Fix !zilog_is_dirty() assert during ztest

ztest occasionally hits an assert that !zilog_is_dirty() during
zil_close(). This is caused by an interaction between 2 threads.
First, ztest_run() waits for each test thread to complete and
closes the associated dataset as soon as the thread joins. At
the same time, the ztest_vdev_add_remove() test may attempt to
remove the slog, which will open, dirty, and reset the logs on
every dataset in the pool (including those of other threads).
This patch simply ensures that we always join all of the test
threads before closing any datasets.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes #8094

diff --git a/cmd/ztest/ztest.c b/cmd/ztest/ztest.c
index 36647c4ff9d00cd8b16df97be4e94c7785db5de4..eab8940fbcf1b3cf5572c29c2a265289983b47a9 100644 (file)
--- a/cmd/ztest/ztest.c
+++ b/cmd/ztest/ztest.c
@@ -6918,11 +6918,17 @@ ztest_run(ztest_shared_t *zs)
        }
 
        /*
-        * Wait for all of the tests to complete.  We go in reverse order
-        * so we don't close datasets while threads are still using them.
+        * Wait for all of the tests to complete.
         */
-       for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
+       for (t = 0; t < ztest_opts.zo_threads; t++)
                VERIFY0(thread_join(run_threads[t]));
+
+       /*
+        * Close all datasets. This must be done after all the threads
+        * are joined so we can be sure none of the datasets are in-use
+        * by any of the threads.
+        */
+       for (t = 0; t < ztest_opts.zo_threads; t++) {
                if (t < ztest_opts.zo_datasets)
                        ztest_dataset_close(t);
        }