Crash Tolerance (see below):

int gdbm_failure_atomic (GDBM_FILE dbf, const char *even, const char *odd);
int gdbm_latest_snapshot (const char *even, const char *odd, const char **result);

Opening a database

A process that opens a gdbm file is designated as a "reader" or a "writer". Only one writer may open a gdbm file and many readers may open the file. Readers and writers can not open the gdbm file at the same time. The procedure for opening a gdbm file is:

GDBM_FILE gdbm_open (const char *name, int block_size, int flags, int mode, void (*fatal_func)(const char *));

Name is the name of the file (the complete name, gdbm does not append any characters to this name).

Block_size is the size of a single transfer from disk to memory. If the value is less than 512, the file system block size is used instead. The size is adjusted so that the block can hold exact number of directory entries, so that the effective block size can be slightly greater than requested. This adjustment is disabled if the GDBM_BSEXACT flag is used.

The flags parameter is a bitmask, composed of the access mode and one or more modifier flags. The access mode bit designates the process as a reader or writer and must be one of the following:

GDBM_READER

reader

GDBM_WRITER

writer

GDBM_WRCREAT

writer - if database does not exist create new one

GDBM_NEWDB

writer - create new database regardless if one exists

Additional flags (modifiers) can be combined with these values by bitwise OR. Not all of them are meaningful with all access modes.

Flags that are valid for any value of access mode are:

GDBM_CLOEXEC

Set the close-on-exec flag on the database file descriptor.

GDBM_NOLOCK

Prevents the library from performing any locking on the database file.

GDBM_NOMMAP

Instructs gdbm_open to disable the use of mmap(2).

GDBM_PREREAD

When mapping GDBM file to memory, read its contents immediately, instead of when needed (prefault reading). This can be advantageous if you open a read-only database and are going to do a lot of look-ups on it. In this case entire database will be read at once and searches will operate on an in-memory copy. In contrast, GDBM_PREREAD should not be used if you open a database (even in read-only mode) only to retrieve a couple of keys.

Finally, never use GDBM_PREREAD when opening a database for updates, especially for inserts: this will degrade performance.

This flag has no effect if GDBM_NOMMAP is given, or if the operating system does not support prefault reading. It is known to work on Linux and FreeBSD kernels.

GDBM_XVERIFY

Enable additional consistency checks. With this flag, eventual corruptions of the database are discovered when opening it, instead of when a corrupted structure is read during normal operation. However, on large databases, it can slow down the opening process.

The following additional flags are valid when the database is opened for writing (GDBM_WRITER, GDBM_WRCREAT, or GDBM_NEWDB):

GDBM_SYNC

Causes all database operations to be synchronized to the disk.

NOTE: this option entails severe performance degradation and does not necessarily ensure that the resulting database state is consistent, therefore we discourage its use. For a discussion of how to ensure database consistency with minimal performance overhead, see CRASH TOLERANCE below.

GDBM_FAST

A reverse of GDBM_SYNC: synchronize writes only when needed. This is the default. This flag is provided only for compatibility with previous versions of GDBM.

The following flags can be used together with GDBM_NEWDB. They also take effect when used with GDBM_WRCREAT, if the requested database file doesn't exist:

GDBM_BSEXACT

If this flag is set and the requested block_size value cannot be used, gdbm_open will refuse to create the database. In this case it will set the gdbm_errno variable to GDBM_BLOCK_SIZE_ERROR and return NULL.

Without this flag, gdbm_open will silently adjust the block_size to a usable value, as described above.

GDBM_NUMSYNC

Create new database in extended database format, a format best suited for effective crash recovery. For a detailed discussion, see the CRASH RECOVERY chapter below.

Mode is the file mode (see chmod(2) and open(2)). It is used if the file is created.

Fatal_func is a function to be called when gdbm if it encounters a fatal error. This parameter is deprecated and must always be NULL.

The return value is the pointer needed by all other routines to access that gdbm file. If the return is the NULL pointer, gdbm_open was not successful. In this case, the reason of the failure can be found in the gdbm_errno variable. If the following call returns true (non-zero value):

  gdbm_check_syserr(gdbm_open)

the system errno variable must be examined in order to obtain more detail about the failure.

GDBM_FILE gdbm_fd_open (int FD, const char *name, int block_size, int flags, int mode, void (*fatal_func)(const char *));

This is an alternative entry point to gdbm_open. FD is a valid file descriptor obtained as a result of a call to open(2) or creat(2). The function opens (or creates) a DBM database this descriptor refers to. The descriptor is not dup'ed, and will be closed when the returned GDBM_FILE is closed. Use dup (2) if that is not desirable.

In case of error, the function behaves like gdbm_open and does not close FD. This can be altered by the following value passed in flags:

GDBM_CLOERROR

Close FD before exiting on error.

The rest of arguments are the same as for gdbm_open.

Calling convention

All GDBM functions take as their first parameter the database handle (GDBM_FILE), returned from gdbm_open or gdbm_fd_open.

Any value stored in the GDBM database is described by datum, an aggregate type defined as:

typedef struct
{


  char *dptr;


  int   dsize;
} datum;

The dptr field points to the actual data. Its type is char * for historical reasons. Actually it should have been typed void *. Programmers are free to store data of arbitrary complexity, both scalar and aggregate, in this field.

The dsize field contains the number of bytes stored in dptr.

The datum type is used to describe both keys and content (values) in the database. Values of this type can be passed as arguments or returned from GDBM function calls.

GDBM functions that return datum indicate failure by setting its dptr field to NULL.

Functions returning integer value, indicate success by returning 0 and failure by returning a non-zero value (the only exception to this rule is gdbm_exists, for which the return value is reversed).

If the returned value indicates failure, the gdbm_errno variable contains an integer value indicating what went wrong. A similar value is associated with the dbf handle and can be accessed using the gdbm_last_errno function. Immediately after return from a function, both values are exactly equal. Subsequent GDBM calls with another dbf as argument may alter the value of the global gdbm_errno, but the value returned by gdbm_last_errno will always indicate the most recent code of an error that occurred for that particular database. Programmers are encouraged to use such per-database error codes.

Sometimes the actual reason of the failure can be clarified by examining the system errno value. To make sure its value is meaningful for a given GDBM error code, use the gdbm_check_syserr function. The function takes error code as argument and returns 1 if the errno is meaningful for that error, or 0 if it is irrelevant.

Similarly to gdbm_errno, the latest errno value associated with a particular database can be obtained using the gdbm_last_syserr function.

The gdbm_clear_error clears the error indicator (both GDBM and system error codes) associated with a database handle.

Some critical errors leave the database in a structurally inconsistent state. If that happens, all subsequent GDBM calls accessing that database will fail with the GDBM error code of GDBM_NEED_RECOVERY (a special function gdbm_needs_recovery is also provided, which returns true if the database handle given as its argument is structurally inconsistent). To return such databases to consistent state, use the gdbm_recover function (see below).

The GDBM_NEED_RECOVERY error cannot be cleared using gdbm_clear_error.

Error functions

This section describes the error handling functions outlined above.

gdbm_error gdbm_last_errno (GDBM_FILE dbf)

Returns the error code of the most recent failure encountered when operating on dbf.

int gdbm_last_syserr (GDBM_FILE dbf)

Returns the value of the system errno variable associated with the most recent failure that occurred on dbf.

Notice that not all gdbm_error codes have a relevant system error code. Use the following function to determine if a given code has.

int gdbm_check_syserr (gdbm_error err)

Returns 1, if system errno value should be checked to get more info on the error described by GDBM code err.

void gdbm_clear_error (GDBM_FILE dbf)

Clears the error state for the database dbf. This function is called implicitly upon entry to any GDBM function that operates on GDBM_FILE.

The GDBM_NEED_RECOVERY error cannot be cleared.

int gdbm_needs_recovery (GDBM_FILE dbf)

Return 1 if the database file dbf is in inconsistent state and needs recovery.

const char *gdbm_strerror (gdbm_error err)

Returns a textual description of the error code err.

const char *gdbm_db_strerror (GDBM_FILE dbf)

Returns a textual description of the recent error in database dbf. This description includes the system errno value, if relevant.

Closing the database

It is important that every database file opened is also closed. This is needed to update the reader/writer count on the file. This is done by:

int gdbm_close (GDBM_FILE dbf);

Database lookups

int gdbm_exists (GDBM_FILE dbf, datum key);

If the key is found within the database, the return value will be true (1). If nothing appropriate is found, false (0) is returned and gdbm_errno set to GDBM_NO_ERROR.

On error, returns 0 and sets gdbm_errno.

datum gdbm_fetch (GDBM_FILE dbf, datum key);

Dbf is the pointer returned by gdbm_open. Key is the key data.

If the dptr element of the return value is NULL, the gdbm_errno variable should be examined. The value of GDBM_ITEM_NOT_FOUND means no data was found for that key. Other value means an error occurred.

Otherwise the return value is a pointer to the found data. The storage space for the dptr element is allocated using malloc(3). GDBM does not automatically free this data. It is the programmer's responsibility to free this storage when it is no longer needed.

Iterating over the database

The following two routines allow for iterating over all items in the database. Such iteration is not key sequential, but it is guaranteed to visit every key in the database exactly once. (The order has to do with the hash values.)

datum gdbm_firstkey (GDBM_FILE dbf);

Returns first key in the database.

datum gdbm_nextkey (GDBM_FILE dbf, datum key);

Given a key, returns the database key that follows it. End of iteration is marked by returning datum with dptr field set to NULL and setting the gdbm_errno value to GDBM_ITEM_NOT_FOUND.

After successful return from both functions, dptr points to data allocated by malloc(3). It is the caller responsibility to free the data when no longer needed.

A typical iteration loop looks like:

datum key, nextkey, content;
key = gdbm_firstkey (dbf);
while (key.dptr)


  {


    content = gdbm_fetch (dbf, key);


    /* Do something with key and/or content */


    nextkey = gdbm_nextkey (dbf, key);


    free (key.dptr);


    key = nextkey;


  }

These functions are intended to visit the database in read-only algorithms. Avoid any database modifications within the iteration loop. File visiting is based on a hash table. The gdbm_delete and, in most cases, gdbm_store, functions rearrange the hash table to make sure that any collisions in the table do not leave some item `un-findable'. Thus, a call to either of these functions changes the order in which the keys are ordered. Therefore, these functions should not be used when iterating over all the keys in the database. For example, the following loop is wrong: it is possible that some keys will not be visited or will be visited twice if it is executed:

key = gdbm_firstkey (dbf);
while (key.dptr)


  {


    nextkey = gdbm_nextkey (dbf, key);


    if (some condition)


      gdbm_delete ( dbf, key );


    free (key.dptr);


    key = nextkey;


  }

Updating the database

int gdbm_store (GDBM_FILE dbf, datum key, datum content, int flag);

Dbf is the pointer returned by gdbm_open. Key is the key data. Content is the data to be associated with the key. Flag can have one of the following values:

The function returns 0 on success and -1 on failure. If the key already exists in the database and the flag is GDBM_INSERT, the function does not modify the database. It sets gdbm_errno to GDBM_CANNOT_REPLACE and returns 1.

int gdbm_delete (GDBM_FILE dbf, datum key);

Looks up and deletes the given key from the database dbf.

The return value is 0 if there was a successful delete or -1 on error. In the latter case, the gdbm_errno value GDBM_ITEM_NOT_FOUND indicates that the key is not present in the database. Other gdbm_errno values indicate failure.

Recovering structural consistency

If a function leaves the database in structurally inconsistent state, it can be recovered using the gdbm_recover function.

int gdbm_recover (GDBM_FILE dbf, gdbm_recovery * rcvr, int flags)

Check the database file DBF and fix eventual inconsistencies. The rcvr argument can be used both to control the recovery and to return additional statistics about the process, as indicated by flags. For a detailed discussion of these arguments and their usage, see the GDBM Manual, chapter Recovering structural consistency.

You can pass NULL as rcvr and 0 as flags, if no such control is needed.

By default, this function first checks the database for inconsistencies and attempts recovery only if some were found. The special flags bit GDBM_RCVR_FORCE instructs gdbm_recovery to skip this check and to perform database recovery unconditionally.

Export and import

GDBM database files can be exported (dumped) to so called flat files or imported (loaded) from them. A flat file contains exactly the same data as the original database, but it cannot be used for searches or updates. Its purpose is to keep the data from the database for restoring it when the need arrives. As such, flat files are used for backup purposes, and for sending databases over the wire.

As of GDBM version 1.21, there are two flat file formats. The ASCII file format encodes all data in Base64 and stores not only key/data pairs, but also the original database file metadata, such as file name, mode and ownership. Files in this format can be sent without additional encapsulation over transmission channels that normally allow only ASCII data, such as, e.g. SMTP. Due to additional metadata they allow for restoring an exact copy of the database, including file ownership and privileges, which is especially important if the database in question contained some security-related data. This is the preferred format.

Another flat file format is the binary format. It stores only key/data pairs and does not keep information about the database file itself. It cannot be used to copy databases between different architectures. The binary format was introduced in GDBM version 1.9.1 and is retained mainly for backward compatibility.

The following functions are used to export or import GDBM database files.

int gdbm_dump (GDBM_FILE dbf, const char *filename,

The value of open_flag tells gdbm_dump what to do if filename already exists. If it is GDBM_NEWDB, the function will create a new output file, replacing it if it already exists. If its value is GDBM_WRCREAT, the file will be created if it does not exist. If it does exist, gdbm_dump will return error.

The file mode to use when creating the output file is defined by the mode parameter. Its meaning is the same as for open(2).

int gdbm_load (GDBM_FILE *pdbf, const char *filename,

Otherwise, if pdbf points to an already open GDBM_FILE, the function will load data from filename into that database.

The flag parameter controls the function behavior if a key from the dump file already exists in the database. See the gdbm_store function for its possible values.

The meta_mask parameter can be used to disable restoring certain bits of file's meta-data from the information in the input dump file. It is a binary OR of zero or more of the following:

Other functions

int gdbm_reorganize (GDBM_FILE dbf);

If you have had a lot of deletions and would like to shrink the space used by the GDBM file, this routine will reorganize the database.

int gdbm_sync (GDBM_FILE dbf);

Synchronizes the changes in dbf with its disk file.

It will not return until the disk file state is synchronized with the in-memory state of the database.

int gdbm_setopt (GDBM_FILE dbf, int option, void *value, int size);

Query or change some parameter of an already opened database. The option argument defines what parameter to set or retrieve. If the set operation is requested, value points to the new value. Its actual data type depends on option. If the get operation is requested, value points to a memory region where to store the return value. In both cases, size contains the actual size of the memory pointed to by value.

Possible values of option are:

int gdbm_fdesc (GDBM_FILE dbf);

Returns the file descriptor of the database dbf.

DBM compatibility routines

In dbm compatibility mode only one file may be opened at a time. All users are assumed to be writers. If the database file is read only, it will fail as a writer, but will be opened as a reader. All returned pointers in datum structures point to data that the compatibility library will free. They should be treated as static pointers (as standard UNIX dbm does).

The following interfaces are provided:

#include <dbm.h>

int dbminit (const char *name);
int store (datum key, datum content);
datum fetch (datum key);
int delete (datum key);
datum firstkey (void);
datum nextkey (datum key);
int dbmclose (void);

NDBM Compatibility routines:

In this mode, multiple databases can be opened. Each database is identified by a handle of type DBM *. As in the original NDBM, all returned pointers in datum structures point to data that will be freed by the compatibility library. They should be treated as static pointers.

The following interfaces are provided:

#include <ndbm.h>

DBM *dbm_open (const char *name, int flags, int mode);
void dbm_close (DBM *file);
datum dbm_fetch (DBM *file, datum key);
int dbm_store (DBM *file, datum key, datum content, int flags);
int dbm_delete (DBM *file, datum key);
datum dbm_firstkey (DBM *file);
datum dbm_nextkey (DBM *file, datum key);
int dbm_error (DBM *file);
int dbm_clearerr (DBM *file);
int dbm_pagfno (DBM *file);
int dbm_dirfno (DBM *file);
int dbm_rdonly (DBM *file);