[OpenAFS] RE: [OpenAFS-announce] OpenAFS client crashes on RHEL 5.10 and RHEL
J. Bruce Fields
Tue, 11 Mar 2014 11:45:47 -0400
By the way, do you support NFS exports of OpenAFS filesystems from
If not, another fix on your side may be to replace calls to
d_splice_alias by calls to d_materialise_unique.
We have two alternative helper functions for filesystems to call from
their lookup methods:
- d_splice_alias: which was originally designed for NFS servers,
to handle the case where the lookup encounters an inode which
was previously only known by filehandle, and
- d_materialise_unique: which was originally designed which
distributed filesystems in mind, which have the special issue
that the directory tree can be modified by other nodes without
their knowledge, so they need to be ble to deal with lookup
encountering an inode which was previously known under some
The root issue in some sense is that it's become more common for
filesystems to need both, and neither is completely correct for that.
(For example GFS2 is both a distributed filesystem and can be
But if you don't care about exportability then d_materialise_unique
might be better.
(And even if you do d_materialise_unique will probably be the better
choice for you on recent upstream kernels.)
PS: feel free to redirect me to the correct list if this isn't it.