On Thu, May 02 2019, Amir Goldstein wrote:
> On Thu, May 2, 2019 at 12:35 AM NeilBrown <neilb@xxxxxxxx> wrote:
>>
>>
>> If the upper and lower layers use incompatible ACL formats, it is not
>> possible to copy the ACL xttr from one to the other, so overlayfs
>> cannot work with them.
>> This happens particularly with NFSv4 which uses system.nfs4_acl, and
>> ext4 which uses system.posix_acl_access.
>>
>> If all ACLs actually make to Unix permissions, then there is no need
>> to copy up the ACLs, but overlayfs cannot determine this.
>>
>> So allow the sysadmin it assert that ACLs are not needed with a mount
>> option
>> honoracl=off
>> This causes the ACLs to not be copied, so filesystems with different
>> ACL formats can be overlaid together.
>>
>> Signed-off-by: NeilBrown <neilb@xxxxxxxx>
>> ---
>> Documentation/filesystems/overlayfs.txt | 24 ++++++++++++++++++++++++
>> fs/overlayfs/copy_up.c | 9 +++++++--
>> fs/overlayfs/dir.c | 2 +-
>> fs/overlayfs/overlayfs.h | 2 +-
>> fs/overlayfs/ovl_entry.h | 1 +
>> fs/overlayfs/super.c | 15 +++++++++++++++
>> 6 files changed, 49 insertions(+), 4 deletions(-)
>>
>> diff --git a/Documentation/filesystems/overlayfs.txt b/Documentation/filesystems/overlayfs.txt
>> index eef7d9d259e8..7ad675940c93 100644
>> --- a/Documentation/filesystems/overlayfs.txt
>> +++ b/Documentation/filesystems/overlayfs.txt
>> @@ -245,6 +245,30 @@ filesystem - future operations on the file are barely noticed by the
>> overlay filesystem (though an operation on the name of the file such as
>> rename or unlink will of course be noticed and handled).
>>
>> +ACL copy-up
>> +-----------
>> +
>> +When a file that only exists on the lower layer is modified it needs
>> +to be copied up to the upper layer. This means copying the metadata
>> +and (usually) the data (though see "Metadata only copy up" below).
>> +One part of the metadata can be problematic: the ACLs.
>> +
>> +Now all filesystems support ACLs, and when they do they don't all use
>> +the same format. A significant conflict appears between POSIX acls
>> +used on many local filesystems, and NFSv4 ACLs used with NFSv4. There
>> +two formats are, in general, not inter-convertible.
>> +
>> +If a site only uses regular Unix permissions (Read, Write, eXecute by
>> +User, Group and Other), then as these permissions are compatible with
>> +all ACLs, there is no need to copy ACLs. overlayfs cannot determine
>> +if this is the case itself.
>> +
>> +For this reason, overlayfs supports a mount option "honoracl=off"
>> +which causes ACLs, any "system." extended attribute, on the lower
>> +layer to be ignored and, particularly, not copied to the upper later.
>> +This allows NFSv4 to be overlaid with a local filesystem, but should
>> +only be used if the only access controls used on the filesystem are
>> +Unix permission bits.
>>
>
> I don't know. On the one hand "system." is not only ACLs.
Isn't it? What else goes in "system." "??
"man xattr" says:
Extended system attributes
Extended system attributes are used by the kernel to store system
objects such as Access Control Lists. Read and write access
permissions to system attributes depend on the policy implemented
for each system attribute implemented by filesystems in the
kernel.
so it *allows* things other than ACLs, but doesn't confirm that there
are any.
In the kernel source, "XATTR_SYSTEM_PREFIX" is only used with POSIX acls
and "system.sockprotoname" - which is socket specific and no likely to
be found on a filesystem.
"system.
also appears in
CIFS_XATTR_CIFS_ACL
SMB3_XATTR_CIFS_ACL
F2FS_SYSTEM_ADVISE_NAME
XATTR_NAME_NFSV4_ACL
SYSTEM_ORANGEFS_KEY
which should all use XATTR_SYSTEM_PREFIX ...
So yes, I guess they aren't (quite) all ACLs. Bother.
> On the other hand, "honoracl=off" is not the same as -o noacl,
> but it sure sounds the same.
>
> I'd be a lot more comfortable with "ignore_xattrs=system.nfs4_acl"
> argument takes a comma separated list of xattr prefixes to ignore.
That requires the sysadmin to know a lot more about the internals of the
relevant filesystems.... Maybe that is a good idea, but it feels rather
clunky.
In each of these cases, except maybe POSIX_ACLs, it doesn't make sense
to copy-up the "system." xattr unless it is the exact same filesystem
type.
So if given a "noacl" flag (or similar), ignoring copy-up failure for
all "system." attributes is probably the right thing to do, as ACLs are
the only system. attribute for which it can make any sense at all to
copy them.
Thanks,
NeilBrown
>
> ovl_is_private_xattr() can be generalized to ovl_is_ignored_xattr(),
> going over a blacklist of N>=1 which will also be called from
> ovl_can_list(), because there is no point in listing the ACLs that
> are ignored. right?
>
> Thanks,
> Amir.
Attachment:
signature.asc
Description: PGP signature
