|
|
|
|
|
|
|
|
HP-UX Reference Volume 5 of 5 > a
autochanger(7) |
|
NAMEautochanger — SCSI interfaces for medium changer device and magneto-optical autochanger surface device DESCRIPTIONAn autochanger is a SCSI mass storage device, consisting of a mechanical changer device, one or more data transfer devices (such as optical disk drives), and media (such as optical disks) for data storage. The mechanical changer moves media between storage and usage locations within the autochanger. Depending on system architecture, one of two medium changer drivers (schgr or autox0) provides access to the medium changer device; a module (ssrfc) provides access to the surfaces of the optical disks. Two levels of functionality are provided by the medium changer drivers. The mechanical changer device can be accessed directly to move media within the autochanger. Alternatively, media surfaces can be accessed as unique devices, causing the changer driver to move the media into a drive to perform an I/O request. The schgr and autox0 medium changer device drivers follow the SCSI specification for medium changer devices to provide a generic medium changer interface, making it feasible to construct an application level driver for any mechanical changer, jukebox, library, or autochanger device (MO, tape, CD-ROM). However, the ssrfc module is provided specifically to support Hewlett-Packard magneto-optical disk autochanger products. Device Naming ConventionThe device naming convention for the autochanger driver enables accessing the changer device, as well as individual media surfaces. Block devices for autochangers reside in /dev/ac, character devices reside in /dev/rac. Within these directories, names are derived from the "c#t#d#" device naming convention (explained in intro(7)), with the surface descriptor appended at the end. Unique device names are determined by the card instance, target address of the SCSI changer device, LUN of the SCSI changer device, and the surface descriptor. The surface descriptor can be zero or non-specified for the changer device. Also, there is no block special file for the changer itself. For example, /dev/rac/c1t5d0 is the character special file for the changer at SCSI target address 5 and LUN 0, attached to SCSI card instance 1, and is equivalent to /dev/rac/c1t5d0_0. Any given surface is described by the card instance, SCSI target address and SCSI LUN of the changer, and then appended with a surface descriptor for the slot number and side. For example, /dev/ac/c1t5d0_1a is the block special file for surface 1a of the autochanger just mentioned and /dev/rac/c1t5d0_1a is the character special file for the same surface 1a. Major and Minor Number DescriptionsThe following shows the bit assignments (dev_t format) used by the changer drivers to access the changer device and each surface within an autochanger:
MAJOR is the major number of the appropriate driver, INSTANCE is the card instance of the SCSI interface to which the changer device is attached, TARGET is the SCSI target address of the changer device, LUN is the SCSI LUN of the changer device, and SURFACE is the unique descriptor of each surface in the autochanger, as described in the following table. (Note, the surface descriptors refer to bits 23-31.)
All fields in the device number are specified in hexadecimal notation. Note that there is no support for hard partitions (sections) in this minor number. If desired, partitioning can be achieved via LVM soft-partitioning schemes. The major numbers used by the changer drivers are:
Following are long listings showing the major and minor numbers associated with the device special file names of the first surface and the changer: schgr: brw-rw-rw- 1 root sys 29 0x015001 Apr 22 10:22 /dev/ac/c1t5d0_1a crw-rw-rw- 1 root sys 231 0x015001 Apr 22 10:22 /dev/rac/c1t5d0_1a crw-rw-rw- 1 root sys 231 0x015000 Apr 22 10:22 /dev/rac/c1t5d0 autox0: brw-rw-rw- 1 root sys 30 0x015001 Apr 24 11:35 /dev/ac/c1t5d0_1a crw-rw-rw- 1 root sys 230 0x015001 Apr 24 11:35 /dev/rac/c1t5d0_1a crw-rw-rw- 1 root sys 230 0x015000 Apr 24 11:35 /dev/rac/c1t5d0 MAGNETO-OPTICAL AUTOCHANGER SURFACE DEVICE ACCESSTo access disk surfaces within HP magneto-optical libraries, it is necessary to include the entry for the surface module, ssrfc, in the system configuration file /stand/system, as well as an entry for the appropriate SCSI changer driver, schgr or autox0, depending on architecture. The ssrfc module enables accessing a magneto-optical disk surface much like a disk device. The disk is moved into an idle drive by the changer, then the requested disk I/O operation is performed. Upon completion of the request, the disk is returned to its storage location within the autochanger. The surface module allows concurrent access to as many disks as there are drives in the autochanger product. Requests for I/O on additional disks within the autochanger are blocked awaiting an available drive resource. By default, some commands (such as mount, newfs, and mediainit) open the device with the O_NDELAY flag set. Invocations of these commands on an autochanger surface do not wait for a drive resource to become available. Instead, these requests return with EBUSY if no drive is available. Developers using the surface module functionality to access autochanger disks can invoke the open system call with the O_NDELAY flag to achieve this same "non-blocking" behavior: error = open("/dev/rac/c1t5d0_1a",O_RDWR | O_NDELAY); If it is acceptable to block waiting for an available drive resource, the O_NDELAY flag is unnecessary. Here is a sample script to access multiple disk surfaces in an autochanger that has 2 drives, minimizing blocking: dd if=/dev/rdsk/c0t0d0 of=/dev/rac/c1t5d0_1a bs=64k & dd if=/dev/rdsk/c0t1d0 of=/dev/rac/c1t5d0_2a bs=64k & wait dd if=/dev/rdsk/c0t2d0 of=/dev/rac/c1t5d0_1b bs=64k & dd if=/dev/rdsk/c0t3d0 of=/dev/rac/c1t5d0_2b bs=64k & wait ... For developers, the ioctl functions available for accessing magneto-optical disk surfaces are described in the manual pages for SCSI disk drivers. Several ioctl functions provided specifically for magneto-optical disks will be described here briefly. Included from <sys/scsi.h>: #define SIOC_WRITE_WOE _IOW('S', 17, int) #define SIOC_VERIFY_WRITES _IOW('S', 18, int) #define SIOC_ERASE _IOW('S', 19, struct scsi_erase) #define SIOC_VERIFY_BLANK _IOW('S', 20, struct scsi_verify) #define SIOC_VERIFY _IOW('S', 21, struct scsi_verify) SIOC_ERASE (erase) and SIOC_WRITE_WOE (write without erase) can be used together on character special devices. By performing a pre-erase pass of magneto-optical disks, then later setting the SCSI disk driver in write-without-erase mode, improved write performance can be achieved, eliminating the two-pass erase-then-write which is normally necessary on magneto-optical devices. SIOC_VERIFY_WRITES (write and verify) performs a verification pass on any writes to magneto-optical disks. This is a good safeguard for data integrity. However, write operations performed with the verification pass exhibits a decrease in performance. When used with pre-erase and write-without-erase, write and verify provide increased reliability of data without decreased performance. HP recommends operating in write-and-verify mode if also performing write-without-erase. The following are additional ioctl functions that might be desirable for some magneto-optical products, included from <sys/scsi.h>: #define SIOC_GET_IR _IOR('S', 14, int) #define SIOC_SET_IR _IOW('S', 15, int) #define SIOC_SYNC_CACHE _IOW('S', 70, int) SIOC_GET_IR determines the current state of immediate reporting (write caching) on the device. SIOC_SET_IR enables or disables immediate reporting on the device. If SIOC_SET_IR is used to enable write caching, it may be desirable to flush the write cache using the SIOC_SYNC_CACHE ioctl function. The command /usr/sbin/scsictl may be used to perform the pre-erase of magneto-optical disks, set and check the status of immediate reporting. With the surface module configured, several ioctl functions to get status and information from the changer device are also available. These are SIOC_ELEMENT_ADDRESSES, SIOC_ELEMENT_STATUS, and SIOC_INQUIRY; they are explained further in the following section on the changer driver. Functions that modify the state of the autochanger are not allowed when the surface module is configured into the kernel. SCSI MEDIUM CHANGER DEVICE DRIVERThe SCSI medium changer device driver performs moves between different media locations within an autochanger. Each potential media location has a specific element address and is one of the following element types:
Changer Control RequestsThe following ioctl functions are included from <sys/chgrio.h>: #define CHGR_SSRFC_IS_PRESENT _IOR('X', 1, int) #define CHGR_CLEAR_RESET _IO('X', 2)
The following ioctl functions and structure definitions are included from <sys/scsi.h>: #define SIOC_INIT_ELEM_STAT _IO('S', 51) #define SIOC_ELEMENT_ADDRESSES _IOW('S', 52, struct element_addresses) #define SIOC_ELEMENT_STATUS _IOWR('S', 53, struct element_status) #define SIOC_RESERVE _IOW('S', 54, struct reservation_parms) #define SIOC_RELEASE _IOW('S', 55, struct reservation_parms) #define SIOC_MOVE_MEDIUM _IOW('S', 56, struct move_medium_parms) #define SIOC_EXCHANGE_MEDIUM _IOW('S', 57, struct exchange_medium_parms) /* structure for SIOC_ELEMENT_ADDRESSES ioctl */ struct element_addresses { unsigned short first_transport; unsigned short num_transports; unsigned short first_storage; unsigned short num_storages; unsigned short first_import_export; unsigned short num_import_exports; unsigned short first_data_transfer; unsigned short num_data_transfers; }; /* structure for SIOC_ELEMENT_STATUS ioctl */ struct element_status { unsigned short element; /* element address */ unsigned int resv1:2; unsigned int import_enable:1; /* allows media insertion (load) */ unsigned int export_enable:1; /* allows media removal (eject) */ unsigned int access:1; /* transport element accessible */ unsigned int except:1; /* is in an abnormal state */ unsigned int operatr:1; /* medium positioned by operator */ unsigned int full:1; /* holds a a unit of media */ unsigned char resv2; unsigned char sense_code; /* info. about abnormal state */ unsigned char sense_qualifier; /* info. about abnormal state */ unsigned int not_bus:1; /* transfer device SCSI bus differs */ unsigned int resv3:1; unsigned int id_valid:1; /* bus_address is valid */ unsigned int lu_valid:1; /* lun is valid */ unsigned int sublu_valid:1; /* sub_lun is valid */ unsigned int lun:3; /* transfer device SCSI LUN */ unsigned char bus_address; /* transfer device SCSI address */ unsigned char sub_lun; /* sub-logical unit number */ unsigned int source_valid:1; /* source_element is valid */ unsigned int invert:1; /* media in element was inverted */ unsigned int resv4:6; unsigned short source_element; /* last storage medium location */ char pri_vol_tag[36]; /* volume tag (device optional) */ char alt_vol_tag[36]; /* volume tag (device optional) */ unsigned char misc_bytes[168]; /* device specific */ }; /* structure for SIOC_RESERVE and SIOC_RELEASE ioctls */ struct reservation_parms { unsigned short element; unsigned char identification; unsigned char all_elements; }; /* structure for SIOC_MOVE_MEDIUM ioctl */ struct move_medium_parms { unsigned short transport; unsigned short source; unsigned short destination; unsigned char invert; }; /* structure for SIOC_EXCHANGE_MEDIUM ioctl */ struct exchange_medium_parms { unsigned short transport; unsigned short source; unsigned short first_destination; unsigned short second_destination; unsigned char invert_first; unsigned char invert_second; };
DEPENDENCIESTo obtain access to disk surfaces within HP magneto-optical libraries, the ssrfc module must be specified in the system configuration file. The ssrfc module depends on either the schgr driver, or the autox0 driver. If ssrfc is to be included, then one or both of schgr or autox0 must also be included. DEFAULT CONFIGURATIONSBy default, ssrfc, schgr, and autox0 are not included in the system configuration (/stand/system) file. EXAMPLESThe following example uses the SIOC_ELEMENT_ADDRESSES and SIOC_ELEMENT_STATUS ioctl functions to get bus address information about the drives in an HP magneto-optical autochanger: int last_drive_el; struct element_addresses el_addrs; struct element_status el_stat; /* * Changer attached to card instance 1, with SCSI target id 5, lun 0. */ fd = open("/dev/rac/c1t5d0",O_RDWR); if ((error = ioctl(fd, SIOC_ELEMENT_ADDRESSES, &el_addrs)) != 0) { syserr("ioctl: SIOC_ELEMENT_ADDRESSES"); return -1; } else { last_drive_el = el_addrs.first_data_transfer + el_addrs.num_data_transfers - 1; for (i = el_addrs.first_data_transfer; i <= last_drive_el; i++) { el_stat.element = i; if ((error = ioctl(fd, SIOC_ELEMENT_STATUS, &el_stat)) != 0) { syserr("ioctl: SIOC_ELEMENT_ADDRESSES"); return -1; } else { /* * You may wish to also check some of the other fields * in the el_stat structure to verify that the data is * valid. Fields: el_stat.access (ac accessible), * el_stat.except (exception). */ if (! el_stat.not_bus && el_stat.id_valid) { drive[i].addr = el_stat.bus_address; if (! el_stat.lu_valid) { drive[i].lun = 0; } else { drive[i].lun = el_stat.lun; } } } } } WARNINGSDo not use LVM to configure multiple autochanger surfaces as one large file system. LVM was designed for on-line volumes. In an autochanger, only the disks actualy in the drives are on-line, while the disks stored in their slots are off-line. If LVM is not carefully configured, thrashing of the autochanger disks result in undesirable I/O performance. Plan carefully for best performance. Some non-HP media changer devices do not support the SIOC_INIT_ELEM_STAT and SIOC_ELEMENT_STATUS ioctls. Some older media changer devices do not support the SIOC_EXCHANGE_MEDIUM ioctl. For these devices, multiple SIOC_MOVE_MEDIUM ioctl operations may be used to accomplish the same results, provided a suitable temporary element address may be found. |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||
|
|
|
|||||||||||||
|
|||||||||||||
