Security 2

Notes: Comp 3000 Operating Systems Nov 21st 07

-The total test mark will be the highest mark between the 2 tests

-VFS - Abstraction between the os and the file system - keeps the os from being tied into the filesystem.

-NFS - Old file system.

-What is the appropriate level of abstaction for a networked file system? Direct access to disk -> block level network filesystem

SCSI over IP
BAD IDEA. No file permissions/file control. No security.
Not good to have more then one OS accessing the disk at one time.
There are uses for this kind of system: SAN - Storage area network.
Bunch of computers in network. (IMAGE-> NETWORK RING)
Bunch of disks pretty much removed from the computing aspect. (IMAGE -> NETWORK *RING AND STORAGE RING)
Seperate channel to communicate between computers and storage.
Fast, fast connection
Look at trust boundries in networked storage. Who do I trust? Connected to internet?

-Not block access, file access. -NFS -> File server is stateless, in old versions.

Stateless is weird... VFS has OPEN, CLOSE, WRITE, SEEK, CLOSE, STAT etc.
OPEN and CLOSE don't work. Cannot keep track of this in stateless.
Advantage: File server can reboot and not effect clients,
SEEK doesn't work either. No file pointer.
You have to specify EVERYTHING, File, how much data, offset etc. at
Open only converts a pathname to a file.
What happens if you open a random number?
NFS = NO FILE SECURITY.
All security is on the client side. NFS server can choose workstations that can access read, read-write. That's it.
Only works in a trusted environment.

-The internet is a type of networked file system. (Read only).

-WebDAV lets you do writes and reads over the internet.

-Best distributed file system: Sets the bar... AFS.

-Allow someone to access their files from thousands of computers around the world securely and with high performance.

-AFS allows file control lists. Looks like local storage.

-So, problems: Authentication. Namespace. Performance. Kerberos. -> Single signon without trusting all clients.

-Login, then you have a ticket to authenticate yourself after logging on to the Kerberos system. Ticket is time limited. (GOOD!)

-AFS also uses a DNS like service to resolve which machine matches the name of the machine you want.

-Performance? You want to cache! (with NFS , you can't do this. stateless, can't keep track of who is accessing what) AFS is the opposite. It keeps as much state as possible. (AFS is stateful)

-Callbacks - Contact the client to update it's cached file.

-Close is crazy. (Save works like normal, saves to local disk)

-What if file server goes down? Close fails! You cannot update the file on the server. You must check the return value of close. Close = commit to server.

-For standard cases, local and networked disks work pretty much the same. It's the error cases that don't work in similar ways. In local systems, things are working fine, or not working at all. In distributed systems, things work alright all the time.

-Tickets expire, kinda weird. What happens? File loss?

-RPCs, remote procedure calls. Wrappers cause problems. All you see is a procedure call. what you are actually doing is network communication. You don't program properly.

-AFS doesn't encrypt, less them optimal integrity checks.

FIRST DRAFT ONLY These notes are a bit rough... I'll be fixing them later tonight.

Posted By Kevin Bowrin 2:45 PM Nov 22nd