OS Interface Object

*** DRAFT ***

SQLite C Interface

OS Interface Object

typedef struct sqlite3_vfs sqlite3_vfs;
struct sqlite3_vfs {
  int iVersion;            /* Structure version number (currently 2) */
  int szOsFile;            /* Size of subclassed sqlite3_file */
  int mxPathname;          /* Maximum file pathname length */
  sqlite3_vfs *pNext;      /* Next registered VFS */
  const char *zName;       /* Name of this virtual file system */
  void *pAppData;          /* Pointer to application-specific data */
  int (*xOpen)(sqlite3_vfs*, const char *zName, sqlite3_file*,
               int flags, int *pOutFlags);
  int (*xDelete)(sqlite3_vfs*, const char *zName, int syncDir);
  int (*xAccess)(sqlite3_vfs*, const char *zName, int flags, int *pResOut);
  int (*xFullPathname)(sqlite3_vfs*, const char *zName, int nOut, char *zOut);
  void *(*xDlOpen)(sqlite3_vfs*, const char *zFilename);
  void (*xDlError)(sqlite3_vfs*, int nByte, char *zErrMsg);
  void (*(*xDlSym)(sqlite3_vfs*,void*, const char *zSymbol))(void);
  void (*xDlClose)(sqlite3_vfs*, void*);
  int (*xRandomness)(sqlite3_vfs*, int nByte, char *zOut);
  int (*xSleep)(sqlite3_vfs*, int microseconds);
  int (*xCurrentTime)(sqlite3_vfs*, double*);
  int (*xGetLastError)(sqlite3_vfs*, int, char *);
  /*
  ** The methods above are in version 1 of the sqlite_vfs object
  ** definition.  Those that follow are added in version 2 or later
  */
  int (*xCurrentTimeInt64)(sqlite3_vfs*, sqlite3_int64*);
  /*
  ** The methods above are in versions 1 and 2 of the sqlite_vfs object.
  ** New fields may be appended in figure versions.  The iVersion
  ** value will increment whenever this happens. 
  */
};

An instance of the sqlite3_vfs object defines the interface between the SQLite core and the underlying operating system. The "vfs" in the name of the object stands for "virtual file system".

The value of the iVersion field is initially 1 but may be larger in future versions of SQLite. Additional fields may be appended to this object when the iVersion value is increased. Note that the structure of the sqlite3_vfs object changes in the transaction between SQLite version 3.5.9 and 3.6.0 and yet the iVersion field was not modified.

The szOsFile field is the size of the subclassed sqlite3_file structure used by this VFS. mxPathname is the maximum length of a pathname in this VFS.

Registered sqlite3_vfs objects are kept on a linked list formed by the pNext pointer. The sqlite3_vfs_register() and sqlite3_vfs_unregister() interfaces manage this list in a thread-safe way. The sqlite3_vfs_find() interface searches the list. Neither the application code nor the VFS implementation should use the pNext pointer.

The pNext field is the only field in the sqlite3_vfs structure that SQLite will ever modify. SQLite will only access or modify this field while holding a particular static mutex. The application should never modify anything within the sqlite3_vfs object once the object has been registered.

The zName field holds the name of the VFS module. The name must be unique across all VFS modules.

SQLite will guarantee that the zFilename parameter to xOpen is either a NULL pointer or string obtained from xFullPathname(). SQLite further guarantees that the string will be valid and unchanged until xClose() is called. Because of the previous sentence, the sqlite3_file can safely store a pointer to the filename if it needs to remember the filename for some reason. If the zFilename parameter is xOpen is a NULL pointer then xOpen must invent its own temporary name for the file. Whenever the xFilename parameter is NULL it will also be the case that the flags parameter will include SQLITE_OPEN_DELETEONCLOSE.

The flags argument to xOpen() includes all bits set in the flags argument to sqlite3_open_v2(). Or if sqlite3_open() or sqlite3_open16() is used, then flags includes at least SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE. If xOpen() opens a file read-only then it sets *pOutFlags to include SQLITE_OPEN_READONLY. Other bits in *pOutFlags may be set.

SQLite will also add one of the following flags to the xOpen() call, depending on the object being opened:

The file I/O implementation can use the object type flags to change the way it deals with files. For example, an application that does not care about crash recovery or rollback might make the open of a journal file a no-op. Writes to this journal would also be no-ops, and any attempt to read the journal would return SQLITE_IOERR. Or the implementation might recognize that a database file will be doing page-aligned sector reads and writes in a random order and set up its I/O subsystem accordingly.

SQLite might also add one of the following flags to the xOpen method:

The SQLITE_OPEN_DELETEONCLOSE flag means the file should be deleted when it is closed. The SQLITE_OPEN_DELETEONCLOSE will be set for TEMP databases, journals and for subjournals.

The SQLITE_OPEN_EXCLUSIVE flag is always used in conjunction with the SQLITE_OPEN_CREATE flag, which are both directly analogous to the O_EXCL and O_CREAT flags of the POSIX open() API. The SQLITE_OPEN_EXCLUSIVE flag, when paired with the SQLITE_OPEN_CREATE, is used to indicate that file should always be created, and that it is an error if it already exists. It is not used to indicate the file should be opened for exclusive access.

At least szOsFile bytes of memory are allocated by SQLite to hold the sqlite3_file structure passed as the third argument to xOpen. The xOpen method does not have to allocate the structure; it should just fill it in. Note that the xOpen method must set the sqlite3_file.pMethods to either a valid sqlite3_io_methods object or to NULL. xOpen must do this even if the open fails. SQLite expects that the sqlite3_file.pMethods element will be valid after xOpen returns regardless of the success or failure of the xOpen call.

The flags argument to xAccess() may be SQLITE_ACCESS_EXISTS to test for the existence of a file, or SQLITE_ACCESS_READWRITE to test whether a file is readable and writable, or SQLITE_ACCESS_READ to test whether a file is at least readable. The file can be a directory.

SQLite will always allocate at least mxPathname+1 bytes for the output buffer xFullPathname. The exact size of the output buffer is also passed as a parameter to both methods. If the output buffer is not large enough, SQLITE_CANTOPEN should be returned. Since this is handled as a fatal error by SQLite, vfs implementations should endeavor to prevent this by setting mxPathname to a sufficiently large value.

The xRandomness(), xSleep(), xCurrentTime(), and xCurrentTimeInt64() interfaces are not strictly a part of the filesystem, but they are included in the VFS structure for completeness. The xRandomness() function attempts to return nBytes bytes of good-quality randomness into zOut. The return value is the actual number of bytes of randomness obtained. The xSleep() method causes the calling thread to sleep for at least the number of microseconds given. The xCurrentTime() method returns a Julian Day Number for the current date and time as a floating point value. The xCurrentTimeInt64() method returns, as an integer, the Julian Day Number multipled by 86400000 (the number of milliseconds in a 24-hour day). SQLite will use the xCurrentTimeInt64() method to get the current date and time if that method is available (if iVersion is 2 or greater and the function pointer is not NULL) and will fall back to xCurrentTime() if xCurrentTimeInt64() is unavailable.

See also lists of Objects, Constants, and Functions.

*** DRAFT ***