From: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>

Job (especially mirror) may call block_job_error_action several
times before actual pause if it has several in-flight requests.

block_job_error_action will call job_pause more than once in this case,
which lead to following block-job-resume qmp command can't actually
resume the job.

Fix it by do not increase pause level in block_job_error_action if
user_paused already set.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 blockjob.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/blockjob.c b/blockjob.c
index 58de8cb024..730101d282 100644
--- a/blockjob.c
+++ b/blockjob.c
@@ -501,9 +501,11 @@ BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
                                         action);
     }
     if (action == BLOCK_ERROR_ACTION_STOP) {
-        job_pause(&job->job);
-        /* make the pause user visible, which will be resumed from QMP. */
-        job->job.user_paused = true;
+        if (!job->job.user_paused) {
+            job_pause(&job->job);
+            /* make the pause user visible, which will be resumed from QMP. */
+            job->job.user_paused = true;
+        }
         block_job_iostatus_set_err(job, error);
     }
     return action;
-- 
2.20.1