fakefs Data Operations

The data operations create and destroy physical copies of virtual files in a cache in a physical filesystem on the host OS. These operations are known as `checking out' and `checking in' files, respectively.

When fakefs `checks out' a file, it creates a physical file in the cache area (see section ) whose contents are identical to the corresponding virtual file inside fakefs. This is further subdivided into two modes for read-write and read-only access to the physical file:

• For read-only access, the physical file may be shared with other Inodes or even fakefs permanent storage. Modification of the physical file is prohibited, and may damage the fakefs virtual filesystem if it occurs.

• For read-write access, exactly one physical file exists for each distinct open Inode. The normal Unix semantics of a file hard-linked to multiple different names apply--if the file is opened under two names, the two file descriptors correspond to the same physical file as well as the same virtual one. Note that the physical file will have only one hard link.

When fakefs `checks in' a file, it analyzes a physical file in the cache and computes an efficient representation of that file (see chapter 8) to store in the database.

• Fast comparison of files by the Hash of their contents.

• Retrieval of Inodes by Hash (returns list of Inodes that are identical) or by $(Inode, Epoch)$ pair (returns a Hash).

• Retrieve a file by Inode. In read-write mode, files with distinct Inode numbers get distinct physical files, while files with identical Inode numbers get references to a single physical file. Physical files retrieved from Inodes with non-current Epoch values are always read-only.

• Retrieve a file by Hash. The physical files are read-only. Only one physical file exists no matter how many times the same Hash is retrieved by clients. This physical file may be shared with files accessed read-only by Inode.

• Insert a physical file into fakefs by Inode or by filename (which creates a new Inode or looks up an existing one). Insert-by-Inode can only use existing Inodes.