Re: [PATCH v3 00/12] fs/ceph: optimize struct layouts

Hi Max,

Most of the series is clean — the field removals, bool conversions and
padding reordering are all welcome. Three blockers:

Patch 10 replaces the CEPH_SNAPDIR check in ceph_d_prune() with
ceph_in_snap(), which also matches ordinary snapshot subdirectories.
Those lose their __ceph_dir_clear_complete() call on prune, so the
dcache completeness claim is wrong for them after VFS reclaim. Keep the
exact ceph_snap(...) == CEPH_SNAPDIR check there.

Patch 12 declares vino but only uses it inside #if BITS_PER_LONG >= 64.
The 32-bit path never references it, so -Wunused-variable breaks the
build with CONFIG_WERROR=y.

struct ceph_file_layout is shared with RBD which works with 64-bit OSD
pool IDs. Narrowing pool_id to u32 silently truncates any pool above
U32_MAX in the RBD path. The MDS wire format being __le32 is fine; the
in-memory field needs to stay 64-bit.

Could you respin?

--
Alex Markuze