Lecture Outline

• Different Organisations of Storage Space

• Sharing & Security Issues

• Filing System Integrity

Contiguous Allocation

• Each block comprising the file will occupy adjacent blocks of secondary storage

• Each file is identified by its start address and its length
  + Logically close records in the file will occupy physically close addresses. In a buffered and paged system, 1 disk i/o will read in possibly several records that the process will require next.

  - Internal fragmentation

  - External fragmentation, recovered by compaction

  - Difficult for files to grow dynamically. File size specified at file creation. Usually, user has to create second, larger file, copy contents of old file to new one, and delete the original.

  - If file size is over-estimated at file creation, unused space is wasted, and cannot be allocated to another file.

  + Random access possible using policy similar to virtual address translation mechanism

• Free space is either maintained in a separate list, or when a file is deleted, the space is marked as free in the UFD.

• Better policy for files that can shrink and grow over time

• Three different methods of organising files, all falling under the category of block allocation

• In block allocation, files are again segmented into fixed-size blocks, but blocks do not have to occupy adjacent addresses in SS.

Block Chaining (Block Linkage, in L&E)

• UFD points to first block. Each block points to the next block, until a block with a nil pointer is reached

• Searching is sequential - have to traverse each block until record is found - good for sequentially accessed files. Bad for others.

• Separate free block list maintained
  - If a pointer is corrupted, then the rest of the file cannot be accessed (protected by having redundant files containing pointers)

  + Small overhead (approx. 4 bytes per block for pointer)

Index Block Chaining (Index Blocks in L&E)

• Pointers to data blocks held in Index Blocks instead of the data blocks

• UFD points to first index block, and index block contains record key and pointers to the physical address of the record. Index blocks can be chained.

• When record is located, entire data block containing record can be read into PM / Buffer

• Searching takes place in index blocks (each will contain more keys than data block can contain records, so 1 index block can point to several data blocks) - retrieval is faster

• `Bit Map' can be maintained to indicate whether a block is free or allocated
  - Space overhead for index blocks

• A special file contains a file map of all blocks.

• The UFD points to an entry in the file map. This indicates the physical block number of the first physical block. The entry contains the block number of the next block, etc. Last block entry for file contains a null.

• The physical address of each block can be calculated (block number x block size)

• Have to search through each block sequentially, but file map can be extended to include record ranges for each block.

• No need to maintain separate free list. All free space can be referred to as a file, or each free block entry can contain a special symbol.

• To insert a block, look for free block nearest to logical block (to reduce seek time), and change block pointers

• To delete a block, mark file map entry as free

• What might users want to do?

• Individual files / directories might need combinations of R, W, X permissions.

• Users can be grouped into classes: Owner, Group, World.

• How can filing system maintain this information?

Access Control Lists

• ACLs are a 3-dimensional matrix listing all files by user class.

• For each class / file combination, the access rights are given for R, W, and E.

• Each user access attempt is filtered through the access control list. If the access attempt is legal, the request is granted, otherwise it is denied.

• There is also a need to protect computer system and data from damage from fire, flooding, power surges, etc.

• Also to protect against damage caused by hardware malfunction (e.g., disk crash), to prevent data being lost.

• Managed by taking regular backups of the system, and storing the backups off-site, and in fire- & water-proof rooms / safes

• A full backup of a large system would take too long to run. Usually perform, say, once a week.

• Incremental backups are taken nightly. Only data that has changed since the full backup, or the last incremental backup is backed up.

• System can be recovered either by recovering individual files, or by copying the backed up files in reverse chronological order.

• A full day's work can still be lost, because incremental backup is performed nightly. Journalling (transaction logging) of individual files provides up-to-date states of files

Next Lecture ...