From c933e15569c9d9b3db7031d8c54712f819816f39 Mon Sep 17 00:00:00 2001
From: Kevin Wolf <kwolf@redhat.com>
Date: Fri, 14 Sep 2018 10:55:39 +0200
Subject: [PATCH 48/49] job: Avoid deadlocks in job_completed_txn_abort()

RH-Author: Kevin Wolf <kwolf@redhat.com>
Message-id: <20180914105540.18077-42-kwolf@redhat.com>
Patchwork-id: 82194
O-Subject: [RHV-7.6 qemu-kvm-rhev PATCH 41/42] job: Avoid deadlocks in job_completed_txn_abort()
Bugzilla: 1601212
RH-Acked-by: John Snow <jsnow@redhat.com>
RH-Acked-by: Max Reitz <mreitz@redhat.com>
RH-Acked-by: Fam Zheng <famz@redhat.com>

Amongst others, job_finalize_single() calls the .prepare/.commit/.abort
callbacks of the individual job driver. Recently, their use was adapted
for all block jobs so that they involve code calling AIO_WAIT_WHILE()
now. Such code must be called under the AioContext lock for the
respective job, but without holding any other AioContext lock.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
---
 job.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/job.c b/job.c
index 42af9e2..5b53e43 100644
--- a/job.c
+++ b/job.c
@@ -713,6 +713,7 @@ static void job_cancel_async(Job *job, bool force)
 
 static void job_completed_txn_abort(Job *job)
 {
+    AioContext *outer_ctx = job->aio_context;
     AioContext *ctx;
     JobTxn *txn = job->txn;
     Job *other_job;
@@ -726,23 +727,26 @@ static void job_completed_txn_abort(Job *job)
     txn->aborting = true;
     job_txn_ref(txn);
 
-    /* We are the first failed job. Cancel other jobs. */
-    QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
-        ctx = other_job->aio_context;
-        aio_context_acquire(ctx);
-    }
+    /* We can only hold the single job's AioContext lock while calling
+     * job_finalize_single() because the finalization callbacks can involve
+     * calls of AIO_WAIT_WHILE(), which could deadlock otherwise. */
+    aio_context_release(outer_ctx);
 
     /* Other jobs are effectively cancelled by us, set the status for
      * them; this job, however, may or may not be cancelled, depending
      * on the caller, so leave it. */
     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
         if (other_job != job) {
+            ctx = other_job->aio_context;
+            aio_context_acquire(ctx);
             job_cancel_async(other_job, false);
+            aio_context_release(ctx);
         }
     }
     while (!QLIST_EMPTY(&txn->jobs)) {
         other_job = QLIST_FIRST(&txn->jobs);
         ctx = other_job->aio_context;
+        aio_context_acquire(ctx);
         if (!job_is_completed(other_job)) {
             assert(job_is_cancelled(other_job));
             job_finish_sync(other_job, NULL, NULL);
@@ -751,6 +755,8 @@ static void job_completed_txn_abort(Job *job)
         aio_context_release(ctx);
     }
 
+    aio_context_acquire(outer_ctx);
+
     job_txn_unref(txn);
 }
 
-- 
1.8.3.1