search

mwri / elixir-userfs

Elixir FUSE (Filesystem in Userspace) interface
10 stars 3 forks source link

readme

userfs Build Status Coverage Status

This is an Elixir FUSE (Filesystem in Userspace) interface. You can use it to create additional file structure in your filesystem, defined in Elixir.

To do so, you must to write an implementation module (a module that implements the 'Userfs.Fs' behaviour). All this really involves is answering FUSE's questions about your files, for example fuse asks you what files are in a given directory and you return a list of names, or fuse asks you what a given file is, and you reply that it is a file with N bytes of content, P permissions, etc.

Example filesystems

Three example file system come with the userfs app for you to peruse.

Userfs.Fs.Hello

More or less the simplest possible case, it has a file and a sym link in the root directory.

Userfs.Fs.Example

Not a lot more interesting than Userfs.Fs.Hello but with some more objects.

Userfs.Fs.Elixir

Much more interesting, more than just a static set of objects, necessitating a more creative implementation. This filesystem allows access to aspects of Elixir's run time state via the filesystem (for example, in the "pids" directory you will find a directory for every process currently running.

To try out the example 'Elixir FS' filesystem create a mount point, "/tmp/elixirfs" say, start the userfs app, and mount the FS like this:

iex> userfs.mount("/tmp/elixirfs", Userfs.Fs.Elixir, [])
{:ok, #PID<0.40.0>}

Now, look inside the /tmp/elixirfs directory and you should find it populated.

The best and cleanest way to unmount the filesystem is to run the OS 'umount' shell command, but you can unmount the filesystem like this:

iex> Userfs.umount("/tmp/elixirfs")
{:ok, {:stopping, #PID<0.40.0>}}

Your filesystem experience should then be something like this:

$ cd /tmp/elixirfs/
$ ls -l
total 0
drwxr-xr-x 2 root root 0 Jan  1  1970 apps
drwxr-xr-x 2 root root 0 Jan  1  1970 code
drwxr-xr-x 2 root root 0 Jan  1  1970 names
drwxr-xr-x 2 root root 0 Jan  1  1970 nodes
drwxr-xr-x 2 root root 0 Jan  1  1970 pids
$ ls -l apps
total 0
drwxr-xr-x 2 root root 0 Jan  1  1970 userfs
drwxr-xr-x 2 root root 0 Jan  1  1970 kernel
drwxr-xr-x 2 root root 0 Jan  1  1970 stdlib
$ ls -l apps/userfs/
total 0
lrwxr-xr-x 1 root root  0 Jan  1  1970 app_proc -> /tmp/elixirfs/pids/<0.36.0>
-rw-r--r-- 1 root root 38 Jan  1  1970 descr
drwxr-xr-x 2 root root  0 Jan  1  1970 env
lrwxr-xr-x 1 root root  0 Jan  1  1970 top_sup -> /tmp/elixirfs/pids/<0.38.0>
-rw-r--r-- 1 root root  6 Jan  1  1970 vsn
$ cat apps/userfs/descr
Erlang FUSE (Filesystem in Userspace)
$ ls -l apps/userfs/app_proc/
total 0
drwxr-xr-x 2 root root 0 Jan  1  1970 linked
drwxr-xr-x 2 root root 0 Jan  1  1970 process_info
$ ls -l apps/userfs/app_proc/linked/
total 0
lrwxr-xr-x 1 root root 0 Jan  1  1970 <0.37.0> -> /tmp/elixirfs/pids/<0.37.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 <0.7.0> -> /tmp/elixirfs/pids/<0.7.0>
$ cat code/modules/userfs/file
/home/mjw/dev/erlang/userfs-1.0.0/ebin/userfs.beam
$ ls -l names/local/
total 0
lrwxr-xr-x 1 root root 0 Jan  1  1970 application_controller -> /tmp/elixirfs/pids/<0.7.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 code_server -> /tmp/elixirfs/pids/<0.20.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 userfs_mount_sup -> /tmp/elixirfs/pids/<0.39.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 userfs_sup -> /tmp/elixirfs/pids/<0.38.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 erl_prim_loader -> /tmp/elixirfs/pids/<0.3.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 error_logger -> /tmp/elixirfs/pids/<0.6.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 file_server_2 -> /tmp/elixirfs/pids/<0.19.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 global_group -> /tmp/elixirfs/pids/<0.18.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 global_name_server -> /tmp/elixirfs/pids/<0.13.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 inet_db -> /tmp/elixirfs/pids/<0.16.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 init -> /tmp/elixirfs/pids/<0.0.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 kernel_safe_sup -> /tmp/elixirfs/pids/<0.29.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 kernel_sup -> /tmp/elixirfs/pids/<0.11.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 rex -> /tmp/elixirfs/pids/<0.12.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 standard_error -> /tmp/elixirfs/pids/<0.22.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 standard_error_sup -> /tmp/elixirfs/pids/<0.21.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 user -> /tmp/elixirfs/pids/<0.25.0>
lrwxr-xr-x 1 root root 0 Jan  1  1970 user_drv -> /tmp/elixirfs/pids/<0.24.0>
$ ls -la pids/\<0.41.0\>/process_info/
total 0
drwxr-xr-x 2 root root  0 Jan  1  1970 .
drwxr-xr-x 2 root root  0 Jan  1  1970 ..
-rw-r--r-- 1 root root 33 Jan  1  1970 current_function
-rw-r--r-- 1 root root 74 Jan  1  1970 dictionary
-rw-r--r-- 1 root root 14 Jan  1  1970 error_handler
-rw-r--r-- 1 root root 93 Jan  1  1970 garbage_collection
-rw-r--r-- 1 root root  9 Jan  1  1970 group_leader
-rw-r--r-- 1 root root  4 Jan  1  1970 heap_size
-rw-r--r-- 1 root root 20 Jan  1  1970 initial_call
-rw-r--r-- 1 root root 19 Jan  1  1970 links
-rw-r--r-- 1 root root  2 Jan  1  1970 message_queue_len
-rw-r--r-- 1 root root  3 Jan  1  1970 messages
-rw-r--r-- 1 root root  7 Jan  1  1970 priority
-rw-r--r-- 1 root root  4 Jan  1  1970 reductions
-rw-r--r-- 1 root root  2 Jan  1  1970 stack_size
-rw-r--r-- 1 root root  8 Jan  1  1970 status
-rw-r--r-- 1 root root  3 Jan  1  1970 suspending
-rw-r--r-- 1 root root  4 Jan  1  1970 total_heap_size
-rw-r--r-- 1 root root  5 Jan  1  1970 trap_exit
$ cat pids/\<0.41.0\>/process_info/garbage_collection
[{min_bin_vheap_size,46422},
 {min_heap_size,233},
 {fullsweep_after,65535},
 {minor_gcs,1}]
$

Supervision behaviour

Filesystems, once mounted, are supervised, so a crash should result in a reinstatement. Since a filesystem can be terminated in a normal way though you should be aware that this might not be true for every circumstance.

As you might expect, if you unmount the file system by calling Userfs.umount/1 the filesystem will be unmounted and not reinstated.

An OS umount call made by a system administrator, from the shell or by other means, will also cause the filesystem to be unmounted and not reinstated. This is less clear, but some cooperation is obviously necessary between userfs and the OS, and it seems a bit unfair on the system administrator if the execution of a perfectly normal and deliberate activity is frustrated.

More controversial, if you kill the port process from the system with a TERM signal, the filesystem is also not reinstated. However a request to TERMinate is still an administrative action ultimately.

If the port receives a signal which is not obviously administrative, or it crashes, then the supervisor (userfs_fs_sup) will reinstate the filesystem. You can kill the port with signal 9 to invoke this action.

Writing your own userfs implementation module

Any filesystem callback implementation should use the Userfs.Fs module:

use Userfs.Fs

This will require the filesystem behaviour and import the attributes.

Your module's userfs_init/2 function will be called when the filesystem is mounted, it receives the mount point and the options/config (the third parameter passed to Userfs.mount/3) as parameters, and returns {:ok, state} where state is passed subsequently to the other FS callbacks. None of the example filesystems use this state but some filesystems wll require it.

Next, when you list the files in a directory in your filesystem userfs_readdir/2 will be called. Obviously the first call is likely to be for the root of your filesystem, and Userfs.Fs.Hello handles this as follows:

def userfs_readdir(state, "/") do
  {:ok, ["hello", "world"], state}
end
def userfs_readdir(state, _) do
  {:error, @error_noent, state}
end

Since there are no directories in this filesystem apart from the root, it's quite simple, if it's the root it returns a list of two objects, and if not it returns a @error_noent error (note this is really a soft error for the end user, noent is POSIX speak for no entity). The return list, as you can see, is a list of strings/binaries.

As soon as the user does something like ls -l instead of just ls, the file system will suddenly have to answer details such as what type of objects 'hello' and 'world' are, and for this your callback module will receive a userfs_getattr/2 call. Here is the userfs_hellofs implementation:

def userfs_getattr(state, "/") do
  {:ok, {0o0755, @attr_dir, 0}, state}
end
def userfs_getattr(state, "/hello") do
  {:ok, {0o0644, @attr_file, byte_size("Hello world!\n")}, state}
end
def userfs_getattr(state, "/world") do
  {:ok, {0o0755, @attr_symlink, String.length("hello")}, state}
end
def userfs_getattr(state, _) do
  {:error, @error_noent, state}
end

Our filesystem has three objects in it in total, the root directory, a file and a symbolic link, and the userfs_getattr function matches each individually and returns a response {ok, {perms, type, size}}. The permissions is the octal value you can use with UNIX's chmod, the type is directory, file or sym link (use the attributes, as above) and the size, for files is the length of the content, and for sym links is the length of the path.

Anything else gets the not found response.

When a file is read, the content must be provided. There is just one file in this case and userfs_getattr is implementated like so:

def userfs_read(state, "/hello") do
  {:ok, "Hello world!\n", state}
end
def userfs_read(state, _) do
  {:error, @error_noent, state}
end

So that the destination of the sym link can be found, the userfs_readlink/2 is implemented:

def userfs_readlink(state, "/world") do
  {:ok, "hello", state}
end
def userfs_readlink(state, _) do
  {:error, @error_noent, state}
end

Licensing

Copyright 2018 Michael Wright mjw@methodanalysis.com

'userfs' is free software, you can redistribute it and/or modify it under the terms of the MIT license.