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 November 30, 2006 .Dt PUFFS 3 .Os .Sh NAME .Nm libpuffs .Nd Pass-to-Userspace Framework File System development interface .Sh LIBRARY .Lb libpuffs .Sh SYNOPSIS n 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.
p .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.
p 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 PUFFSFLAG_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.
p 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.
p 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.
p .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.
p 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.
p All callbacks can be prototyped with the file system name and operation name using the macro .Fn PUFFSVFS_PROTOS fsname .
p .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 .
p .Fn puffs_vfs_statvfs should fill in the following fields of .Fa sbp : d -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 .
p 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.
p 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.
p The callbacks can be prototyped according to file system name by using the macro .Fn PUFFSVN_PROTOS fsname .
p 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 .
p .Fn puffs_vn_getattr fills out a struct vattr pointed to by .Fa va .
p .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.
p 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.
p .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 .
p 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.
p 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).
p 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.
p 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.
p 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 .
p 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 .
p 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.
p .Fn puffs_vn_close releases all the resources allocated by .Fn puffs_vn_open .
p 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 .
p .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.
p .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.
p .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.
p .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.
p Operations which require arbitrary blocking periods are not well handled by the library in this version.