.\"     $NetBSD: ukfs.3,v 1.10 2009/11/22 18:14:49 pooka Exp $
.\"
.\" Copyright (c) 2008 Antti Kantee.  All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.Dd November 22, 2009
.Dt UKFS 3
.Os
.Sh NAME
.Nm ukfs
.Nd user kernel file system library interface
.Sh LIBRARY
ukfs Library (libukfs, \-lukfs)
.Sh SYNOPSIS
.In rump/ukfs.h
.Sh DESCRIPTION
The
.Nm
library provides direct access to file systems without having to
specially mount a file system.
Therefore, accessing a file system through
.Nm
requires no special kernel support apart from standard POSIX functionality.
As
.Nm
is built upon
.Xr rump 3 ,
all kernel file systems which are supported by rump are available.
It allows to write utilities for accessing file systems without having
to duplicate file system internals knowledge already present in kernel
file system drivers.
.Pp
.Nm
provides a high-level pathname based interface for accessing file systems.
If a lower level interface it desired,
.Xr rump 3
should be used directly.
However, much like system calls, the interfaces of
.Nm ,
are self-contained and require no tracking and release of resources.
The only exception is the file system handle
.Ft struct ukfs
which should be released after use.
.Sh INITIALIZATION
.Bl -ohang
.It Ft int
.Fn ukfs_init
.It Ft int
.Fn ukfs_modload "const char *fname"
.It Ft int
.Fn ukfs_modload_dir "const char *dirname"
.It Ft ssize_t
.Fn ukfs_vfstypes "char *buf" "size_t buflen"
.It Ft struct ukfs *
.Fn ukfs_mount "const char *vfsname" "const char *devpath" \
"const char *mountpath"  "int mntflags" "void *arg" "size_t alen"
.It Ft struct ukfs *
.Fn ukfs_mount_disk "const char *vfsname" "const char *devpath" \
"int partition" "const char *mountpath"  "int mntflags" \
"void *arg" "size_t alen"
.It Ft int
.Fn ukfs_release "struct ukfs *ukfs" "int flags"
.El
.Pp
.Fn ukfs_init
intializes the library and must be called once per process using
.Nm .
.Pp
.Fn ukfs_modload
is used at runtime to dynamically load a library which contains a
file system module.
For this to succeed, the
.Xr rump 3
library and the module targetted must be compiled with compatible kernel
versions and the application must be dynamically linked.
Additionally, since this routine does not handle dependencies, all the
dependencies of the library must be loaded beforehand.
The routine returns \-1 for fatal error, 0 for dependency failure and 1
for success.
.Pp
.Fn ukfs_modload_dir
loads all
.Xr rump 3
file system modules in directory
.Fa dirname .
It looks for libraries which begin with
.Pa librumpfs_
and end in
.Pa .so .
The routine tries to handle dependencies by retrying to load libraries
which failed due to dependencies.
.Fn ukfs_modload_dir
returns the number of vfs modules loaded or sets errno and
returns \-1 in case of a fatal error in directory searching.
In case a fatal error occurs after some modules have already been
loaded, the number of loaded module is returned.
Fatal errors in loading the modules themselves are ignored and
.Fn ukfs_modload
should be used directly if finegrained error reporting is desired.
.Pp
It should be noted that the above routines affect the whole process,
not just a specific instance of
.Nm .
It is preferable to call them from only one thread, as the underlying
dynamic library interfaces may not be threadsafe.
.Pp
.Fn ukfs_vfstypes
queries the available file system types and returns a nul-terminated
list of types separated by spaces in
.Fa buf .
The format of the list is equivalent to the one returned by
.Xr sysctl 3
on the name
.Pa vfs.generic.fstypes .
The function returns the length of the string without the trailing nul
or \-1 for error.
Notably, the return value 0 means there are no file systems available.
If there is not enough room in the caller's buffer for all file system
types, as many as fit will be returned.
.Pp
.Fn ukfs_mount
intializes a file system image.
The handle resulting from the operation is passed to all other routines
and identifies the instance of the mount analoguous to what a pathname
specifies in a normally mounted file system.
The parameters are the following:
.Bl -tag -width XXX -offset indent
.It vfsname
Name of the file system to be used, e.g.
.Dv MOUNT_FFS .
.It devpath
Path of file system image.
It can be either a regular file, device or, if the file system does
not support the concept of a device, an abitrary string, e.g. network
address.
.It mountpath
Path where the file system is mounted to.
This parameter is used only by the file system being mounted.
Most of the time
.Dv UKFS_DEFAULTMP
is the correct path.
.It mntflags
Flags as passed to the
.Xr mount 2
system call, for example
.Dv MNT_RDONLY .
In addition to generic parameters, file system specific parameters such as
.Dv MNT_LOG
(ffs) may be passed here.
.It arg
File system private argument structure.
This is passed directly to the file system.
It must match what
.Fa vfsname
expects.
.It alen
Size of said structure.
.El
.Pp
The
.Fn ukfs_mount_disk
function must be used to mount disk-based file systems.
It takes the same arguments as
.Fn ukfs_mount ,
except for an additional argument signifying the
.Fa partition
number.
If the image
.Fa devpath
contains a disklabel, this value specifies the number of the partition
within the image used as the file system backend.
If
.Fa devpath
does not contain a disklabel, the value
.Dv UKFS_PARTITION_NONE
must be used to signal that the file system backend is the entire
image.
.Pp
.Fn ukfs_release
unmounts the file system and releases the resources associated with
.Fa ukfs .
The return value signals the return value of the unmount operation.
If non-zero,
.Fa ukfs
will continue to remain valid.
The possible values for flags are:
.Bl -tag -width XUKFS_RELFLAG_NOUNMOUT -offset indent
.It Dv UKFS_RELFLAG_NOUNMOUNT
Do not unmount file system, just release ukfs handle.
Release always succeeds.
.It Dv UKFS_RELFLAG_FORCE
Forcefully unmount the file system.
This means that any busy nodes (due to e.g.
.Fn ukfs_chdir )
will be ignored.
Release always succeeds.
.El
.Sh OPERATION
.Bl -ohang
.It Ft int
.Fn ukfs_chdir "struct ukfs *ukfs" "const char *path"
.It Ft int
.Fn ukfs_getdents "struct ukfs *ukfs" "const char *dirname" "off_t *off" \
"uint8_t *buf" "size_t bufsize"
.It Ft ssize_t
.Fn ukfs_read "struct ukfs *ukfs" "const char *filename" "off_t off" \
"uint8_t *buf" "size_t bufsize"
.It Ft ssize_t
.Fn ukfs_write "struct ukfs *ukfs" "const char *filename" "off_t off" \
"uint8_t *buf" "size_t bufsize"
.It Ft int
.Fn ukfs_create "struct ukfs *ukfs" "const char *filename" "mode_t mode"
.It Ft int
.Fn ukfs_mknod "struct ukfs *ukfs" "const char *path" "mode_t mode" "dev_t dev"
.It Ft int
.Fn ukfs_mkfifo "struct ukfs *ukfs" "const char *path" "mode_t mode"
.It Ft int
.Fn ukfs_mkdir "struct ukfs *ukfs" "const char *filename" "mode_t mode"
.It Ft int
.Fn ukfs_remove "struct ukfs *ukfs" "const char *filename"
.It Ft int
.Fn ukfs_rmdir "struct ukfs *ukfs" "const char *filename"
.It Ft int
.Fn ukfs_link "struct ukfs *ukfs" "const char *filename" "const char *f_create"
.It Ft int
.Fn ukfs_symlink "struct ukfs *ukfs" "const char *filename" \
"const char *linkname"
.It Ft ssize_t
.Fn ukfs_readlink "struct ukfs *ukfs" "const char *filename" \
"char *linkbuf" "size_t buflen"
.It Ft int
.Fn ukfs_rename "struct ukfs *ukfs" "const char *from" "const char *to"
.It Ft int
.Fn ukfs_stat "struct ukfs *ukfs" "const char *filename" \
"struct stat *file_stat"
.It Ft int
.Fn ukfs_lstat "struct ukfs *ukfs" "const char *filename" \
"struct stat *file_stat"
.It Ft int
.Fn ukfs_chmod "struct ukfs *ukfs" "const char *filename" "mode_t mode"
.It Ft int
.Fn ukfs_lchmod "struct ukfs *ukfs" "const char *filename" "mode_t mode"
.It Ft int
.Fn ukfs_chown "struct ukfs *ukfs" "const char *filename" "uid_t uid" \
"gid_t gid"
.It Ft int
.Fn ukfs_lchown "struct ukfs *ukfs" "const char *filename" "uid_t uid" \
"gid_t gid"
.It Ft int
.Fn ukfs_chflags "struct ukfs *ukfs" "const char *filename" "u_long flags"
.It Ft int
.Fn ukfs_lchflags "struct ukfs *ukfs" "const char *filename" "u_long flags"
.It Ft int
.Fn ukfs_utimes "struct ukfs *ukfs" "const char *filename" \
"const struct timeval *tptr"
.It Ft int
.Fn ukfs_lutimes "struct ukfs *ukfs" "const char *filename" \
"const struct timeval *tptr"
.El
.Pp
The above routines operate like their system call counterparts and the
system call manual pages without the ukfs_ prefix should be referred to
for further information on the parameters.
.Pp
The only call which modifies
.Fa ukfs
state is
.Fn ukfs_chdir .
It works like
.Xr chdir 2
in the sense that it affects the interpretation of relative paths.
If succesful, all relative pathnames will be resolved starting from the
current directory.
Currently the call affects all accesses to that particular
.Fa ,
but it might be later changed to be thread private.
.Sh UTILITIES
.Bl -ohang
.It Ft int
.Fn ukfs_util_builddirs "struct ukfs *ukfs" "const char *pathname" "mode_t mode"
.El
.Pp
Builds a directory hierarchy.
Unlike mkdir, the
.Fa pathname
argument may contain multiple levels of hierarchy.
It is not considered an error if any of the directories specified exist
already.
.Sh SEE ALSO
.Xr rump 3
.Sh HISTORY
.Nm
first appeared in
.Nx 5.0 .
.Sh AUTHORS
.An Antti Kantee Aq pooka@cs.hut.fi
.Sh NOTES
.Nm
should be considered experimental technology and may change without warning.
.Sh BUGS
On Linux, dynamically linked binaries can include support for only
one file system due to restrictions with the dynamic linker.
If more are desired, they must be loaded at runtime using
.Fn ukfs_modload .
Even though
.Nx
does not have this restriction, portable programs should load all
file system drivers dynamically.