File System: Access

1
File System Access

• Unix mounting
• Operation makes a file system visible in the name
  space
• Root file system mounted by ad hoc means at
  boot-time and contains programs and data needed
  to mount other file systems
• Mount point
• Directory on which file system is mounted
• Previous contents of directory are no longer
  accessible
• Syntax
• /sbin/mount ltdevice or remote fsgt ltdirectorygt
• /etc/fstab
• file describing the file system and swapping
  partitions
• Sysadmin can modify it with text editor
• Also read by mount and un-mount commands
• To un-mount, file system must have no users no
  open files or users whose current working
  directory is in the file system
• Windows analog mapping file systems

2
File System Access

• Why is mounting useful?
• Allows addition of new file systems
• Ability to un-mount and re-mount means that file
  system names can be re-used
• Late binding server not forced to be ready when
  client comes up
• Reasons for un-mounting
• Backing up file system
• Significant change of disk data structures or
  important files or changes to many files
• Changing name of file system

3
File System Access

• After file system is mounted, read/write from/to
  files enabled through system calls
• System call is slower than memory access, so map
  files to memory locations
• Memory-mapped access
• Map portion of file to memory region designated
  by user or chosen by OS
• Access to Nth byte of memory region is access to
  Nth byte of file
• Updated pages are written back to disk eventually
• Syntax 2 new system calls
• Map
• Unmap
• When to write-back to disk?
• When instructed by user
• When process exits
• When paged out
• When unmap is called

4
File System Memory-mapped Access

• Advantages
• Simpler programming interface, but data typing
  can rarely be applied to memory regions filled at
  runtime
• Since pages are faulted into memory, read exactly
  those pages that are referenced and not more
• Eliminate double copying and double buffering
• Copy file block into reserved OS buffer
• Copy again into process virtual memory space
• Disadvantages
• Hard for system to know exact length of file
  since write command (which specifies number of
  bytes written) is not used
• Consistency physical file may not be updated
  causing consistency problems when there are
  writers and readers
• File may be larger than segment or virtual
  address space, so need to map only a portion of
  the file

5
File System Interface

• File descriptors
• Returned by open, creat, dup, pipe, socket,
  socketpair
• Manipulated by fcntl
• File descriptor is a small integer
• Conventional meanings
• 0) stdin
• stdout
• stderr
• Why couldnt hash of file name serve as
  descriptor?
• May have several descriptors per file
• May want to reuse a descriptor for several
  different files over time

6
File System Interface

• Per-process descriptor table
• File descriptor indexes into descriptor table
• An entry contains
• Open mode (read, read/write, etc.)
• special behaviour flags no-delay, asynch,
  close-on-exec, append, etc.
• Pointer to global open file table
• Unix global file table
• Multiple file descriptors can point to same
  global file table entry created by
• Dup within a single process
• Fork across separate processes
• 2nd open of same file in same process actually
  creates a new descriptor and 2nd entry in global
  table

7
File System Interface

• Unix global file table entry contains
• Type of object
• Pointer to object-specific record inode, mode,
  sock (function pointers for operations)
• File offset pointer for sequential access
• Reference count
• Incremented on dup, fork, etc.
• Decremented on close, exit (implicit close) and
  exec if marked close-on-exec
• Entry is destroyed when reference count goes to 0

8
File System Interface

• Why 2 tables per-process and global?
• Want separate processes to be able to share file
  pointer, so need a global table
• Want to allow a process to use the same
  descriptor number for different purposes
  (different files or different open-modes of same
  file), so need a per-process table

9
File Allocation

• Contiguous allocation
• Not really used anymore on hard-drive, but used
  for read-only media (CDROMs and DVDs)
• Each file stored in contiguous space
• Store starting block and length of each file in a
  table
• Linked-list allocation
• File not in contiguous space
• Each block of file points to the next block
• For efficiency, store pointers to next block in a
  table instead of with file block File
  Allocation Table (FAT)
• But table may get very large if disk is large, so
  store only open file block pointers in table
  (stored as an inode)

10
File Allocation

• Inodes
• Each file has an inode with file attributes and
  disk addresses of file blocks
• Only inodes for open files are stored in memory
• When files grow, access the additional free disk
  blocks pointed to in the last disk address of
  inode