1. File Systems
Kanwar Gill July 7, 2015

2. File Systems Provides long-term storage Requirements: Two views:
Store large amounts of information
Information must be persistent
Multiple processes should be able to access it concurrently
Two views:
User view: logical organization of information
OS view: managing actual physical storage media
File systems then essentially:
implement an abstraction for secondary storage
Organize files logically
Permit data sharing b/w processes/users/machines
Provide protection of the data

3. File operations File sharing File access
Read, write, open, close, seek, etc.
File sharing
Two keys issues:
Semantics of concurrent access
Protection
Two levels of internal tables:
Per process open file table
System wide open file table
File access
Sequential Access: read bytes one at a time
Direct/Random Access: random access given block/byte number

4. Disk Layout Files span multiple disk blocks.
Problem: How can we find all the blocks for a file?
Contiguous allocation
Fast but causes fragmentation
Linked structure
Works for sequential access but not for others
Indexed structure (inode)
Handles random access better

5. Question:
Say we have a file system with a block size of 4096 bytes. Assume that each time a disk block is read, it is cached in memory.
Suppose a user program issues the following system calls for accessing the file "data". This file is 50KB (51200 bytes) in size, and is stored in the root (/) directory. Assume we start with no data or metadata cached in memory.
Linked allocation
For the first set of statements, assume you are using the linked allocation strategy, and that each operation is executed in sequence. (In other words, memory is not reset in between each operation.) In this system, the disk block containing the root directory is in a well-known location and can be read directly. Directory entries hold the name of a file or directory and the block number of the first block in the file. (For this question, we will ignore where meta data is stored.) How many disk blocks are read (from disk) on each system call?
open
read 50 bytes at offset 100
read 60 bytes at offset 12250
read 120 bytes at offset 49300
read 80 bytes at offset 22010

6. Hybrid indexed allocation
For the following statements, assume you are using a hybrid indexed allocation strategy as with UNIX: the first 12 blocks of the file are linked directly from the inode, followed by a single-indirect link, a double-indirect link, and a triple-indirect link. The root inode is known, and can be read directly. Assume each inode that needs to be read in the question below is stored in a different disk block. Remember that you are starting with no data or metadata cached in memory. How many disk blocks are read (from disk) on each system call?
open
read 50 bytes at offset 100
read 60 bytes at offset 12250
read 120 bytes at offset 49300
read 80 bytes at offset 22010