rmt(8)
NAME
rmt - remote magnetic tape protocol server
SYNOPSIS
/opt/schily/sbin/rmt /etc/rmt
DESCRIPTION
- This is the description of the enhanced Schily version of
- the rmt remote tape server program. rmt is a program used by
- programs like star and ufsdump that like to access remote magnet
- ic tape drives and files through an interprocess communication
- connection. rmt is normally started up with an rexec(3) or
- rcmd(3) call.
- The rmt program accepts open, close, read, write and seek
- requests as well as requests that are specific to magnetic tapes.
- rmt performs the commands and then responds with a status indica
- tion.
- This version of the rmt server gives full compatibility to
- the original BSD version, the enhanced Sun version and the en
- hanced GNU version. In addition to the Sun and GNU enhancements,
- it implements further abstractions for better cross platform com
- pliance. It supports best speed and best compliance even when
- server and client code are running on different platforms. It is
- prepared to be installed as a user shell in the passwd file to
- create remote tape specific logins and security checking. To use
- the enhanced compatibility features, you need to either use the
- remote tape client code from star which is available in librmt or
- reimplement it's features.
- All responses are send back in ASCII and in one of the
- following two forms.
- Successful commands have responses of
Anumber_n- where number is the ASCII representation of a decimal num
- ber that usually is the return code of the corresponding system
- call. Unsuccessful commands are responded to with
Eerror-numbernerror-messagen- where error-number is one of the possible error numbers
- described in intro(2), and error-message is the corresponding er
- ror string as retrieved by strerror(3).
- The protocol implements the following commands:
Odevicenmoden- Odevicenmode symbolic_moden
Open the specified device or fileusing the indicated mode. device is a full path name, and modeis an ASCII representation of a decimal number suitable for beingpassed as second parameter to open(2). A variant of the opencommand includes the symbolic_mode string which is a GNU extension. If both, mode and symbolic_mode are present, they are separated by a space character; symbolic_mode appears on the sameline as the numeric mode. It is send using the same notation asused in a C source (e.g. O_RDWR|O_CREAT). If the symbolic_modeis send to the server, the numeric mode is ignored. The symbolicnotation allows to send the expected open mode over the wire, using a system independent method. This is needed because different operating systems usually define all bits in a different way.An exception are the lowest two bits. The lowest two bits allowto code O_RCONLY,O_WRONLY and O_RDWR. To prevent unexpected behavior, rmt masks the numeric open mode with 0x03 before using itas argument to the open(2) call. If you need more bits in thesecond parameter ot open(2), you need to use the symbolic mode.If no file /etc/default/rmt exists,only filenames starting with /dev/ are accepted for security reasons.If a device is already open, it isclosed before a new open is performed.A RMT protocol VERSION 1 clientshould issue a
I-1n0_n
command just after opening a file ordevice. This is needed to tell the server that the client isaware of the official order of the mt_op codes in the range 0..7and that is maps deviating values to the official ones. - Cdevicen Close the currently open device or
- file. The argument device is ignored.
- Rcountn Read count bytes of data from the open
- device or file. rmt performs the requested read(2) operation and
- responds with Acount-read_n if the read operation was successful;
- otherwise an error in standard format is returned. If the read
- operation was successful, the data read is sent directly after
- the response described above.
- Wcountn Write data to the open device or file.
- After reading the command specification, rmt reads count bytes
- from the network connection and aborts if a premature EOF is en
- countered. The return value from the write(2) operation is re
- turned as reply.
- Lwhencenoffsetn
- Perform an lseek(2) operation on the
- open device or file using the specified parameters. The return
- value from the lseek(2) operation is returned as reply.
- On large file aware operating sys
- tems, rmt will correctly handle large lseek(2) requests.
- S The old non-portable status call.
- This call should not be used anymore, it has been replaced by the
- new RMT protocol version 1 extended status call below. If the
- currently open device is a magnetic tape, return the magnetic
- tape status, as obtained with a MTIOCGET ioctl call. If the open
- device is not a magnetic tape, an error is returned. If the
MTIOCGET
- of the status buffer, then the status buffer is sent. As the
status buffer is sent in binary, this command it considered outdated. Please use the extended status command instead. This command is not terminated by a new-line.
ssub-command The new portable status call. This - command is part of the RMT protocol version 1. If the currently
- open device is a magnetic tape, return a single specified member
- of the magnetic tape status structure, as obtained with a
MTIOCGET ioctl
- an error is returned. If the MTIOCGET operation was successful,
the numerical value of the structure member is returned in decimal. The following sub commands are supported:
T return the content of the - structure member mt_type which contains the type of the magnetic
- tape device.
- D return the content of the
- structure member mt_dsreg which contains the "drive status regis
- ter".
- E return the content of the
- structure member mt_erreg which contains the "error register".
This structure member must be- retrieved first because it is cleared after each MTIOCGET ioctl
- call. The librmt will always retrieve the member mt_erreg first
- when it is told to retrieve a complete status structure.
- R return the content of the
- structure member mt_resid which contains the residual count of
- the last I/O.
- F return the content of the
- structure member mt_fileno which contains the block number of the
- current tape position.
- B return the content of the
- structure member mt_blkno which contains the block number of the
- current tape position.
- f return the content of the
- structure member mt_flags which contains MTF_ flags from the
- driver.
- b return the content of the
- structure member mt_bf which contains the optimum blocking fac
- tor.
- This command is not terminated with
- a new-line.
- Ioperationncountn
- Perform a MTIOCOP ioctl(2) command
- using the specified parameters. The parameters are interpreted
- as the ASCII representations of the decimal values to place in
- the mt_op and mt_count fields of the structure used in the ioctl
- call. When the operation is successful the return value is the
- count parameter. Only Opcodes 0..7 are unique across different
- architectures. But as in many cases Linux does not even follow
- this rule. If we know that we have been called by a RMT protocol
VERSION 1
- ing Linux mapping over the wire but the standard mapping described below:
-1 Retrieve the version number - of the rmt server and tell the server that the client is aware of
- the official order of the MTIOCOP ioctl(2) opcodes in the range
- 0..7. Local mt_op codes must be remapped to the official values
- before sending them over the wire.
The answer of the current- version of rmt is 1. Old rmt implementations send an error code
- back when this command is used. Future rmt implementations with
- further enhancements will send an answer with a value > 1.
- 0 Issue a MTWEOF command (write
- count end-of-file records).
- 1 Issue a MTFSF command (for
- ward space over count file marks).
- 2 Issue a MTBSF command (back
- ward space over count file marks).
- 3 Issue a MTFSR command (for
- ward space count inter-record gaps).
- 4 Issue a MTBSR command (back
- ward space count inter-record gaps).
- 5 Issue a MTREW command
- (rewind).
- 6 Issue a MTOFFL command
- (rewind and put the drive off-line).
- 7 Issue a MTNOP command (no op
- eration, set status only).
- ioperationncountn
- Perform a MTIOCOP ioctl(2) command
- using the specified parameters. This command is a RMT protocol
VERSION 1
- MTWEOF..MTNOP (0..7). The parameters are interpreted as the
ASCII representations of the decimal values described below.
They are converted into the local values mt_op and mt_count fields of the structure used in the ioctl call according to the
actual values found in <sys/mtio.h>. When the operation is successful the return value is the count parameter.
0 Issue a MTCACHE command - (switch cache on).
- 1 Issue a MTNOCACHE command
- (switch cache off).
- 2 Issue a MTRETEN command
- (retension the tape).
- 3 Issue a MTERASE command
- (erase the entire tape).
- 4 Issue a MTEOM command (posi
- tion to end of media).
- 5 Issue a MTNBSF command (back
- ward space count files to BOF).
- v_n Return the version of the rmt server.
- This is currently the decimal number 1.
- Any other command causes rmt to exit.
FILES
- /etc/default/rmt
- Default values can be set for the following options
- in /etc/default/rmt. For example:
- DEBUG=/tmp/rmt.debug
USER=tape
ACCESS=tape myhost.mydomain.org /dev/rmt/* - All keywords must be on the beginning of a line.
- DEBUG If you like to get debug information, set
- this to a file name where rmt should put debug information.
- USER The name of a user (local to the magnetic
- tape server) that may use the services of the rmt server. More
- than one USER=name line is possible. A line USER=* grants access
- to all users.
- ACCESS This keyword is followed by three parameters
- separated by a TAB. The name of a user (local to the magnetic
- tape server host) that may use the services of the rmt server
- followed by the name of a host from where operation is granted
- and a file specifier pattern for a file or file sub tree that may
- be accessed if this ACCESS line matches. More than one
ACCESS=name host path
- If standard input of rmt is not a socket
- from a remote host, rmt will compare the host entry from
- /etc/default/rmt with the following strings:
- PIPE If stdin is a UNIX pipe.
If you like to allow remote con- nections that use the ssh protocol, you need to use the word PIPE
- instead of thr real hostname in the matching ACCESS= line.
- ILLEGAL_SOCKET
- If getpeername() does not work for
- stdin.
- NOT_IP If getpeername() works for stdin
- but is not connected to an internet socket.
SEE ALSO
- star(1), ufsdump(1), ufsrestore(1), intro(2), open(2),
- close(2), read(2), write(2), ioctl(2), lseek(2), getpeername(3)
- rcmd(3), rexec(3), strerror(3), mtio(7)
DIAGNOSTICS
- All responses are send to the network connection. They
- use the form described above.
NOTES
- To use rmt as a remote file access protocol you need to
- use the symbolic open modes as e.g. the O_CREAT flag is not
- unique between different architectures.
- In order to allow this implementation to be used as a re
- mote file access protocol, it accepts file names up to 4096 bytes
- with the open command. Other rmt implementations allow no more
- than 64 bytes.
- The possibility to create a debug file by calling rmt file
- has been disabled for security reasons. If you like to debug rmt
- edit /etc/default/rmt and insert a DEBUG entry.
- This implementation of rmt adds some security features to
- the server that make it behave slightly different from older im
- plementations. Read the above documentation about the file
- /etc/default/rmt to make sure your local installation is config
- ured for your needs.
- To grant the same permissions as with old rmt servers,
- create a file /etc/default/rmt and add the following lines to
- this file:
- USER=*
ACCESS=* * * - Note that the three fields in the ACCESS= line need to be
- separated by a TAB character.
- Be very careful when designing patterns to match path
- names that may be accepted for open. If a pattern would allow to
- include /../ a possible intruder could vitually access any path
- on your system.
BUGS
None known.
HISTORY
- The rmt command first appeared on BSD in march 1981. This
- rmt server is a new implementation that tries to be compatible to
- all existing implementations. It is the only known implementa
- tion that in addition tries to fix the data exchange problems be
- tween different architectures.
AUTHOR
Joerg Schilling
Seestr. 110
D-13353 Berlin
Germany
Mail bugs and suggestions to:
- schilling@fokus.fhg.de or js@cs.tu-berlin.de or
- joerg@schily.isdn.cs.tu-berlin.de
- Joerg Schilling Release 1.1