blob: 4450b7a770b0331b594ea02a6202c9bbe7db767d [file] [log] [blame]
Chih-Hung Hsiehcc6e6352020-10-26 19:08:21 -07001/*
2** 2010 April 7
3**
4** The author disclaims copyright to this source code. In place of
5** a legal notice, here is a blessing:
6**
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.
10**
11*************************************************************************
12**
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
17** platforms.
18**
19** OVERVIEW
20**
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:
24**
25** File-system: access(), unlink(), getcwd()
26** File IO: open(), read(), write(), fsync(), close(), fstat()
27** Other: sleep(), usleep(), time()
28**
29** The following VFS features are omitted:
30**
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.
35**
36** 2. The loading of dynamic extensions (shared libraries).
37**
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
40** compile with:
41**
42** -DSQLITE_TEMP_STORE=3
43**
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.
48**
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.
53**
54** JOURNAL WRITE-BUFFERING
55**
56** To commit a transaction to the database, SQLite first writes rollback
57** information into the journal file. This usually consists of 4 steps:
58**
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.
64**
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,
71** as follows:
72**
73** Write offset | Bytes written
74** ----------------------------
75** 0 512
76** 512 4
77** 516 4096
78** 4612 4
79** 4616 4
80** 4620 4096
81** 8716 4
82** 8720 4
83** 8724 4096
84** 12820 4
85** ++++++++++++SYNC+++++++++++
86** 0 12
87** ++++++++++++SYNC+++++++++++
88**
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
93** start of a sector.
94**
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:
103**
104** Write offset | Bytes written
105** ----------------------------
106** 0 8192
107** 8192 4632
108** ++++++++++++SYNC+++++++++++
109** 0 12
110** ++++++++++++SYNC+++++++++++
111**
112** Much more efficient if the underlying OS is not caching write
113** operations.
114*/
115
116#if !defined(SQLITE_TEST) || SQLITE_OS_UNIX
117
118#include "sqlite3.h"
119
120#include <assert.h>
121#include <string.h>
122#include <sys/types.h>
123#include <sys/stat.h>
124#include <sys/file.h>
125#include <sys/param.h>
126#include <unistd.h>
127#include <time.h>
128#include <errno.h>
129#include <fcntl.h>
130
131/*
132** Size of the write buffer used by journal files in bytes.
133*/
134#ifndef SQLITE_DEMOVFS_BUFFERSZ
135# define SQLITE_DEMOVFS_BUFFERSZ 8192
136#endif
137
138/*
139** The maximum pathname length supported by this VFS.
140*/
141#define MAXPATHNAME 512
142
143/*
144** When using this VFS, the sqlite3_file* handles that SQLite uses are
145** actually pointers to instances of type DemoFile.
146*/
147typedef struct DemoFile DemoFile;
148struct DemoFile {
149 sqlite3_file base; /* Base class. Must be first. */
150 int fd; /* File descriptor */
151
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] */
155};
156
157/*
158** Write directly to the file passed as the first argument. Even if the
159** file has a write-buffer (DemoFile.aBuffer), ignore it.
160*/
161static 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 */
166){
167 off_t ofst; /* Return value from lseek() */
168 size_t nWrite; /* Return value from write() */
169
170 ofst = lseek(p->fd, iOfst, SEEK_SET);
171 if( ofst!=iOfst ){
172 return SQLITE_IOERR_WRITE;
173 }
174
175 nWrite = write(p->fd, zBuf, iAmt);
176 if( nWrite!=iAmt ){
177 return SQLITE_IOERR_WRITE;
178 }
179
180 return SQLITE_OK;
181}
182
183/*
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.
187*/
188static int demoFlushBuffer(DemoFile *p){
189 int rc = SQLITE_OK;
190 if( p->nBuffer ){
191 rc = demoDirectWrite(p, p->aBuffer, p->nBuffer, p->iBufferOfst);
192 p->nBuffer = 0;
193 }
194 return rc;
195}
196
197/*
198** Close a file.
199*/
200static int demoClose(sqlite3_file *pFile){
201 int rc;
202 DemoFile *p = (DemoFile*)pFile;
203 rc = demoFlushBuffer(p);
204 sqlite3_free(p->aBuffer);
205 close(p->fd);
206 return rc;
207}
208
209/*
210** Read data from a file.
211*/
212static int demoRead(
213 sqlite3_file *pFile,
214 void *zBuf,
215 int iAmt,
216 sqlite_int64 iOfst
217){
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() */
222
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.
228 */
229 rc = demoFlushBuffer(p);
230 if( rc!=SQLITE_OK ){
231 return rc;
232 }
233
234 ofst = lseek(p->fd, iOfst, SEEK_SET);
235 if( ofst!=iOfst ){
236 return SQLITE_IOERR_READ;
237 }
238 nRead = read(p->fd, zBuf, iAmt);
239
240 if( nRead==iAmt ){
241 return SQLITE_OK;
242 }else if( nRead>=0 ){
243 return SQLITE_IOERR_SHORT_READ;
244 }
245
246 return SQLITE_IOERR_READ;
247}
248
249/*
250** Write data to a crash-file.
251*/
252static int demoWrite(
253 sqlite3_file *pFile,
254 const void *zBuf,
255 int iAmt,
256 sqlite_int64 iOfst
257){
258 DemoFile *p = (DemoFile*)pFile;
259
260 if( p->aBuffer ){
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 */
264
265 while( n>0 ){
266 int nCopy; /* Number of bytes to copy into buffer */
267
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.
271 */
272 if( p->nBuffer==SQLITE_DEMOVFS_BUFFERSZ || p->iBufferOfst+p->nBuffer!=i ){
273 int rc = demoFlushBuffer(p);
274 if( rc!=SQLITE_OK ){
275 return rc;
276 }
277 }
278 assert( p->nBuffer==0 || p->iBufferOfst+p->nBuffer==i );
279 p->iBufferOfst = i - p->nBuffer;
280
281 /* Copy as much data as possible into the buffer. */
282 nCopy = SQLITE_DEMOVFS_BUFFERSZ - p->nBuffer;
283 if( nCopy>n ){
284 nCopy = n;
285 }
286 memcpy(&p->aBuffer[p->nBuffer], z, nCopy);
287 p->nBuffer += nCopy;
288
289 n -= nCopy;
290 i += nCopy;
291 z += nCopy;
292 }
293 }else{
294 return demoDirectWrite(p, zBuf, iAmt, iOfst);
295 }
296
297 return SQLITE_OK;
298}
299
300/*
301** Truncate a file. This is a no-op for this VFS (see header comments at
302** the top of the file).
303*/
304static int demoTruncate(sqlite3_file *pFile, sqlite_int64 size){
305#if 0
306 if( ftruncate(((DemoFile *)pFile)->fd, size) ) return SQLITE_IOERR_TRUNCATE;
307#endif
308 return SQLITE_OK;
309}
310
311/*
312** Sync the contents of the file to the persistent media.
313*/
314static int demoSync(sqlite3_file *pFile, int flags){
315 DemoFile *p = (DemoFile*)pFile;
316 int rc;
317
318 rc = demoFlushBuffer(p);
319 if( rc!=SQLITE_OK ){
320 return rc;
321 }
322
323 rc = fsync(p->fd);
324 return (rc==0 ? SQLITE_OK : SQLITE_IOERR_FSYNC);
325}
326
327/*
328** Write the size of the file in bytes to *pSize.
329*/
330static 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 */
334
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.
339 */
340 rc = demoFlushBuffer(p);
341 if( rc!=SQLITE_OK ){
342 return rc;
343 }
344
345 rc = fstat(p->fd, &sStat);
346 if( rc!=0 ) return SQLITE_IOERR_FSTAT;
347 *pSize = sStat.st_size;
348 return SQLITE_OK;
349}
350
351/*
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.
356*/
357static int demoLock(sqlite3_file *pFile, int eLock){
358 return SQLITE_OK;
359}
360static int demoUnlock(sqlite3_file *pFile, int eLock){
361 return SQLITE_OK;
362}
363static int demoCheckReservedLock(sqlite3_file *pFile, int *pResOut){
364 *pResOut = 0;
365 return SQLITE_OK;
366}
367
368/*
369** No xFileControl() verbs are implemented by this VFS.
370*/
371static int demoFileControl(sqlite3_file *pFile, int op, void *pArg){
372 return SQLITE_OK;
373}
374
375/*
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.
379*/
380static int demoSectorSize(sqlite3_file *pFile){
381 return 0;
382}
383static int demoDeviceCharacteristics(sqlite3_file *pFile){
384 return 0;
385}
386
387/*
388** Open a file handle.
389*/
390static int demoOpen(
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) */
396){
397 static const sqlite3_io_methods demoio = {
398 1, /* iVersion */
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 */
411 };
412
413 DemoFile *p = (DemoFile*)pFile; /* Populate this structure */
414 int oflags = 0; /* flags to pass to open() call */
415 char *aBuf = 0;
416
417 if( zName==0 ){
418 return SQLITE_IOERR;
419 }
420
421 if( flags&SQLITE_OPEN_MAIN_JOURNAL ){
422 aBuf = (char *)sqlite3_malloc(SQLITE_DEMOVFS_BUFFERSZ);
423 if( !aBuf ){
424 return SQLITE_NOMEM;
425 }
426 }
427
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;
432
433 memset(p, 0, sizeof(DemoFile));
434 p->fd = open(zName, oflags, 0600);
435 if( p->fd<0 ){
436 sqlite3_free(aBuf);
437 return SQLITE_CANTOPEN;
438 }
439 p->aBuffer = aBuf;
440
441 if( pOutFlags ){
442 *pOutFlags = flags;
443 }
444 p->base.pMethods = &demoio;
445 return SQLITE_OK;
446}
447
448/*
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.
452*/
453static int demoDelete(sqlite3_vfs *pVfs, const char *zPath, int dirSync){
454 int rc; /* Return code */
455
456 rc = unlink(zPath);
457 if( rc!=0 && errno==ENOENT ) return SQLITE_OK;
458
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 */
463
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++);
468 zDir[i] = '\0';
469
470 /* Open a file-descriptor on the directory. Sync. Close. */
471 dfd = open(zDir, O_RDONLY, 0);
472 if( dfd<0 ){
473 rc = -1;
474 }else{
475 rc = fsync(dfd);
476 close(dfd);
477 }
478 }
479 return (rc==0 ? SQLITE_OK : SQLITE_IOERR_DELETE);
480}
481
482#ifndef F_OK
483# define F_OK 0
484#endif
485#ifndef R_OK
486# define R_OK 4
487#endif
488#ifndef W_OK
489# define W_OK 2
490#endif
491
492/*
493** Query the file-system to see if the named file exists, is readable or
494** is both readable and writable.
495*/
496static int demoAccess(
497 sqlite3_vfs *pVfs,
498 const char *zPath,
499 int flags,
500 int *pResOut
501){
502 int rc; /* access() return code */
503 int eAccess = F_OK; /* Second argument to access() */
504
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) */
508 );
509
510 if( flags==SQLITE_ACCESS_READWRITE ) eAccess = R_OK|W_OK;
511 if( flags==SQLITE_ACCESS_READ ) eAccess = R_OK;
512
513 rc = access(zPath, eAccess);
514 *pResOut = (rc==0);
515 return SQLITE_OK;
516}
517
518/*
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.
523**
524** This function assumes that paths are UNIX style. Specifically, that:
525**
526** 1. Path components are separated by a '/'. and
527** 2. Full paths begin with a '/' character.
528*/
529static 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 */
534){
535 sqlite3_snprintf(nPathOut, zPathOut, "%s", zPath);
536 zPathOut[nPathOut-1] = '\0';
537
538 return SQLITE_OK;
539}
540
541/*
542** The following four VFS methods:
543**
544** xDlOpen
545** xDlError
546** xDlSym
547** xDlClose
548**
549** are supposed to implement the functionality needed by SQLite to load
550** extensions compiled as shared objects. This simple VFS does not support
551** this functionality, so the following functions are no-ops.
552*/
553static void *demoDlOpen(sqlite3_vfs *pVfs, const char *zPath){
554 return 0;
555}
556static void demoDlError(sqlite3_vfs *pVfs, int nByte, char *zErrMsg){
557 sqlite3_snprintf(nByte, zErrMsg, "Loadable extensions are not supported");
558 zErrMsg[nByte-1] = '\0';
559}
560static void (*demoDlSym(sqlite3_vfs *pVfs, void *pH, const char *z))(void){
561 return 0;
562}
563static void demoDlClose(sqlite3_vfs *pVfs, void *pHandle){
564 return;
565}
566
567/*
568** Parameter zByte points to a buffer nByte bytes in size. Populate this
569** buffer with pseudo-random data.
570*/
571static int demoRandomness(sqlite3_vfs *pVfs, int nByte, char *zByte){
572 return SQLITE_OK;
573}
574
575/*
576** Sleep for at least nMicro microseconds. Return the (approximate) number
577** of microseconds slept for.
578*/
579static int demoSleep(sqlite3_vfs *pVfs, int nMicro){
580 sleep(nMicro / 1000000);
581 usleep(nMicro % 1000000);
582 return nMicro;
583}
584
585/*
586** Set *pTime to the current UTC time expressed as a Julian day. Return
587** SQLITE_OK if successful, or an error code otherwise.
588**
589** http://en.wikipedia.org/wiki/Julian_day
590**
591** This implementation is not very good. The current time is rounded to
592** an integer number of seconds. Also, assuming time_t is a signed 32-bit
593** value, it will stop working some time in the year 2038 AD (the so-called
594** "year 2038" problem that afflicts systems that store time this way).
595*/
596static int demoCurrentTime(sqlite3_vfs *pVfs, double *pTime){
597 time_t t = time(0);
598 *pTime = t/86400.0 + 2440587.5;
599 return SQLITE_OK;
600}
601
602/*
603** This function returns a pointer to the VFS implemented in this file.
604** To make the VFS available to SQLite:
605**
606** sqlite3_vfs_register(sqlite3_demovfs(), 0);
607*/
608sqlite3_vfs *sqlite3_demovfs(void){
609 static sqlite3_vfs demovfs = {
610 1, /* iVersion */
611 sizeof(DemoFile), /* szOsFile */
612 MAXPATHNAME, /* mxPathname */
613 0, /* pNext */
614 "demo", /* zName */
615 0, /* pAppData */
616 demoOpen, /* xOpen */
617 demoDelete, /* xDelete */
618 demoAccess, /* xAccess */
619 demoFullPathname, /* xFullPathname */
620 demoDlOpen, /* xDlOpen */
621 demoDlError, /* xDlError */
622 demoDlSym, /* xDlSym */
623 demoDlClose, /* xDlClose */
624 demoRandomness, /* xRandomness */
625 demoSleep, /* xSleep */
626 demoCurrentTime, /* xCurrentTime */
627 };
628 return &demovfs;
629}
630
631#endif /* !defined(SQLITE_TEST) || SQLITE_OS_UNIX */
632
633
634#ifdef SQLITE_TEST
635
636#if defined(INCLUDE_SQLITE_TCL_H)
637# include "sqlite_tcl.h"
638#else
639# include "tcl.h"
640# ifndef SQLITE_TCLAPI
641# define SQLITE_TCLAPI
642# endif
643#endif
644
645#if SQLITE_OS_UNIX
646static int SQLITE_TCLAPI register_demovfs(
647 ClientData clientData, /* Pointer to sqlite3_enable_XXX function */
648 Tcl_Interp *interp, /* The TCL interpreter that invoked this command */
649 int objc, /* Number of arguments */
650 Tcl_Obj *CONST objv[] /* Command arguments */
651){
652 sqlite3_vfs_register(sqlite3_demovfs(), 1);
653 return TCL_OK;
654}
655static int SQLITE_TCLAPI unregister_demovfs(
656 ClientData clientData, /* Pointer to sqlite3_enable_XXX function */
657 Tcl_Interp *interp, /* The TCL interpreter that invoked this command */
658 int objc, /* Number of arguments */
659 Tcl_Obj *CONST objv[] /* Command arguments */
660){
661 sqlite3_vfs_unregister(sqlite3_demovfs());
662 return TCL_OK;
663}
664
665/*
666** Register commands with the TCL interpreter.
667*/
668int Sqlitetest_demovfs_Init(Tcl_Interp *interp){
669 Tcl_CreateObjCommand(interp, "register_demovfs", register_demovfs, 0, 0);
670 Tcl_CreateObjCommand(interp, "unregister_demovfs", unregister_demovfs, 0, 0);
671 return TCL_OK;
672}
673
674#else
675int Sqlitetest_demovfs_Init(Tcl_Interp *interp){ return TCL_OK; }
676#endif
677
678#endif /* SQLITE_TEST */
679
680// Register sqlite3_demovfs
681int sqlite3_os_init()
682{
683 sqlite3_vfs_register(sqlite3_demovfs(), 0);
684 return 0;
685}
686
687int sqlite3_os_end()
688{
689 return 0;
690}