CI: Fix the ThreadSanitizer build

.cirrus.yml

@@ -355,18 +355,21 @@ ubsan_sanitizer_task:
     ZEEK_TAILORED_UB_CHECKS: 1
     UBSAN_OPTIONS: print_stacktrace=1
 
-# tsan_sanitizer_task:
+tsan_sanitizer_task:
-#   container:
+  container:
-#     # Just uses a recent/common distro to run memory error/leak checks.
+    # Just uses a recent/common distro to run memory error/leak checks.
-#     dockerfile: ci/ubuntu-20.04/Dockerfile
+    dockerfile: ci/ubuntu-20.04/Dockerfile
-#     << : *SANITIZERS_RESOURCE_TEMPLATE
+    << : *SANITIZERS_RESOURCE_TEMPLATE
-#
+
-#   << : *CI_TEMPLATE
+  << : *CI_TEMPLATE
-#   test_fuzzers_script: ./ci/test-fuzzers.sh
+  << : *SKIP_TASK_ON_PR
-#   env:
+  env:
-#     CXXFLAGS: -DZEEK_DICT_DEBUG
+    ZEEK_CI_CONFIGURE_FLAGS: *TSAN_SANITIZER_CONFIG
-#     ZEEK_CI_CONFIGURE_FLAGS: *TSAN_SANITIZER_CONFIG
+    ZEEK_CI_DISABLE_SCRIPT_PROFILING: 1
-#     ZEEK_CI_DISABLE_SCRIPT_PROFILING: 1
+    # If this is defined directly in the environment, configure fails to find
+    # OpenSSL. Instead we define it with a different name and then give it
+    # the correct name in the testing scripts.
+    ZEEK_TSAN_OPTIONS: suppressions=/zeek/ci/tsan_suppressions.txt
 
 windows_task:
   # 2 hour timeout just for potential of building Docker image taking a while

ci/test.sh

@@ -19,6 +19,10 @@ fi
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" &>/dev/null && pwd)"
 . ${SCRIPT_DIR}/common.sh
 
+if [ -n "${ZEEK_TSAN_OPTIONS}" ]; then
+    export TSAN_OPTIONS=${ZEEK_TSAN_OPTIONS}
+fi
+
 function pushd {
     command pushd "$@" >/dev/null || exit 1
 }

ci/tsan_suppressions.txt

@@ -0,0 +1,34 @@
+# This is a list of suppressions for ThreadSanitizer. Anything listed here will be
+# ignored during testing. See https://github.com/google/sanitizers/wiki/ThreadSanitizerSuppressions
+# for documentation on how this file works.
+
+# There's a bug in libstdc++ that causes ThreadSanitizer to flag this as a data race.
+# See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77704. Ignore any reports from this
+# as it's really really noisy, and there's not much we can do to fix it.
+race:std::ctype<char>::narrow
+
+# =====================================================================
+# Everything below here are known failures in Zeek. These are here until they
+# can be fixed, just so we can get the ThreadSanitizer builds running on Cirrus
+# and catch anything new. If we can't fix something in this list (possibly the
+# sqlite ones?) split them out into a separate block above here with a comment
+# as to why.
+
+race:broker::internal::connector::run_impl
+race:caf::net::multiplexer::set_thread_id
+race:caf::action::run
+
+# This one causes supervisor.config-bare-mode to fail occasionally but not always
+signal:caf::actor_control_block::enqueue
+
+# There's a bunch of failures down inside the sqlite code itself, mostly
+# around opening the database in the SQLite input reader and the SQLite
+# logging writer.
+race:sqlite3MutexInit
+race:sqlite3Malloc
+race:sqlite3_mutex_enter
+race:sqlite3_initialize
+
+# This one isn't actually in sqlite code, but some StringVal object gets ref'd by
+# zeek::id::find_const and throws a data race.
+race:zeek::logging::writer::detail::SQLite::DoInit