Technologies like RDMA benefit from the ability to map file-backed pages
into memory. This benefit extends to persistent-memory devices, where the
backing store for the file can be mapped directly without the need to go
through the kernel’s page cache. There is a fundamental conflict, though,
between mapping a file’s backing store directly and letting the filesystem
code modify that file’s on-disk layout, especially when the mapping is held
in place for a long time (as RDMA is wont to do). The problem seems
intractable, but there may yet be a solution in the form of this
patch set (marked “V1,000,002”) from Ira Weiny.

Source: LWN.net – [$] Long-term get_user_pages() and truncate(): solved at last?