[CRIU] [PATCH 0/7] irmap: Introduce the on-disk cache for irmap data

[Pavel Emelyanov]

The irmap generation is slow. Although we typically scan only a small
subset from the fs tree, it's anyway slow. In the end, without proper
hints, we may end up scanning for the whole sub-tree.

One of the ways to solve this is to generate the irmap cache on pre-dump
while tasks continue running. This cache will be effective, as tasks
rarely change the inodes they fsnotify.

Signed-off-by: Pavel Emelyanov <xemul at parallels.com>