Re: Implementing Meta File information in Linux

Albert D. Cahalan (acahalan@cs.uml.edu)
Tue, 1 Sep 1998 22:41:26 -0400 (EDT)


Chris Wedgwood writes:
> On Tue, Sep 01, 1998 at 07:07:59PM -0400, Albert D. Cahalan wrote:

>> Then we must forget about ACLs, privs, the immutable bit...
>> We should be using 14-character unix filenames too. No, make
>> that 8.3 case-insensitive filenames. We must be portable!
>
> All of the above gives us something and cost very little (ACLs are a
> little more complex here I suspect, I don't know them well enough to
> comment).

Long filenames (over 14 characters) broke stuff everywhere.
I don't see anyone complaining about them now though.

>> All our current tools (NFS, tar, cp, mv etc) fail already. They
>> don't handle all the ext2 flags. They certainly don't handle Coda
>> or FAT. What else is new? Userspace was, is, and will be broken.
>
> Bullshit.
>
> For the most part, NFS, tar, cp, mv do work. Yes, they don't handle
> certain ext2fs flags - and arguably neither should they. If I mark a
> file as immutable, it means I don't want it unlinked, and I don't
> care how fucken smart rm ever becomes, I don't want is unlinking
> files marked immutable.

What about tar though? If you must restore from a backup,
wouldn't you want to keep the immutable bits? How would you
copy a directory tree with ext2 attributes intact?

>> Windows has a FileCopyEx call that could be stolen. It deals with
>> all the extra data, including both security attributes and strange
>> forks. It can avoid network traffic when the server supports the
>> call.
>
> This can be done in userspace.

No it can't. Example:

/home/bob is a Windows share, mounted from an NT server.
Bob does "cp foo bar" to make a copy of his file.
That file has:
* ACLs
* multiple forks
* extended attributes
* several gigabytes of data
* holes (compression)

Now, you want to haul several gigabytes of data over 10baseT
and lose all the information that Linux can't understand???
You'd have the server uncompress and recompress all the data too.

That looks like a 30-minute operation, plus data loss
which causes a security hole. No, that is not OK.

Perhaps Bob should bypass the kernel filesystem. He could do raw
network I/O to reach the server using a setuid /bin/cp.

>> Windows has a backup API that handles all the details. Microsoft
>> could add weird new features to the filesystem without breaking
>> backup tools. The backup API requires special privilege and lets
>> the backup admin avoid disturbing _any_ of the time stamps.
>
> This can be done in userspace.

Sure: unmount the filesystem and hit the raw device. That kind of
downtime is simply not acceptable for many business uses.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html