File System API

File System

• refers to a collection of files and the code in the operating system that manages the files

File

• an array of persistence bytes

Types of Names

1. UID (Unique ID); called inode numbers in Unix/Linux
   • Each file has its own inode number
   • Files deleted will have their inode number recycled
   • Only has to be unique for a given file system
   • Can see inode numbers by running: ls -i
   • 1% of storage is inodes
2. Path Name
3. File Descriptor

inode Structure

• location: pointer to where the file is on disk
• holds a set of meta-data for the file
• Implementing inodes with read(int inode, void* buf) is hard to keep track of since an integer can easily be changed, misplaced, or just wrong.

Paths

• Instead of one level we can build a tree. So only the string name within a given directory has to be unique
• Directories are actually files with a list of file names mapping to inode numbers

mkdir: gets a new inode and creates an empty file for the directory

• the filesystem can tell if it's a directory by the d in drwxr-xr-x.

File Descriptor (FD)

• the inode only has to be fetched once leading to less directory traversals
• can track the offset in the file

Deleting Files

• There is no system call for deleting files
• When there are no references to a file the filesystem will recycle the inode and garbage collect the data blocks used to store the file
• Linux is limited to 16 file descriptors at a time
• If a file has an open file descriptor and the directory for the file is deleted the file block will not be deleted

Links

Hard Link ln

• Another path name we can use to access the file
• Each file is associated with the same inode number
• The system has to keep track of the hardlinks so the system knows when to garbage collect
• No hard links to directory (this would create a cycle)
• This is basically a (string, inode) pair in a directory file

Soft Link (Symbolic Link) ln -s

• This is essentially a file (it has it's own inode) that contains a path to another file
• If the file it points to is deleted, accessing the soft link will show an error

mount

• Attaches the file system we are mounting to the pre-exosting file system
• File systems will have unique inode numbers in themselves
• Command: mount {device} {dir} mount the device to the directory.

fsync

mv: rename a file

• delete's one link to a file and creates a new link to a file
• This command is atomic so that a crash in the middle would either give you either:
  1. the old state of the file
  2. the new state of the file

Atomic File Update

1. Make a temporary file
2. Rename the temporary file