source: rtems-docs/filesystem/fileystem_implmentation.rst @ d389819

Filesystem Implementation Requirements
######################################

This chapter details the behavioral requirements that all filesystem
implementations must adhere to.

General
=======

The RTEMS filesystem framework was intended to be compliant with the
POSIX Files and Directories interface standard. The following filesystem
characteristics resulted in a functional switching layer.
.. code:: c

    Figure of the Filesystem Functional Layering goes here.
    This figure includes networking and disk caching layering.

# Application programs are presented with a standard set of POSIX
  compliant functions that allow them to interface with the files, devices
  and directories in the filesystem. The interfaces to these routines do
  not reflect the type of subordinate filesystem implementation in which
  the file will be found.

# The filesystem framework developed under RTEMS allows for mounting
  filesystem of different types under the base filesystem.

# The mechanics of locating file information may be quite different
  between filesystem types.

# The process of locating a file may require crossing filesystem
  boundaries.

# The transitions between filesystem and the processing required to
  access information in different filesystem is not visible at the level
  of the POSIX function call.

# The POSIX interface standard provides file access by character
  pathname to the file in some functions and through an integer file
  descriptor in other functions.

# The nature of the integer file descriptor and its associated
  processing is operating system and filesystem specific.

# Directory and device information must be processed with some of the
  same routines that apply to files.

# The form and content of directory and device information differs
  greatly from that of a regular file.

# Files, directories and devices represent elements (nodes) of a tree
  hierarchy.

# The rules for processing each of the node types that exist under the
  filesystem are node specific but are still not reflected in the POSIX
  interface routines.

.. code:: c

    Figure of the Filesystem Functional Layering goes here.
    This figure focuses on the Base Filesystem and IMFS.

.. code:: c

    Figure of the IMFS Memfile control blocks


File and Directory Removal Constraints
======================================

The following POSIX constraints must be honored by all filesystems.

- If a node is a directory with children it cannot be removed.

- The root node of any filesystem, whether the base filesystem or a
  mounted filesystem, cannot be removed.

- A node that is a directory that is acting as the mount point of a file
  system cannot be removed.

- On filesystems supporting hard links, a link count is maintained.
  Prior to node removal, the node's link count is decremented by one.  The
  link count must be less than one to allow for removal of the node.

API Layering
============

Mapping of Generic System Calls to Filesystem Specific Functions
----------------------------------------------------------------

The list of generic system calls includes the routines open(), read(),
write(), close(), etc..

The Files and Directories section of the POSIX Application Programs
Interface specifies a set of functions with calling arguments that are
used to gain access to the information in a filesystem. To the
application program, these functions allow access to information in any
mounted filesystem without explicit knowledge of the filesystem type or
the filesystem mount configuration. The following are functions that are
provided to the application:

# access()

# chdir()

# chmod()

# chown()

# close()

# closedir()

# fchmod()

# fcntl()

# fdatasync()

# fpathconf()

# fstat()

# fsync()

# ftruncate()

# link()

# lseek()

# mkdir()

# mknod()

# mount()

# open()

# opendir()

# pathconf()

# read()

# readdir()

# rewinddir()

# rmdir()

# rmnod()

# scandir()

# seekdir()

# stat()

# telldir()

# umask()

# unlink()

# unmount()

# utime()

# write()

The filesystem's type as well as the node type within the filesystem
determine the nature of the processing that must be performed for each of
the functions above. The RTEMS filesystem provides a framework that
allows new filesystem to be developed and integrated without alteration
to the basic framework.

To provide the functional switching that is required, each of the POSIX
file and directory functions have been implemented as a shell function.
The shell function adheres to the POSIX interface standard. Within this
functional shell, filesystem and node type information is accessed which
is then used to invoke the appropriate filesystem and node type specific
routine to process the POSIX function call.

File/Device/Directory function access via file control block - rtems_libio_t structure
--------------------------------------------------------------------------------------

The POSIX open() function returns an integer file descriptor that is used
as a reference to file control block information for a specific file. The
file control block contains information that is used to locate node, file
system, mount table and functional handler information. The diagram in
Figure 8 depicts the relationship between and among the following
components.

# File Descriptor Table
  This is an internal RTEMS structure that tracks all currently defined file
  descriptors in the system. The index that is returned by the file open()
  operation references a slot in this table. The slot contains a pointer to
  the file descriptor table entry for this file. The rtems_libio_t structure
  represents the file control block.

# Allocation of entry in the File Descriptor Table
  Access to the file descriptor table is controlled through a semaphore that
  is implemented using the rtems_libio_allocate() function. This routine
  will grab a semaphore and then scan the file control blocks to determine
  which slot is free for use. The first free slot is marked as used and the
  index to this slot is returned as the file descriptor for the open()
  request. After the alterations have been made to the file control block
  table, the semaphore is released to allow further operations on the table.

# Maximum number of entries in the file descriptor table is
  configurable through the src/exec/sapi/headers/confdefs.h file. If the
  CONFIGURE_LIBIO_MAXIMUM_FILE_DESCRIPTORS constant is defined its value
  will represent the maximum number of file descriptors that are allowed.
  If CONFIGURE_LIBIO_MAXIMUM_FILE_DESCRIPTORS is not specified a default
  value of 20 will be used as the maximum number of file descriptors
  allowed.

# File control block - rtems_libio_t structure

  .. code:: c

      struct rtems_libio_tt {
      rtems_driver_name_t              \*driver;
      off_t                             size;
      off_t                             offset;
      unsigned32                        flags;
      rtems_filesystem_location_info_t  pathinfo;
      Objects_Id                        sem;
      unsigned32                        data0;
      void                              data1;
      void                              file_info;
      rtems_filesystem_file_handlers_r  handlers;
      };

  A file control block can exist for regular files, devices and directories.
  The following fields are important for regular file and directory access:

  - Size - For a file this represents the number of bytes currently
    stored in a file. For a directory this field is not filled in.

  - Offset - For a file this is the byte file position index relative to
    the start of the file. For a directory this is the byte offset into a
    sequence of dirent structures.

  - Pathinfo - This is a structure that provides a pointer to node
    information, OPS table functions, Handler functions and the mount table
    entry associated with this node.

  - file_info - A pointer to node information that is used by Handler
    functions

  - handlers - A pointer to a table of handler functions that operate on
    a file, device or directory through a file descriptor index

File/Directory function access via rtems_filesystem_location_info_t structure
-----------------------------------------------------------------------------

The rtems_filesystem_location_info_tt structure below provides sufficient
information to process nodes under a mounted filesystem.

.. code:: c

    struct rtems_filesystem_location_info_tt {
    void                                     \*node_access;
    rtems_filesystem_file_handlers_r         \*handlers;
    rtems_filesystem_operations_table        \*ops;
    rtems_filesystem_mount_table_entry_t     \*mt_entry;
    };

It contains a void pointer to filesystem specific nodal structure,
pointers to the OPS table for the filesystem that contains the node, the
node type specific handlers for the node and a reference pointer to the
mount table entry associated with the filesystem containing the node

Operation Tables
================

Filesystem specific operations are invoked indirectly.  The set of
routines that implement the filesystem are configured into two tables.
The Filesystem Handler Table has routines that are specific to a
filesystem but remain constant regardless of the actual file type.
The File Handler Table has routines that are both filesystem and file type
specific.

Filesystem Handler Table Functions
----------------------------------

OPS table functions are defined in a ``rtems_filesystem_operations_table``
structure.  It defines functions that are specific to a given filesystem.
One table exists for each filesystem that is supported in the RTEMS
configuration. The structure definition appears below and is followed by
general developmental information on each of the functions contained in this
function management structure.

.. code:: c

    typedef struct {
    rtems_filesystem_evalpath_t        evalpath;
    rtems_filesystem_evalmake_t        evalformake;
    rtems_filesystem_link_t            link;
    rtems_filesystem_unlink_t          unlink;
    rtems_filesystem_node_type_t       node_type;
    rtems_filesystem_mknod_t           mknod;
    rtems_filesystem_rmnod_t           rmnod;
    rtems_filesystem_chown_t           chown;
    rtems_filesystem_freenode_t        freenod;
    rtems_filesystem_mount_t           mount;
    rtems_filesystem_fsmount_me_t      fsmount_me;
    rtems_filesystem_unmount_t         unmount;
    rtems_filesystem_fsunmount_me_t    fsunmount_me;
    rtems_filesystem_utime_t           utime;
    rtems_filesystem_evaluate_link_t   eval_link;
    rtems_filesystem_symlink_t         symlink;
    } rtems_filesystem_operations_table;

.. COMMENT: @page

evalpath Handler
~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

evalpath

**Arguments:**

.. code:: c

    const char                        \*pathname,      /* IN     \*/
    int                                flags,         /* IN     \*/
    rtems_filesystem_location_info_t  \*pathloc        /* IN/OUT \*/

**Description:**

This routine is responsible for evaluating the pathname passed in
based upon the flags and the valid ``rthems_filesystem_location_info_t``.
Additionally, it must make any changes to pathloc necessary to identify
the pathname node.  This should include calling the evalpath for a mounted
filesystem, if the given filesystem supports the mount command.

This routine returns a 0 if the evaluation was successful.
Otherwise, it returns a -1 and sets errno to the correct error.

This routine is required and should NOT be set to NULL.

.. COMMENT: @page

evalformake Handler
~~~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

evalformake

**Arguments:**

.. code:: c

    const char                       \*path,       /* IN \*/
    rtems_filesystem_location_info_t \*pathloc,    /* IN/OUT \*/
    const char                      \**name        /* OUT    \*/

**Description:**

This method is given a path to evaluate and a valid start location.  It
is responsible for finding the parent node for a requested make command,
setting pathloc information to identify the parent node, and setting
the name pointer to the first character of the name of the new node.
Additionally, if the filesystem supports the mount command, this method
should call the evalformake routine for the mounted filesystem.

This routine returns a 0 if the evaluation was successful.  Otherwise, it
returns a -1 and sets errno to the correct error.

This routine is required and should NOT be set to NULL.  However, if
the filesystem does not support user creation of a new node, it may
set errno to ENOSYS and return -1.

.. COMMENT: @page

link Handler
~~~~~~~~~~~~

**Corresponding Structure Element:**

link

**Arguments:**

.. code:: c

    rtems_filesystem_location_info_t    \*to_loc,      /* IN \*/
    rtems_filesystem_location_info_t    \*parent_loc,  /* IN \*/
    const char                          \*token        /* IN \*/

**Description:**

This routine is used to create a hard-link.

It will first examine the st_nlink count of the node that we are trying to.
If the link count exceeds LINK_MAX an error will be returned.

The name of the link will be normalized to remove extraneous separators from
the end of the name.

This routine is not required and may be set to NULL.

.. COMMENT: @page

unlink Handler
~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

node_type Handler
~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

node_type()

**Arguments:**

.. code:: c

    rtems_filesystem_location_info_t    \*pathloc        /* IN \*/

**Description:**

XXX

.. COMMENT: @page

mknod Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

mknod()

**Arguments:**

.. code:: c

    const char                          \*token,        /* IN \*/
    mode_t                               mode,         /* IN \*/
    dev_t                                dev,          /* IN \*/
    rtems_filesystem_location_info_t    \*pathloc       /* IN/OUT \*/

**Description:**

XXX

.. COMMENT: @page

rmnod Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

chown Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

chown()

**Arguments:**

.. code:: c

    rtems_filesystem_location_info_t    \*pathloc        /* IN \*/
    uid_t                                owner          /* IN \*/
    gid_t                                group          /* IN \*/

**Description:**

XXX

.. COMMENT: @page

freenod Handler
~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

freenod()

**Arguments:**

.. code:: c

    rtems_filesystem_location_info_t      \*pathloc       /* IN \*/

**Description:**

This routine is used by the generic code to allow memory to be allocated
during the evaluate routines, and set free when the generic code is finished
accessing a node.  If the evaluate routines allocate memory to identify
a node this routine should be utilized to free that memory.

This routine is not required and may be set to NULL.

.. COMMENT: @page

mount Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

mount()

**Arguments:**

.. code:: c

    rtems_filesystem_mount_table_entry_t   \*mt_entry

**Description:**

XXX

.. COMMENT: @page

fsmount_me Handler
~~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

.. code:: c

    rtems_filesystem_mount_table_entry_t   \*mt_entry

**Description:**

This function is provided with a filesystem to take care of the internal
filesystem management details associated with mounting that filesystem
under the RTEMS environment.

It is not responsible for the mounting details associated the filesystem
containing the mount point.

The rtems_filesystem_mount_table_entry_t structure contains the key elements
below:

rtems_filesystem_location_info_t         \*mt_point_node,

This structure contains information about the mount point. This
allows us to find the ops-table and the handling functions
associated with the filesystem containing the mount point.

rtems_filesystem_location_info_t         \*fs_root_node,

This structure contains information about the root node in the file
system to be mounted. It allows us to find the ops-table and the
handling functions associated with the filesystem to be mounted.

rtems_filesystem_options_t                 options,

Read only or read/write access

void                                         \*fs_info,

This points to an allocated block of memory the will be used to
hold any filesystem specific information of a global nature. This
allocated region if important because it allows us to mount the
same filesystem type more than once under the RTEMS system.
Each instance of the mounted filesystem has its own set of global
management information that is separate from the global
management information associated with the other instances of the
mounted filesystem type.

rtems_filesystem_limits_and_options_t    pathconf_info,

The table contains the following set of values associated with the
mounted filesystem:

- link_max

- max_canon

- max_input

- name_max

- path_max

- pipe_buf

- posix_async_io

- posix_chown_restrictions

- posix_no_trunc

- posix_prio_io

- posix_sync_io

- posix_vdisable

These values are accessed with the pathconf() and the fpathconf ()
functions.

const char                                   \*dev

The is intended to contain a string that identifies the device that contains
the filesystem information. The filesystems that are currently implemented
are memory based and don't require a device specification.

If the mt_point_node.node_access is NULL then we are mounting the base file
system.

The routine will create a directory node for the root of the IMFS file
system.

The node will have read, write and execute permissions for owner, group and
others.

The node's name will be a null string.

A filesystem information structure(fs_info) will be allocated and
initialized for the IMFS filesystem. The fs_info pointer in the mount table
entry will be set to point the filesystem information structure.

The pathconf_info element of the mount table will be set to the appropriate
table of path configuration constants (LIMITS_AND_OPTIONS).

The fs_root_node structure will be filled in with the following:

- pointer to the allocated root node of the filesystem

- directory handlers for a directory node under the IMFS filesystem

- OPS table functions for the IMFS

A 0 will be returned to the calling routine if the process succeeded,
otherwise a 1 will be returned.

.. COMMENT: @page

unmount Handler
~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

fsunmount_me Handler
~~~~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

imfs_fsunmount_me()

**Arguments:**

.. code:: c

    rtems_filesystem_mount_table_entry_t   \*mt_entry

**Description:**

XXX

.. COMMENT: @page

utime Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

eval_link Handler
~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

symlink Handler
~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

.. COMMENT: @page

File Handler Table Functions
----------------------------

Handler table functions are defined in a ``rtems_filesystem_file_handlers_r``
structure. It defines functions that are specific to a node type in a given
filesystem. One table exists for each of the filesystem's node types. The
structure definition appears below. It is followed by general developmental
information on each of the functions associated with regular files contained
in this function management structure.
.. code:: c

    typedef struct {
    rtems_filesystem_open_t           open;
    rtems_filesystem_close_t          close;
    rtems_filesystem_read_t           read;
    rtems_filesystem_write_t          write;
    rtems_filesystem_ioctl_t          ioctl;
    rtems_filesystem_lseek_t          lseek;
    rtems_filesystem_fstat_t          fstat;
    rtems_filesystem_fchmod_t         fchmod;
    rtems_filesystem_ftruncate_t      ftruncate;
    rtems_filesystem_fpathconf_t      fpathconf;
    rtems_filesystem_fsync_t          fsync;
    rtems_filesystem_fdatasync_t      fdatasync;
    rtems_filesystem_fcntl_t          fcntl;
    } rtems_filesystem_file_handlers_r;

.. COMMENT: @page

open Handler
~~~~~~~~~~~~

**Corresponding Structure Element:**

open()

**Arguments:**

.. code:: c

    rtems_libio_t   \*iop,
    const char      \*pathname,
    unsigned32       flag,
    unsigned32       mode

**Description:**

XXX

.. COMMENT: @page

close Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

close()

**Arguments:**

.. code:: c

    rtems_libio_t     \*iop

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

read Handler
~~~~~~~~~~~~

**Corresponding Structure Element:**

read()

**Arguments:**

.. code:: c

    rtems_libio_t     \*iop,
    void              \*buffer,
    unsigned32         count

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

write Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

ioctl Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

.. code:: c

    rtems_libio_t     \*iop,
    unsigned32       command,
    void              \*buffer

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

lseek Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

lseek()

**Arguments:**

.. code:: c

    rtems_libio_t     \*iop,
    off_t              offset,
    int                whence

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

fstat Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

fstat()

**Arguments:**

.. code:: c

    rtems_filesystem_location_info_t   \*loc,
    struct stat                        \*buf

**Description:**

The following information is extracted from the filesystem
specific node and placed in the ``stat`` structure:

- st_mode

- st_nlink

- st_ino

- st_uid

- st_gid

- st_atime

- st_mtime

- st_ctime

**NOTES:**

Both the ``stat()`` and ``lstat()`` services are
implemented directly using the ``fstat()`` handler.  The
difference in behavior is determined by how the path is evaluated
prior to this handler being called on a particular
file entity.

The ``fstat()`` system call is implemented directly
on top of this filesystem handler.

.. COMMENT: @page

fchmod Handler
~~~~~~~~~~~~~~

**Corresponding Structure Element:**

fchmod()

**Arguments:**

.. code:: c

    rtems_libio_t     \*iop
    mode_t              mode

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

ftruncate Handler
~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

fpathconf Handler
~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

fsync Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

fdatasync Handler
~~~~~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: @page

fcntl Handler
~~~~~~~~~~~~~

**Corresponding Structure Element:**

XXX

**Arguments:**

XXX

**Description:**

XXX

**NOTES:**

XXX

.. COMMENT: COPYRIGHT (c) 1988-2002.

.. COMMENT: On-Line Applications Research Corporation (OAR).

.. COMMENT: All rights reserved.