haiku/docs/user/drivers/fs_interface.dox
Ingo Weinhold 8338b05a30 More FS interface documentation. All FS hooks required for writing a basic
read-only FS are documented, now. Be encouraged to review and add more. :-)


git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@20568 a95241bf-73f2-0310-859d-f6bbb57e9c96
2007-04-04 23:46:05 +00:00

569 lines
22 KiB
Plaintext

/*!
\file fs_interface.h
\ingroup drivers
*/
/*!
\fn status_t new_vnode(mount_id mountID, vnode_id vnodeID, fs_vnode privateNode)
\brief Creates the vnode with ID \a vnodeID and associates it with the private
data handle \a privateNode, but leaves is in an unpublished state.
The effect of the function is similar to publish_vnode(), but the vnode
remains in an unpublished state, with the effect that a subsequent
remove_vnode() will just delete the vnode and not invoke the file system's
\link file_system_module_info::remove_vnode remove_vnode() \endlink is not
invoked.
If the vnode shall be kept, publish_vnode() has to be invoked afterwards to
mark the vnode published. The combined effect is the same as only invoking
publish_vnode().
The function fails, if the vnode does already exist.
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\param privateNode The private data handle to be associated with the node.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t publish_vnode(mount_id mountID, vnode_id vnodeID,
fs_vnode privateNode)
\brief Creates the vnode with ID \a vnodeID and associates it with the private
data handle \a privateNode or just marks it published.
If the vnode does already exist and has been published, the function fails.
If it has not been published yet (i.e. after a successful new_vnode()), the
function just marks the vnode published. If the vnode did not exist at all
before, it is created and published.
If the function is successful, the caller owns a reference to the vnode. A
sequence of new_vnode() and publish_vnode() results in just one reference as
well. The reference can be surrendered by calling put_vnode().
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\param privateNode The private data handle to be associated with the node.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t get_vnode(mount_id mountID, vnode_id vnodeID,
fs_vnode *_privateNode)
\brief Retrieves the private data handle for the node with the given ID.
If the function is successful, the caller owns a reference to the vnode. The
reference can be surrendered by calling put_vnode().
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\param _privateNode Pointer to a pre-allocated variable the private data
handle shall be written to.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t put_vnode(mount_id mountID, vnode_id vnodeID)
\brief Surrenders a reference to the specified vnode.
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t remove_vnode(mount_id mountID, vnode_id vnodeID)
\brief Marks the specified vnode removed.
The caller must own a reference to the vnode or at least ensure that a
reference to the vnode exists. The function does not surrender a reference,
though.
As soon as the last reference to the vnode has been surrendered, the VFS
invokes the file system's
\link file_system_module_info::remove_vnode remove_vnode() \endlink
hook.
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t unremove_vnode(mount_id mountID, vnode_id vnodeID);
\brief Clears the "removed" mark of the specified vnode.
The caller must own a reference to the vnode or at least ensure that a
reference to the vnode exists.
The function is usually called when the caller, who has invoked remove_vnode()
before realizes that it is not possible to remove the node (e.g. due to an
error).
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t get_vnode_removed(mount_id mountID, vnode_id vnodeID,
bool* removed);
\brief Returns whether the specified vnode is marked removed.
The caller must own a reference to the vnode or at least ensure that a
reference to the vnode exists.
\param mountID The ID of the volume.
\param vnodeID The ID of the node.
\param removed Pointer to a pre-allocated variable set to \c true, if the
node is marked removed, to \c false otherwise.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\class file_system_module_info
\brief Kernel module interface for file systems.
*/
/*!
\fn status_t (*file_system_module_info::mount)(mount_id id, const char *device,
uint32 flags, const char *args, fs_volume *_fs, vnode_id *_rootVnodeID)
\brief Mount a volume according to the specified parameters.
Invoked by the VFS when it has been requested to mount the volume. The FS is
supposed to perform whatever one-time initialization is necessary for the
volume. It is required to create a volume handle for the volume and pass it
back in \a _fs. Moreover it must invoke publish_vnode() for the root node
of the volume and pass the ID of the volume back in \a _rootVnodeID.
A disk-based FS will need to check whether \a device is not \c NULL, open
it, and analyze whether the device or image file actually represents a volume
of that FS type.
If mounting the volume fails for whatever reason, the hook must return an
error code other than \c B_OK. In this case all resources allocated by the
hook must be freed before returning. If and only if \c B_OK is returned, the
unmount() hook will be invoked at a later point when unmounting the volume.
\param id The ID of the volume to be mounted. It should be saved in the FS's
volume private data (volume handle).
\param device The path to the device (or image file) representing the volume
to be mounted. Can be \c NULL.
\param flags Flags:
- \c B_MOUNT_READ_ONLY: Mount the volume read-only.
\param args Null-terminated string in driver settings format, containing FS
specific parameters.
\param _fs Pointer to a pre-allocated variable the volume handle shall be
written to.
\param _rootVnodeID Pointer to a pre-allocated variable the ID of the volume's
root directory shall be written to.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::unmount)(fs_volume fs)
\brief Unmounts the given volume.
Invoked by the VFS when it is asked to unmount the volume. The function must
free all resources associated with the mounted volume, including the volume
handle. Although the mount() hook called publish_vnode() for the root node
of the volume, unmount() must not invoke put_vnode().
\param fs The volume handle.
\return \c B_OK if everything went fine, another error code otherwise. The
error code will be ignored, though.
*/
/*!
\fn status_t (*file_system_module_info::read_fs_info)(fs_volume fs,
struct fs_info *info)
\brief Retrieves general information about the volume.
The following fields of the \c fs_info structure need to be filled in:
- \c flags: Flags applying to the volume, e.g. \c B_FS_IS_READONLY,
\c B_FS_HAS_ATTR, etc.
- \c block_size: The size of blocks the volume data are organized in.
Meaningful mainly for disk-based FSs, other FSs should use some reasonable
value for computing \c total_blocks and \c free_blocks.
- \c io_size: Preferred size of the buffers passed to read() and write().
- \c total_blocks: Total number of blocks the volume contains.
- \c free_blocks: Number of free blocks on the volume.
- \c total_nodes: Maximal number of nodes the volume can contain. If there is
no such limitation use \c LONGLONG_MAX.
- \c free_nodes: Number of additional nodes the volume could contain. If there
is no such limitation use \c LONGLONG_MAX.
- \c device_name: The name of the device or image file containing the volume.
Non-disk-based FSs shall fill in an empty string.
- \c volume_name: The name of the volume.
The other values are filled in by the VFS.
\param fs The volume handle.
\param info Pointer to a pre-allocated variable the FS info shall be written
to.
\return \c B_OK if everything went fine, another error code otherwise. The
error code will be ignored, though.
*/
/*!
\fn status_t (*file_system_module_info::lookup)(fs_volume fs, fs_vnode dir,
const char *name, vnode_id *_id, int *_type)
\brief Looks up the node a directory entry refers to.
The VFS uses this hook to resolve path names to vnodes. It is used quite often
and should be implemented efficiently.
If the parameter \a dir does not specify a directory, the function shall fail.
It shall also fail, if it is a directory, but does not contain an entry with
the given name \a name. Otherwise the function shall invoke get_vnode() for
the node the entry refers to and pass back the ID and the type of the node
in \a _id and \a _type respectively.
Note that a directory must contain the special entries \c "." and \c "..",
referring to the same directory and the parent directory respectively.
lookup() must resolve the nodes accordingly. \c ".." for the root directory
of the volume shall be resolved to the root directory itself.
\param fs The volume handle.
\param dir The node handle of the directory.
\param name The name of the directory entry.
\param _id Pointer to a pre-allocated variable the ID of the found node
shall be written to.
\param _type Pointer to a pre-allocated variable the type of the found node
shall be written to. The type is encoded as in the \c st_mode field of a
<tt>struct stat</tt> (bitwise anded with \c S_IFMT).
\retval B_OK Everything went fine.
\retval B_ENTRY_NOT_FOUND The given directory does not contain an entry with
the given name.
*/
/*!
\fn status_t (*file_system_module_info::get_vnode)(fs_volume fs, vnode_id id,
fs_vnode *_vnode, bool reenter)
\brief Creates the private data handle to be associated with the node referred
to by \a id.
Invoked by the VFS when it creates the vnode for the respective node.
\param fs The volume handle.
\param id The ID of the node.
\param _vnode Pointer to a pre-allocated variable the node handle shall be
written to.
\param reenter \c true if the hook invocation has been caused by the FS itself,
e.g. by invoking ::get_vnode().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn \fn status_t (*file_system_module_info::put_vnode)(fs_volume fs,
fs_vnode vnode, bool reenter)
\brief Deletes the private data handle associated with the specified node.
Invoked by the VFS when it deletes the vnode for the respective node and the
node is not marked removed.
\param fs The volume handle.
\param vnode The node handle.
\param reenter \c true if the hook invocation has been caused by the FS itself,
e.g. by invoking ::put_vnode().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::remove_vnode)(fs_volume fs,
fs_vnode vnode, bool reenter)
\brief Deletes the private data handle associated with the specified node.
Invoked by the VFS when it deletes the vnode for the respective node and the
node is marked removed.
\param fs The volume handle.
\param vnode The node handle.
\param reenter \c true if the hook invocation has been caused by the FS itself,
e.g. by invoking ::put_vnode().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::read_symlink)(fs_volume fs,
fs_vnode link, char *buffer, size_t *_bufferSize)
\brief Reads the value of a symlink.
If the function is successful, the string written to the buffer shall be
null-terminated and the variable \a _bufferSize points to shall be set to
the length of that string, including the terminating null character.
\param fs The volume handle.
\param link The node handle.
\param buffer Pointer to a pre-allocated buffer the link value shall be written
to.
\param buffer Pointer to a pre-allocated variable containing the size of the
buffer supplied to the function. Upon successful completion the hook shall
store the number of bytes actually written into the buffer in the variable.
\retval B_OK Everything went fine.
\retval B_BAD_VALUE \a link does not identify a symbolic link.
\retval B_BUFFER_OVERFLOW The supplied buffer is not big enough to contain the
complete link value.
*/
/*!
\fn status_t (*file_system_module_info::access)(fs_volume fs, fs_vnode vnode,
int mode)
\brief Checks whether the current user is allowed to access the node in the
specified way.
\a mode is a bitwise combination of:
- \c R_OK: Read access.
- \c W_OK: Write access.
- \c X_OK: Execution.
If the current user does not have any of the access permissions represented
by the set bits, the function shall return \c B_NOT_ALLOWED. As a special case,
if the volume is read-only and write access is requested,
\c B_READ_ONLY_DEVICE shall be returned. If the requested access mode complies
with the user's access permissions, the function shall return \c B_OK.
For most FSs the permissions a user has are defined by the \c st_mode,
\c st_uid, and \c st_gid fields of the node's stat data. As a special
exception, the root user (<tt>geteuid() == 0<\tt>) does always have
read and write permissions, execution permission only when at least one of the
execution permission bits are set.
\param fs The volume handle.
\param vnode The node handle.
\param mode The access mode mask.
\retval B_OK The user has the permissions to access the node in the requested
way.
\retval B_READ_ONLY_DEVICE The volume is read-only, but the write access has
been requested.
\retval B_NOT_ALLOWED The user does not have all permissions to access the node
in the requested way.
*/
/*!
\fn status_t (*file_system_module_info::read_stat)(fs_volume fs,
fs_vnode vnode, struct stat *stat)
\brief Retrieves the stat data for a given node.
All values of the <tt>struct stat</tt> save \c st_dev, \c st_ino, \c st_rdev,
and \c st_type need to be filled in.
\param fs The volume handle.
\param vnode The node handle.
\param stat Pointer to a pre-allocated variable the stat data shall be written
to.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::open)(fs_volume fs, fs_vnode vnode,
int openMode, fs_cookie *_cookie)
\brief Opens the given node.
The function shall check whether it is possible to open the node according to
the mode specified by \c openMode (also considering the user's access
permissions), create a node cookie, and store it in the variable
\a _cookie points to.
The open mode \a openMode is encoded in the same way as the parameter of the
POSIX function \c open(), i.e. it is either \c O_RDONLY, \c O_WRONLY, or
\c O_RDWR, bitwise or'ed with flags. The only relevant flags for this hook are
\c O_TRUNC and \c O_NONBLOCK.
\param fs The volume handle.
\param vnode The node handle.
\param openMode The open mode.
\param _cookie Pointer to a pre-allocated variable the node cookie shall be
written to.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::close)(fs_volume fs, fs_vnode vnode,
fs_cookie cookie)
\brief Closes the given node cookie.
The hook is invoked, when closing the node has been requested. At this point
other threads might still use the cookie, i.e. still execute hooks to which
the cookie has been passed. If the FS supports blocking I/O operations, this
hook should make sure to unblock all currently blocking threads performing
an operation using the cookie, and mark the cookie such that no further
threads will block using it.
For many FSs this hook is a no-op.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The node cookie as returned by open().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::free_cookie)(fs_volume fs,
fs_vnode vnode, fs_cookie cookie)
\brief Frees the given node cookie.
The hook is invoked after close(), when no other thread uses or is going to
use the cookie. All resources associated with the cookie must be freed.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The node cookie as returned by open().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::read)(fs_volume fs, fs_vnode vnode,
fs_cookie cookie, off_t pos, void *buffer, size_t *length)
\brief Reads data from a file.
The function shall failed if
- the node is not a file,
- the cookie has not been opened for reading,
- \a pos is negative, or
- some other error occurs while trying to read the data, and no data have been
read at all.
The number of bytes to be read is stored in the variable pointed to by
\a length. If less data are available at file position \a pos, or if \a pos
if greater than the size of the file, only as many data as available shall
be read, the function shall store the number of bytes actually read into the
variable pointed to by \a length, and return \c B_OK.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The node cookie as returned by open().
\param pos The file position where to start reading data.
\param buffer Pointer to a pre-allocated buffer the read data shall be
written to.
\param length Pointer to a pre-allocated variable containing the size of the
buffer when invoked, and into which the size of the data actually read
shall be written.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::open_dir)(fs_volume fs, fs_vnode vnode,
fs_cookie *_cookie)
\brief Opens the given directory node.
If the specified node is not a directory, or if the current user does not have
the permissions to read the directory, the function shall fail. Otherwise it
shall allocate a directory cookie and store it in the variable \a _cookie
points to. A subsequent read_dir() using the cookie shall start reading the
first entry of the directory.
\param fs The volume handle.
\param vnode The node handle.
\param _cookie Pointer to a pre-allocated variable the directory cookie shall
be written to.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::close_dir)(fs_volume fs,
fs_vnode vnode, fs_cookie cookie)
\brief Closes the given directory cookie.
Generally the situation is similar to the one described for close(). In
practice it is a bit, though, since directory cookies are exclusively used
for directory iteration, and it normally doesn't make sense to have multiple
threads read the same directory concurrently. Furthermore reading a directory
should not block. Therefore for most FSs this hook is a no-op.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The directory cookie as returned by open_dir().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::free_dir_cookie)(fs_volume fs,
fs_vnode vnode, fs_cookie cookie)
\brief Frees the given directory cookie.
The hook is invoked after close_dir(), when no other thread uses or is going
to use the cookie. All resources associated with the cookie must be freed.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The directory cookie as returned by open_dir().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::read_dir)(fs_volume fs, fs_vnode vnode,
fs_cookie cookie, struct dirent *buffer, size_t bufferSize, uint32 *_num)
\brief Reads the next one or more directory entries.
The number of entries to be read at maximum is stored in the variable \a _num
points to.
Per read \c dirent the following fields have to be filled in:
- \c d_dev: The volume ID.
- \c d_ino: The ID of the node the entry refers to.
- \c d_name: The null-terminated name of the entry.
- \c d_reclen: The size of the \c dirent structure in bytes, starting from
the beginning of the structure, counting all bytes up to and including
the null-termination char of the name stored in \c d_name.
If more than one entry is read, the corresponding \c dirent structures are
tightly packed, i.e. the second entry begins directly after the end of the
first one (i.e. \c d_reclen bytes after the beginning of the first one).
Most FSs read only one entry at a time though, even if more are requested.
When the function is invoked after the end of the directory has been reached,
it shall set the variable \a _num points to to \c 0 and return \c B_OK. If the
provided buffer is too small to contain even the single next entry,
\c B_BUFFER_OVERFLOW shall be returned. It shall not fail, if at least one
entry has been read, and the buffer is just too small to hold as many entries
as requested.
Note that a directory is expected to contain the special entries \c "." and
\c "..", referring to the same directory and the parent directory respectively.
The \c dirent structure returned for the \c ".." entry of the volume's root
directory shall refer to the root node itself.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The directory cookie as returned by open_dir().
\param buffer Pointer to a pre-allocated buffer the directory entries shall be
written to.
\param bufferSize The size of \a buffer in bytes.
\param _num Pointer to a pre-allocated variable, when invoked, containing the
number of directory entries to be read, and into which the number of
entries actually read shall be written.
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::rewind_dir)(fs_volume fs,
fs_vnode vnode, fs_cookie cookie)
\brief Resets the directory cookie to the first entry of the directory.
\param fs The volume handle.
\param vnode The node handle.
\param cookie The directory cookie as returned by open_dir().
\return \c B_OK if everything went fine, another error code otherwise.
*/
/*!
\fn status_t (*file_system_module_info::read_query)(fs_volume fs,
fs_cookie cookie, struct dirent *buffer, size_t bufferSize, uint32 *_num)
\brief Reads the next one or more entries matching the query.
This hook function works pretty much the same way as read_dir(), with the
difference that it doesn't read the entries of a directory, but the entries
matching the given query.
\param fs The volume handle.
\param cookie The query cookie as returned by open_query().
\param buffer Pointer to a pre-allocated buffer the directory entries shall be
written to.
\param bufferSize The size of \a buffer in bytes.
\param _num Pointer to a pre-allocated variable, when invoked, containing the
number of entries to be read, and into which the number of entries
actually read shall be written.
\return \c B_OK if everything went fine, another error code otherwise.
*/