test: Dump coredumps from journal in the integration test wrapper

Fixes #35277

.github/workflows/linter.yml

@@ -37,7 +37,7 @@ jobs:
           VALIDATE_GITHUB_ACTIONS: true
 
       - name: Check that tabs are not used in Python code
-        run: sh -c '! git grep -P "\\t" -- src/ukify/ukify.py'
+        run: sh -c '! git grep -P "\\t" -- src/ukify/ukify.py test/integration-test-wrapper.py'
 
       - name: Install ruff and mypy
         run: |
@@ -47,14 +47,14 @@ jobs:
       - name: Run mypy
         run: |
           python3 -m mypy --version
-          python3 -m mypy src/ukify/ukify.py
+          python3 -m mypy src/ukify/ukify.py test/integration-test-wrapper.py
 
       - name: Run ruff check
         run: |
           ruff --version
-          ruff check src/ukify/ukify.py
+          ruff check src/ukify/ukify.py test/integration-test-wrapper.py
 
       - name: Run ruff format
         run: |
           ruff --version
-          ruff format --check src/ukify/ukify.py
+          ruff format --check src/ukify/ukify.py test/integration-test-wrapper.py

man/coredumpctl.xml

@@ -391,7 +391,7 @@
     <title>Exit status</title>
     <para>On success, 0 is returned; otherwise, a non-zero failure
     code is returned. Not finding any matching core dumps is treated as
-    failure.
+    failure unless JSON output is enabled.
     </para>
   </refsect1>
 

mkosi.images/build/mkosi.conf.d/centos-fedora/mkosi.conf

@@ -13,6 +13,7 @@ Environment=
 
 [Content]
 Packages=
+        clangd
         compiler-rt
         gdb
         git-core

mkosi.images/build/mkosi.conf.d/debian-ubuntu/mkosi.conf

@@ -15,6 +15,7 @@ Environment=
 [Content]
 Packages=
         apt
+        clangd
         erofs-utils
         git-core
         libclang-rt-dev

mkosi.images/build/mkosi.conf.d/opensuse/mkosi.conf

@@ -12,6 +12,7 @@ Environment=
 
 [Content]
 Packages=
+        clang
         diffutils
         erofs-utils
         gcc-c++

mkosi.sanitizers/mkosi.postinst

@@ -57,6 +57,8 @@ wrap=(
     delv
     dhcpd
     dig
+    dnf
+    dnf5
     dmsetup
     dnsmasq
     findmnt

src/coredump/coredumpctl.c

@@ -965,6 +965,8 @@ static int dump_list(int argc, char **argv, void *userdata) {
                 if (!arg_field && n_found <= 0) {
                         if (!arg_quiet)
                                 log_notice("No coredumps found.");
+
+                        if (!sd_json_format_enabled(arg_json_format_flags))
                                 return -ESRCH;
                 }
         }

src/ukify/ukify.py

@@ -467,7 +467,7 @@ class SignTool:
         raise NotImplementedError()
 
     @staticmethod
-    def from_string(name) -> type['SignTool']:
+    def from_string(name: str) -> type['SignTool']:
         if name == 'pesign':
             return PeSign
         elif name == 'sbsign':

test/TEST-02-UNITTESTS/meson.build

@@ -3,6 +3,7 @@
 integration_tests += [
         integration_test_template + {
                 'name' : fs.name(meson.current_source_dir()),
+                'coredump-exclude-regex' : '/(bash|python3.[0-9]+|systemd-executor)$',
                 'cmdline' : integration_test_template['cmdline'] + [
                         '''
 

test/TEST-16-EXTEND-TIMEOUT/meson.build

@@ -4,7 +4,7 @@ integration_tests += [
         integration_test_template + {
                 'name' : fs.name(meson.current_source_dir()),
                 'unit' : files('TEST-16-EXTEND-TIMEOUT.service'),
-                'coredump-exclude-regex' : '/(bash|sleep),
+                'coredump-exclude-regex' : '/(bash|sleep)$',
         },
 ]
 

test/TEST-17-UDEV/meson.build

@@ -4,5 +4,6 @@ integration_tests += [
         integration_test_template + {
                 'name' : fs.name(meson.current_source_dir()),
                 'vm' : true,
+                'coredump-exclude-regex' : '/(sleep|udevadm)$',
         },
 ]

test/TEST-59-RELOADING-RESTART/meson.build

@@ -3,5 +3,6 @@
 integration_tests += [
         integration_test_template + {
                 'name' : fs.name(meson.current_source_dir()),
+                'coredump-exclude-regex' : '/(sleep|bash|systemd-notify)$',
         },
 ]

test/TEST-74-AUX-UTILS/meson.build

@@ -5,6 +5,7 @@ integration_tests += [
                 'name' : fs.name(meson.current_source_dir()),
                 'storage': 'persistent',
                 'vm' : true,
+                'coredump-exclude-regex' : '/(test-usr-dump|test-dump|bash)$',
         },
 ]
 

test/integration-test-wrapper.py

@@ -1,19 +1,18 @@
 #!/usr/bin/python3
 # SPDX-License-Identifier: LGPL-2.1-or-later
 
-'''Test wrapper command for driving integration tests.
+"""Test wrapper command for driving integration tests."""
-'''
 
 import argparse
 import json
 import os
+import re
 import shlex
 import subprocess
 import sys
 import textwrap
 from pathlib import Path
 
-
 EMERGENCY_EXIT_DROPIN = """\
 [Unit]
 Wants=emergency-exit.service
@@ -34,7 +33,61 @@ ExecStart=false
 """
 
 
-def main():
+def dump_coredumps(args: argparse.Namespace, journal_file: Path) -> bool:
+    # Collect executable paths of all coredumps and filter out the expected ones.
+    # The following are excluded:
+    #   sleep/bash - intentional SIGABRT caused by TEST-57
+    #   systemd-notify - intermittent (and intentional) SIGABRT caused by TEST-59
+    #   test-execute - intentional coredump in TEST-02
+    #   test(-usr)?-dump - intentional coredumps from systemd-coredump tests in TEST-74
+    if args.coredump_exclude_regex:
+        exclude_regex = re.compile(args.coredump_exclude_regex)
+    else:
+        exclude_regex = None
+
+    coredumps = json.loads(
+        subprocess.run(
+            [
+                args.mkosi,
+                '--directory', os.fspath(args.meson_source_dir),
+                '--extra-search-path', os.fspath(args.meson_build_dir),
+                'sandbox',
+                'coredumpctl',
+                '--file', journal_file,
+                '--json=short',
+            ],
+            check=True,
+            stdout=subprocess.PIPE,
+            text=True,
+        ).stdout
+    )  # fmt: skip
+
+    coredumps = [
+        coredump for coredump in coredumps if not exclude_regex or not exclude_regex.search(coredump['exe'])
+    ]
+
+    if not coredumps:
+        return False
+
+    subprocess.run(
+        [
+            args.mkosi,
+            '--directory', os.fspath(args.meson_source_dir),
+            '--extra-search-path', os.fspath(args.meson_build_dir),
+            'sandbox',
+            'coredumpctl',
+            '--file', journal_file,
+            '--no-pager',
+            'info',
+            *(coredump['exe'] for coredump in coredumps),
+        ],
+        check=True,
+    )  # fmt: skip
+
+    return True
+
+
+def main() -> None:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument('--mkosi', required=True)
     parser.add_argument('--meson-source-dir', required=True, type=Path)
@@ -46,34 +99,44 @@ def main():
     parser.add_argument('--slow', action=argparse.BooleanOptionalAction)
     parser.add_argument('--vm', action=argparse.BooleanOptionalAction)
     parser.add_argument('--exit-code', required=True, type=int)
-    parser.add_argument('mkosi_args', nargs="*")
+    parser.add_argument('--coredump-exclude-regex', required=True)
+    parser.add_argument('mkosi_args', nargs='*')
     args = parser.parse_args()
 
-    if not bool(int(os.getenv("SYSTEMD_INTEGRATION_TESTS", "0"))):
+    if not bool(int(os.getenv('SYSTEMD_INTEGRATION_TESTS', '0'))):
-        print(f"SYSTEMD_INTEGRATION_TESTS=1 not found in environment, skipping {args.name}", file=sys.stderr)
+        print(
+            f'SYSTEMD_INTEGRATION_TESTS=1 not found in environment, skipping {args.name}',
+            file=sys.stderr,
+        )
         exit(77)
 
-    if args.slow and not bool(int(os.getenv("SYSTEMD_SLOW_TESTS", "0"))):
+    if args.slow and not bool(int(os.getenv('SYSTEMD_SLOW_TESTS', '0'))):
-        print(f"SYSTEMD_SLOW_TESTS=1 not found in environment, skipping {args.name}", file=sys.stderr)
+        print(
+            f'SYSTEMD_SLOW_TESTS=1 not found in environment, skipping {args.name}',
+            file=sys.stderr,
+        )
         exit(77)
 
-    if args.vm and bool(int(os.getenv("TEST_NO_QEMU", "0"))):
+    if args.vm and bool(int(os.getenv('TEST_NO_QEMU', '0'))):
-        print(f"TEST_NO_QEMU=1, skipping {args.name}", file=sys.stderr)
+        print(f'TEST_NO_QEMU=1, skipping {args.name}', file=sys.stderr)
         exit(77)
 
-    for s in os.getenv("TEST_SKIP", "").split():
+    for s in os.getenv('TEST_SKIP', '').split():
         if s in args.name:
-            print(f"Skipping {args.name} due to TEST_SKIP", file=sys.stderr)
+            print(f'Skipping {args.name} due to TEST_SKIP', file=sys.stderr)
             exit(77)
 
-    keep_journal = os.getenv("TEST_SAVE_JOURNAL", "fail")
+    keep_journal = os.getenv('TEST_SAVE_JOURNAL', 'fail')
-    shell = bool(int(os.getenv("TEST_SHELL", "0")))
+    shell = bool(int(os.getenv('TEST_SHELL', '0')))
 
     if shell and not sys.stderr.isatty():
-        print(f"--interactive must be passed to meson test to use TEST_SHELL=1", file=sys.stderr)
+        print(
+            '--interactive must be passed to meson test to use TEST_SHELL=1',
+            file=sys.stderr,
+        )
         exit(1)
 
-    name = args.name + (f"-{i}" if (i := os.getenv("MESON_TEST_ITERATION")) else "")
+    name = args.name + (f'-{i}' if (i := os.getenv('MESON_TEST_ITERATION')) else '')
 
     dropin = textwrap.dedent(
         """\
@@ -84,14 +147,14 @@ def main():
 
     if not shell:
         dropin += textwrap.dedent(
-            f"""
+            """
             [Unit]
             SuccessAction=exit
             SuccessActionExitStatus=123
             """
         )
 
-    if os.getenv("TEST_MATCH_SUBTEST"):
+    if os.getenv('TEST_MATCH_SUBTEST'):
         dropin += textwrap.dedent(
             f"""
             [Service]
@@ -99,7 +162,7 @@ def main():
             """
         )
 
-    if os.getenv("TEST_MATCH_TESTCASE"):
+    if os.getenv('TEST_MATCH_TESTCASE'):
         dropin += textwrap.dedent(
             f"""
             [Service]
@@ -107,7 +170,9 @@ def main():
             """
         )
 
-    journal_file = None
+    journal_file = (args.meson_build_dir / (f'test/journal/{name}.journal')).absolute()
+    journal_file.unlink(missing_ok=True)
+
     if not sys.stderr.isatty():
         dropin += textwrap.dedent(
             """
@@ -115,9 +180,6 @@ def main():
             FailureAction=exit
             """
         )
-
-        journal_file = (args.meson_build_dir / (f"test/journal/{name}.journal")).absolute()
-        journal_file.unlink(missing_ok=True)
     elif not shell:
         dropin += textwrap.dedent(
             """
@@ -136,87 +198,93 @@ def main():
         *(['--forward-journal', journal_file] if journal_file else []),
         *(
             [
-                '--credential',
+                '--credential', f'systemd.extra-unit.emergency-exit.service={shlex.quote(EMERGENCY_EXIT_SERVICE)}',  # noqa: E501
-                f"systemd.extra-unit.emergency-exit.service={shlex.quote(EMERGENCY_EXIT_SERVICE)}",
+                '--credential', f'systemd.unit-dropin.emergency.target={shlex.quote(EMERGENCY_EXIT_DROPIN)}',
-                '--credential',
-                f"systemd.unit-dropin.emergency.target={shlex.quote(EMERGENCY_EXIT_DROPIN)}",
             ]
             if not sys.stderr.isatty()
             else []
         ),
-        '--credential',
+        '--credential', f'systemd.unit-dropin.{args.unit}={shlex.quote(dropin)}',
-        f"systemd.unit-dropin.{args.unit}={shlex.quote(dropin)}",
         '--runtime-network=none',
         '--runtime-scratch=no',
         *args.mkosi_args,
-        '--qemu-firmware', args.firmware,
+        '--qemu-firmware',
-        *(['--qemu-kvm', 'no'] if int(os.getenv("TEST_NO_KVM", "0")) else []),
+        args.firmware,
+        *(['--qemu-kvm', 'no'] if int(os.getenv('TEST_NO_KVM', '0')) else []),
         '--kernel-command-line-extra',
-        ' '.join([
+        ' '.join(
+            [
                 'systemd.hostname=H',
-            f"SYSTEMD_UNIT_PATH=/usr/lib/systemd/tests/testdata/{args.name}.units:/usr/lib/systemd/tests/testdata/units:",
+                f'SYSTEMD_UNIT_PATH=/usr/lib/systemd/tests/testdata/{args.name}.units:/usr/lib/systemd/tests/testdata/units:',
-            *([f"systemd.unit={args.unit}"] if not shell else []),
+                *([f'systemd.unit={args.unit}'] if not shell else []),
                 'systemd.mask=systemd-networkd-wait-online.service',
                 *(
                     [
-                    "systemd.mask=serial-getty@.service",
+                        'systemd.mask=serial-getty@.service',
-                    "systemd.show_status=error",
+                        'systemd.show_status=error',
-                    "systemd.crash_shell=0",
+                        'systemd.crash_shell=0',
-                    "systemd.crash_action=poweroff",
+                        'systemd.crash_action=poweroff',
                     ]
                     if not sys.stderr.isatty()
                     else []
                 ),
-        ]),
+            ]
+        ),
         '--credential', f"journal.storage={'persistent' if sys.stderr.isatty() else args.storage}",
         *(['--runtime-build-sources=no'] if not sys.stderr.isatty() else []),
         'qemu' if args.vm or os.getuid() != 0 else 'boot',
-    ]
+    ]  # fmt: skip
 
     result = subprocess.run(cmd)
 
     # On Debian/Ubuntu we get a lot of random QEMU crashes. Retry once, and then skip if it fails again.
     if args.vm and result.returncode == 247 and args.exit_code != 247:
+        if journal_file:
             journal_file.unlink(missing_ok=True)
         result = subprocess.run(cmd)
         if args.vm and result.returncode == 247 and args.exit_code != 247:
-            print(f"Test {args.name} failed due to QEMU crash (error 247), ignoring", file=sys.stderr)
+            print(
+                f'Test {args.name} failed due to QEMU crash (error 247), ignoring',
+                file=sys.stderr,
+            )
             exit(77)
 
-    if journal_file and (keep_journal == "0" or (result.returncode in (args.exit_code, 77) and keep_journal == "fail")):
+    coredumps = dump_coredumps(args, journal_file)
+
+    if keep_journal == '0' or (
+        keep_journal == 'fail' and result.returncode in (args.exit_code, 77) and not coredumps
+    ):
         journal_file.unlink(missing_ok=True)
 
-    if shell or result.returncode in (args.exit_code, 77):
+    if shell or (result.returncode in (args.exit_code, 77) and not coredumps):
         exit(0 if shell or result.returncode == args.exit_code else 77)
 
-    if journal_file:
     ops = []
 
-        if os.getenv("GITHUB_ACTIONS"):
+    if os.getenv('GITHUB_ACTIONS'):
-            id = os.environ["GITHUB_RUN_ID"]
+        id = os.environ['GITHUB_RUN_ID']
-            iteration = os.environ["GITHUB_RUN_ATTEMPT"]
+        iteration = os.environ['GITHUB_RUN_ATTEMPT']
         j = json.loads(
             subprocess.run(
                 [
                     args.mkosi,
-                        "--directory", os.fspath(args.meson_source_dir),
+                    '--directory', os.fspath(args.meson_source_dir),
-                        "--json",
+                    '--json',
-                        "summary",
+                    'summary',
                 ],
                 stdout=subprocess.PIPE,
                 text=True,
             ).stdout
-            )
+        )  # fmt: skip
-            distribution = j["Images"][-1]["Distribution"]
+        distribution = j['Images'][-1]['Distribution']
-            release = j["Images"][-1]["Release"]
+        release = j['Images'][-1]['Release']
-            artifact = f"ci-mkosi-{id}-{iteration}-{distribution}-{release}-failed-test-journals"
+        artifact = f'ci-mkosi-{id}-{iteration}-{distribution}-{release}-failed-test-journals'
-            ops += [f"gh run download {id} --name {artifact} -D ci/{artifact}"]
+        ops += [f'gh run download {id} --name {artifact} -D ci/{artifact}']
-            journal_file = Path(f"ci/{artifact}/test/journal/{name}.journal")
+        journal_file = Path(f'ci/{artifact}/test/journal/{name}.journal')
 
-        ops += [f"journalctl --file {journal_file} --no-hostname -o short-monotonic -u {args.unit} -p info"]
+    ops += [f'journalctl --file {journal_file} --no-hostname -o short-monotonic -u {args.unit} -p info']
 
-        print("Test failed, relevant logs can be viewed with: \n\n"
+    print("Test failed, relevant logs can be viewed with: \n\n" f"{(' && '.join(ops))}\n", file=sys.stderr)
-              f"{(' && '.join(ops))}\n", file=sys.stderr)
 
     # 0 also means we failed so translate that to a non-zero exit code to mark the test as failed.
     exit(result.returncode or 1)

test/meson.build

@@ -297,6 +297,7 @@ integration_test_template = {
         'qemu-args' : [],
         'exit-code' : 123,
         'vm' : false,
+        'coredump-exclude-regex' : '',
 }
 testdata_subdirs = [
         'auxv',
@@ -391,6 +392,7 @@ foreach integration_test : integration_tests
                 '--storage', integration_test['storage'],
                 '--firmware', integration_test['firmware'],
                 '--exit-code', integration_test['exit-code'].to_string(),
+                '--coredump-exclude-regex', integration_test['coredump-exclude-regex'],
         ]
 
         if 'unit' in integration_test