4 ** The author disclaims copyright to this source code. In place of
5 ** a legal notice, here is a blessing:
7 ** May you do good and not evil.
8 ** May you find forgiveness for yourself and forgive others.
9 ** May you share freely, never taking more than you give.
11 *************************************************************************
13 ** This file implements an example of a simple VFS implementation that
14 ** omits complex features often not required or not possible on embedded
15 ** platforms. Code is included to buffer writes to the journal file,
16 ** which can be a significant performance improvement on some embedded
21 ** The code in this file implements a minimal SQLite VFS that can be
22 ** used on Linux and other posix-like operating systems. The following
23 ** system calls are used:
25 ** File-system: access(), unlink(), getcwd()
26 ** File IO: open(), read(), write(), fsync(), close(), fstat()
27 ** Other: sleep(), usleep(), time()
29 ** The following VFS features are omitted:
31 ** 1. File locking. The user must ensure that there is at most one
32 ** connection to each database when using this VFS. Multiple
33 ** connections to a single shared-cache count as a single connection
34 ** for the purposes of the previous statement.
36 ** 2. The loading of dynamic extensions (shared libraries).
38 ** 3. Temporary files. The user must configure SQLite to use in-memory
39 ** temp files when using this VFS. The easiest way to do this is to
42 ** -DSQLITE_TEMP_STORE=3
44 ** 4. File truncation. As of version 3.6.24, SQLite may run without
45 ** a working xTruncate() call, providing the user does not configure
46 ** SQLite to use "journal_mode=truncate", or use both
47 ** "journal_mode=persist" and ATTACHed databases.
49 ** It is assumed that the system uses UNIX-like path-names. Specifically,
50 ** that '/' characters are used to separate path components and that
51 ** a path-name is a relative path unless it begins with a '/'. And that
52 ** no UTF-8 encoded paths are greater than 512 bytes in length.
54 ** JOURNAL WRITE-BUFFERING
56 ** To commit a transaction to the database, SQLite first writes rollback
57 ** information into the journal file. This usually consists of 4 steps:
59 ** 1. The rollback information is sequentially written into the journal
60 ** file, starting at the start of the file.
61 ** 2. The journal file is synced to disk.
62 ** 3. A modification is made to the first few bytes of the journal file.
63 ** 4. The journal file is synced to disk again.
65 ** Most of the data is written in step 1 using a series of calls to the
66 ** VFS xWrite() method. The buffers passed to the xWrite() calls are of
67 ** various sizes. For example, as of version 3.6.24, when committing a
68 ** transaction that modifies 3 pages of a database file that uses 4096
69 ** byte pages residing on a media with 512 byte sectors, SQLite makes
70 ** eleven calls to the xWrite() method to create the rollback journal,
73 ** Write offset | Bytes written
74 ** ----------------------------
85 ** ++++++++++++SYNC+++++++++++
87 ** ++++++++++++SYNC+++++++++++
89 ** On many operating systems, this is an efficient way to write to a file.
90 ** However, on some embedded systems that do not cache writes in OS
91 ** buffers it is much more efficient to write data in blocks that are
92 ** an integer multiple of the sector-size in size and aligned at the
95 ** To work around this, the code in this file allocates a fixed size
96 ** buffer of SQLITE_DEMOVFS_BUFFERSZ using sqlite3_malloc() whenever a
97 ** journal file is opened. It uses the buffer to coalesce sequential
98 ** writes into aligned SQLITE_DEMOVFS_BUFFERSZ blocks. When SQLite
99 ** invokes the xSync() method to sync the contents of the file to disk,
100 ** all accumulated data is written out, even if it does not constitute
101 ** a complete block. This means the actual IO to create the rollback
102 ** journal for the example transaction above is this:
104 ** Write offset | Bytes written
105 ** ----------------------------
108 ** ++++++++++++SYNC+++++++++++
110 ** ++++++++++++SYNC+++++++++++
112 ** Much more efficient if the underlying OS is not caching write
116 #if !defined(SQLITE_TEST) || SQLITE_OS_UNIX
122 #include <sys/types.h>
123 #include <sys/stat.h>
124 #include <sys/file.h>
125 #include <sys/param.h>
132 ** Size of the write buffer used by journal files in bytes.
134 #ifndef SQLITE_DEMOVFS_BUFFERSZ
135 # define SQLITE_DEMOVFS_BUFFERSZ 8192
139 ** The maximum pathname length supported by this VFS.
141 #define MAXPATHNAME 512
144 ** When using this VFS, the sqlite3_file* handles that SQLite uses are
145 ** actually pointers to instances of type DemoFile.
147 typedef struct DemoFile DemoFile
;
149 sqlite3_file base
; /* Base class. Must be first. */
150 int fd
; /* File descriptor */
152 char *aBuffer
; /* Pointer to malloc'd buffer */
153 int nBuffer
; /* Valid bytes of data in zBuffer */
154 sqlite3_int64 iBufferOfst
; /* Offset in file of zBuffer[0] */
158 ** Write directly to the file passed as the first argument. Even if the
159 ** file has a write-buffer (DemoFile.aBuffer), ignore it.
161 static int demoDirectWrite(
162 DemoFile
*p
, /* File handle */
163 const void *zBuf
, /* Buffer containing data to write */
164 int iAmt
, /* Size of data to write in bytes */
165 sqlite_int64 iOfst
/* File offset to write to */
167 off_t ofst
; /* Return value from lseek() */
168 size_t nWrite
; /* Return value from write() */
170 ofst
= lseek(p
->fd
, iOfst
, SEEK_SET
);
172 return SQLITE_IOERR_WRITE
;
175 nWrite
= write(p
->fd
, zBuf
, iAmt
);
177 return SQLITE_IOERR_WRITE
;
184 ** Flush the contents of the DemoFile.aBuffer buffer to disk. This is a
185 ** no-op if this particular file does not have a buffer (i.e. it is not
186 ** a journal file) or if the buffer is currently empty.
188 static int demoFlushBuffer(DemoFile
*p
){
191 rc
= demoDirectWrite(p
, p
->aBuffer
, p
->nBuffer
, p
->iBufferOfst
);
200 static int demoClose(sqlite3_file
*pFile
){
202 DemoFile
*p
= (DemoFile
*)pFile
;
203 rc
= demoFlushBuffer(p
);
204 sqlite3_free(p
->aBuffer
);
210 ** Read data from a file.
218 DemoFile
*p
= (DemoFile
*)pFile
;
219 off_t ofst
; /* Return value from lseek() */
220 int nRead
; /* Return value from read() */
221 int rc
; /* Return code from demoFlushBuffer() */
223 /* Flush any data in the write buffer to disk in case this operation
224 ** is trying to read data the file-region currently cached in the buffer.
225 ** It would be possible to detect this case and possibly save an
226 ** unnecessary write here, but in practice SQLite will rarely read from
227 ** a journal file when there is data cached in the write-buffer.
229 rc
= demoFlushBuffer(p
);
234 ofst
= lseek(p
->fd
, iOfst
, SEEK_SET
);
236 return SQLITE_IOERR_READ
;
238 nRead
= read(p
->fd
, zBuf
, iAmt
);
242 }else if( nRead
>=0 ){
243 return SQLITE_IOERR_SHORT_READ
;
246 return SQLITE_IOERR_READ
;
250 ** Write data to a crash-file.
252 static int demoWrite(
258 DemoFile
*p
= (DemoFile
*)pFile
;
261 char *z
= (char *)zBuf
; /* Pointer to remaining data to write */
262 int n
= iAmt
; /* Number of bytes at z */
263 sqlite3_int64 i
= iOfst
; /* File offset to write to */
266 int nCopy
; /* Number of bytes to copy into buffer */
268 /* If the buffer is full, or if this data is not being written directly
269 ** following the data already buffered, flush the buffer. Flushing
270 ** the buffer is a no-op if it is empty.
272 if( p
->nBuffer
==SQLITE_DEMOVFS_BUFFERSZ
|| p
->iBufferOfst
+p
->nBuffer
!=i
){
273 int rc
= demoFlushBuffer(p
);
278 assert( p
->nBuffer
==0 || p
->iBufferOfst
+p
->nBuffer
==i
);
279 p
->iBufferOfst
= i
- p
->nBuffer
;
281 /* Copy as much data as possible into the buffer. */
282 nCopy
= SQLITE_DEMOVFS_BUFFERSZ
- p
->nBuffer
;
286 memcpy(&p
->aBuffer
[p
->nBuffer
], z
, nCopy
);
294 return demoDirectWrite(p
, zBuf
, iAmt
, iOfst
);
301 ** Truncate a file. This is a no-op for this VFS (see header comments at
302 ** the top of the file).
304 static int demoTruncate(sqlite3_file
*pFile
, sqlite_int64 size
){
306 if( ftruncate(((DemoFile
*)pFile
)->fd
, size
) ) return SQLITE_IOERR_TRUNCATE
;
312 ** Sync the contents of the file to the persistent media.
314 static int demoSync(sqlite3_file
*pFile
, int flags
){
315 DemoFile
*p
= (DemoFile
*)pFile
;
318 rc
= demoFlushBuffer(p
);
324 return (rc
==0 ? SQLITE_OK
: SQLITE_IOERR_FSYNC
);
328 ** Write the size of the file in bytes to *pSize.
330 static int demoFileSize(sqlite3_file
*pFile
, sqlite_int64
*pSize
){
331 DemoFile
*p
= (DemoFile
*)pFile
;
332 int rc
; /* Return code from fstat() call */
333 struct stat sStat
; /* Output of fstat() call */
335 /* Flush the contents of the buffer to disk. As with the flush in the
336 ** demoRead() method, it would be possible to avoid this and save a write
337 ** here and there. But in practice this comes up so infrequently it is
338 ** not worth the trouble.
340 rc
= demoFlushBuffer(p
);
345 rc
= fstat(p
->fd
, &sStat
);
346 if( rc
!=0 ) return SQLITE_IOERR_FSTAT
;
347 *pSize
= sStat
.st_size
;
352 ** Locking functions. The xLock() and xUnlock() methods are both no-ops.
353 ** The xCheckReservedLock() always indicates that no other process holds
354 ** a reserved lock on the database file. This ensures that if a hot-journal
355 ** file is found in the file-system it is rolled back.
357 static int demoLock(sqlite3_file
*pFile
, int eLock
){
360 static int demoUnlock(sqlite3_file
*pFile
, int eLock
){
363 static int demoCheckReservedLock(sqlite3_file
*pFile
, int *pResOut
){
369 ** No xFileControl() verbs are implemented by this VFS.
371 static int demoFileControl(sqlite3_file
*pFile
, int op
, void *pArg
){
376 ** The xSectorSize() and xDeviceCharacteristics() methods. These two
377 ** may return special values allowing SQLite to optimize file-system
378 ** access to some extent. But it is also safe to simply return 0.
380 static int demoSectorSize(sqlite3_file
*pFile
){
383 static int demoDeviceCharacteristics(sqlite3_file
*pFile
){
388 ** Open a file handle.
391 sqlite3_vfs
*pVfs
, /* VFS */
392 const char *zName
, /* File to open, or 0 for a temp file */
393 sqlite3_file
*pFile
, /* Pointer to DemoFile struct to populate */
394 int flags
, /* Input SQLITE_OPEN_XXX flags */
395 int *pOutFlags
/* Output SQLITE_OPEN_XXX flags (or NULL) */
397 static const sqlite3_io_methods demoio
= {
399 demoClose
, /* xClose */
400 demoRead
, /* xRead */
401 demoWrite
, /* xWrite */
402 demoTruncate
, /* xTruncate */
403 demoSync
, /* xSync */
404 demoFileSize
, /* xFileSize */
405 demoLock
, /* xLock */
406 demoUnlock
, /* xUnlock */
407 demoCheckReservedLock
, /* xCheckReservedLock */
408 demoFileControl
, /* xFileControl */
409 demoSectorSize
, /* xSectorSize */
410 demoDeviceCharacteristics
/* xDeviceCharacteristics */
413 DemoFile
*p
= (DemoFile
*)pFile
; /* Populate this structure */
414 int oflags
= 0; /* flags to pass to open() call */
421 if( flags
&SQLITE_OPEN_MAIN_JOURNAL
){
422 aBuf
= (char *)sqlite3_malloc(SQLITE_DEMOVFS_BUFFERSZ
);
428 if( flags
&SQLITE_OPEN_EXCLUSIVE
) oflags
|= O_EXCL
;
429 if( flags
&SQLITE_OPEN_CREATE
) oflags
|= O_CREAT
;
430 if( flags
&SQLITE_OPEN_READONLY
) oflags
|= O_RDONLY
;
431 if( flags
&SQLITE_OPEN_READWRITE
) oflags
|= O_RDWR
;
433 memset(p
, 0, sizeof(DemoFile
));
434 p
->fd
= open(zName
, oflags
, 0600);
437 return SQLITE_CANTOPEN
;
444 p
->base
.pMethods
= &demoio
;
449 ** Delete the file identified by argument zPath. If the dirSync parameter
450 ** is non-zero, then ensure the file-system modification to delete the
451 ** file has been synced to disk before returning.
453 static int demoDelete(sqlite3_vfs
*pVfs
, const char *zPath
, int dirSync
){
454 int rc
; /* Return code */
457 if( rc
!=0 && errno
==ENOENT
) return SQLITE_OK
;
459 if( rc
==0 && dirSync
){
460 int dfd
; /* File descriptor open on directory */
461 int i
; /* Iterator variable */
462 char zDir
[MAXPATHNAME
+1]; /* Name of directory containing file zPath */
464 /* Figure out the directory name from the path of the file deleted. */
465 sqlite3_snprintf(MAXPATHNAME
, zDir
, "%s", zPath
);
466 zDir
[MAXPATHNAME
] = '\0';
467 for(i
=strlen(zDir
); i
>1 && zDir
[i
]!='/'; i
++);
470 /* Open a file-descriptor on the directory. Sync. Close. */
471 dfd
= open(zDir
, O_RDONLY
, 0);
479 return (rc
==0 ? SQLITE_OK
: SQLITE_IOERR_DELETE
);
493 ** Query the file-system to see if the named file exists, is readable or
494 ** is both readable and writable.
496 static int demoAccess(
502 int rc
; /* access() return code */
503 int eAccess
= F_OK
; /* Second argument to access() */
505 assert( flags
==SQLITE_ACCESS_EXISTS
/* access(zPath, F_OK) */
506 || flags
==SQLITE_ACCESS_READ
/* access(zPath, R_OK) */
507 || flags
==SQLITE_ACCESS_READWRITE
/* access(zPath, R_OK|W_OK) */
510 if( flags
==SQLITE_ACCESS_READWRITE
) eAccess
= R_OK
|W_OK
;
511 if( flags
==SQLITE_ACCESS_READ
) eAccess
= R_OK
;
513 rc
= access(zPath
, eAccess
);
519 ** Argument zPath points to a nul-terminated string containing a file path.
520 ** If zPath is an absolute path, then it is copied as is into the output
521 ** buffer. Otherwise, if it is a relative path, then the equivalent full
522 ** path is written to the output buffer.
524 ** This function assumes that paths are UNIX style. Specifically, that:
526 ** 1. Path components are separated by a '/'. and
527 ** 2. Full paths begin with a '/' character.
529 static int demoFullPathname(
530 sqlite3_vfs
*pVfs
, /* VFS */
531 const char *zPath
, /* Input path (possibly a relative path) */
532 int nPathOut
, /* Size of output buffer in bytes */
533 char *zPathOut
/* Pointer to output buffer */
535 char zDir
[MAXPATHNAME
+1];
539 if( getcwd(zDir
, sizeof(zDir
))==0 ) return SQLITE_IOERR
;
541 zDir
[MAXPATHNAME
] = '\0';
543 sqlite3_snprintf(nPathOut
, zPathOut
, "%s/%s", zDir
, zPath
);
544 zPathOut
[nPathOut
-1] = '\0';
550 ** The following four VFS methods:
557 ** are supposed to implement the functionality needed by SQLite to load
558 ** extensions compiled as shared objects. This simple VFS does not support
559 ** this functionality, so the following functions are no-ops.
561 static void *demoDlOpen(sqlite3_vfs
*pVfs
, const char *zPath
){
564 static void demoDlError(sqlite3_vfs
*pVfs
, int nByte
, char *zErrMsg
){
565 sqlite3_snprintf(nByte
, zErrMsg
, "Loadable extensions are not supported");
566 zErrMsg
[nByte
-1] = '\0';
568 static void (*demoDlSym(sqlite3_vfs
*pVfs
, void *pH
, const char *z
))(void){
571 static void demoDlClose(sqlite3_vfs
*pVfs
, void *pHandle
){
576 ** Parameter zByte points to a buffer nByte bytes in size. Populate this
577 ** buffer with pseudo-random data.
579 static int demoRandomness(sqlite3_vfs
*pVfs
, int nByte
, char *zByte
){
584 ** Sleep for at least nMicro microseconds. Return the (approximate) number
585 ** of microseconds slept for.
587 static int demoSleep(sqlite3_vfs
*pVfs
, int nMicro
){
588 sleep(nMicro
/ 1000000);
589 usleep(nMicro
% 1000000);
594 ** Set *pTime to the current UTC time expressed as a Julian day. Return
595 ** SQLITE_OK if successful, or an error code otherwise.
597 ** http://en.wikipedia.org/wiki/Julian_day
599 ** This implementation is not very good. The current time is rounded to
600 ** an integer number of seconds. Also, assuming time_t is a signed 32-bit
601 ** value, it will stop working some time in the year 2038 AD (the so-called
602 ** "year 2038" problem that afflicts systems that store time this way).
604 static int demoCurrentTime(sqlite3_vfs
*pVfs
, double *pTime
){
606 *pTime
= t
/86400.0 + 2440587.5;
611 ** This function returns a pointer to the VFS implemented in this file.
612 ** To make the VFS available to SQLite:
614 ** sqlite3_vfs_register(sqlite3_demovfs(), 0);
616 sqlite3_vfs
*sqlite3_demovfs(void){
617 static sqlite3_vfs demovfs
= {
619 sizeof(DemoFile
), /* szOsFile */
620 MAXPATHNAME
, /* mxPathname */
624 demoOpen
, /* xOpen */
625 demoDelete
, /* xDelete */
626 demoAccess
, /* xAccess */
627 demoFullPathname
, /* xFullPathname */
628 demoDlOpen
, /* xDlOpen */
629 demoDlError
, /* xDlError */
630 demoDlSym
, /* xDlSym */
631 demoDlClose
, /* xDlClose */
632 demoRandomness
, /* xRandomness */
633 demoSleep
, /* xSleep */
634 demoCurrentTime
, /* xCurrentTime */
639 #endif /* !defined(SQLITE_TEST) || SQLITE_OS_UNIX */
647 static int register_demovfs(
648 ClientData clientData
, /* Pointer to sqlite3_enable_XXX function */
649 Tcl_Interp
*interp
, /* The TCL interpreter that invoked this command */
650 int objc
, /* Number of arguments */
651 Tcl_Obj
*CONST objv
[] /* Command arguments */
653 sqlite3_vfs_register(sqlite3_demovfs(), 1);
656 static int unregister_demovfs(
657 ClientData clientData
, /* Pointer to sqlite3_enable_XXX function */
658 Tcl_Interp
*interp
, /* The TCL interpreter that invoked this command */
659 int objc
, /* Number of arguments */
660 Tcl_Obj
*CONST objv
[] /* Command arguments */
662 sqlite3_vfs_unregister(sqlite3_demovfs());
667 ** Register commands with the TCL interpreter.
669 int Sqlitetest_demovfs_Init(Tcl_Interp
*interp
){
670 Tcl_CreateObjCommand(interp
, "register_demovfs", register_demovfs
, 0, 0);
671 Tcl_CreateObjCommand(interp
, "unregister_demovfs", unregister_demovfs
, 0, 0);
676 int Sqlitetest_demovfs_Init(Tcl_Interp
*interp
){ return TCL_OK
; }
679 #endif /* SQLITE_TEST */