NetBSD/lib/libpuffs/puffs.3
2006-12-01 15:31:25 +00:00

548 lines
17 KiB
Groff

.\" $NetBSD: puffs.3,v 1.7 2006/12/01 15:31:25 pooka Exp $
.\"
.\" Copyright (c) 2006 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 December 1, 2006
.Dt PUFFS 3
.Os
.Sh NAME
.Nm libpuffs
.Nd Pass-to-Userspace Framework File System development interface
.Sh LIBRARY
.Lb libpuffs
.Sh SYNOPSIS
.In puffs.h
.Ft struct puffs_usermount *
.Fo puffs_mount
.Fa "struct puffs_vfsops *pvfs" "struct puffs_vnops *pvn"
.Fa "const char *dir" "int mntflags" "const char *puffsname"
.Fa "uint32_t pflags" "size_t maxreqlen"
.Fc
.Ft int
.Fn puffs_mainloop "struct puffs_usermount *pu" "int flags"
.Ft int
.Fn puffs_oneop "struct puffs_usermount *pu" "uint8_t *buf" "size_t buflen"
.Ft int
.Fn puffs_getselectable "struct puffs_usermount *pu"
.Ft int
.Fn puffs_setblockingmode "struct puffs_usermount *pu" "int mode"
.Ft int
.Fn puffs_vfs_mount "struct puffs_usermount *pu" "void **rootcookie"
.Ft int
.Fn puffs_vfs_unmount "struct puffs_usermount *pu" "int flags" "pid_t pid"
.Ft int
.Fo puffs_vfs_statvfs
.Fa "struct puffs_usermount *pu" "struct statvfs *sbp" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vfs_sync
.Fa "struct puffs_usermount *pu" "int waitfor" "const struct puffs_cred *cred"
.Fa "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_lookup
.Fa "struct puffs_usermount *pu" "void *opc" "void **newnode"
.Fa "enum vtype *newtype" "voff_t *newsize" "dev_t *newrdev"
.Fa "const struct puffs_cn *pcn"
.Fc
.Ft int
.Fo puffs_vn_getattr
.Fa "struct puffs_usermount *pu" "void *opc" "struct vattr *va"
.Fa "const struct puffs_cred *pcr" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_setattr
.Fa "struct puffs_usermount *pu" "void *opc" "const struct vattr *va"
.Fa "const struct puffs_cred *pcr" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_create
.Fa "struct puffs_usermount *pu" "void *opc" "void **newnode"
.Fa "const struct puffs_cn *pcn" "const struct vattr *va"
.Fc
.Ft int
.Fo puffs_vn_remove
.Fa "struct puffs_usermount *pu" "void *opc" "void *targ"
.Fa "const struct puffs_cn *pcn"
.Fc
.Ft int
.Fo puffs_vn_mkdir
.Fa "struct puffs_usermount *pu" "void *opc" "void **newnode"
.Fa "const struct puffs_cn *pcn" "const struct vattr *va"
.Fc
.Ft int
.Fo puffs_vn_rmdir
.Fa "struct puffs_usermount *pu" "void *opc" "void *targ"
.Fa "const struct puffs_cn *pcn"
.Fc
.Ft int
.Fo puffs_vn_readdir
.Fa "struct puffs_usermount *pu" "void *opc" "struct dirent *dent"
.Fa "const struct puffs_cred *pcr" "off_t *readoff" "size_t *reslen"
.Fc
.Ft int
.Fo puffs_vn_rename
.Fa "struct puffs_usermount *pu" "void *opc" "void *src"
.Fa "const struct puffs_cn *pcn_src" "void *targ_dir" "void *targ"
.Fa "const struct puffs_cn *pcn_targ"
.Fc
.Ft int
.Fo puffs_vn_link
.Fa "struct puffs_usermount *pu" "void *opc" "void *targ"
.Fa "const struct puffs_cn *pcn"
.Fc
.Ft int
.Fo puffs_vn_symlink
.Fa "struct puffs_usermount *pu" "void *opc" "void **newnode"
.Fa "const struct puffs_cn *pcn_src" "const struct vattr *va"
.Fa "const char *link_target"
.Fc
.Ft int
.Fo puffs_vn_readlink
.Fa "struct puffs_usermount *pu" "void *opc" "const struct puffs_cred *cred"
.Fa "char *link" "size_t *linklen"
.Fc
.Ft int
.Fo puffs_vn_mknod
.Fa "struct puffs_usermount *pu" "void *opc" "void **newnode"
.Fa "const struct puffs_cn *pcn" "const struct vattr *va"
.Fc
.Ft int
.Fo puffs_vn_open
.Fa "struct puffs_usermount *pu" "void *opc" "int flags"
.Fa "const struct puffs_cred *pcr" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_close
.Fa "struct puffs_usermount *pu" "void *opc" "int flags"
.Fa "const struct puffs_cred *pcr" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_access
.Fa "struct puffs_usermount *pu" "void *opc" "int mode"
.Fa "struct puffs_cred *pcr" "pid_t pid"
.Fc
.Ft int
.Fo puffs_vn_read
.Fa "struct puffs_usermount *pu" "void *opc" "uint8_t *buf"
.Fa "off_t offset" "size_t *resid" "const struct puffs_cred *pcr" "int ioflag"
.Fc
.Ft int
.Fo puffs_vn_write
.Fa "struct puffs_usermount *pu" "void *opc" "uint8_t *buf"
.Fa "off_t offset" "size_t *resid" "const struct puffs_cred *pcr" "int ioflag"
.Fc
.Ft int
.Fn puffs_vn_reclaim "struct puffs_usermount *pu" "void *opc" "pid_t pid"
.Ft int
.Fo puffs_vn_inactive
.Fa "struct puffs_usermount *pu" "void *opc" "pid_t pid" "int *refcount"
.Fc
.Sh DESCRIPTION
.Em IMPORTANT NOTE!
This document describes interfaces which are not yet guaranteed to be
stable.
In case you update your system sources, please recompile everything
and fix complation errors.
If your sources are out-of-sync, incorrect operation may result.
The interfaces in this document will most likely be hugely simplified
in later versions or made transparent to the implementation.
.Pp
.Nm
provides a framework for creating file systems as userspace servers.
Operations are transported from the kernel virtual file system layer
to the concrete implementation behind
.Nm ,
where they are processed and results are sent back to the kernel.
.Pp
It is possible to use
.Nm
in two different ways.
Calling
.Fn puffs_mainloop
takes execution context away from the caller and automatically handles
all requests by using the callbacks.
Alternatively, control can be kept with the caller and operations
handled manually.
In the latter case the function
.Fn puffs_oneop
can be helpful.
.Ss Library operation
The file system is mounted using
.Fn puffs_vfs_mount .
The callbacks are passed as the fields in the structures
.Fa pvfs
and
.Fa pvn .
The ones intended to be used should initialized to point to the correct
routines before calling and the rest should be set to zero.
All of the VFS routines are mandatory, but all of the node operations
with the exception of
.Fn puffs_vfs_lookup
are optional.
However, leaving operations blank will naturally have an effect on the
features available from the file system implementation.
The argument
.Fa dir
signifies the mount point,
.Fa mntflags
is the flagset given to
.Xr mount 2 ,
and
.Fa puffsname
is the name of the file system implementation.
Flags for
.Nm
can be given via
.Fa pflags .
Currently
.Dv PUFFS_FLAG_OPDUMP
is supported, this dumps each received operation to stdout before handling it.
Finally, the maximum operation buffer length is requested by
.Fa maxreqlen .
The field
.Va pu_maxreqlen
from the returned mount structure is the kernel sanity-checked value and
should always be consulted after the mount call returns.
Supplying 0 as this parameter will make the kernel choose the longest
possible buffer length.
In case of success,
.Fn puffs_vfs_mount
returns the address of the user mount instance.
Otherwise,
.Dv NULL
is returned and errno is set to specify the error.
.Pp
To handle all requests automatically until the file system is
unmounted,
.Fn puffs_mainloop
should be used.
It returns 0 if the file system was succesfully unmounted or \-1 if it
was killed in action.
Unless
.Fa flags
is used to pass
.Dv PUFFSLOOP_NODAEMON ,
.Fn puffs_mainloop
will also detach from the terminal.
.Pp
To handle a single operation,
.Fn puffs_oneop
can be used.
The buffer for the request should be supplied by the caller and, if
possible, should match the buffer length gotten from mount.
In case the request was succefully handled (orthogonal to if the
request itself was a success from the file system point of view),
0 is returned.
Otherwise, \-1 is returned and errno is set.
.Pp
.Fn puffs_getselectable
can be used to query a handle to do I/O multiplexing with:
.Xr select 2 ,
.Xr poll 2 ,
and
.Xr kqueue 2
are all examples of acceptable operations.
.Pp
The library can be set in blocking or non-blocking mode using
.Fn puffs_setblockingmode .
Acceptable values for the argument are
.Dv PUFFSDEV_BLOCK
and
.Dv PUFFSDEV_NONBLOCK .
.Ss Cookies
Every file (regular file, directory, device node, ...) instance is
attached to the kernel using a cookie.
A cookie should uniquely map to a file during its lifetime.
If file instances are kept in memory, a simple strategy is to use
the virtual address of the structure describing the file.
The cookie can be recycled when
.Fn puffs_vn_reclaim
is called for a node.
.Ss File system callbacks
The callbacks do all the actual work in implementing the file system.
Currently they are fairly close to the vfs and vnode operations in
the kernel but with simplified operation.
This section describes the calls which relate to the file system
itself.
.Pp
All callbacks can be prototyped with the file system name and operation
name using the macro
.Fn PUFFSVFS_PROTOS fsname .
.Pp
.Fn puffs_vfs_mount
should handle all operations which are necessary to mount the file
system, e.g. open backing storage, check magic numbers, open a network
connection, authenticate, etc.
It returns the file system root directory cookie in
.Fa rootcookie .
.Pp
.Fn puffs_vfs_statvfs
should fill in the following fields of
.Fa sbp :
.Bd -literal
* unsigned long f_bsize; file system block size
* unsigned long f_frsize; fundamental file system block size
* fsblkcnt_t f_blocks; number of blocks in file system,
* (in units of f_frsize)
*
* fsblkcnt_t f_bfree; free blocks avail in file system
* fsblkcnt_t f_bavail; free blocks avail to non-root
* fsblkcnt_t f_bresvd; blocks reserved for root
*
* fsfilcnt_t f_files; total file nodes in file system
* fsfilcnt_t f_ffree; free file nodes in file system
* fsfilcnt_t f_favail; free file nodes avail to non-root
* fsfilcnt_t f_fresvd; file nodes reserved for root
.Ed
The process requiring this information is given by
.Fa pid .
.Pp
The file system should be sychronized to storage when
.Fn puffs_vfs_sync
is called.
The
.Fa waitfor
parameter should handled similarly as inside the kernel.
.Pp
The file system should be unmounted when
.Fn puffs_vfs_unmount
is called.
If the flag
.Dv MNT_FORCE
is not honored, the kernel will proceed to forcibly unmount the file system
despite this.
.Ss Node callbacks
These operations operate in the level of individual files.
The file cookie is always provided as the second argument
.Fa opc .
If the operation is for a file, it will be the cookie of the file.
The case the operation involves a directory (such as
.Dq create file in directory ) ,
the cookie will be for the directory.
Some operations take additional cookies to describe the rest of
the operands.
The return value 0 signals success, else an appropriate errno value
should be returned.
Please note that neither this list nor the descriptions are complete.
.Pp
The callbacks can be prototyped according to file system name by
using the macro
.Fn PUFFSVN_PROTOS fsname .
.Pp
The
.Fn puffs_vn_lookup
function is used to locate nodes.
The implementation should match the name in
.Fa pcn
against the existing entries in the directory provided by the cookie.
If found, the cookie for the located node should be returned in
.Fa newnode .
Additionally, the type and size (latter applicable to regular files only)
should be returned in
.Fa newtype
and
.Fa newsize ,
respectively.
If the located entry is a block device or character device file,
the dev_t for the entry should be returned in
.Fa newrdev .
Otherwise, 0 signals a found node and a nonzero value signals an errno.
As a special case,
.Er ENOENT
signals success for cases where the lookup operation is
.Dv PUFFSLOOKUP_CREATE
or
.Dv PUFFSLOOKUP_RENAME .
Failure in these cases can be signalled by returning another appropriate
error code, for example
.Er EACCESS .
.Pp
.Fn puffs_vn_getattr
fills out a struct vattr pointed to by
.Fa va .
.Pp
.Fn puffs_vn_setattr
sets the attributes in
.Fa va .
Instead of setting everything according to that file, only fields which
are not marked
.Dv VNOVAL
should be set.
.Pp
A file node is created in the directory specified by the cookie when
.Fn puffs_vn_create
is called.
The attributes are specified by
.Fa va
and the cookie for the newly created node should be returned in
.Fa newnode .
Similarly,
.Fn puffs_vn_mkdir
creates a directory.
.Pp
.Fn puffs_vn_remove
removes the file
.Fa targ
from the directory indicated by the cookie.
Similarly,
.Fn puffs_vn_rmdir
removes a directory.
The name of the directory entry to remove is described by
.Fa pcn .
.Pp
To read directory entries,
.Fn puffs_vn_readdir
is called.
It should store directories as struct dirents in the space pointed to by
.Fa dent .
The amount of space available is given by
.Fa reslen
and before returning it should be set to the amount of space
.Em remaining
in the buffer.
The argument
.Fa offset
is used to specify the offset to the directory.
Its intepretation is up to the file systme and it should be set to
signal the continuation point when there is no more room for the next
entry in
.Fa dent .
It is most performant to return the maximal amount of directory
entries each call.
In case the directory was exhausted, the parameters should not be
modified to signal end-of-directory.
.Pp
A node rename is done by calling
.Fn puffs_vn_rename .
If the destination file cookie is non-null, it must be removed
and the new entry overwritten atomically.
The directory entry names to be used are described by the
struct puffs_cn's (cf. create and remove).
.Pp
A hard link is created by
.Fn puffs_vn_link .
In practice this means adding a directory entry described by
.Fa pcn
to the cookied directory and the entry pointing to the target node.
.Pp
A symbolic link in turn is created by
.Fn puffs_vn_symlink .
It is similar to creating a regular file, except that
.Fa link_target
specifies the target of the link which should be set for the link.
.Pp
To read the target of a symbolic link,
.Fa puffs_vn_readlink
is called.
The path in the link target should be copied to
.Fa link
and the length without the terminating nul set in
.Fa linklen .
.Pp
A device node is created using
.Fn puffs_vn_mknod .
The only difference to creating a normal file is that the attribute
struct contains the device identifier in
.Fa va-\*[Gt]va_rdev .
.Pp
Files are opened with a call to
.Fn puffs_vn_open .
Most of the time this can be left unimplemented, unless special
resource allocation is required.
.Pp
.Fn puffs_vn_close
releases all the resources allocated by
.Fn puffs_vn_open .
.Pp
To check if access of type
.Va mode
to a file is allowed,
.Fn puffs_vn_access
is called.
This controls file access, not e.g.
.Fn puffs_vn_open .
.Pp
.Fn puffs_vn_read
reads the contents of a file.
It will gather the data from
.Fa offset
in the file and read the number
.Fa resid
octets.
The buffer is guaranteed to have this much space.
The amount of data requested by
.Fa resid
should be read, except in the case of eof-of-file or an error.
The parameter
.Fa resid
should be set to indicate the amount of request NOT completed.
In the normal case this should be 0.
.Pp
.Fn puffs_vn_write
writes data to a file at
.Fa offset
extending the file if necessary.
The number of octets written is indicated by
.Fa resid ;
everything must be written or an error will be generated.
The parameter must be set to indicate the amount of data NOT written.
In case the flag
.Dv PUFFS_IO_APPEND
is specified, the data should be appended to the end of the file.
.Pp
.Fn puffs_vn_reclaim
signals that the cookie will no longer be referenced without a further
call to
.Fn puffs_vn_lookup .
This information can be used to free resources and specifically release
a file for which no directory entries remain.
.Pp
.Fn puffs_vn_inactive
signals that the kernel has released its last reference to the node.
However, the cookie must still remain valid until
.Fn puffs_vn_reclaim
is called.
The file system should return its internal reference count on the file
(usually number of links to the file) in
.Fa refcount .
If this is zero, the kernel will call reclaim immediately.
.Sh SEE ALSO
.Xr puffs 4
.Sh HISTORY
An unsupported experimental version of
.Nm
first appeared in
.Nx 4.0 .
.Sh AUTHORS
.An Antti Kantee Aq pooka@iki.fi
.Sh BUGS
struct puffs_node is of questionable content.
Especially the use of
.Va pn_va
should be avoided.
.Pp
Operations which require arbitrary blocking periods are not well handled
by the library in this version.