On Tue, Mar 21, 2017 at 10:53:09AM +0900, Satoru Takeuchi wrote:

There have been some discussions about inconsistent device between /proc/pid/maps and stat(2).

http://thr3ads.net/btrfs-devel/2011/05/2346176-RFC-PATCH-0-2-btrfs-vfs-Return-same-device-in-stat-2-and-proc-pid-maps
https://www.spinics.net/lists/linux-btrfs/msg09044.html
https://patchwork.kernel.org/patch/2825842/
https://patchwork.kernel.org/patch/2831525/

It's important since it breaks user programs like lsof(8). There was a patche by Mark to fix this problem.
However, unfortunately, that patch is not merged so far.

And no variant of the get_map_dev will ever be merged. Reworking this
requires extensions to the superblock and subvolume structures, making
it more generic and suitable for VFS.