nvme/fio_plugin: fix race during startup

We can crash with spdk_nvme_qpair_process_completions(qpair=0x0) when
called from spdk_fio_getevents(). This was observed when passing more
than two namespaces to an fio job.

This is because this callback can be called concurrently with
spdk_fio_open(), which assigns ->qpair. We'll just skip any
non-initialized qpairs in the processing loop, as eventually
spdk_fio_open() will set them.

Fixes: "f69367c78 fio_nvme: defer qpair allocation to file_open callback"

Signed-off-by: John Levon <john.levon@nutanix.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/12338 (master)

(cherry picked from commit a6aaf848a5)
Change-Id: Ie8f1ac37726e202bb971ffeb497f9e32656392aa
Signed-off-by: Krzysztof Karas <krzysztof.karas@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/12488
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This commit is contained in:
John Levon 2022-04-22 11:36:14 +01:00 committed by Keith Lucas
parent 2d8c9c571e
commit 5a3769e871

View File

@ -1165,6 +1165,15 @@ static int spdk_fio_getevents(struct thread_data *td, unsigned int min,
} }
while (fio_qpair != NULL) { while (fio_qpair != NULL) {
/*
* We can be called while spdk_fio_open()s are still
* ongoing, in which case, ->qpair can still be NULL.
*/
if (fio_qpair->qpair == NULL) {
fio_qpair = TAILQ_NEXT(fio_qpair, link);
continue;
}
spdk_nvme_qpair_process_completions(fio_qpair->qpair, max - fio_thread->iocq_count); spdk_nvme_qpair_process_completions(fio_qpair->qpair, max - fio_thread->iocq_count);
if (fio_thread->iocq_count >= min) { if (fio_thread->iocq_count >= min) {