[PATCH] memtx: cancel checkpoint thread at exit

[Vladimir Davydov]

If a tarantool instance exits while checkpointing is in progress, the
memtx checkpoint thread, which writes the snap file, can access already
freed data resulting in a crash. Let's fix this the same way we did for
relay and vinyl threads - simply cancel the thread forcefully and wait
for it to terminate.

Closes #4170
---
https://github.com/tarantool/tarantool/issues/4170
https://github.com/tarantool/tarantool/commits/dv/gh-4170-memtx-cancel-checkpoint-thread-at-exit

 src/box/memtx_engine.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/src/box/memtx_engine.c b/src/box/memtx_engine.c
index 48f700a0..2b9bef24 100644
--- a/src/box/memtx_engine.c
+++ b/src/box/memtx_engine.c
@@ -52,6 +52,9 @@
 /* sync snapshot every 16MB */
 #define SNAP_SYNC_INTERVAL	(1 << 24)
 
+static void
+checkpoint_cancel(struct checkpoint *ckpt);
+
 /*
  * Memtx yield-in-transaction trigger: roll back the effects
  * of the transaction and mark the transaction as aborted.
@@ -179,6 +182,8 @@ static void
 memtx_engine_shutdown(struct engine *engine)
 {
 	struct memtx_engine *memtx = (struct memtx_engine *)engine;
+	if (memtx->checkpoint != NULL)
+		checkpoint_cancel(memtx->checkpoint);
 	mempool_destroy(&memtx->iterator_pool);
 	if (mempool_is_initialized(&memtx->rtree_iterator_pool))
 		mempool_destroy(&memtx->rtree_iterator_pool);
@@ -590,6 +595,11 @@ checkpoint_new(const char *snap_dirname, uint64_t snap_io_rate_limit)
 	xdir_create(&ckpt->dir, snap_dirname, SNAP, &INSTANCE_UUID, &opts);
 	vclock_create(&ckpt->vclock);
 	ckpt->touch = false;
+	/*
+	 * Reset the checkpoint thread id so that checkpoint_cancel()
+	 * doesn't attempt to cancel the thread if it isn't running.
+	 */
+	ckpt->cord.id = 0;
 	return ckpt;
 }
 
@@ -605,6 +615,19 @@ checkpoint_delete(struct checkpoint *ckpt)
 	free(ckpt);
 }
 
+static void
+checkpoint_cancel(struct checkpoint *ckpt)
+{
+	/*
+	 * Cancel the checkpoint thread if it's running and wait
+	 * for it to terminate so as to eliminate the possibility
+	 * of use-after-free.
+	 */
+	if (ckpt->cord.id != 0 &&
+	    tt_pthread_cancel(ckpt->cord.id) != ESRCH)
+		tt_pthread_join(ckpt->cord.id, NULL);
+	checkpoint_delete(ckpt);
+}
 
 static int
 checkpoint_add_space(struct space *sp, void *data)
-- 
2.11.0