Validate shard id is less than shard count before indexing pools

[dkropachev]

The connection pool trusts the shard id reported by the server and later indexes the per-shard connection array with it.

Code permalink:

scylla-rust-driver/scylla/src/network/connection_pool.rs

Line 826 in e04a12a

let shard_info = connection.get_shard_info().as_ref();

Shard metadata validation ensures the shard count is nonzero, but it does not reject a shard id greater than or equal to the shard count. A malformed or buggy server response can therefore panic the pool worker instead of being treated as invalid sharding information.

Suggested direction: validate that the reported shard id is strictly less than the reported shard count before resizing/indexing the per-shard pool state.

[Lorak-mmk]

This is true, but it is worth noting that a misbehaving server can do other things that we have no way to protect against.

For example, if it sends a different number of shards, then we'll perform resharding and drop all connections in the pool.
This issue is still true and should be fixed - self.conns[shard_id] can panic even after resharding.