Hi Andrew,
On Fri, 23 Feb 2001, Andrew Morgan wrote:
> [...]
> I'm still concerned by the proposed CAP_TRUSTED thing. As with
> capabilities, ACLs already have a 'POSIX.1e' restriction on being
> manipulated. Adding CAP_TRUSTED to the capabilities needed to manipulate
> any 'trusted' EA seems superfluous.
ACLs should not live in the trusted namespace. The trusted namespace was
intended for one very specific purpose, namely, to allow some user-space
processes that are "trusted" by the kernel/system to have extended
attributes which no untrusted processes have any access to. The idea was
conceived by Casey Schaufler of SGI. With this, it becomes possible to
introduce new secure services based on extended attributes without writing
kernel code.
The other namespaces have other restictions:
user = everybody who has access to the file data
inode = owners only (I'm thinking of renaming this to owner,
as Robert suggested.)
system (or acl_posix etc.) = defined in the kernel
trusted = defined by trusted user-space processes
As far as CAP_TRUSTED is concerned, I just didn't find any capability that
would fit the purpose well, so I used CAP_SYS_MODULE and gave it the name
CAP_TRUSTED. Perhaps what CAP_SYS_MODULE does isn't far off anyway. If you
have any better ideas, I'd be pleased to hear them.
> I'd probably be less concerned if this was just a 'default' behavior in
> cases where the kernel didn't actually know what a "trusted:foo.bar" EA
> was. But in the cases of ACL or full capability support, the presence of
> specific kernel code to enforce the symantics of these EAs would replace
> a capable(CAP_TRUSTED) test.
The purpose is not to combine ACLs and trusted EA's. ACL's are another
thing. The purpose was to allow trusted user-space processes to introduce
new EA's in the trusted namespace without needing to hack the kernel.
--Andreas.
------------------------------------------------------------------------
Andreas Gruenbacher, [EMAIL PROTECTED]
Contact information: http://www.bestbits.at/~ag/
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]