From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 From: Stefan Reiter Date: Mon, 4 Jan 2021 14:49:14 +0100 Subject: [PATCH] PVE: fix aborting multiple 'CREATED' jobs in sequential transaction Deadlocks could occur in the AIO_WAIT_WHILE loop in job_finish_sync, which would wait for CREATED but not running jobs to complete, even though job_enter is a no-op in that scenario. Mark offending jobs as ABORTING immediately via job_update_rc if required. Manifested itself in cancelling or failing backups with more than 2 drives. Signed-off-by: Stefan Reiter Tested-by: Mira Limbeck Signed-off-by: Wolfgang Bumiller --- job.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/job.c b/job.c index 8f06e05fbf..05b7797e82 100644 --- a/job.c +++ b/job.c @@ -1035,6 +1035,13 @@ int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp) return -EBUSY; } + /* in a sequential transaction jobs with status CREATED can appear at time + * of cancelling, these have not begun work so job_enter won't do anything, + * let's ensure they are marked as ABORTING if required */ + if (job->status == JOB_STATUS_CREATED && job->txn->sequential) { + job_update_rc(job); + } + AIO_WAIT_WHILE(job->aio_context, (job_enter(job), !job_is_completed(job)));