Content-type: text/html Man page of MM_DATA


Section: File Formats (5)
Updated: Dec 11, 08
Index Return to Main Contents


mm_data - NetWorker media multiplexor data (tape and disk) format  


This documents the data format that the NetWorker media multiplexor daemon, nsrmmd(8), writes to long term storage media such as tapes and optical disks. See nsr_device(5) and nsrmm(8) for a discussion of supported device families and types. The format described here applies to any fixed record device, such as raw disks, or fixed record tape devices with file marks. NetWorker uses the eXternal Data Representation (XDR) standard to write media which can be interchanged among a wide variety of machines. Only the mechanism used to multiplex save set streams onto the storage media is described here; the formats of save set streams depend on the type of NetWorker client, and are described in nsr_data(5).

A volume is one physical piece of media such as a tape reel or disk cartridge. A tape volume is made up of multiple media files, and each media file may contain several media records. These media files and records should not be confused with a client's (for example UNIX or DOS) user files or records; the two do not necessarily correspond. For example, a given media file or even a single media record may contain many small client user files. On the other hand, a single large client file may be split across several media files, and even across several volumes. Media files do not span volume boundaries. Save sets may span media files and even volumes.

On most tapes, media files can be skipped very quickly by the device's hardware or associated device driver software, and the hardware can detect when an end of a file has been reached. On some tapes, records can also be quickly skipped forward. Otherwise, access to the media is sequential.

Media records are described by the mrecord structure. Label records are fixed in size, MINMRECSIZE bytes. Other records can potentially be a larger size that must be some constant for the rest of the volume. NetWorker always writes, reads and skips data in units of full-sized media records. Each mrecord contains zero or more mchunks These mrecords are used for storing one or more client save sessions or used by NetWorker for synchronization and labelling. The XDR format of a media file's mrecords and mchunks are as follows:

const MINMRECSIZE = 32768;      /* minimum media record size */
const MMAXCHK = 2048;           /* maximum number of chunks in record */
const MHNDLEN = 120;            /* private area length for handlers */

enum mrec_version {             /* mrecord version */
    MREC_VER5 = 0,              /* older format mrecord */
    MREC_VER6 = 6               /* current format mrecord */

 * For media record format version 5, the data types lgui_t, lg_off64_t,
 * and lg_time64_t are defined as:
typedef struct lgui_t unsigned long;
typedef struct lg_off64_t unsigned long;
typedef struct lg_time64_t unsigned long;

 * For media record format version 6, the data types lgui_t, lg_off64_t,
 * and lg_time64_t are defined as:
typedef struct lgui_t {         /* XDR encoded Unique Id. */
    char _bytes[20];
} lgui_t;
typedef struct lg_off64_t unsigned long long;
typedef struct lg_time64_t unsigned long long;

typedef lgui_t ssid_t;          /* save set id */
typedef lgui_t volid_t;         /* key for the volume database */

struct  mchunk {
    ssid_t mc_ssid;             /* owning save set id */
    lg_off64_t mc_low;          /* 1st byte, relative to save stream */
    opaque mc_data<MINMRECSIZE>;/* chunk's data */

struct  mrecord {
    opaque mr_handler[MHNDLEN]; /* private to media handler */
    mrec_version mr_version;    /* Media record version number */
    u_long mr_orec;             /* record size */
    volid_t mr_volid;           /* encompassing volume's id */
    u_long mr_fn;               /* encompassing file number */
    u_long mr_rn;               /* record number within the file */
    u_long mr_len;              /* record byte length */
    mchunk mr_chunk<MMAXCHK>;   /* chunks of save streams */

The first field of an mrecord, mr_handler, is reserved for media-specific data (currently it is not used by any implementation). The mr_version field is the version number of the media record format. The size of the rest of the fields in the media record depends on the version number. The mr_orec field is the size of the current record. A media record's header fields, mr_volid, mr_fn, and mr_rn, are used to check the tape position and the data read from the record. The file numbers and record numbers start at zero and increment sequentially. The record number is reset each time the file number is incremented. On disks, file numbers are always zero. The mr_len field is the actual number of valid bytes in this record, as opposed to the size of the device's read or write request.

If file or record skipping is unreliable, NetWorker can still recover from isolated errors, at worst by rewinding and reading the tape from the start. If a volume can be physically unmounted or mounted without notice to the media management daemon, then the volume identifier in each record provides a quick way of verifying when this happens, without the need for a full rewind and reading of the label in most cases.

The mchunks within an mrecord contain client data from one or more save sessions. The mc_ssid and mc_low values are used to reconstruct the save streams from the chunks within the records. The mc_data field holds the actual data of each chunk. For a given save set, mc_low plus the length of mc_data should equal the following chunk's value for mc_low. Save sets may by intermingled arbitrarily within media records.

The first chunk of the first record of the first media file on the volume encapsulates the volume label information; for some media, the second chunk contains additional volume information, for example, the media pool the volume belongs to: Subsequent data in the first file is reserved for future expansion. The label may be duplicated in a second file for redundancy, in case the first copy of the label gets accidentally overwritten. The formats of the volume label and additional label information are described by the following XDR data structures:

const MVOLMAGIC = 0x070460;     /* volume magic number */
const NSR_LENGTH = 64;          /* length of several strings */
const RAP_MAXNAMELEN = 64;      /* maximum length of attribute name */

struct mvollabel {
    u_long  mvl_magic;          /* medium volume verification number */
    lg_time64_t mvl_createtime; /* time at which volume labeled */
    lg_time64_t mvl_expiretime; /* time for volume to expire */
    u_long  mvl_recsize;        /* expected size of mrecords */
    volid_t mvl_volid;          /* medium volume id */
    string  mvl_volname<NSR_LENGTH>;/* medium volume name */

struct vallist {
    vallist *next;
    string value<>;             /* attribute value */

struct attrlist {
    attrlist *next;
    string name<RAP_MAXNAMELEN>;/* attribute name */
    vallist *values;            /* attribute values */

 * Additional information may includes the following attributes
 * (listed by the name they are stored with):
 * "volume pool" : the media pool
struct mvolinfo {
    struct attrlist *mvi_attributes;/* any other information */

The mvl_magic field must be equal to MVOLMAGIC in order for the chunk to represent a valid volume label. If the volume label changes in the future, the new format will have another ``magic'' number, but the format described here must still be allowed. The mvl_volid is an internal identifier assigned and managed by the media manager. The mvl_volname is the volume name that is assigned when the media is first labeled. The time fields are in UST format - the number of seconds elapsed since 00:00 GMT, January 1, 1970. The mvl_recsize is the size of all subsequent media records found on the tape.

The mvp_pool is the pool name that is assigned when the media is first labeled. Different media pools allow administrators to segregate their data onto sets of volumes. Media cannot be reassigned from one media pool to another. Pool names are a maximum of NSR_LENGTH characters long.

Synchronization marks, called schunks, are also written periodically to the media for each save set. Synchronization chunks are used by scanner(8) when verifying or extracting directly from a volume. They are also used by nsrmmd when trying to recover from media errors during file recovery. The following XDR data structure describes a synchronization chunk:

typedef lgui_t clientid_t;

struct                             ssclone_t {
    lg_time64_t ssc_cloneid;       /* unique UST stamp wrt ss_ssid */
    u_long ssc_flag;               /* lots of status buried here*/
    u_long ssc_frag;               /* not used, always 0 */

 * Synchronization chunk of the newer MREC_VER6 media format.
struct schunk {
    u_long ssi_gen;                /* Not used. */
    ssid_t ssi_ssid;               /* save set identifier */
    ssid_t ssi_prev;               /* non-zero iff continuation */
    u_long ssi_level;              /* backup level*/
    lg_time64_t ssi_time;          /* save time on client */
    lg_time64_t ssi_create;        /* creation time on server */
    lg_time64_t ssi_insert;        /* insertion time on server */
    lg_time64_t ssi_complete;      /* completion time on server */
    clientid_t ssi_clientid;       /* client name identifier */
    u_long ssi_flags;              /* more details about this ss */
    string ssi_host<>;             /* client name - save set owner */
    string ssi_name<>;             /* symbolic name, for example "/usr" */
    uint64_t ssi_size;             /* actual number of bytes saved */
    uint64_t ssi_nfiles;           /* number of client files saved */
    u_long ssi_browse;             /* browse time offset */
    u_long ssi_recycle;            /* recycle time offset */
    struct attrlist *ssi_al;       /* generic RAP attribute list */
    ssclone_t ssi_clones<>;        /* information about this clone */

 * Synchronization chunk of the older MREC_VER5 media format.
struct old_schunk {
    opaque ssi_host[NSR_LENGTH];   /* save set host */
    opaque ssi_name[NSR_LENGTH];   /* symbolic name */
    u_long ssi_time;               /* save time */
    u_long ssi_expiry;             /* expiration date */
    u_long ssi_size;               /* actual size saved */
    u_long ssi_nfiles;             /* number of files */
    ssid_t ssi_ssid;               /* ssid for this save set */
    u_long ssi_flag;               /* various flags, see below */
    u_long ssi_info;               /* volid or ssid, see below */

#define SSI_START       1          /* start of a save set */
#define SSI_SYNC        2          /* synchronization point */
#define SSI_CONT        3          /* continued from another volume */
#define SSI_END         4          /* end of this save set */
#define SSI_SSMASK      0x0000000f /* save set sync chunk type */
#define SSI_LBIAS       0x10000000 /* the level is included in the flags */
#define SSI_LMASK       0xff000000 /* mask to cover bits for level */
#define SSI_LSHIFT      24         /* shift amount for the level */
#define SSI_INCOMPLETE  0x00010000 /* not finished (aborted) */
#define SSI_CONTINUED   0x00800000 /* continued save set series */

The ssi_ssid is the save set identifier of this save set. The ssi_time field contains the create time of the save set in UST based on the client's clock. The ssi_create field contains the create time of the save set in UST based on the server's clock. The ssi_insert field contains the time the save set was inserted into the media database in UST based on the server's clock. The ssi_complete field contains the completion time of the save set in UST based on the server's clock. The ssi_clientid and ssi_host are the client identifier and name of the index which contains this save set. Traditionally this is the client identifier and name of the client where the save set originated. The ssi_name is the save set name to be presented to the user. These are both null-terminated strings, even though the fields are fixed length in the older version media records. The ssi_size and ssi_nfiles are the number of bytes and number of files saved so far for this save set. The ssi_browse is the time offset in seconds from the save set insertion time to the time this save set is no longer browsable. The ssi_recycle is the time offset in seconds from the save set insertion time to the time this save set becomes recyclable. The ssi_al is the generic save set attribute.

The ssi_flag indicates the type of this synchronization chunk and other information about the save set. In the older version synchronization chunk, this field also contains the level of this save set. There are four basic types of synchronization marks that can be found from examining ssi_flag & SSI_SSMASK. SSI_START is used to mark the beginning of a save set. SSI_SYNC marks a periodic synchronization point and is only written at an exact file boundary in the save set. SSI_CONT indicates that this is the continuation of a save set that started on a different volume. When ssi_flag & SSI_SSMASK is SSI_CONT, ssi_prev or ssi_info contains the volume identifier for the save set's preceding volume. These synchronization chunks are used when a save set spans a volume boundary. SSI_END marks the end of a save set.

On the new version of synchronization chunk, the ssi_level field contains the save set backup level. On the older version of synchronization chunk. Should the SSI_LBIAS bit be set then ssi_flag & SSI_LMASK shifted to the right by the value of SSI_LSHIFT specifies the level of the save set. The SSI_INCOMPLETE bit indicates that this save set did not finish properly. This could be caused by a user interrupting an in progress save.

The SSI_CONTINUED bit indicates that this save set is logically continued to or from another save set. These continued save sets are used to handle very large save sets. If the SSI_CONTINUED bit is set and ssi_flag & SSI_SSMASK is SSI_START, then ssi_prev or ssi_info gives the previous save set id that this save set was continued from. If the SSI_CONTINUED bit is set and ssi_flag & SSI_SSMASK is SSI_END, then ssi_prev or ssi_info gives the next save set id that this save set is continued to.

The ssi_expiry field is the expiration date, in UST, for this save set. This field is zero if an explicit save set expiration time was not specified when the save set was created. This field no longer exists in the new synchronization chunk.  


nsr_device(5), nsr_data(5), nsrmm(8), nsrmmd(8), nsrmmdbd(8), nsr(8), scanner(8).
RFC 1014 XDR: External Data Representation Specification




This document was created by man2html, using the manual pages.
Time: 02:39:47 GMT, October 02, 2010