Reading

• Peterson and Silbershatz: 8.3-8.7

• Peterson, Silbershatz and Galvin: 9.3-9.10

File Structure
• Silbershatz and Galvin: 11.1-11.3, 11.5, 12.1-12.6, 13.1-
13.3
Steve Martin
Dave Latham • Dietel, pp. 305-316, 322-336

• Tannenbaum: 6.1-6.3

Copyright note: figures and some elements from slides by Hank Levy, CSE451, University of
Washington. Thank you, Hank!
• Silbershatz, Galvin, and Gagne: 11, 12

Review: physical disk structure What is a File?

• Disk components • A named collection of bits, usually stored on a disk, with some
properties
– platters
– contents, size, owner, last read/write time, protection, type…
– surfaces
sector
– tracks track • From the OS’s standpoint, a file is a bunch of disk blocks on a
– sectors secondary storage device.
– cylinders – Even if the file is used at a higher level of abstraction (records,
surface
bytestreams), this doesn’t matter to the file management system.
– arm
– heads • A file’s type can be encoded in the file’s name or contents
cylinder
– Windows encodes type in name
• .com, .exe, .bat, .dll, .jpg, .mov, .mp3, …
platter – Unix has a smattering of both
arm
• in content via magic numbers or initial characters (e.g., #!)
head
• Files are managed by the file system

Basic file operations File access methods

Unix Windows NT • Some file systems provide different access methods that specify
• create(name) • CreateFile(name, CREATE) ways the application will access data
• open(name, mode) • CreateFile(name, OPEN)
• Sequential access
• read(fd, buf, len) • ReadFile(handle, …) – Read bytes one at a time, in order
• write(fd, buf, len) • WriteFile(handle, …) – This is the most common mode.

• sync(fd) • FlushFileBuffers(handle, …)
• Random access
• seek(fd, pos) • SetFilePointer(handle, …) – Random access given a block/byte #
• close(fd) • CloseHandle(handle, …) – Examples: data set for demand paging, libraries, databases. . .
• unlink(name) • DeleteFile(name)
• Keyed/Indexed access
• rename(old, new) • CopyFile(name)
– FS contains an index to a particular field of each record in a file
• MoveFile(name) – Apps can find a file based on value in that record (similar to DB)
– Can be considered a form of random access

1
 Directories Directories (2)
• Directories provide: • Let’s say you want to open “/one/two/three” in Unix
fd = open(“/one/two/three”, O_RDWR);
– A way for users to organize their files
– A convenient file name space for both users and FS’s
• What goes on inside the Unix file system?
– Open directory “/” (well known, can always find)
• Most file systems support multi-level directories – Search the directory for “one”, get location of “one”
– Naming hierarchies (/, /usr, /usr/local, /usr/local/bin, …) – Open directory “one”, search for “two”, get location of “two”
– Open directory “two”, search for “three”, get loc. of “three”
• Most file systems support the notion of current directory – Open file “three”
– Permissions are checked at each step

• A directory is typically just a file that contains special metadata

• Unix FS spends lots of time walking down directory paths
– Directory = list of (name of file, file attributes)
– This is why open is separate from read/write (session state)
– Attributes include such things as:
– OS will cache prefix lookups to enhance performance
• Size, protection, location on disk, creation time, access time, … • /a/b, /a/bb, /a/bbb all share the “/a” prefix

File System Issues Disk Management Issues

• Disk Management • How should the blocks of a file be placed on disk?
– Make efficient use of disk space – How should files be organized for best performance?
– Fast access to files
– Provide hardware independent view of disk to user (and somewhat, to
OS also) • How should the map to find and access the blocks in a file look?
• Naming
– How do users refer to files? (directories, links, etc) • Must take into account user and file characteristics
– Most files are small, especially in Unix
• Protection – A large proportion of the disk is allocated to a few large files.
– Want to protect users from each other – Many I/O operations are made to large files
– Want to have files from various users on same disk
– Most file I/O operations (60-80%) are reads
– Want to permit controlled sharing
– Most file I/O operations are sequential
• Reliability
– Information must be persistent—must last safely for long periods of • From the above characteristics, per-file cost must be low but large
time.
files must have good performance.

File Block Layout and Access Contiguous Allocation

• File Descriptors • Contiguous allocation
– Data structure that stores:
– Allocate file in contiguous set of blocks or tracks
• file attributes
• map which tells you where the blocks of your file are. • When creating a file, make user specify length and allocate
– Stored on disk along with the files all at once, record in file descriptor.
– In Unix, called inodes – Keep a ‘free list’ of unused areas of the disk
– Don’t get confused from Nachos!
• What are some advantages of this scheme?
• File block layout and access
– Contiguous
– Linked • Disadvantages?
– Index or tree structured

• This is just standard data structures stuff, only on disk

– Pointers now point to blocks instead of addresses in memory

2
 Contiguous Allocation (2) Linked Allocation
• Contiguous allocation advantages • Linked Allocation
– Sequential and random access are easy
– Link the blocks of the file together as a linked list.
– Low overhead
– Simple – In file descriptor, just keep pointer to first block.
– Few seeks – In each block of file keep pointer to next block.
– Very good sequential access performance
• Advantages?
• Disadvantages:
– Horrible fragmentation!
– Hard to predict needs at file creation time • Disadvantages?
– May over-allocate
– Hard to enlarge files

• Can improve this scheme by allocating in fixed contiguous blocks

called extents—i.e. if one block isn't enough, ask for another
– Does this sort of fix sound familiar?

Linked Allocation (2) Indexed Allocation

• Advantages of linked file allocation: • (Simple) Indexed allocation
– Files can be extended easily – Keep array of block pointers for each file
– No external fragmentation problem – File maximum length must be declared at creation time
– Sequential access is easy—just chase links! – Allocate an array to hold pointers to all blocks
• But don’t allocate blocks
• Disadvantages: – Fill in the pointers dynamically using a free list of open blocks.
– How can we do random access?
– Lots of seeking, blocks are likely not laid on the disk right next to • Advantages?
one another
• Why is this bad?
• Disadvantages?
– Some overhead in each block for link.

Indexed Allocation (2) Multi-level Indexed Allocation

• Advantages of indexed file allocation: • This is the VAX Unix solution.
– Not as much space wasted by over predicting – In general, any sort of multi-level treelike structure. Specifically,
• Why? what we will describe is what Berkeley 4.3 BSD Unix does.
– Sequential and random access are easy
– Only waste space in index • Each file descriptor contains file properties and 15
pointers.
• Disadvantages – First 12 point to data blocks
– Still have to set maximum file size – Next three to indirect, doubly-indirect, and triply-indirect blocks
• Can have overflow scheme if file is larger than predicted • 256 pointers in each block
– Blocks are probably allocated all over disk, so will have lots of – Descriptor space isn’t allocated until needed
seeks.
– Index array may be large
• File is a tree of pointers to blocks.
• Requires a large file descriptor, more overhead.

3
 Indexed allocation example Multi-level Indexed Allocation (2)
• Create a ‘tree’ of block pointers to find all the blocks of a • Advantages of Multi-level Indexed Allocation
– Simple
given file: – Easy to implement
– Incremental expansion
– Easy access to small files
0 • Why?
…
1 – Good random access to blocks

…
– Easy to insert blocks in the middle of file
…

– Easy to append to file

12
– Small file map
…

13
14

…
…

• Disadvantages
…

– Maximum file size is still fixed, although it is large

– Indirect mechanism isn’t efficient for large files
• Three descriptor ops for each operation
– File usually isn’t allocated continuously, so need to seek between
blocks.

How do we find a free block? How do we find a free block? (2)

• If all blocks are the same size, can use bit map • A more efficient solution
– One bit per disk block, cache parts of map in memory – Allocate groups of sequential blocks
– Use multi-level index scheme, except each pointer is to a sequence of
• If blocks are variable size, use free list free blocks on disk.
– Requires free storage area management
• Fragmentation and compaction • When we need a free block, try to allocate next physical block on
track (or in cylinder)
• In Unix, blocks are grouped for efficiency
– Each block in free list contains pointers to many free blocks, plus pointer
to next block on list. • If we have detected a sequence of sequential writes, grab a group of
– Aren’t many references involved in allocation or de-allocation blocks at a time, and release them later if unused
– Size of group depends on number of sequential writes we’ve seen so
• Organization of free list means files get spread to different blocks all far.
over disk.
– Why is this bad? • Keep part of disk unallocated always to maximize probability of
finding a sequential block to allocate.