[Devel,3/3] sunrpc: do not place task to sleep if has to be killed

Submitted by Stanislav Kinsburskiy on July 10, 2017, 4:41 p.m.

Details

Message ID 20170710164151.15570.78060.stgit@localhost.localdomain
State New
Series "SUNRPC: fix races in tasks killing"
Headers show

Commit Message

Stanislav Kinsburskiy July 10, 2017, 4:41 p.m.
There might be a situation, when task is expected to be killed, but it's
execution already passed this check in __rpc_execute() and the task is not yet
sleeping. In this case it will fall into sleep. I.e. it's a race.
This patch add check for per-net "kill_tasks" is enabled before going to
sleep.
What's good here, is that this check is done with queue->lock being held.
And tasks killing is also performed with the same lock being held.
IOW, this lock is a synchronization point, which guarantees us that task will
either die by itself or killed.

https://jira.sw.ru/browse/PSBM-68214

Signed-off-by: Stanislav Kinsburskiy <skinsbursky@virtuozzo.com>
---
 net/sunrpc/sched.c |    6 ++++++
 1 file changed, 6 insertions(+)

Patch hide | download patch | download mbox

diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index cd41946..a551111 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -354,6 +354,12 @@  static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 		rpc_action action,
 		unsigned char queue_priority)
 {
+	if (rpc_abort_task(task)) {
+		task->tk_flags |= RPC_TASK_KILLED;
+		rpc_exit(task, -EIO);
+		return;
+	}
+
 	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 			task->tk_pid, rpc_qname(q), jiffies);