README

January 23, 2016 · View on GitHub

OVERVIEW

nullfs is FUSE filesystem driver which discards all files' data written to it.

Reading from any nullfs file returns EOF.

Regarding directories, different strategies exist and several implementations ar provided (see below).

DETAILED DESCRIPTION

Proper implementation requires nullfs driver to store metadata for files, directories and special files (devices/sockets/fifos), and this may not be desirable in some cases because of memory footprint.

Therefore 3 implementations are to be provided:

  1. nul1fs

nul1fs implements flat filesystem without support for directories except the "/" one. Readdir (ls) show only "." and ".." entries in "/".

Any file can be created, read or written, because lookup for any filename returns TRUE and reports root:root owner and 0666 permissions.

Building and mounting:

xrgtn@ux280p:/jff/nullfsmakecleanrmfnul1fsnullfsnulnfs.oxrgtn@ux280p: /jff/nullfs make clean rm -f nul1fs nullfs nulnfs *.o xrgtn@ux280p:~/jff/nullfs make cc "-lfuse" nul1fs.c -o nul1fs g++ "-lfuse" -o nullfs nullfs.c++ cc "-lfuse" nulnfs.c -o nulnfs xrgtn@ux280p:/jff/nullfsmkdirmntxrgtn@ux280p: /jff/nullfs mkdir mnt xrgtn@ux280p:~/jff/nullfs ./nul1fs ./mnt

Using:

xrgtn@xrgtn-q40:/jff/nullfslsal./mnttotal4drwxrwxrwx2rootroot02010081212:09.drwxrxrx3xrgtnxrgtn40962010081212:09..xrgtn@xrgtnq40: /jff/nullfs ls -al ./mnt total 4 drwxrwxrwx 2 root root 0 2010-08-12 12:09 . drwxr-xr-x 3 xrgtn xrgtn 4096 2010-08-12 12:09 .. xrgtn@xrgtn-q40:~/jff/nullfs echo foo > ./mnt/bar xrgtn@xrgtn-q40:/jff/nullfscat./mnt/barxrgtn@xrgtnq40: /jff/nullfs cat ./mnt/bar xrgtn@xrgtn-q40:~/jff/nullfs cat ./mnt/foo xrgtn@xrgtn-q40:~/jff/nullfslsl./mnt/bazrwrwrw1rootroot02010081212:10./mnt/bazxrgtn@xrgtnq40: /jff/nullfs ls -l ./mnt/baz -rw-rw-rw- 1 root root 0 2010-08-12 12:10 ./mnt/baz xrgtn@xrgtn-q40:~/jff/nullfs

  1. nulnfs

nulnfs implements nullfs with limited number of available inodes. When upper limit is reached, old inodes are forgotten if possible (here's example when forgetting is impossible: cd /tmp/nullfs/ ; while mkdir d ; do cd d ; done). If nulnfs cannot free some inodes, it returns ENOSPC in response to mkdir/mknod/create.

NOTE: nulnfs hasn't been finished yet (it crashes on use) and I have no plans to continue working on it at the moment. But two other implementations (minimalistic nul1fs one and C++ nullfs) work just fine as advertised.

  1. nullfs

nullfs permits to create files/directories until it gets OOM killed or malloc()/new() stop working (in the later case ot responds with ENOMEM).