On Tue, Jun 08, 2021 at 03:48:11PM +0200, Laurent Bigonville wrote:
There is definitely something boggus here
Yes, just tested it now and it's still happening with the kernel currently
in unstable (5.10.40-1)
Why do you think this numbers are wrong? "fstrim" initially requests
discard of all unused blocks, so a large number. But why should it do
the same on subsequent calls while it still knows what it did the last
time?
ext4 caches the information if a particular block group was trimmed, but
this information is not persistent. (Using bit
EXT4_GROUP_INFO_WAS_TRIMMED_BIT in ext4_group_info.bb_state.)