQNUM=3588 killed
[   24.265141] systemd[1]: Created slice system-systemd\x2dcoredump.slice.
[   24.284960] systemd[1]: Started systemd-coredump@0-1707-0.service.
[   27.545120] systemd-journald[1225]: Failed to write entry to /var/log/journal/6da99a97048e4f08abd4ddabcf92bbdd/system.journal (51 items, 89252196 bytes) despite vacuuming, ignoring: Argument list too long
[   27.551759] systemd-coredump[1709]: ==1709==LeakSanitizer has encountered a fatal error.
[   27.551759] systemd-coredump[1709]: ==1709==HINT: For debugging, try setting environment variable LSAN_OPTIONS=verbosity=1:log_threads=1
[   27.551759] systemd-coredump[1709]: ==1709==HINT: LeakSanitizer does not work under ptrace (strace, gdb, etc)

The upstream version of this test doesn't suffer from this because it
was recently-ish rewritten to not check for the crash event
(5592608bdcb, but we're missing some udevd prerequisites for that to
work in RHEL9), and we also started instructing ASan to allow coredumps
after that change, so the issue was never encountered there in the first
place.

Since we don't really care about the actual coredump in this case, let's
just temporarily override the core rlimit to 0 for the udevd process.

Related: RHEL-30372
rhel-only
---
 test/units/testsuite-17.03.sh | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/test/units/testsuite-17.03.sh b/test/units/testsuite-17.03.sh
index 318afdcb5a..8fa58e2f08 100755
--- a/test/units/testsuite-17.03.sh
+++ b/test/units/testsuite-17.03.sh
@@ -17,9 +17,16 @@ event_timeout=10
 timeout_signal=SIGABRT
 EOF
 
+    mkdir -p /run/systemd/system/systemd-udevd.service.d/
+    cat >/run/systemd/system/systemd-udevd.service.d/99-disable-coredumps.conf <<EOF
+[Service]
+LimitCORE=0
+EOF
+    systemctl daemon-reload
     systemctl restart systemd-udevd.service
 }
 
+# shellcheck disable=SC2317
 teardown() {
     set +e
 
@@ -27,10 +34,11 @@ teardown() {
         kill "$KILL_PID"
     fi
 
-    rm -rf "$TMPDIR"
+    rm -rf "$TMPDIR" /run/systemd/system/systemd-udevd.service.d
 
     mv -f /etc/udev/udev.conf.bckp /etc/udev/udev.conf
     rm -f "$test_rule"
+    systemctl daemon-reload
     systemctl restart systemd-udevd.service
 }