* Eric Whitney <enwlinux@xxxxxxxxx>:
> Ext4 can make gross overestimates of the number of reserved clusters
> required to handle bigalloc write requests under delayed allocation,
> resulting in premature ENOSPC and quota limit failures.
>
> These patches fix the implementation of reserved cluster accounting for
> bigalloc file systems configured to use delayed allocation, and fix the
> specific bugs described in bugzilla #151491.
>
> These patches also supersede a previous RFC posting from May 2018
> ("ext4: rework delayed allocated cluster accounting"), as well as a more
> recent RFC posting from August 2018 with the same title as this series.
>
> Eric Whitney (6):
> ext4: generalize extents status tree search functions
> ext4: add new pending reservation mechanism
> ext4: fix reserved cluster accounting at delayed write time
> ext4: reduce reserved cluster count by number of allocated clusters
> ext4: adjust reserved cluster count when removing extents
> ext4: fix reserved cluster accounting at page invalidation time
>
> fs/ext4/ext4.h | 10 +-
> fs/ext4/ext4_extents.h | 13 +
> fs/ext4/extents.c | 602 ++++++++++++++++++++--------------------
> fs/ext4/extents_status.c | 662 ++++++++++++++++++++++++++++++++++++++++++--
> fs/ext4/extents_status.h | 84 +++++-
> fs/ext4/inode.c | 113 +++++---
> fs/ext4/mballoc.c | 14 +-
> fs/ext4/super.c | 8 +
> include/trace/events/ext4.h | 99 +++++--
> 9 files changed, 1207 insertions(+), 398 deletions(-)
>
> --
> 2.11.0
>
An FYI for reviewers -
I've got a v2 for this series in testing. It fixes some bugs that lead to
infrequent test failures for generic/112 when running the kvm-xfstests
test appliance on the bigalloc and bigalloc_1k test cases. Basically,
fallocated extents weren't always being handled correctly. Also, I've
corrected some misleading nomenclature and a few comments. These changes
are small and limited to patches 2, 4, and 6 - review effort invested in
v1 should be largely applicable to v2. The patches look very solid at this
point.
I expect to post v2 tomorrow once a complete test appliance run has
finished.
Thanks,
Eric
