[PATCH v1 0/2] nfs_common: Encode an ACL with fewer than three entries as empty

[Chuck Lever <cel@kernel.org>]

Both ACL encoders in fs/nfs_common report at least four wire entries
for any ACL that has entries, then walk a_entries[] up to that count.
An ACL with one or two entries sends the walk past the end of the
posix_acl allocation.

nfsacl_encode() builds only the NFS client's SETACL arguments, and
nfs_stream_encode_acl() builds only NFSD's GETACL replies. The fixes
are separate patches because a combined one does not apply to the
stable trees that predate nfs_stream_encode_acl().

The NFS client still caches a GETACL result without validating it.
This series does not change that.

Neither over-read has been observed at runtime. Both paths were
traced from the code, and the patches are build-tested only.

Patch 1 changes what the NFS client sends in SETACL, so it needs an
Acked-by from the NFS client maintainers before it goes through the
nfsd tree.

Chuck Lever (2):
  nfs_common: Do not encode an ACL with fewer than three entries
  nfs_common: Do not stream-encode an ACL with fewer than three entries

 fs/nfs_common/nfsacl.c | 6 ++++--
 include/linux/nfsacl.h | 5 +++--
 2 files changed, 7 insertions(+), 4 deletions(-)

-- 
2.55.0