The FNS File System Namespace

Files may be named relative to users, hosts, organizations, and sites in FNS by appending the fs namespace identifier to the name of the object, and following this with the name of the file. For example, an engineering organization's tools directory might be named org/engineering/fs/tools.

The initial context is located under /xfn in the root directory. Thus a user might access the tools directory by typing

% cd /xfn/org/engineering/fs/tools

Existing applications can access this directory just as they would any other directory. Applications do not need to be modified in any way or use the XFN API.

NFS File Servers

NFS is Sun's distributed file system. The files associated with an object will generally reside on one or more remote NFS file servers. In the simplest case, the namespace identifier fs corresponds to the root of an exported NFS file system, as shown in Figure 16-1.

Figure 16-1

In contrast, an object's file system may be composed of multiple-- and possibly overlapping--remote mounts, woven together into a "virtual" directory structure managed by FNS.

Figure 16-2 on page 293 illustrates how this capability might be used to piece together an organization's file system from three separate file servers. The project directory, along with its lib subdirectory, resides on one file server, while the src subdirectory resides on another. Users and applications need not be aware of the use of multiple servers; they see a single, seamless namespace.

Figure 16-2

The Automounter

For efficiency, the automounter is used to mount FNS directories on demand. The default /etc/auto_master configuration file contains the line:

/xfn -xfn

which tells the automounter that the FNS namespace is "mounted" under /xfn, as specified by XFN.

Since the automounter is used to mount directories named through FNS, the subdirectories of an FNS directory cannot be listed until they have been mounted. For example, the command

% ls /xfn/org/engineering/fs

shows only those subdirectories that are currently mounted. To see the entire listing, use the fnlist command.

Creating File Contexts

The fncreate_fs command creates file contexts for organizations and sites. It may also be used to override the default file contexts for users and hosts that are created by the fncreate command.

There are two methods of using the fncreate_fs command. The context bindings may be provided by either the

• Input file - See "Creating the Input File" on page 295.
• Command line - See "Using Command-line Input" on page 297.

The two methods of fncreate_fs have the following syntax:

fncreate_fs -f input_file composite_name fncreate_fs [-v] [-r] composite_name [options][location...]

The fncreate_fs command-line options -v and -r are described in Table 16-1.

Table 16-1 fncreate

Option	Description
-v	Sets verbose output, displaying information about the contexts being created and modified.
-r	Replaces the bindings in the context named by composite_name-- and all of its subcontexts--with only those specified in the input. This is equivalent to destroying the context (and, recursively, its subcontexts), and then running fncreate_fs without this option. The -r option should be used with care.

The fncreate_fs command manipulates FNS contexts and bindings of the onc_fn_fs reference type. It uses an address of type onc_fn_fs_mount to represent each remote mount point. The data associated with an address of this type are the corresponding mount options and locations in a single string, XDR-encoded.

Creating the Input File

The input file supplies the names and values to be bound in the context of composite_name. Its format is based upon and similar, but not identical, to the format of indirect automount maps. The input file contains an entry with the form:

name [-options] [location...]

For each entry a reference to the location(s) and the corresponding options is bound to the name, composite_name/name.

The name field may be a simple atomic name or a slash-separated hierarchical name. It may also be "." (dot), in which case the reference is bound directly to composite_name.

The location field specifies the host or hosts that serve the files for composite_name/name. In a simple NFS mount, location takes the form:

host:path

where host is the name of the server from which to mount the file system, and path is the path name of the directory to mount.

The options field is a comma-separated list of the mount options to use when mounting the directory. These options also apply to any subcontexts of composite_name/name that do not specify mount options of their own.

If options and location are both omitted, then no reference is bound to composite_name/name. Any existing reference is unbound.

Using the example from Figure 16-1 on page 292, suppose you want jsmith's file system to be an NFS mount of the directory /export/home/jsmith from host svr1. The command would be run as follows:

% fncreate_fs -f infile user/jsmith/fs

with infile containing

. svr1:/export/home/jsmith

To set up the file system illustrated in Figure 16-2 on page 293, run the command

% fncreate_fs -f infile org/engineering/fs

with infile containing

tools/db svr1:/export/db project svr1:/export/proj project/src svr2:/export/src

To change the NFS mounts for project and its subcontext to be read-only, you can change infile as follows:

tools/db svr1:/export/db project -ro svr1:/export/proj project/src svr2:/export/src

The -ro is unnecessary in the third line. Since src is a subcontext of project, it will inherit the -ro mount option from above.

The following input file would make all of the mounts read-only except for org/engineering/fs/project/src.

. -ro tools/db svr1:/export/db project svr1:/export/proj project/src -rw svr2:/export/src

Using Command-line Input

The fncreate_fs command also allows the binding description to be provided on the command line:

fncreate_fs composite_name [options] [location ...]

This is equivalent to using the first form of the command and providing a one-line input file containing "." in the name field, and the given mount options and locations. The previous example in which jsmith's file system was set could be set from the command line as follows:

% fncreate_fs user/jsmith/fs svr1:/export/home/jsmith

Similarly, the hierarchy in Figure 16-2 on page 293 could have been set up by running the sequence of commands:

% fncreate_fs org/engineering/fs/tools/db svr1:/export/db % fncreate_fs org/engineering/fs/project svr1:/export/proj % fncreate_fs org/engineering/fs/project/src svr2:/export/src

To make all three of the mounts read-only, you would run this command:

% fncreate_fs org/engineering/fs -ro

Advanced Input Formats

The following two sections apply to both input file and command-line input formats.

Multiple Locations

Multiple location fields may be specified for NFS file systems that are exported from multiple, functionally equivalent locations:

% fncreate_fs org/sales/fs svr1:/sales svr2:/sales

The automounter will attempt to choose the best server from among the alternatives provided. If several locations in the list share the same path name, they may be combined using a comma-separated list of host names:

% fncreate_fs org/sales/fs svr1,svr2:/sales

The hosts may be weighted, with the weighting factor appended to the host name as a nonnegative integer in parentheses: the lower the number, the more desirable the server. The default weighting factor is zero (most desirable).

The following example illustrates one way to indicate that svr2 is the preferred server:

% fncreate_fs org/sales/fs svr1(2),svr2(1):/sales

Variable Substitution

Variable names, prefixed by $, may be used in the options or location fields of fncreate_fs. For example, a location may be given as:

svr1:/export/$CPU

The automounter will substitute client-specific values for these variables when mounting the corresponding file systems. In the above example, $CPU is replaced by the output of uname -p; for example, sparc.

Backward Compatibility Input Format

For additional compatibilty with automount maps, the following input file format is also accepted by fncreate_fs:

name [options] [location ...] \ /offset1 [options1] location1 ... \ /offset2 [options2] location2 ... \ ...

where each offset field is a slash-separated hierarchy. The backslash ("\") indicates the continuation of a single long line. This is interpreted as being equivalent to:

name [options] [location ...] name/offset1 [options1] location1 ... name/offset2 [options2] location2 ... ...

The first line is omitted if both options and location are omitted. This format is for compatibility only. It provides no additional functionality, and its use is discouraged.

Administering File Contexts

File contexts may be inspected using the fnlist and fnlookup commands, and may be pruned or destroyed using fnunbind and fndestroy. These commands and sample output are described in Chapter 14, "Administering FNS in NIS+." Refer also to the man page for each command.