The OMVS File SystemThe OMVS file system structure resembles an upside-down tree. The first directory is called the root directory.
From an MVS perspective, OMVS files are stored in HFS (Hierarchical File System) datasets. The root HFS, the first file system mounted, is specified in the OMVS initialisation parameters. The root HFS contains the root directory.
Additional HFS datasets can be mounted at specified locations within already-mounted HFS's. These locations must be directories and are called "mount points". HFS's can be mounted either by an explicit mount or through a mechanism called "automount", which allows HFS datasets to be mounted when required. The MVS command "D OMVS,F" can be used to display mounted file systems and associated mount points.
DFDSS can be used to dump or restore an entire HFS dataset. To copy an HFS dataset, dump it to a sequential dataset and then restore to a new HFS dataset from that sequential dataset. A "super user" account is required to dump an active HFS. You can also use DFHSM to backup or restore an HFS dataset.
File ownership information for an OMVS file is stored as a numeric uid value. If this value is the same as your uid, then your user name will appear as the owner. Otherwise, the security product will be called to map the uid value to a user name. See ACF2 Security for OMVS - Access to Data for information about access to data within an HFS. See zSeries File System for information about another type of OMVS file system.
HFS files have multiple timestamps associated with them:
For HFS datasets that are mounted at IPL time, the date of last reference and dataset changed indicator are only reset when the file is closed, which happens when the HFS is unmounted. Normally, this will only happen for such HFS datasets at system shutdown. Also note that, because the time that an OMVS file was last accessed is stored in the HFS dataset, a read access for any file in an HFS is enough to set the dataset changed indicator. The end result is that, although DFHSM does support HFS datasets in theory, in practice this support is not as well-integrated as one might have hoped.