Dataset and HFS File System Access
SSH Tectia Server for IBM z/OS has not introduced any extensions to the SFTP protocol and thus supports all old SFTP clients. The differences between z/OS and other operating systems (Unix, Windows, etc.) introduce some difficulties in accessing data in the z/OS system.
z/OS has both MVS datasets, and Hierarchical File System (HFS) files. As both types must be accessed by the SFTP server, there must be a mechanism for distinguishing between them. Traditionally, MVS datasets in z/OS are accessed using the filename format
//NAME.OF.MVS.DATASET, while HFS files are accessed using the filename format
/path/to/hfs/file. Old SFTP clients regard
/ (slash) as a directory separator, which requires the SFTP server to use some "tricks" to fool the clients. It must be noted that the SFTP client generally does not have any inherent knowledge of MVS datasets and accesses all files as they were part of a hierarchical file system.
If an SFTP client tries to open the file
//FILE.NAME, the MVS dataset is actually opened. Some SFTP clients remove all consecutive slashes so that the file to be accessed becomes
/FILE.NAME. The SFTP server will interpret this as the HFS file
FILE.NAME, located at the HFS root. In order to avoid this client behavior, the SFTP server has other ways of naming MVS datasets. If a client tries to access the file
__FILE.NAME, the SFTP server interprets these as the dataset
//FILE.NAME. All these combinations are needed as it may be necessary to make the SFTP clients pass the correct file or dataset name to the server.
In z/OS, if a dataset name is not enclosed in single quotes, the user prefix is added in front of the dataset name. For example, if user
USER1 has a dataset
DATASET.NAME1, the user can access it using the dataset name
//DATASET.NAME1. It is also possible to use an absolute prefixed name
//'USER1.DATASET.NAME1'. Because single quotes are special characters in most environments, the SFTP server introduces a new convention to use triple slashes to access absolute prefixed dataset names. The name becomes
///USER1.DATASET.NAME1. All combinations of slashes and underscores can be used. For example,
USER1 can access his/her dataset using one of the following names:
It must be noted that the choice of name type depends on the client. If consecutive slashes are not sent to the server, "
/_" (or "
/__" for a prefixed absolute dataset) is probably the best choice. The SFTP client interprets the file path to be absolute and does not add any directories in front of the name. If a file transfer advice string (see Section Advice String) is added in front of the file name, "
__" (or "
___" for a prefixed absolute dataset) is a safe choice. The dataset name becomes
/FTADV:X=BIN/__DATASET.NAME1. The client interprets it as an absolute path to the file and does not change it. The SFTP server on the other hand recognizes the advice string
/FTADV:X=BIN/ and interprets
__DATASET.NAME1 to be the real name of the dataset
z/OS has library datasets, whose members are usually accessed using the dataset name
//DATASET.NAME1(MEMBER1). Again, parentheses may be special characters in some environments. The SFTP server introduces a new convention for accessing library members using a single slash as a member separator. The name becomes
Because of the new conventions for dataset names there are some exceptions in the HFS filenames. If an HFS file or directory name starts with an underscore and the file or directory is located in the root directory, the HFS root must be written as "
/___". The name thus becomes
/____path/to/hfs/file. Also, if an advice string is used with absolute HFS names, the HFS root must be written either as "
/___", or "
____". If the client does not allow two consecutive slashes or if the HFS file or directory name starts with an underscore and is located in the root directory, the HFS root must be written as "
____" (four underscores).
HFS file names are case-sensitive. For example,
/tmp/myfile result in two different files.
MVS dataset names are case-insensitive. For example
//_user1.dataset.name1 are handled the same way.