|
1 /* |
|
2 ** 2001 September 15 |
|
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 ** This is the implementation of the page cache subsystem or "pager". |
|
13 ** |
|
14 ** The pager is used to access a database disk file. It implements |
|
15 ** atomic commit and rollback through the use of a journal file that |
|
16 ** is separate from the database file. The pager also implements file |
|
17 ** locking to prevent two processes from writing the same database |
|
18 ** file simultaneously, or one process from reading the database while |
|
19 ** another is writing. |
|
20 ** |
|
21 ** @(#) $Id: pager.c,v 1.469 2008/08/02 03:50:39 drh Exp $ |
|
22 */ |
|
23 #ifndef SQLITE_OMIT_DISKIO |
|
24 #include "sqliteInt.h" |
|
25 #include <assert.h> |
|
26 #include <string.h> |
|
27 |
|
28 /* |
|
29 ** Macros for troubleshooting. Normally turned off |
|
30 */ |
|
31 #if 0 |
|
32 #define sqlite3DebugPrintf printf |
|
33 #define PAGERTRACE1(X) sqlite3DebugPrintf(X) |
|
34 #define PAGERTRACE2(X,Y) sqlite3DebugPrintf(X,Y) |
|
35 #define PAGERTRACE3(X,Y,Z) sqlite3DebugPrintf(X,Y,Z) |
|
36 #define PAGERTRACE4(X,Y,Z,W) sqlite3DebugPrintf(X,Y,Z,W) |
|
37 #define PAGERTRACE5(X,Y,Z,W,V) sqlite3DebugPrintf(X,Y,Z,W,V) |
|
38 #else |
|
39 #define PAGERTRACE1(X) |
|
40 #define PAGERTRACE2(X,Y) |
|
41 #define PAGERTRACE3(X,Y,Z) |
|
42 #define PAGERTRACE4(X,Y,Z,W) |
|
43 #define PAGERTRACE5(X,Y,Z,W,V) |
|
44 #endif |
|
45 |
|
46 /* |
|
47 ** The following two macros are used within the PAGERTRACEX() macros above |
|
48 ** to print out file-descriptors. |
|
49 ** |
|
50 ** PAGERID() takes a pointer to a Pager struct as its argument. The |
|
51 ** associated file-descriptor is returned. FILEHANDLEID() takes an sqlite3_file |
|
52 ** struct as its argument. |
|
53 */ |
|
54 #define PAGERID(p) ((int)(p->fd)) |
|
55 #define FILEHANDLEID(fd) ((int)fd) |
|
56 |
|
57 /* |
|
58 ** The page cache as a whole is always in one of the following |
|
59 ** states: |
|
60 ** |
|
61 ** PAGER_UNLOCK The page cache is not currently reading or |
|
62 ** writing the database file. There is no |
|
63 ** data held in memory. This is the initial |
|
64 ** state. |
|
65 ** |
|
66 ** PAGER_SHARED The page cache is reading the database. |
|
67 ** Writing is not permitted. There can be |
|
68 ** multiple readers accessing the same database |
|
69 ** file at the same time. |
|
70 ** |
|
71 ** PAGER_RESERVED This process has reserved the database for writing |
|
72 ** but has not yet made any changes. Only one process |
|
73 ** at a time can reserve the database. The original |
|
74 ** database file has not been modified so other |
|
75 ** processes may still be reading the on-disk |
|
76 ** database file. |
|
77 ** |
|
78 ** PAGER_EXCLUSIVE The page cache is writing the database. |
|
79 ** Access is exclusive. No other processes or |
|
80 ** threads can be reading or writing while one |
|
81 ** process is writing. |
|
82 ** |
|
83 ** PAGER_SYNCED The pager moves to this state from PAGER_EXCLUSIVE |
|
84 ** after all dirty pages have been written to the |
|
85 ** database file and the file has been synced to |
|
86 ** disk. All that remains to do is to remove or |
|
87 ** truncate the journal file and the transaction |
|
88 ** will be committed. |
|
89 ** |
|
90 ** The page cache comes up in PAGER_UNLOCK. The first time a |
|
91 ** sqlite3PagerGet() occurs, the state transitions to PAGER_SHARED. |
|
92 ** After all pages have been released using sqlite_page_unref(), |
|
93 ** the state transitions back to PAGER_UNLOCK. The first time |
|
94 ** that sqlite3PagerWrite() is called, the state transitions to |
|
95 ** PAGER_RESERVED. (Note that sqlite3PagerWrite() can only be |
|
96 ** called on an outstanding page which means that the pager must |
|
97 ** be in PAGER_SHARED before it transitions to PAGER_RESERVED.) |
|
98 ** PAGER_RESERVED means that there is an open rollback journal. |
|
99 ** The transition to PAGER_EXCLUSIVE occurs before any changes |
|
100 ** are made to the database file, though writes to the rollback |
|
101 ** journal occurs with just PAGER_RESERVED. After an sqlite3PagerRollback() |
|
102 ** or sqlite3PagerCommitPhaseTwo(), the state can go back to PAGER_SHARED, |
|
103 ** or it can stay at PAGER_EXCLUSIVE if we are in exclusive access mode. |
|
104 */ |
|
105 #define PAGER_UNLOCK 0 |
|
106 #define PAGER_SHARED 1 /* same as SHARED_LOCK */ |
|
107 #define PAGER_RESERVED 2 /* same as RESERVED_LOCK */ |
|
108 #define PAGER_EXCLUSIVE 4 /* same as EXCLUSIVE_LOCK */ |
|
109 #define PAGER_SYNCED 5 |
|
110 |
|
111 /* |
|
112 ** If the SQLITE_BUSY_RESERVED_LOCK macro is set to true at compile-time, |
|
113 ** then failed attempts to get a reserved lock will invoke the busy callback. |
|
114 ** This is off by default. To see why, consider the following scenario: |
|
115 ** |
|
116 ** Suppose thread A already has a shared lock and wants a reserved lock. |
|
117 ** Thread B already has a reserved lock and wants an exclusive lock. If |
|
118 ** both threads are using their busy callbacks, it might be a long time |
|
119 ** be for one of the threads give up and allows the other to proceed. |
|
120 ** But if the thread trying to get the reserved lock gives up quickly |
|
121 ** (if it never invokes its busy callback) then the contention will be |
|
122 ** resolved quickly. |
|
123 */ |
|
124 #ifndef SQLITE_BUSY_RESERVED_LOCK |
|
125 # define SQLITE_BUSY_RESERVED_LOCK 0 |
|
126 #endif |
|
127 |
|
128 /* |
|
129 ** This macro rounds values up so that if the value is an address it |
|
130 ** is guaranteed to be an address that is aligned to an 8-byte boundary. |
|
131 */ |
|
132 #define FORCE_ALIGNMENT(X) (((X)+7)&~7) |
|
133 |
|
134 typedef struct PgHdr PgHdr; |
|
135 |
|
136 /* |
|
137 ** Each pager stores all currently unreferenced pages in a list sorted |
|
138 ** in least-recently-used (LRU) order (i.e. the first item on the list has |
|
139 ** not been referenced in a long time, the last item has been recently |
|
140 ** used). An instance of this structure is included as part of each |
|
141 ** pager structure for this purpose (variable Pager.lru). |
|
142 ** |
|
143 ** Additionally, if memory-management is enabled, all unreferenced pages |
|
144 ** are stored in a global LRU list (global variable sqlite3LruPageList). |
|
145 ** |
|
146 ** In both cases, the PagerLruList.pFirstSynced variable points to |
|
147 ** the first page in the corresponding list that does not require an |
|
148 ** fsync() operation before its memory can be reclaimed. If no such |
|
149 ** page exists, PagerLruList.pFirstSynced is set to NULL. |
|
150 */ |
|
151 typedef struct PagerLruList PagerLruList; |
|
152 struct PagerLruList { |
|
153 PgHdr *pFirst; /* First page in LRU list */ |
|
154 PgHdr *pLast; /* Last page in LRU list (the most recently used) */ |
|
155 PgHdr *pFirstSynced; /* First page in list with PgHdr.needSync==0 */ |
|
156 }; |
|
157 |
|
158 /* |
|
159 ** The following structure contains the next and previous pointers used |
|
160 ** to link a PgHdr structure into a PagerLruList linked list. |
|
161 */ |
|
162 typedef struct PagerLruLink PagerLruLink; |
|
163 struct PagerLruLink { |
|
164 PgHdr *pNext; |
|
165 PgHdr *pPrev; |
|
166 }; |
|
167 |
|
168 /* |
|
169 ** Each in-memory image of a page begins with the following header. |
|
170 ** This header is only visible to this pager module. The client |
|
171 ** code that calls pager sees only the data that follows the header. |
|
172 ** |
|
173 ** Client code should call sqlite3PagerWrite() on a page prior to making |
|
174 ** any modifications to that page. The first time sqlite3PagerWrite() |
|
175 ** is called, the original page contents are written into the rollback |
|
176 ** journal and PgHdr.inJournal and PgHdr.needSync are set. Later, once |
|
177 ** the journal page has made it onto the disk surface, PgHdr.needSync |
|
178 ** is cleared. The modified page cannot be written back into the original |
|
179 ** database file until the journal pages has been synced to disk and the |
|
180 ** PgHdr.needSync has been cleared. |
|
181 ** |
|
182 ** The PgHdr.dirty flag is set when sqlite3PagerWrite() is called and |
|
183 ** is cleared again when the page content is written back to the original |
|
184 ** database file. |
|
185 ** |
|
186 ** Details of important structure elements: |
|
187 ** |
|
188 ** needSync |
|
189 ** |
|
190 ** If this is true, this means that it is not safe to write the page |
|
191 ** content to the database because the original content needed |
|
192 ** for rollback has not by synced to the main rollback journal. |
|
193 ** The original content may have been written to the rollback journal |
|
194 ** but it has not yet been synced. So we cannot write to the database |
|
195 ** file because power failure might cause the page in the journal file |
|
196 ** to never reach the disk. It is as if the write to the journal file |
|
197 ** does not occur until the journal file is synced. |
|
198 ** |
|
199 ** This flag is false if the page content exactly matches what |
|
200 ** currently exists in the database file. The needSync flag is also |
|
201 ** false if the original content has been written to the main rollback |
|
202 ** journal and synced. If the page represents a new page that has |
|
203 ** been added onto the end of the database during the current |
|
204 ** transaction, the needSync flag is true until the original database |
|
205 ** size in the journal header has been synced to disk. |
|
206 ** |
|
207 ** inJournal |
|
208 ** |
|
209 ** This is true if the original page has been written into the main |
|
210 ** rollback journal. This is always false for new pages added to |
|
211 ** the end of the database file during the current transaction. |
|
212 ** And this flag says nothing about whether or not the journal |
|
213 ** has been synced to disk. For pages that are in the original |
|
214 ** database file, the following expression should always be true: |
|
215 ** |
|
216 ** inJournal = sqlite3BitvecTest(pPager->pInJournal, pgno) |
|
217 ** |
|
218 ** The pPager->pInJournal object is only valid for the original |
|
219 ** pages of the database, not new pages that are added to the end |
|
220 ** of the database, so obviously the above expression cannot be |
|
221 ** valid for new pages. For new pages inJournal is always 0. |
|
222 ** |
|
223 ** dirty |
|
224 ** |
|
225 ** When true, this means that the content of the page has been |
|
226 ** modified and needs to be written back to the database file. |
|
227 ** If false, it means that either the content of the page is |
|
228 ** unchanged or else the content is unimportant and we do not |
|
229 ** care whether or not it is preserved. |
|
230 ** |
|
231 ** alwaysRollback |
|
232 ** |
|
233 ** This means that the sqlite3PagerDontRollback() API should be |
|
234 ** ignored for this page. The DontRollback() API attempts to say |
|
235 ** that the content of the page on disk is unimportant (it is an |
|
236 ** unused page on the freelist) so that it is unnecessary to |
|
237 ** rollback changes to this page because the content of the page |
|
238 ** can change without changing the meaning of the database. This |
|
239 ** flag overrides any DontRollback() attempt. This flag is set |
|
240 ** when a page that originally contained valid data is added to |
|
241 ** the freelist. Later in the same transaction, this page might |
|
242 ** be pulled from the freelist and reused for something different |
|
243 ** and at that point the DontRollback() API will be called because |
|
244 ** pages taken from the freelist do not need to be protected by |
|
245 ** the rollback journal. But this flag says that the page was |
|
246 ** not originally part of the freelist so that it still needs to |
|
247 ** be rolled back in spite of any subsequent DontRollback() calls. |
|
248 ** |
|
249 ** needRead |
|
250 ** |
|
251 ** This flag means (when true) that the content of the page has |
|
252 ** not yet been loaded from disk. The in-memory content is just |
|
253 ** garbage. (Actually, we zero the content, but you should not |
|
254 ** make any assumptions about the content nevertheless.) If the |
|
255 ** content is needed in the future, it should be read from the |
|
256 ** original database file. |
|
257 */ |
|
258 struct PgHdr { |
|
259 Pager *pPager; /* The pager to which this page belongs */ |
|
260 Pgno pgno; /* The page number for this page */ |
|
261 PgHdr *pNextHash, *pPrevHash; /* Hash collision chain for PgHdr.pgno */ |
|
262 PagerLruLink free; /* Next and previous free pages */ |
|
263 PgHdr *pNextAll; /* A list of all pages */ |
|
264 u8 inJournal; /* TRUE if has been written to journal */ |
|
265 u8 dirty; /* TRUE if we need to write back changes */ |
|
266 u8 needSync; /* Sync journal before writing this page */ |
|
267 u8 alwaysRollback; /* Disable DontRollback() for this page */ |
|
268 u8 needRead; /* Read content if PagerWrite() is called */ |
|
269 short int nRef; /* Number of users of this page */ |
|
270 PgHdr *pDirty, *pPrevDirty; /* Dirty pages */ |
|
271 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
272 PgHdr *pPrevAll; /* A list of all pages */ |
|
273 PagerLruLink gfree; /* Global list of nRef==0 pages */ |
|
274 #endif |
|
275 #ifdef SQLITE_CHECK_PAGES |
|
276 u32 pageHash; |
|
277 #endif |
|
278 void *pData; /* Page data */ |
|
279 /* Pager.nExtra bytes of local data appended to this header */ |
|
280 }; |
|
281 |
|
282 /* |
|
283 ** For an in-memory only database, some extra information is recorded about |
|
284 ** each page so that changes can be rolled back. (Journal files are not |
|
285 ** used for in-memory databases.) The following information is added to |
|
286 ** the end of every EXTRA block for in-memory databases. |
|
287 ** |
|
288 ** This information could have been added directly to the PgHdr structure. |
|
289 ** But then it would take up an extra 8 bytes of storage on every PgHdr |
|
290 ** even for disk-based databases. Splitting it out saves 8 bytes. This |
|
291 ** is only a savings of 0.8% but those percentages add up. |
|
292 */ |
|
293 typedef struct PgHistory PgHistory; |
|
294 struct PgHistory { |
|
295 u8 *pOrig; /* Original page text. Restore to this on a full rollback */ |
|
296 u8 *pStmt; /* Text as it was at the beginning of the current statement */ |
|
297 PgHdr *pNextStmt, *pPrevStmt; /* List of pages in the statement journal */ |
|
298 u8 inStmt; /* TRUE if in the statement subjournal */ |
|
299 }; |
|
300 |
|
301 /* |
|
302 ** A macro used for invoking the codec if there is one |
|
303 */ |
|
304 #ifdef SQLITE_HAS_CODEC |
|
305 # define CODEC1(P,D,N,X) if( P->xCodec!=0 ){ P->xCodec(P->pCodecArg,D,N,X); } |
|
306 # define CODEC2(P,D,N,X) ((char*)(P->xCodec!=0?P->xCodec(P->pCodecArg,D,N,X):D)) |
|
307 #else |
|
308 # define CODEC1(P,D,N,X) /* NO-OP */ |
|
309 # define CODEC2(P,D,N,X) ((char*)D) |
|
310 #endif |
|
311 |
|
312 /* |
|
313 ** Convert a pointer to a PgHdr into a pointer to its data |
|
314 ** and back again. |
|
315 */ |
|
316 #define PGHDR_TO_DATA(P) ((P)->pData) |
|
317 #define PGHDR_TO_EXTRA(G,P) ((void*)&((G)[1])) |
|
318 #define PGHDR_TO_HIST(P,PGR) \ |
|
319 ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->nExtra]) |
|
320 |
|
321 /* |
|
322 ** A open page cache is an instance of the following structure. |
|
323 ** |
|
324 ** Pager.errCode may be set to SQLITE_IOERR, SQLITE_CORRUPT, or |
|
325 ** or SQLITE_FULL. Once one of the first three errors occurs, it persists |
|
326 ** and is returned as the result of every major pager API call. The |
|
327 ** SQLITE_FULL return code is slightly different. It persists only until the |
|
328 ** next successful rollback is performed on the pager cache. Also, |
|
329 ** SQLITE_FULL does not affect the sqlite3PagerGet() and sqlite3PagerLookup() |
|
330 ** APIs, they may still be used successfully. |
|
331 */ |
|
332 struct Pager { |
|
333 sqlite3_vfs *pVfs; /* OS functions to use for IO */ |
|
334 u8 journalOpen; /* True if journal file descriptors is valid */ |
|
335 u8 journalStarted; /* True if header of journal is synced */ |
|
336 u8 useJournal; /* Use a rollback journal on this file */ |
|
337 u8 noReadlock; /* Do not bother to obtain readlocks */ |
|
338 u8 stmtOpen; /* True if the statement subjournal is open */ |
|
339 u8 stmtInUse; /* True we are in a statement subtransaction */ |
|
340 u8 stmtAutoopen; /* Open stmt journal when main journal is opened*/ |
|
341 u8 noSync; /* Do not sync the journal if true */ |
|
342 u8 fullSync; /* Do extra syncs of the journal for robustness */ |
|
343 u8 sync_flags; /* One of SYNC_NORMAL or SYNC_FULL */ |
|
344 u8 state; /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */ |
|
345 u8 tempFile; /* zFilename is a temporary file */ |
|
346 u8 readOnly; /* True for a read-only database */ |
|
347 u8 needSync; /* True if an fsync() is needed on the journal */ |
|
348 u8 dirtyCache; /* True if cached pages have changed */ |
|
349 u8 alwaysRollback; /* Disable DontRollback() for all pages */ |
|
350 u8 memDb; /* True to inhibit all file I/O */ |
|
351 u8 setMaster; /* True if a m-j name has been written to jrnl */ |
|
352 u8 doNotSync; /* Boolean. While true, do not spill the cache */ |
|
353 u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */ |
|
354 u8 journalMode; /* On of the PAGER_JOURNALMODE_* values */ |
|
355 u8 dbModified; /* True if there are any changes to the Db */ |
|
356 u8 changeCountDone; /* Set after incrementing the change-counter */ |
|
357 u32 vfsFlags; /* Flags for sqlite3_vfs.xOpen() */ |
|
358 int errCode; /* One of several kinds of errors */ |
|
359 int dbSize; /* Number of pages in the file */ |
|
360 int origDbSize; /* dbSize before the current change */ |
|
361 int stmtSize; /* Size of database (in pages) at stmt_begin() */ |
|
362 int nRec; /* Number of pages written to the journal */ |
|
363 u32 cksumInit; /* Quasi-random value added to every checksum */ |
|
364 int stmtNRec; /* Number of records in stmt subjournal */ |
|
365 int nExtra; /* Add this many bytes to each in-memory page */ |
|
366 int pageSize; /* Number of bytes in a page */ |
|
367 int nPage; /* Total number of in-memory pages */ |
|
368 int nRef; /* Number of in-memory pages with PgHdr.nRef>0 */ |
|
369 int mxPage; /* Maximum number of pages to hold in cache */ |
|
370 Pgno mxPgno; /* Maximum allowed size of the database */ |
|
371 Bitvec *pInJournal; /* One bit for each page in the database file */ |
|
372 Bitvec *pInStmt; /* One bit for each page in the database */ |
|
373 char *zFilename; /* Name of the database file */ |
|
374 char *zJournal; /* Name of the journal file */ |
|
375 char *zDirectory; /* Directory hold database and journal files */ |
|
376 sqlite3_file *fd, *jfd; /* File descriptors for database and journal */ |
|
377 sqlite3_file *stfd; /* File descriptor for the statement subjournal*/ |
|
378 BusyHandler *pBusyHandler; /* Pointer to sqlite.busyHandler */ |
|
379 PagerLruList lru; /* LRU list of free pages */ |
|
380 PgHdr *pAll; /* List of all pages */ |
|
381 PgHdr *pStmt; /* List of pages in the statement subjournal */ |
|
382 PgHdr *pDirty; /* List of all dirty pages */ |
|
383 i64 journalOff; /* Current byte offset in the journal file */ |
|
384 i64 journalHdr; /* Byte offset to previous journal header */ |
|
385 i64 stmtHdrOff; /* First journal header written this statement */ |
|
386 i64 stmtCksum; /* cksumInit when statement was started */ |
|
387 i64 stmtJSize; /* Size of journal at stmt_begin() */ |
|
388 int sectorSize; /* Assumed sector size during rollback */ |
|
389 #ifdef SQLITE_TEST |
|
390 int nHit, nMiss; /* Cache hits and missing */ |
|
391 int nRead, nWrite; /* Database pages read/written */ |
|
392 #endif |
|
393 void (*xDestructor)(DbPage*,int); /* Call this routine when freeing pages */ |
|
394 void (*xReiniter)(DbPage*,int); /* Call this routine when reloading pages */ |
|
395 #ifdef SQLITE_HAS_CODEC |
|
396 void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */ |
|
397 void *pCodecArg; /* First argument to xCodec() */ |
|
398 #endif |
|
399 int nHash; /* Size of the pager hash table */ |
|
400 PgHdr **aHash; /* Hash table to map page number to PgHdr */ |
|
401 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
402 Pager *pNext; /* Doubly linked list of pagers on which */ |
|
403 Pager *pPrev; /* sqlite3_release_memory() will work */ |
|
404 volatile int iInUseMM; /* Non-zero if unavailable to MM */ |
|
405 volatile int iInUseDB; /* Non-zero if in sqlite3_release_memory() */ |
|
406 #endif |
|
407 char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */ |
|
408 char dbFileVers[16]; /* Changes whenever database file changes */ |
|
409 i64 journalSizeLimit; /* Size limit for persistent journal files */ |
|
410 }; |
|
411 |
|
412 /* |
|
413 ** The following global variables hold counters used for |
|
414 ** testing purposes only. These variables do not exist in |
|
415 ** a non-testing build. These variables are not thread-safe. |
|
416 */ |
|
417 #ifdef SQLITE_TEST |
|
418 int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */ |
|
419 int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */ |
|
420 int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */ |
|
421 int sqlite3_pager_pgfree_count = 0; /* Number of cache pages freed */ |
|
422 # define PAGER_INCR(v) v++ |
|
423 #else |
|
424 # define PAGER_INCR(v) |
|
425 #endif |
|
426 |
|
427 /* |
|
428 ** The following variable points to the head of a double-linked list |
|
429 ** of all pagers that are eligible for page stealing by the |
|
430 ** sqlite3_release_memory() interface. Access to this list is |
|
431 ** protected by the SQLITE_MUTEX_STATIC_MEM2 mutex. |
|
432 */ |
|
433 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
434 static Pager *sqlite3PagerList = 0; |
|
435 static PagerLruList sqlite3LruPageList = {0, 0, 0}; |
|
436 #endif |
|
437 |
|
438 |
|
439 /* |
|
440 ** Journal files begin with the following magic string. The data |
|
441 ** was obtained from /dev/random. It is used only as a sanity check. |
|
442 ** |
|
443 ** Since version 2.8.0, the journal format contains additional sanity |
|
444 ** checking information. If the power fails while the journal is begin |
|
445 ** written, semi-random garbage data might appear in the journal |
|
446 ** file after power is restored. If an attempt is then made |
|
447 ** to roll the journal back, the database could be corrupted. The additional |
|
448 ** sanity checking data is an attempt to discover the garbage in the |
|
449 ** journal and ignore it. |
|
450 ** |
|
451 ** The sanity checking information for the new journal format consists |
|
452 ** of a 32-bit checksum on each page of data. The checksum covers both |
|
453 ** the page number and the pPager->pageSize bytes of data for the page. |
|
454 ** This cksum is initialized to a 32-bit random value that appears in the |
|
455 ** journal file right after the header. The random initializer is important, |
|
456 ** because garbage data that appears at the end of a journal is likely |
|
457 ** data that was once in other files that have now been deleted. If the |
|
458 ** garbage data came from an obsolete journal file, the checksums might |
|
459 ** be correct. But by initializing the checksum to random value which |
|
460 ** is different for every journal, we minimize that risk. |
|
461 */ |
|
462 static const unsigned char aJournalMagic[] = { |
|
463 0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7, |
|
464 }; |
|
465 |
|
466 /* |
|
467 ** The size of the header and of each page in the journal is determined |
|
468 ** by the following macros. |
|
469 */ |
|
470 #define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8) |
|
471 |
|
472 /* |
|
473 ** The journal header size for this pager. In the future, this could be |
|
474 ** set to some value read from the disk controller. The important |
|
475 ** characteristic is that it is the same size as a disk sector. |
|
476 */ |
|
477 #define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize) |
|
478 |
|
479 /* |
|
480 ** The macro MEMDB is true if we are dealing with an in-memory database. |
|
481 ** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set, |
|
482 ** the value of MEMDB will be a constant and the compiler will optimize |
|
483 ** out code that would never execute. |
|
484 */ |
|
485 #ifdef SQLITE_OMIT_MEMORYDB |
|
486 # define MEMDB 0 |
|
487 #else |
|
488 # define MEMDB pPager->memDb |
|
489 #endif |
|
490 |
|
491 /* |
|
492 ** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is |
|
493 ** reserved for working around a windows/posix incompatibility). It is |
|
494 ** used in the journal to signify that the remainder of the journal file |
|
495 ** is devoted to storing a master journal name - there are no more pages to |
|
496 ** roll back. See comments for function writeMasterJournal() for details. |
|
497 */ |
|
498 /* #define PAGER_MJ_PGNO(x) (PENDING_BYTE/((x)->pageSize)) */ |
|
499 #define PAGER_MJ_PGNO(x) ((PENDING_BYTE/((x)->pageSize))+1) |
|
500 |
|
501 /* |
|
502 ** The maximum legal page number is (2^31 - 1). |
|
503 */ |
|
504 #define PAGER_MAX_PGNO 2147483647 |
|
505 |
|
506 /* |
|
507 ** The pagerEnter() and pagerLeave() routines acquire and release |
|
508 ** a mutex on each pager. The mutex is recursive. |
|
509 ** |
|
510 ** This is a special-purpose mutex. It only provides mutual exclusion |
|
511 ** between the Btree and the Memory Management sqlite3_release_memory() |
|
512 ** function. It does not prevent, for example, two Btrees from accessing |
|
513 ** the same pager at the same time. Other general-purpose mutexes in |
|
514 ** the btree layer handle that chore. |
|
515 */ |
|
516 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
517 static void pagerEnter(Pager *p){ |
|
518 p->iInUseDB++; |
|
519 if( p->iInUseMM && p->iInUseDB==1 ){ |
|
520 #ifndef SQLITE_MUTEX_NOOP |
|
521 sqlite3_mutex *mutex; |
|
522 mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM2); |
|
523 #endif |
|
524 p->iInUseDB = 0; |
|
525 sqlite3_mutex_enter(mutex); |
|
526 p->iInUseDB = 1; |
|
527 sqlite3_mutex_leave(mutex); |
|
528 } |
|
529 assert( p->iInUseMM==0 ); |
|
530 } |
|
531 static void pagerLeave(Pager *p){ |
|
532 p->iInUseDB--; |
|
533 assert( p->iInUseDB>=0 ); |
|
534 } |
|
535 #else |
|
536 # define pagerEnter(X) |
|
537 # define pagerLeave(X) |
|
538 #endif |
|
539 |
|
540 /* |
|
541 ** Add page pPg to the end of the linked list managed by structure |
|
542 ** pList (pPg becomes the last entry in the list - the most recently |
|
543 ** used). Argument pLink should point to either pPg->free or pPg->gfree, |
|
544 ** depending on whether pPg is being added to the pager-specific or |
|
545 ** global LRU list. |
|
546 */ |
|
547 static void listAdd(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){ |
|
548 pLink->pNext = 0; |
|
549 pLink->pPrev = pList->pLast; |
|
550 |
|
551 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
552 assert(pLink==&pPg->free || pLink==&pPg->gfree); |
|
553 assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList); |
|
554 #endif |
|
555 |
|
556 if( pList->pLast ){ |
|
557 int iOff = (char *)pLink - (char *)pPg; |
|
558 PagerLruLink *pLastLink = (PagerLruLink *)(&((u8 *)pList->pLast)[iOff]); |
|
559 pLastLink->pNext = pPg; |
|
560 }else{ |
|
561 assert(!pList->pFirst); |
|
562 pList->pFirst = pPg; |
|
563 } |
|
564 |
|
565 pList->pLast = pPg; |
|
566 if( !pList->pFirstSynced && pPg->needSync==0 ){ |
|
567 pList->pFirstSynced = pPg; |
|
568 } |
|
569 } |
|
570 |
|
571 /* |
|
572 ** Remove pPg from the list managed by the structure pointed to by pList. |
|
573 ** |
|
574 ** Argument pLink should point to either pPg->free or pPg->gfree, depending |
|
575 ** on whether pPg is being added to the pager-specific or global LRU list. |
|
576 */ |
|
577 static void listRemove(PagerLruList *pList, PagerLruLink *pLink, PgHdr *pPg){ |
|
578 int iOff = (char *)pLink - (char *)pPg; |
|
579 |
|
580 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
581 assert(pLink==&pPg->free || pLink==&pPg->gfree); |
|
582 assert(pLink==&pPg->gfree || pList!=&sqlite3LruPageList); |
|
583 #endif |
|
584 |
|
585 if( pPg==pList->pFirst ){ |
|
586 pList->pFirst = pLink->pNext; |
|
587 } |
|
588 if( pPg==pList->pLast ){ |
|
589 pList->pLast = pLink->pPrev; |
|
590 } |
|
591 if( pLink->pPrev ){ |
|
592 PagerLruLink *pPrevLink = (PagerLruLink *)(&((u8 *)pLink->pPrev)[iOff]); |
|
593 pPrevLink->pNext = pLink->pNext; |
|
594 } |
|
595 if( pLink->pNext ){ |
|
596 PagerLruLink *pNextLink = (PagerLruLink *)(&((u8 *)pLink->pNext)[iOff]); |
|
597 pNextLink->pPrev = pLink->pPrev; |
|
598 } |
|
599 if( pPg==pList->pFirstSynced ){ |
|
600 PgHdr *p = pLink->pNext; |
|
601 while( p && p->needSync ){ |
|
602 PagerLruLink *pL = (PagerLruLink *)(&((u8 *)p)[iOff]); |
|
603 p = pL->pNext; |
|
604 } |
|
605 pList->pFirstSynced = p; |
|
606 } |
|
607 |
|
608 pLink->pNext = pLink->pPrev = 0; |
|
609 } |
|
610 |
|
611 /* |
|
612 ** Add page pPg to the list of free pages for the pager. If |
|
613 ** memory-management is enabled, also add the page to the global |
|
614 ** list of free pages. |
|
615 */ |
|
616 static void lruListAdd(PgHdr *pPg){ |
|
617 listAdd(&pPg->pPager->lru, &pPg->free, pPg); |
|
618 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
619 if( !pPg->pPager->memDb ){ |
|
620 sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
621 listAdd(&sqlite3LruPageList, &pPg->gfree, pPg); |
|
622 sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
623 } |
|
624 #endif |
|
625 } |
|
626 |
|
627 /* |
|
628 ** Remove page pPg from the list of free pages for the associated pager. |
|
629 ** If memory-management is enabled, also remove pPg from the global list |
|
630 ** of free pages. |
|
631 */ |
|
632 static void lruListRemove(PgHdr *pPg){ |
|
633 listRemove(&pPg->pPager->lru, &pPg->free, pPg); |
|
634 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
635 if( !pPg->pPager->memDb ){ |
|
636 sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
637 listRemove(&sqlite3LruPageList, &pPg->gfree, pPg); |
|
638 sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
639 } |
|
640 #endif |
|
641 } |
|
642 |
|
643 /* |
|
644 ** This function is called just after the needSync flag has been cleared |
|
645 ** from all pages managed by pPager (usually because the journal file |
|
646 ** has just been synced). It updates the pPager->lru.pFirstSynced variable |
|
647 ** and, if memory-management is enabled, the sqlite3LruPageList.pFirstSynced |
|
648 ** variable also. |
|
649 */ |
|
650 static void lruListSetFirstSynced(Pager *pPager){ |
|
651 pPager->lru.pFirstSynced = pPager->lru.pFirst; |
|
652 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
653 if( !pPager->memDb ){ |
|
654 PgHdr *p; |
|
655 sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
656 for(p=sqlite3LruPageList.pFirst; p && p->needSync; p=p->gfree.pNext); |
|
657 assert(p==pPager->lru.pFirstSynced || p==sqlite3LruPageList.pFirstSynced); |
|
658 sqlite3LruPageList.pFirstSynced = p; |
|
659 sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
660 } |
|
661 #endif |
|
662 } |
|
663 |
|
664 /* |
|
665 ** Return true if page *pPg has already been written to the statement |
|
666 ** journal (or statement snapshot has been created, if *pPg is part |
|
667 ** of an in-memory database). |
|
668 */ |
|
669 static int pageInStatement(PgHdr *pPg){ |
|
670 Pager *pPager = pPg->pPager; |
|
671 if( MEMDB ){ |
|
672 return PGHDR_TO_HIST(pPg, pPager)->inStmt; |
|
673 }else{ |
|
674 return sqlite3BitvecTest(pPager->pInStmt, pPg->pgno); |
|
675 } |
|
676 } |
|
677 |
|
678 /* |
|
679 ** Change the size of the pager hash table to N. N must be a power |
|
680 ** of two. |
|
681 */ |
|
682 static void pager_resize_hash_table(Pager *pPager, int N){ |
|
683 PgHdr **aHash, *pPg; |
|
684 assert( N>0 && (N&(N-1))==0 ); |
|
685 #ifdef SQLITE_MALLOC_SOFT_LIMIT |
|
686 if( N*sizeof(aHash[0])>SQLITE_MALLOC_SOFT_LIMIT ){ |
|
687 N = SQLITE_MALLOC_SOFT_LIMIT/sizeof(aHash[0]); |
|
688 } |
|
689 if( N==pPager->nHash ) return; |
|
690 #endif |
|
691 pagerLeave(pPager); |
|
692 if( pPager->aHash!=0 ) sqlite3BeginBenignMalloc(); |
|
693 aHash = sqlite3MallocZero( sizeof(aHash[0])*N ); |
|
694 if( pPager->aHash!=0 ) sqlite3EndBenignMalloc(); |
|
695 pagerEnter(pPager); |
|
696 if( aHash==0 ){ |
|
697 /* Failure to rehash is not an error. It is only a performance hit. */ |
|
698 return; |
|
699 } |
|
700 sqlite3_free(pPager->aHash); |
|
701 pPager->nHash = N; |
|
702 pPager->aHash = aHash; |
|
703 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ |
|
704 int h; |
|
705 if( pPg->pgno==0 ){ |
|
706 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 ); |
|
707 continue; |
|
708 } |
|
709 h = pPg->pgno & (N-1); |
|
710 pPg->pNextHash = aHash[h]; |
|
711 if( aHash[h] ){ |
|
712 aHash[h]->pPrevHash = pPg; |
|
713 } |
|
714 aHash[h] = pPg; |
|
715 pPg->pPrevHash = 0; |
|
716 } |
|
717 } |
|
718 |
|
719 /* |
|
720 ** Read a 32-bit integer from the given file descriptor. Store the integer |
|
721 ** that is read in *pRes. Return SQLITE_OK if everything worked, or an |
|
722 ** error code is something goes wrong. |
|
723 ** |
|
724 ** All values are stored on disk as big-endian. |
|
725 */ |
|
726 static int read32bits(sqlite3_file *fd, i64 offset, u32 *pRes){ |
|
727 unsigned char ac[4]; |
|
728 int rc = sqlite3OsRead(fd, ac, sizeof(ac), offset); |
|
729 if( rc==SQLITE_OK ){ |
|
730 *pRes = sqlite3Get4byte(ac); |
|
731 } |
|
732 return rc; |
|
733 } |
|
734 |
|
735 /* |
|
736 ** Write a 32-bit integer into a string buffer in big-endian byte order. |
|
737 */ |
|
738 #define put32bits(A,B) sqlite3Put4byte((u8*)A,B) |
|
739 |
|
740 /* |
|
741 ** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK |
|
742 ** on success or an error code is something goes wrong. |
|
743 */ |
|
744 static int write32bits(sqlite3_file *fd, i64 offset, u32 val){ |
|
745 char ac[4]; |
|
746 put32bits(ac, val); |
|
747 return sqlite3OsWrite(fd, ac, 4, offset); |
|
748 } |
|
749 |
|
750 /* |
|
751 ** If file pFd is open, call sqlite3OsUnlock() on it. |
|
752 */ |
|
753 static int osUnlock(sqlite3_file *pFd, int eLock){ |
|
754 if( !pFd->pMethods ){ |
|
755 return SQLITE_OK; |
|
756 } |
|
757 return sqlite3OsUnlock(pFd, eLock); |
|
758 } |
|
759 |
|
760 /* |
|
761 ** This function determines whether or not the atomic-write optimization |
|
762 ** can be used with this pager. The optimization can be used if: |
|
763 ** |
|
764 ** (a) the value returned by OsDeviceCharacteristics() indicates that |
|
765 ** a database page may be written atomically, and |
|
766 ** (b) the value returned by OsSectorSize() is less than or equal |
|
767 ** to the page size. |
|
768 ** |
|
769 ** If the optimization cannot be used, 0 is returned. If it can be used, |
|
770 ** then the value returned is the size of the journal file when it |
|
771 ** contains rollback data for exactly one page. |
|
772 */ |
|
773 #ifdef SQLITE_ENABLE_ATOMIC_WRITE |
|
774 static int jrnlBufferSize(Pager *pPager){ |
|
775 int dc; /* Device characteristics */ |
|
776 int nSector; /* Sector size */ |
|
777 int szPage; /* Page size */ |
|
778 sqlite3_file *fd = pPager->fd; |
|
779 |
|
780 if( fd->pMethods ){ |
|
781 dc = sqlite3OsDeviceCharacteristics(fd); |
|
782 nSector = sqlite3OsSectorSize(fd); |
|
783 szPage = pPager->pageSize; |
|
784 } |
|
785 |
|
786 assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); |
|
787 assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); |
|
788 |
|
789 if( !fd->pMethods || |
|
790 (dc & (SQLITE_IOCAP_ATOMIC|(szPage>>8)) && nSector<=szPage) ){ |
|
791 return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager); |
|
792 } |
|
793 return 0; |
|
794 } |
|
795 #endif |
|
796 |
|
797 /* |
|
798 ** This function should be called when an error occurs within the pager |
|
799 ** code. The first argument is a pointer to the pager structure, the |
|
800 ** second the error-code about to be returned by a pager API function. |
|
801 ** The value returned is a copy of the second argument to this function. |
|
802 ** |
|
803 ** If the second argument is SQLITE_IOERR, SQLITE_CORRUPT, or SQLITE_FULL |
|
804 ** the error becomes persistent. Until the persisten error is cleared, |
|
805 ** subsequent API calls on this Pager will immediately return the same |
|
806 ** error code. |
|
807 ** |
|
808 ** A persistent error indicates that the contents of the pager-cache |
|
809 ** cannot be trusted. This state can be cleared by completely discarding |
|
810 ** the contents of the pager-cache. If a transaction was active when |
|
811 ** the persistent error occured, then the rollback journal may need |
|
812 ** to be replayed. |
|
813 */ |
|
814 static void pager_unlock(Pager *pPager); |
|
815 static int pager_error(Pager *pPager, int rc){ |
|
816 int rc2 = rc & 0xff; |
|
817 assert( |
|
818 pPager->errCode==SQLITE_FULL || |
|
819 pPager->errCode==SQLITE_OK || |
|
820 (pPager->errCode & 0xff)==SQLITE_IOERR |
|
821 ); |
|
822 if( |
|
823 rc2==SQLITE_FULL || |
|
824 rc2==SQLITE_IOERR || |
|
825 rc2==SQLITE_CORRUPT |
|
826 ){ |
|
827 pPager->errCode = rc; |
|
828 if( pPager->state==PAGER_UNLOCK && pPager->nRef==0 ){ |
|
829 /* If the pager is already unlocked, call pager_unlock() now to |
|
830 ** clear the error state and ensure that the pager-cache is |
|
831 ** completely empty. |
|
832 */ |
|
833 pager_unlock(pPager); |
|
834 } |
|
835 } |
|
836 return rc; |
|
837 } |
|
838 |
|
839 /* |
|
840 ** If SQLITE_CHECK_PAGES is defined then we do some sanity checking |
|
841 ** on the cache using a hash function. This is used for testing |
|
842 ** and debugging only. |
|
843 */ |
|
844 #ifdef SQLITE_CHECK_PAGES |
|
845 /* |
|
846 ** Return a 32-bit hash of the page data for pPage. |
|
847 */ |
|
848 static u32 pager_datahash(int nByte, unsigned char *pData){ |
|
849 u32 hash = 0; |
|
850 int i; |
|
851 for(i=0; i<nByte; i++){ |
|
852 hash = (hash*1039) + pData[i]; |
|
853 } |
|
854 return hash; |
|
855 } |
|
856 static u32 pager_pagehash(PgHdr *pPage){ |
|
857 return pager_datahash(pPage->pPager->pageSize, |
|
858 (unsigned char *)PGHDR_TO_DATA(pPage)); |
|
859 } |
|
860 |
|
861 /* |
|
862 ** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES |
|
863 ** is defined, and NDEBUG is not defined, an assert() statement checks |
|
864 ** that the page is either dirty or still matches the calculated page-hash. |
|
865 */ |
|
866 #define CHECK_PAGE(x) checkPage(x) |
|
867 static void checkPage(PgHdr *pPg){ |
|
868 Pager *pPager = pPg->pPager; |
|
869 assert( !pPg->pageHash || pPager->errCode || MEMDB || pPg->dirty || |
|
870 pPg->pageHash==pager_pagehash(pPg) ); |
|
871 } |
|
872 |
|
873 #else |
|
874 #define pager_datahash(X,Y) 0 |
|
875 #define pager_pagehash(X) 0 |
|
876 #define CHECK_PAGE(x) |
|
877 #endif |
|
878 |
|
879 /* |
|
880 ** When this is called the journal file for pager pPager must be open. |
|
881 ** The master journal file name is read from the end of the file and |
|
882 ** written into memory supplied by the caller. |
|
883 ** |
|
884 ** zMaster must point to a buffer of at least nMaster bytes allocated by |
|
885 ** the caller. This should be sqlite3_vfs.mxPathname+1 (to ensure there is |
|
886 ** enough space to write the master journal name). If the master journal |
|
887 ** name in the journal is longer than nMaster bytes (including a |
|
888 ** nul-terminator), then this is handled as if no master journal name |
|
889 ** were present in the journal. |
|
890 ** |
|
891 ** If no master journal file name is present zMaster[0] is set to 0 and |
|
892 ** SQLITE_OK returned. |
|
893 */ |
|
894 static int readMasterJournal(sqlite3_file *pJrnl, char *zMaster, int nMaster){ |
|
895 int rc; |
|
896 u32 len; |
|
897 i64 szJ; |
|
898 u32 cksum; |
|
899 u32 u; /* Unsigned loop counter */ |
|
900 unsigned char aMagic[8]; /* A buffer to hold the magic header */ |
|
901 |
|
902 zMaster[0] = '\0'; |
|
903 |
|
904 rc = sqlite3OsFileSize(pJrnl, &szJ); |
|
905 if( rc!=SQLITE_OK || szJ<16 ) return rc; |
|
906 |
|
907 rc = read32bits(pJrnl, szJ-16, &len); |
|
908 if( rc!=SQLITE_OK ) return rc; |
|
909 |
|
910 if( len>=nMaster ){ |
|
911 return SQLITE_OK; |
|
912 } |
|
913 |
|
914 rc = read32bits(pJrnl, szJ-12, &cksum); |
|
915 if( rc!=SQLITE_OK ) return rc; |
|
916 |
|
917 rc = sqlite3OsRead(pJrnl, aMagic, 8, szJ-8); |
|
918 if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc; |
|
919 |
|
920 rc = sqlite3OsRead(pJrnl, zMaster, len, szJ-16-len); |
|
921 if( rc!=SQLITE_OK ){ |
|
922 return rc; |
|
923 } |
|
924 zMaster[len] = '\0'; |
|
925 |
|
926 /* See if the checksum matches the master journal name */ |
|
927 for(u=0; u<len; u++){ |
|
928 cksum -= zMaster[u]; |
|
929 } |
|
930 if( cksum ){ |
|
931 /* If the checksum doesn't add up, then one or more of the disk sectors |
|
932 ** containing the master journal filename is corrupted. This means |
|
933 ** definitely roll back, so just return SQLITE_OK and report a (nul) |
|
934 ** master-journal filename. |
|
935 */ |
|
936 zMaster[0] = '\0'; |
|
937 } |
|
938 |
|
939 return SQLITE_OK; |
|
940 } |
|
941 |
|
942 /* |
|
943 ** Seek the journal file descriptor to the next sector boundary where a |
|
944 ** journal header may be read or written. Pager.journalOff is updated with |
|
945 ** the new seek offset. |
|
946 ** |
|
947 ** i.e for a sector size of 512: |
|
948 ** |
|
949 ** Input Offset Output Offset |
|
950 ** --------------------------------------- |
|
951 ** 0 0 |
|
952 ** 512 512 |
|
953 ** 100 512 |
|
954 ** 2000 2048 |
|
955 ** |
|
956 */ |
|
957 static void seekJournalHdr(Pager *pPager){ |
|
958 i64 offset = 0; |
|
959 i64 c = pPager->journalOff; |
|
960 if( c ){ |
|
961 offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager); |
|
962 } |
|
963 assert( offset%JOURNAL_HDR_SZ(pPager)==0 ); |
|
964 assert( offset>=c ); |
|
965 assert( (offset-c)<JOURNAL_HDR_SZ(pPager) ); |
|
966 pPager->journalOff = offset; |
|
967 } |
|
968 |
|
969 /* |
|
970 ** Write zeros over the header of the journal file. This has the |
|
971 ** effect of invalidating the journal file and committing the |
|
972 ** transaction. |
|
973 */ |
|
974 static int zeroJournalHdr(Pager *pPager, int doTruncate){ |
|
975 int rc = SQLITE_OK; |
|
976 static const char zeroHdr[28]; |
|
977 |
|
978 if( pPager->journalOff ){ |
|
979 i64 iLimit = pPager->journalSizeLimit; |
|
980 |
|
981 IOTRACE(("JZEROHDR %p\n", pPager)) |
|
982 if( doTruncate || iLimit==0 ){ |
|
983 rc = sqlite3OsTruncate(pPager->jfd, 0); |
|
984 }else{ |
|
985 rc = sqlite3OsWrite(pPager->jfd, zeroHdr, sizeof(zeroHdr), 0); |
|
986 } |
|
987 if( rc==SQLITE_OK && !pPager->noSync ){ |
|
988 rc = sqlite3OsSync(pPager->jfd, SQLITE_SYNC_DATAONLY|pPager->sync_flags); |
|
989 } |
|
990 |
|
991 /* At this point the transaction is committed but the write lock |
|
992 ** is still held on the file. If there is a size limit configured for |
|
993 ** the persistent journal and the journal file currently consumes more |
|
994 ** space than that limit allows for, truncate it now. There is no need |
|
995 ** to sync the file following this operation. |
|
996 */ |
|
997 if( rc==SQLITE_OK && iLimit>0 ){ |
|
998 i64 sz; |
|
999 rc = sqlite3OsFileSize(pPager->jfd, &sz); |
|
1000 if( rc==SQLITE_OK && sz>iLimit ){ |
|
1001 rc = sqlite3OsTruncate(pPager->jfd, iLimit); |
|
1002 } |
|
1003 } |
|
1004 } |
|
1005 return rc; |
|
1006 } |
|
1007 |
|
1008 /* |
|
1009 ** The journal file must be open when this routine is called. A journal |
|
1010 ** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the |
|
1011 ** current location. |
|
1012 ** |
|
1013 ** The format for the journal header is as follows: |
|
1014 ** - 8 bytes: Magic identifying journal format. |
|
1015 ** - 4 bytes: Number of records in journal, or -1 no-sync mode is on. |
|
1016 ** - 4 bytes: Random number used for page hash. |
|
1017 ** - 4 bytes: Initial database page count. |
|
1018 ** - 4 bytes: Sector size used by the process that wrote this journal. |
|
1019 ** - 4 bytes: Database page size. |
|
1020 ** |
|
1021 ** Followed by (JOURNAL_HDR_SZ - 28) bytes of unused space. |
|
1022 */ |
|
1023 static int writeJournalHdr(Pager *pPager){ |
|
1024 int rc = SQLITE_OK; |
|
1025 char *zHeader = pPager->pTmpSpace; |
|
1026 int nHeader = pPager->pageSize; |
|
1027 int nWrite; |
|
1028 |
|
1029 if( nHeader>JOURNAL_HDR_SZ(pPager) ){ |
|
1030 nHeader = JOURNAL_HDR_SZ(pPager); |
|
1031 } |
|
1032 |
|
1033 if( pPager->stmtHdrOff==0 ){ |
|
1034 pPager->stmtHdrOff = pPager->journalOff; |
|
1035 } |
|
1036 |
|
1037 seekJournalHdr(pPager); |
|
1038 pPager->journalHdr = pPager->journalOff; |
|
1039 |
|
1040 memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic)); |
|
1041 |
|
1042 /* |
|
1043 ** Write the nRec Field - the number of page records that follow this |
|
1044 ** journal header. Normally, zero is written to this value at this time. |
|
1045 ** After the records are added to the journal (and the journal synced, |
|
1046 ** if in full-sync mode), the zero is overwritten with the true number |
|
1047 ** of records (see syncJournal()). |
|
1048 ** |
|
1049 ** A faster alternative is to write 0xFFFFFFFF to the nRec field. When |
|
1050 ** reading the journal this value tells SQLite to assume that the |
|
1051 ** rest of the journal file contains valid page records. This assumption |
|
1052 ** is dangerous, as if a failure occured whilst writing to the journal |
|
1053 ** file it may contain some garbage data. There are two scenarios |
|
1054 ** where this risk can be ignored: |
|
1055 ** |
|
1056 ** * When the pager is in no-sync mode. Corruption can follow a |
|
1057 ** power failure in this case anyway. |
|
1058 ** |
|
1059 ** * When the SQLITE_IOCAP_SAFE_APPEND flag is set. This guarantees |
|
1060 ** that garbage data is never appended to the journal file. |
|
1061 */ |
|
1062 assert(pPager->fd->pMethods||pPager->noSync); |
|
1063 if( (pPager->noSync) |
|
1064 || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_SAFE_APPEND) |
|
1065 ){ |
|
1066 put32bits(&zHeader[sizeof(aJournalMagic)], 0xffffffff); |
|
1067 }else{ |
|
1068 put32bits(&zHeader[sizeof(aJournalMagic)], 0); |
|
1069 } |
|
1070 |
|
1071 /* The random check-hash initialiser */ |
|
1072 sqlite3_randomness(sizeof(pPager->cksumInit), &pPager->cksumInit); |
|
1073 put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit); |
|
1074 /* The initial database size */ |
|
1075 put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbSize); |
|
1076 /* The assumed sector size for this process */ |
|
1077 put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize); |
|
1078 if( pPager->journalHdr==0 ){ |
|
1079 /* The page size */ |
|
1080 put32bits(&zHeader[sizeof(aJournalMagic)+16], pPager->pageSize); |
|
1081 } |
|
1082 |
|
1083 for(nWrite=0; rc==SQLITE_OK&&nWrite<JOURNAL_HDR_SZ(pPager); nWrite+=nHeader){ |
|
1084 IOTRACE(("JHDR %p %lld %d\n", pPager, pPager->journalHdr, nHeader)) |
|
1085 rc = sqlite3OsWrite(pPager->jfd, zHeader, nHeader, pPager->journalOff); |
|
1086 pPager->journalOff += nHeader; |
|
1087 } |
|
1088 |
|
1089 return rc; |
|
1090 } |
|
1091 |
|
1092 /* |
|
1093 ** The journal file must be open when this is called. A journal header file |
|
1094 ** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal |
|
1095 ** file. See comments above function writeJournalHdr() for a description of |
|
1096 ** the journal header format. |
|
1097 ** |
|
1098 ** If the header is read successfully, *nRec is set to the number of |
|
1099 ** page records following this header and *dbSize is set to the size of the |
|
1100 ** database before the transaction began, in pages. Also, pPager->cksumInit |
|
1101 ** is set to the value read from the journal header. SQLITE_OK is returned |
|
1102 ** in this case. |
|
1103 ** |
|
1104 ** If the journal header file appears to be corrupted, SQLITE_DONE is |
|
1105 ** returned and *nRec and *dbSize are not set. If JOURNAL_HDR_SZ bytes |
|
1106 ** cannot be read from the journal file an error code is returned. |
|
1107 */ |
|
1108 static int readJournalHdr( |
|
1109 Pager *pPager, |
|
1110 i64 journalSize, |
|
1111 u32 *pNRec, |
|
1112 u32 *pDbSize |
|
1113 ){ |
|
1114 int rc; |
|
1115 unsigned char aMagic[8]; /* A buffer to hold the magic header */ |
|
1116 i64 jrnlOff; |
|
1117 int iPageSize; |
|
1118 |
|
1119 seekJournalHdr(pPager); |
|
1120 if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){ |
|
1121 return SQLITE_DONE; |
|
1122 } |
|
1123 jrnlOff = pPager->journalOff; |
|
1124 |
|
1125 rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic), jrnlOff); |
|
1126 if( rc ) return rc; |
|
1127 jrnlOff += sizeof(aMagic); |
|
1128 |
|
1129 if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){ |
|
1130 return SQLITE_DONE; |
|
1131 } |
|
1132 |
|
1133 rc = read32bits(pPager->jfd, jrnlOff, pNRec); |
|
1134 if( rc ) return rc; |
|
1135 |
|
1136 rc = read32bits(pPager->jfd, jrnlOff+4, &pPager->cksumInit); |
|
1137 if( rc ) return rc; |
|
1138 |
|
1139 rc = read32bits(pPager->jfd, jrnlOff+8, pDbSize); |
|
1140 if( rc ) return rc; |
|
1141 |
|
1142 rc = read32bits(pPager->jfd, jrnlOff+16, (u32 *)&iPageSize); |
|
1143 if( rc==SQLITE_OK |
|
1144 && iPageSize>=512 |
|
1145 && iPageSize<=SQLITE_MAX_PAGE_SIZE |
|
1146 && ((iPageSize-1)&iPageSize)==0 |
|
1147 ){ |
|
1148 u16 pagesize = iPageSize; |
|
1149 rc = sqlite3PagerSetPagesize(pPager, &pagesize); |
|
1150 } |
|
1151 if( rc ) return rc; |
|
1152 |
|
1153 /* Update the assumed sector-size to match the value used by |
|
1154 ** the process that created this journal. If this journal was |
|
1155 ** created by a process other than this one, then this routine |
|
1156 ** is being called from within pager_playback(). The local value |
|
1157 ** of Pager.sectorSize is restored at the end of that routine. |
|
1158 */ |
|
1159 rc = read32bits(pPager->jfd, jrnlOff+12, (u32 *)&pPager->sectorSize); |
|
1160 if( rc ) return rc; |
|
1161 |
|
1162 pPager->journalOff += JOURNAL_HDR_SZ(pPager); |
|
1163 return SQLITE_OK; |
|
1164 } |
|
1165 |
|
1166 |
|
1167 /* |
|
1168 ** Write the supplied master journal name into the journal file for pager |
|
1169 ** pPager at the current location. The master journal name must be the last |
|
1170 ** thing written to a journal file. If the pager is in full-sync mode, the |
|
1171 ** journal file descriptor is advanced to the next sector boundary before |
|
1172 ** anything is written. The format is: |
|
1173 ** |
|
1174 ** + 4 bytes: PAGER_MJ_PGNO. |
|
1175 ** + N bytes: length of master journal name. |
|
1176 ** + 4 bytes: N |
|
1177 ** + 4 bytes: Master journal name checksum. |
|
1178 ** + 8 bytes: aJournalMagic[]. |
|
1179 ** |
|
1180 ** The master journal page checksum is the sum of the bytes in the master |
|
1181 ** journal name. |
|
1182 ** |
|
1183 ** If zMaster is a NULL pointer (occurs for a single database transaction), |
|
1184 ** this call is a no-op. |
|
1185 */ |
|
1186 static int writeMasterJournal(Pager *pPager, const char *zMaster){ |
|
1187 int rc; |
|
1188 int len; |
|
1189 int i; |
|
1190 i64 jrnlOff; |
|
1191 i64 jrnlSize; |
|
1192 u32 cksum = 0; |
|
1193 char zBuf[sizeof(aJournalMagic)+2*4]; |
|
1194 |
|
1195 if( !zMaster || pPager->setMaster) return SQLITE_OK; |
|
1196 pPager->setMaster = 1; |
|
1197 |
|
1198 len = strlen(zMaster); |
|
1199 for(i=0; i<len; i++){ |
|
1200 cksum += zMaster[i]; |
|
1201 } |
|
1202 |
|
1203 /* If in full-sync mode, advance to the next disk sector before writing |
|
1204 ** the master journal name. This is in case the previous page written to |
|
1205 ** the journal has already been synced. |
|
1206 */ |
|
1207 if( pPager->fullSync ){ |
|
1208 seekJournalHdr(pPager); |
|
1209 } |
|
1210 jrnlOff = pPager->journalOff; |
|
1211 pPager->journalOff += (len+20); |
|
1212 |
|
1213 rc = write32bits(pPager->jfd, jrnlOff, PAGER_MJ_PGNO(pPager)); |
|
1214 if( rc!=SQLITE_OK ) return rc; |
|
1215 jrnlOff += 4; |
|
1216 |
|
1217 rc = sqlite3OsWrite(pPager->jfd, zMaster, len, jrnlOff); |
|
1218 if( rc!=SQLITE_OK ) return rc; |
|
1219 jrnlOff += len; |
|
1220 |
|
1221 put32bits(zBuf, len); |
|
1222 put32bits(&zBuf[4], cksum); |
|
1223 memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic)); |
|
1224 rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic), jrnlOff); |
|
1225 jrnlOff += 8+sizeof(aJournalMagic); |
|
1226 pPager->needSync = !pPager->noSync; |
|
1227 |
|
1228 /* If the pager is in peristent-journal mode, then the physical |
|
1229 ** journal-file may extend past the end of the master-journal name |
|
1230 ** and 8 bytes of magic data just written to the file. This is |
|
1231 ** dangerous because the code to rollback a hot-journal file |
|
1232 ** will not be able to find the master-journal name to determine |
|
1233 ** whether or not the journal is hot. |
|
1234 ** |
|
1235 ** Easiest thing to do in this scenario is to truncate the journal |
|
1236 ** file to the required size. |
|
1237 */ |
|
1238 if( (rc==SQLITE_OK) |
|
1239 && (rc = sqlite3OsFileSize(pPager->jfd, &jrnlSize))==SQLITE_OK |
|
1240 && jrnlSize>jrnlOff |
|
1241 ){ |
|
1242 rc = sqlite3OsTruncate(pPager->jfd, jrnlOff); |
|
1243 } |
|
1244 return rc; |
|
1245 } |
|
1246 |
|
1247 /* |
|
1248 ** Add or remove a page from the list of all pages that are in the |
|
1249 ** statement journal. |
|
1250 ** |
|
1251 ** The Pager keeps a separate list of pages that are currently in |
|
1252 ** the statement journal. This helps the sqlite3PagerStmtCommit() |
|
1253 ** routine run MUCH faster for the common case where there are many |
|
1254 ** pages in memory but only a few are in the statement journal. |
|
1255 */ |
|
1256 static void page_add_to_stmt_list(PgHdr *pPg){ |
|
1257 Pager *pPager = pPg->pPager; |
|
1258 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
1259 assert( MEMDB ); |
|
1260 if( !pHist->inStmt ){ |
|
1261 assert( pHist->pPrevStmt==0 && pHist->pNextStmt==0 ); |
|
1262 if( pPager->pStmt ){ |
|
1263 PGHDR_TO_HIST(pPager->pStmt, pPager)->pPrevStmt = pPg; |
|
1264 } |
|
1265 pHist->pNextStmt = pPager->pStmt; |
|
1266 pPager->pStmt = pPg; |
|
1267 pHist->inStmt = 1; |
|
1268 } |
|
1269 } |
|
1270 |
|
1271 /* |
|
1272 ** Find a page in the hash table given its page number. Return |
|
1273 ** a pointer to the page or NULL if not found. |
|
1274 */ |
|
1275 static PgHdr *pager_lookup(Pager *pPager, Pgno pgno){ |
|
1276 PgHdr *p; |
|
1277 if( pPager->aHash==0 ) return 0; |
|
1278 p = pPager->aHash[pgno & (pPager->nHash-1)]; |
|
1279 while( p && p->pgno!=pgno ){ |
|
1280 p = p->pNextHash; |
|
1281 } |
|
1282 return p; |
|
1283 } |
|
1284 |
|
1285 /* |
|
1286 ** Clear the in-memory cache. This routine |
|
1287 ** sets the state of the pager back to what it was when it was first |
|
1288 ** opened. Any outstanding pages are invalidated and subsequent attempts |
|
1289 ** to access those pages will likely result in a coredump. |
|
1290 */ |
|
1291 static void pager_reset(Pager *pPager){ |
|
1292 PgHdr *pPg, *pNext; |
|
1293 if( pPager->errCode ) return; |
|
1294 for(pPg=pPager->pAll; pPg; pPg=pNext){ |
|
1295 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno)); |
|
1296 PAGER_INCR(sqlite3_pager_pgfree_count); |
|
1297 pNext = pPg->pNextAll; |
|
1298 lruListRemove(pPg); |
|
1299 sqlite3PageFree(pPg->pData); |
|
1300 sqlite3_free(pPg); |
|
1301 } |
|
1302 assert(pPager->lru.pFirst==0); |
|
1303 assert(pPager->lru.pFirstSynced==0); |
|
1304 assert(pPager->lru.pLast==0); |
|
1305 pPager->pStmt = 0; |
|
1306 pPager->pAll = 0; |
|
1307 pPager->pDirty = 0; |
|
1308 pPager->nHash = 0; |
|
1309 sqlite3_free(pPager->aHash); |
|
1310 pPager->nPage = 0; |
|
1311 pPager->aHash = 0; |
|
1312 pPager->nRef = 0; |
|
1313 } |
|
1314 |
|
1315 /* |
|
1316 ** Unlock the database file. |
|
1317 ** |
|
1318 ** If the pager is currently in error state, discard the contents of |
|
1319 ** the cache and reset the Pager structure internal state. If there is |
|
1320 ** an open journal-file, then the next time a shared-lock is obtained |
|
1321 ** on the pager file (by this or any other process), it will be |
|
1322 ** treated as a hot-journal and rolled back. |
|
1323 */ |
|
1324 static void pager_unlock(Pager *pPager){ |
|
1325 if( !pPager->exclusiveMode ){ |
|
1326 if( !MEMDB ){ |
|
1327 int rc = osUnlock(pPager->fd, NO_LOCK); |
|
1328 if( rc ) pPager->errCode = rc; |
|
1329 pPager->dbSize = -1; |
|
1330 IOTRACE(("UNLOCK %p\n", pPager)) |
|
1331 |
|
1332 /* Always close the journal file when dropping the database lock. |
|
1333 ** Otherwise, another connection with journal_mode=delete might |
|
1334 ** delete the file out from under us. |
|
1335 */ |
|
1336 if( pPager->journalOpen ){ |
|
1337 sqlite3OsClose(pPager->jfd); |
|
1338 pPager->journalOpen = 0; |
|
1339 sqlite3BitvecDestroy(pPager->pInJournal); |
|
1340 pPager->pInJournal = 0; |
|
1341 } |
|
1342 |
|
1343 /* If Pager.errCode is set, the contents of the pager cache cannot be |
|
1344 ** trusted. Now that the pager file is unlocked, the contents of the |
|
1345 ** cache can be discarded and the error code safely cleared. |
|
1346 */ |
|
1347 if( pPager->errCode ){ |
|
1348 if( rc==SQLITE_OK ) pPager->errCode = SQLITE_OK; |
|
1349 pager_reset(pPager); |
|
1350 if( pPager->stmtOpen ){ |
|
1351 sqlite3OsClose(pPager->stfd); |
|
1352 sqlite3BitvecDestroy(pPager->pInStmt); |
|
1353 pPager->pInStmt = 0; |
|
1354 } |
|
1355 pPager->stmtOpen = 0; |
|
1356 pPager->stmtInUse = 0; |
|
1357 pPager->journalOff = 0; |
|
1358 pPager->journalStarted = 0; |
|
1359 pPager->stmtAutoopen = 0; |
|
1360 pPager->origDbSize = 0; |
|
1361 } |
|
1362 } |
|
1363 |
|
1364 if( !MEMDB || pPager->errCode==SQLITE_OK ){ |
|
1365 pPager->state = PAGER_UNLOCK; |
|
1366 pPager->changeCountDone = 0; |
|
1367 } |
|
1368 } |
|
1369 } |
|
1370 |
|
1371 /* |
|
1372 ** Execute a rollback if a transaction is active and unlock the |
|
1373 ** database file. If the pager has already entered the error state, |
|
1374 ** do not attempt the rollback. |
|
1375 */ |
|
1376 static void pagerUnlockAndRollback(Pager *p){ |
|
1377 /* assert( p->state>=PAGER_RESERVED || p->journalOpen==0 ); */ |
|
1378 if( p->errCode==SQLITE_OK && p->state>=PAGER_RESERVED ){ |
|
1379 sqlite3BeginBenignMalloc(); |
|
1380 sqlite3PagerRollback(p); |
|
1381 sqlite3EndBenignMalloc(); |
|
1382 } |
|
1383 pager_unlock(p); |
|
1384 #if 0 |
|
1385 assert( p->errCode || !p->journalOpen || (p->exclusiveMode&&!p->journalOff) ); |
|
1386 assert( p->errCode || !p->stmtOpen || p->exclusiveMode ); |
|
1387 #endif |
|
1388 } |
|
1389 |
|
1390 /* |
|
1391 ** This routine ends a transaction. A transaction is ended by either |
|
1392 ** a COMMIT or a ROLLBACK. |
|
1393 ** |
|
1394 ** When this routine is called, the pager has the journal file open and |
|
1395 ** a RESERVED or EXCLUSIVE lock on the database. This routine will release |
|
1396 ** the database lock and acquires a SHARED lock in its place if that is |
|
1397 ** the appropriate thing to do. Release locks usually is appropriate, |
|
1398 ** unless we are in exclusive access mode or unless this is a |
|
1399 ** COMMIT AND BEGIN or ROLLBACK AND BEGIN operation. |
|
1400 ** |
|
1401 ** The journal file is either deleted or truncated. |
|
1402 ** |
|
1403 ** TODO: Consider keeping the journal file open for temporary databases. |
|
1404 ** This might give a performance improvement on windows where opening |
|
1405 ** a file is an expensive operation. |
|
1406 */ |
|
1407 static int pager_end_transaction(Pager *pPager, int hasMaster){ |
|
1408 PgHdr *pPg; |
|
1409 int rc = SQLITE_OK; |
|
1410 int rc2 = SQLITE_OK; |
|
1411 assert( !MEMDB ); |
|
1412 if( pPager->state<PAGER_RESERVED ){ |
|
1413 return SQLITE_OK; |
|
1414 } |
|
1415 sqlite3PagerStmtCommit(pPager); |
|
1416 if( pPager->stmtOpen && !pPager->exclusiveMode ){ |
|
1417 sqlite3OsClose(pPager->stfd); |
|
1418 pPager->stmtOpen = 0; |
|
1419 } |
|
1420 if( pPager->journalOpen ){ |
|
1421 if( pPager->exclusiveMode |
|
1422 || pPager->journalMode==PAGER_JOURNALMODE_PERSIST |
|
1423 ){ |
|
1424 rc = zeroJournalHdr(pPager, hasMaster); |
|
1425 pager_error(pPager, rc); |
|
1426 pPager->journalOff = 0; |
|
1427 pPager->journalStarted = 0; |
|
1428 }else{ |
|
1429 sqlite3OsClose(pPager->jfd); |
|
1430 pPager->journalOpen = 0; |
|
1431 if( rc==SQLITE_OK && !pPager->tempFile ){ |
|
1432 rc = sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0); |
|
1433 } |
|
1434 } |
|
1435 sqlite3BitvecDestroy(pPager->pInJournal); |
|
1436 pPager->pInJournal = 0; |
|
1437 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ |
|
1438 pPg->inJournal = 0; |
|
1439 pPg->dirty = 0; |
|
1440 pPg->needSync = 0; |
|
1441 pPg->alwaysRollback = 0; |
|
1442 #ifdef SQLITE_CHECK_PAGES |
|
1443 pPg->pageHash = pager_pagehash(pPg); |
|
1444 #endif |
|
1445 } |
|
1446 pPager->pDirty = 0; |
|
1447 pPager->dirtyCache = 0; |
|
1448 pPager->nRec = 0; |
|
1449 }else{ |
|
1450 assert( pPager->pInJournal==0 ); |
|
1451 } |
|
1452 |
|
1453 if( !pPager->exclusiveMode ){ |
|
1454 rc2 = osUnlock(pPager->fd, SHARED_LOCK); |
|
1455 pPager->state = PAGER_SHARED; |
|
1456 }else if( pPager->state==PAGER_SYNCED ){ |
|
1457 pPager->state = PAGER_EXCLUSIVE; |
|
1458 } |
|
1459 pPager->origDbSize = 0; |
|
1460 pPager->setMaster = 0; |
|
1461 pPager->needSync = 0; |
|
1462 lruListSetFirstSynced(pPager); |
|
1463 pPager->dbSize = -1; |
|
1464 pPager->dbModified = 0; |
|
1465 |
|
1466 return (rc==SQLITE_OK?rc2:rc); |
|
1467 } |
|
1468 |
|
1469 /* |
|
1470 ** Compute and return a checksum for the page of data. |
|
1471 ** |
|
1472 ** This is not a real checksum. It is really just the sum of the |
|
1473 ** random initial value and the page number. We experimented with |
|
1474 ** a checksum of the entire data, but that was found to be too slow. |
|
1475 ** |
|
1476 ** Note that the page number is stored at the beginning of data and |
|
1477 ** the checksum is stored at the end. This is important. If journal |
|
1478 ** corruption occurs due to a power failure, the most likely scenario |
|
1479 ** is that one end or the other of the record will be changed. It is |
|
1480 ** much less likely that the two ends of the journal record will be |
|
1481 ** correct and the middle be corrupt. Thus, this "checksum" scheme, |
|
1482 ** though fast and simple, catches the mostly likely kind of corruption. |
|
1483 ** |
|
1484 ** FIX ME: Consider adding every 200th (or so) byte of the data to the |
|
1485 ** checksum. That way if a single page spans 3 or more disk sectors and |
|
1486 ** only the middle sector is corrupt, we will still have a reasonable |
|
1487 ** chance of failing the checksum and thus detecting the problem. |
|
1488 */ |
|
1489 static u32 pager_cksum(Pager *pPager, const u8 *aData){ |
|
1490 u32 cksum = pPager->cksumInit; |
|
1491 int i = pPager->pageSize-200; |
|
1492 while( i>0 ){ |
|
1493 cksum += aData[i]; |
|
1494 i -= 200; |
|
1495 } |
|
1496 return cksum; |
|
1497 } |
|
1498 |
|
1499 /* Forward declaration */ |
|
1500 static void makeClean(PgHdr*); |
|
1501 |
|
1502 /* |
|
1503 ** Read a single page from the journal file opened on file descriptor |
|
1504 ** jfd. Playback this one page. |
|
1505 ** |
|
1506 ** If useCksum==0 it means this journal does not use checksums. Checksums |
|
1507 ** are not used in statement journals because statement journals do not |
|
1508 ** need to survive power failures. |
|
1509 */ |
|
1510 static int pager_playback_one_page( |
|
1511 Pager *pPager, |
|
1512 sqlite3_file *jfd, |
|
1513 i64 offset, |
|
1514 int useCksum |
|
1515 ){ |
|
1516 int rc; |
|
1517 PgHdr *pPg; /* An existing page in the cache */ |
|
1518 Pgno pgno; /* The page number of a page in journal */ |
|
1519 u32 cksum; /* Checksum used for sanity checking */ |
|
1520 u8 *aData = (u8 *)pPager->pTmpSpace; /* Temp storage for a page */ |
|
1521 |
|
1522 /* useCksum should be true for the main journal and false for |
|
1523 ** statement journals. Verify that this is always the case |
|
1524 */ |
|
1525 assert( jfd == (useCksum ? pPager->jfd : pPager->stfd) ); |
|
1526 assert( aData ); |
|
1527 |
|
1528 rc = read32bits(jfd, offset, &pgno); |
|
1529 if( rc!=SQLITE_OK ) return rc; |
|
1530 rc = sqlite3OsRead(jfd, aData, pPager->pageSize, offset+4); |
|
1531 if( rc!=SQLITE_OK ) return rc; |
|
1532 pPager->journalOff += pPager->pageSize + 4; |
|
1533 |
|
1534 /* Sanity checking on the page. This is more important that I originally |
|
1535 ** thought. If a power failure occurs while the journal is being written, |
|
1536 ** it could cause invalid data to be written into the journal. We need to |
|
1537 ** detect this invalid data (with high probability) and ignore it. |
|
1538 */ |
|
1539 if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){ |
|
1540 return SQLITE_DONE; |
|
1541 } |
|
1542 if( pgno>(unsigned)pPager->dbSize ){ |
|
1543 return SQLITE_OK; |
|
1544 } |
|
1545 if( useCksum ){ |
|
1546 rc = read32bits(jfd, offset+pPager->pageSize+4, &cksum); |
|
1547 if( rc ) return rc; |
|
1548 pPager->journalOff += 4; |
|
1549 if( pager_cksum(pPager, aData)!=cksum ){ |
|
1550 return SQLITE_DONE; |
|
1551 } |
|
1552 } |
|
1553 |
|
1554 assert( pPager->state==PAGER_RESERVED || pPager->state>=PAGER_EXCLUSIVE ); |
|
1555 |
|
1556 /* If the pager is in RESERVED state, then there must be a copy of this |
|
1557 ** page in the pager cache. In this case just update the pager cache, |
|
1558 ** not the database file. The page is left marked dirty in this case. |
|
1559 ** |
|
1560 ** An exception to the above rule: If the database is in no-sync mode |
|
1561 ** and a page is moved during an incremental vacuum then the page may |
|
1562 ** not be in the pager cache. Later: if a malloc() or IO error occurs |
|
1563 ** during a Movepage() call, then the page may not be in the cache |
|
1564 ** either. So the condition described in the above paragraph is not |
|
1565 ** assert()able. |
|
1566 ** |
|
1567 ** If in EXCLUSIVE state, then we update the pager cache if it exists |
|
1568 ** and the main file. The page is then marked not dirty. |
|
1569 ** |
|
1570 ** Ticket #1171: The statement journal might contain page content that is |
|
1571 ** different from the page content at the start of the transaction. |
|
1572 ** This occurs when a page is changed prior to the start of a statement |
|
1573 ** then changed again within the statement. When rolling back such a |
|
1574 ** statement we must not write to the original database unless we know |
|
1575 ** for certain that original page contents are synced into the main rollback |
|
1576 ** journal. Otherwise, a power loss might leave modified data in the |
|
1577 ** database file without an entry in the rollback journal that can |
|
1578 ** restore the database to its original form. Two conditions must be |
|
1579 ** met before writing to the database files. (1) the database must be |
|
1580 ** locked. (2) we know that the original page content is fully synced |
|
1581 ** in the main journal either because the page is not in cache or else |
|
1582 ** the page is marked as needSync==0. |
|
1583 ** |
|
1584 ** 2008-04-14: When attempting to vacuum a corrupt database file, it |
|
1585 ** is possible to fail a statement on a database that does not yet exist. |
|
1586 ** Do not attempt to write if database file has never been opened. |
|
1587 */ |
|
1588 pPg = pager_lookup(pPager, pgno); |
|
1589 PAGERTRACE4("PLAYBACK %d page %d hash(%08x)\n", |
|
1590 PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, aData)); |
|
1591 if( pPager->state>=PAGER_EXCLUSIVE && (pPg==0 || pPg->needSync==0) |
|
1592 && pPager->fd->pMethods ){ |
|
1593 i64 offset = (pgno-1)*(i64)pPager->pageSize; |
|
1594 rc = sqlite3OsWrite(pPager->fd, aData, pPager->pageSize, offset); |
|
1595 if( pPg ){ |
|
1596 makeClean(pPg); |
|
1597 } |
|
1598 } |
|
1599 if( pPg ){ |
|
1600 /* No page should ever be explicitly rolled back that is in use, except |
|
1601 ** for page 1 which is held in use in order to keep the lock on the |
|
1602 ** database active. However such a page may be rolled back as a result |
|
1603 ** of an internal error resulting in an automatic call to |
|
1604 ** sqlite3PagerRollback(). |
|
1605 */ |
|
1606 void *pData; |
|
1607 /* assert( pPg->nRef==0 || pPg->pgno==1 ); */ |
|
1608 pData = PGHDR_TO_DATA(pPg); |
|
1609 memcpy(pData, aData, pPager->pageSize); |
|
1610 if( pPager->xReiniter ){ |
|
1611 pPager->xReiniter(pPg, pPager->pageSize); |
|
1612 } |
|
1613 #ifdef SQLITE_CHECK_PAGES |
|
1614 pPg->pageHash = pager_pagehash(pPg); |
|
1615 #endif |
|
1616 /* If this was page 1, then restore the value of Pager.dbFileVers. |
|
1617 ** Do this before any decoding. */ |
|
1618 if( pgno==1 ){ |
|
1619 memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers)); |
|
1620 } |
|
1621 |
|
1622 /* Decode the page just read from disk */ |
|
1623 CODEC1(pPager, pData, pPg->pgno, 3); |
|
1624 } |
|
1625 return rc; |
|
1626 } |
|
1627 |
|
1628 /* |
|
1629 ** Parameter zMaster is the name of a master journal file. A single journal |
|
1630 ** file that referred to the master journal file has just been rolled back. |
|
1631 ** This routine checks if it is possible to delete the master journal file, |
|
1632 ** and does so if it is. |
|
1633 ** |
|
1634 ** Argument zMaster may point to Pager.pTmpSpace. So that buffer is not |
|
1635 ** available for use within this function. |
|
1636 ** |
|
1637 ** |
|
1638 ** The master journal file contains the names of all child journals. |
|
1639 ** To tell if a master journal can be deleted, check to each of the |
|
1640 ** children. If all children are either missing or do not refer to |
|
1641 ** a different master journal, then this master journal can be deleted. |
|
1642 */ |
|
1643 static int pager_delmaster(Pager *pPager, const char *zMaster){ |
|
1644 sqlite3_vfs *pVfs = pPager->pVfs; |
|
1645 int rc; |
|
1646 int master_open = 0; |
|
1647 sqlite3_file *pMaster; |
|
1648 sqlite3_file *pJournal; |
|
1649 char *zMasterJournal = 0; /* Contents of master journal file */ |
|
1650 i64 nMasterJournal; /* Size of master journal file */ |
|
1651 |
|
1652 /* Open the master journal file exclusively in case some other process |
|
1653 ** is running this routine also. Not that it makes too much difference. |
|
1654 */ |
|
1655 pMaster = (sqlite3_file *)sqlite3Malloc(pVfs->szOsFile * 2); |
|
1656 pJournal = (sqlite3_file *)(((u8 *)pMaster) + pVfs->szOsFile); |
|
1657 if( !pMaster ){ |
|
1658 rc = SQLITE_NOMEM; |
|
1659 }else{ |
|
1660 int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MASTER_JOURNAL); |
|
1661 rc = sqlite3OsOpen(pVfs, zMaster, pMaster, flags, 0); |
|
1662 } |
|
1663 if( rc!=SQLITE_OK ) goto delmaster_out; |
|
1664 master_open = 1; |
|
1665 |
|
1666 rc = sqlite3OsFileSize(pMaster, &nMasterJournal); |
|
1667 if( rc!=SQLITE_OK ) goto delmaster_out; |
|
1668 |
|
1669 if( nMasterJournal>0 ){ |
|
1670 char *zJournal; |
|
1671 char *zMasterPtr = 0; |
|
1672 int nMasterPtr = pPager->pVfs->mxPathname+1; |
|
1673 |
|
1674 /* Load the entire master journal file into space obtained from |
|
1675 ** sqlite3_malloc() and pointed to by zMasterJournal. |
|
1676 */ |
|
1677 zMasterJournal = (char *)sqlite3Malloc(nMasterJournal + nMasterPtr); |
|
1678 if( !zMasterJournal ){ |
|
1679 rc = SQLITE_NOMEM; |
|
1680 goto delmaster_out; |
|
1681 } |
|
1682 zMasterPtr = &zMasterJournal[nMasterJournal]; |
|
1683 rc = sqlite3OsRead(pMaster, zMasterJournal, nMasterJournal, 0); |
|
1684 if( rc!=SQLITE_OK ) goto delmaster_out; |
|
1685 |
|
1686 zJournal = zMasterJournal; |
|
1687 while( (zJournal-zMasterJournal)<nMasterJournal ){ |
|
1688 int exists; |
|
1689 rc = sqlite3OsAccess(pVfs, zJournal, SQLITE_ACCESS_EXISTS, &exists); |
|
1690 if( rc!=SQLITE_OK ){ |
|
1691 goto delmaster_out; |
|
1692 } |
|
1693 if( exists ){ |
|
1694 /* One of the journals pointed to by the master journal exists. |
|
1695 ** Open it and check if it points at the master journal. If |
|
1696 ** so, return without deleting the master journal file. |
|
1697 */ |
|
1698 int c; |
|
1699 int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_MAIN_JOURNAL); |
|
1700 rc = sqlite3OsOpen(pVfs, zJournal, pJournal, flags, 0); |
|
1701 if( rc!=SQLITE_OK ){ |
|
1702 goto delmaster_out; |
|
1703 } |
|
1704 |
|
1705 rc = readMasterJournal(pJournal, zMasterPtr, nMasterPtr); |
|
1706 sqlite3OsClose(pJournal); |
|
1707 if( rc!=SQLITE_OK ){ |
|
1708 goto delmaster_out; |
|
1709 } |
|
1710 |
|
1711 c = zMasterPtr[0]!=0 && strcmp(zMasterPtr, zMaster)==0; |
|
1712 if( c ){ |
|
1713 /* We have a match. Do not delete the master journal file. */ |
|
1714 goto delmaster_out; |
|
1715 } |
|
1716 } |
|
1717 zJournal += (strlen(zJournal)+1); |
|
1718 } |
|
1719 } |
|
1720 |
|
1721 rc = sqlite3OsDelete(pVfs, zMaster, 0); |
|
1722 |
|
1723 delmaster_out: |
|
1724 if( zMasterJournal ){ |
|
1725 sqlite3_free(zMasterJournal); |
|
1726 } |
|
1727 if( master_open ){ |
|
1728 sqlite3OsClose(pMaster); |
|
1729 } |
|
1730 sqlite3_free(pMaster); |
|
1731 return rc; |
|
1732 } |
|
1733 |
|
1734 |
|
1735 static void pager_truncate_cache(Pager *pPager); |
|
1736 |
|
1737 /* |
|
1738 ** Truncate the main file of the given pager to the number of pages |
|
1739 ** indicated. Also truncate the cached representation of the file. |
|
1740 ** |
|
1741 ** Might might be the case that the file on disk is smaller than nPage. |
|
1742 ** This can happen, for example, if we are in the middle of a transaction |
|
1743 ** which has extended the file size and the new pages are still all held |
|
1744 ** in cache, then an INSERT or UPDATE does a statement rollback. Some |
|
1745 ** operating system implementations can get confused if you try to |
|
1746 ** truncate a file to some size that is larger than it currently is, |
|
1747 ** so detect this case and write a single zero byte to the end of the new |
|
1748 ** file instead. |
|
1749 */ |
|
1750 static int pager_truncate(Pager *pPager, int nPage){ |
|
1751 int rc = SQLITE_OK; |
|
1752 if( pPager->state>=PAGER_EXCLUSIVE && pPager->fd->pMethods ){ |
|
1753 i64 currentSize, newSize; |
|
1754 rc = sqlite3OsFileSize(pPager->fd, ¤tSize); |
|
1755 newSize = pPager->pageSize*(i64)nPage; |
|
1756 if( rc==SQLITE_OK && currentSize!=newSize ){ |
|
1757 if( currentSize>newSize ){ |
|
1758 rc = sqlite3OsTruncate(pPager->fd, newSize); |
|
1759 }else{ |
|
1760 rc = sqlite3OsWrite(pPager->fd, "", 1, newSize-1); |
|
1761 } |
|
1762 } |
|
1763 } |
|
1764 if( rc==SQLITE_OK ){ |
|
1765 pPager->dbSize = nPage; |
|
1766 pager_truncate_cache(pPager); |
|
1767 } |
|
1768 return rc; |
|
1769 } |
|
1770 |
|
1771 /* |
|
1772 ** Set the sectorSize for the given pager. |
|
1773 ** |
|
1774 ** The sector size is at least as big as the sector size reported |
|
1775 ** by sqlite3OsSectorSize(). The minimum sector size is 512. |
|
1776 */ |
|
1777 static void setSectorSize(Pager *pPager){ |
|
1778 assert(pPager->fd->pMethods||pPager->tempFile); |
|
1779 if( !pPager->tempFile ){ |
|
1780 /* Sector size doesn't matter for temporary files. Also, the file |
|
1781 ** may not have been opened yet, in whcih case the OsSectorSize() |
|
1782 ** call will segfault. |
|
1783 */ |
|
1784 pPager->sectorSize = sqlite3OsSectorSize(pPager->fd); |
|
1785 } |
|
1786 if( pPager->sectorSize<512 ){ |
|
1787 pPager->sectorSize = 512; |
|
1788 } |
|
1789 } |
|
1790 |
|
1791 /* |
|
1792 ** Playback the journal and thus restore the database file to |
|
1793 ** the state it was in before we started making changes. |
|
1794 ** |
|
1795 ** The journal file format is as follows: |
|
1796 ** |
|
1797 ** (1) 8 byte prefix. A copy of aJournalMagic[]. |
|
1798 ** (2) 4 byte big-endian integer which is the number of valid page records |
|
1799 ** in the journal. If this value is 0xffffffff, then compute the |
|
1800 ** number of page records from the journal size. |
|
1801 ** (3) 4 byte big-endian integer which is the initial value for the |
|
1802 ** sanity checksum. |
|
1803 ** (4) 4 byte integer which is the number of pages to truncate the |
|
1804 ** database to during a rollback. |
|
1805 ** (5) 4 byte big-endian integer which is the sector size. The header |
|
1806 ** is this many bytes in size. |
|
1807 ** (6) 4 byte big-endian integer which is the page case. |
|
1808 ** (7) 4 byte integer which is the number of bytes in the master journal |
|
1809 ** name. The value may be zero (indicate that there is no master |
|
1810 ** journal.) |
|
1811 ** (8) N bytes of the master journal name. The name will be nul-terminated |
|
1812 ** and might be shorter than the value read from (5). If the first byte |
|
1813 ** of the name is \000 then there is no master journal. The master |
|
1814 ** journal name is stored in UTF-8. |
|
1815 ** (9) Zero or more pages instances, each as follows: |
|
1816 ** + 4 byte page number. |
|
1817 ** + pPager->pageSize bytes of data. |
|
1818 ** + 4 byte checksum |
|
1819 ** |
|
1820 ** When we speak of the journal header, we mean the first 8 items above. |
|
1821 ** Each entry in the journal is an instance of the 9th item. |
|
1822 ** |
|
1823 ** Call the value from the second bullet "nRec". nRec is the number of |
|
1824 ** valid page entries in the journal. In most cases, you can compute the |
|
1825 ** value of nRec from the size of the journal file. But if a power |
|
1826 ** failure occurred while the journal was being written, it could be the |
|
1827 ** case that the size of the journal file had already been increased but |
|
1828 ** the extra entries had not yet made it safely to disk. In such a case, |
|
1829 ** the value of nRec computed from the file size would be too large. For |
|
1830 ** that reason, we always use the nRec value in the header. |
|
1831 ** |
|
1832 ** If the nRec value is 0xffffffff it means that nRec should be computed |
|
1833 ** from the file size. This value is used when the user selects the |
|
1834 ** no-sync option for the journal. A power failure could lead to corruption |
|
1835 ** in this case. But for things like temporary table (which will be |
|
1836 ** deleted when the power is restored) we don't care. |
|
1837 ** |
|
1838 ** If the file opened as the journal file is not a well-formed |
|
1839 ** journal file then all pages up to the first corrupted page are rolled |
|
1840 ** back (or no pages if the journal header is corrupted). The journal file |
|
1841 ** is then deleted and SQLITE_OK returned, just as if no corruption had |
|
1842 ** been encountered. |
|
1843 ** |
|
1844 ** If an I/O or malloc() error occurs, the journal-file is not deleted |
|
1845 ** and an error code is returned. |
|
1846 */ |
|
1847 static int pager_playback(Pager *pPager, int isHot){ |
|
1848 sqlite3_vfs *pVfs = pPager->pVfs; |
|
1849 i64 szJ; /* Size of the journal file in bytes */ |
|
1850 u32 nRec; /* Number of Records in the journal */ |
|
1851 u32 u; /* Unsigned loop counter */ |
|
1852 Pgno mxPg = 0; /* Size of the original file in pages */ |
|
1853 int rc; /* Result code of a subroutine */ |
|
1854 int res = 1; /* Value returned by sqlite3OsAccess() */ |
|
1855 char *zMaster = 0; /* Name of master journal file if any */ |
|
1856 |
|
1857 /* Figure out how many records are in the journal. Abort early if |
|
1858 ** the journal is empty. |
|
1859 */ |
|
1860 assert( pPager->journalOpen ); |
|
1861 rc = sqlite3OsFileSize(pPager->jfd, &szJ); |
|
1862 if( rc!=SQLITE_OK || szJ==0 ){ |
|
1863 goto end_playback; |
|
1864 } |
|
1865 |
|
1866 /* Read the master journal name from the journal, if it is present. |
|
1867 ** If a master journal file name is specified, but the file is not |
|
1868 ** present on disk, then the journal is not hot and does not need to be |
|
1869 ** played back. |
|
1870 */ |
|
1871 zMaster = pPager->pTmpSpace; |
|
1872 rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1); |
|
1873 if( rc==SQLITE_OK && zMaster[0] ){ |
|
1874 rc = sqlite3OsAccess(pVfs, zMaster, SQLITE_ACCESS_EXISTS, &res); |
|
1875 } |
|
1876 zMaster = 0; |
|
1877 if( rc!=SQLITE_OK || !res ){ |
|
1878 goto end_playback; |
|
1879 } |
|
1880 pPager->journalOff = 0; |
|
1881 |
|
1882 /* This loop terminates either when the readJournalHdr() call returns |
|
1883 ** SQLITE_DONE or an IO error occurs. */ |
|
1884 while( 1 ){ |
|
1885 |
|
1886 /* Read the next journal header from the journal file. If there are |
|
1887 ** not enough bytes left in the journal file for a complete header, or |
|
1888 ** it is corrupted, then a process must of failed while writing it. |
|
1889 ** This indicates nothing more needs to be rolled back. |
|
1890 */ |
|
1891 rc = readJournalHdr(pPager, szJ, &nRec, &mxPg); |
|
1892 if( rc!=SQLITE_OK ){ |
|
1893 if( rc==SQLITE_DONE ){ |
|
1894 rc = SQLITE_OK; |
|
1895 } |
|
1896 goto end_playback; |
|
1897 } |
|
1898 |
|
1899 /* If nRec is 0xffffffff, then this journal was created by a process |
|
1900 ** working in no-sync mode. This means that the rest of the journal |
|
1901 ** file consists of pages, there are no more journal headers. Compute |
|
1902 ** the value of nRec based on this assumption. |
|
1903 */ |
|
1904 if( nRec==0xffffffff ){ |
|
1905 assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ); |
|
1906 nRec = (szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager); |
|
1907 } |
|
1908 |
|
1909 /* If nRec is 0 and this rollback is of a transaction created by this |
|
1910 ** process and if this is the final header in the journal, then it means |
|
1911 ** that this part of the journal was being filled but has not yet been |
|
1912 ** synced to disk. Compute the number of pages based on the remaining |
|
1913 ** size of the file. |
|
1914 ** |
|
1915 ** The third term of the test was added to fix ticket #2565. |
|
1916 */ |
|
1917 if( nRec==0 && !isHot && |
|
1918 pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff ){ |
|
1919 nRec = (szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager); |
|
1920 } |
|
1921 |
|
1922 /* If this is the first header read from the journal, truncate the |
|
1923 ** database file back to its original size. |
|
1924 */ |
|
1925 if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){ |
|
1926 rc = pager_truncate(pPager, mxPg); |
|
1927 if( rc!=SQLITE_OK ){ |
|
1928 goto end_playback; |
|
1929 } |
|
1930 } |
|
1931 |
|
1932 /* Copy original pages out of the journal and back into the database file. |
|
1933 */ |
|
1934 for(u=0; u<nRec; u++){ |
|
1935 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1); |
|
1936 if( rc!=SQLITE_OK ){ |
|
1937 if( rc==SQLITE_DONE ){ |
|
1938 rc = SQLITE_OK; |
|
1939 pPager->journalOff = szJ; |
|
1940 break; |
|
1941 }else{ |
|
1942 goto end_playback; |
|
1943 } |
|
1944 } |
|
1945 } |
|
1946 } |
|
1947 /*NOTREACHED*/ |
|
1948 assert( 0 ); |
|
1949 |
|
1950 end_playback: |
|
1951 if( rc==SQLITE_OK ){ |
|
1952 zMaster = pPager->pTmpSpace; |
|
1953 rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1); |
|
1954 } |
|
1955 if( rc==SQLITE_OK ){ |
|
1956 rc = pager_end_transaction(pPager, zMaster[0]!='\0'); |
|
1957 } |
|
1958 if( rc==SQLITE_OK && zMaster[0] ){ |
|
1959 /* If there was a master journal and this routine will return success, |
|
1960 ** see if it is possible to delete the master journal. |
|
1961 */ |
|
1962 rc = pager_delmaster(pPager, zMaster); |
|
1963 } |
|
1964 |
|
1965 /* The Pager.sectorSize variable may have been updated while rolling |
|
1966 ** back a journal created by a process with a different sector size |
|
1967 ** value. Reset it to the correct value for this process. |
|
1968 */ |
|
1969 setSectorSize(pPager); |
|
1970 return rc; |
|
1971 } |
|
1972 |
|
1973 /* |
|
1974 ** Playback the statement journal. |
|
1975 ** |
|
1976 ** This is similar to playing back the transaction journal but with |
|
1977 ** a few extra twists. |
|
1978 ** |
|
1979 ** (1) The number of pages in the database file at the start of |
|
1980 ** the statement is stored in pPager->stmtSize, not in the |
|
1981 ** journal file itself. |
|
1982 ** |
|
1983 ** (2) In addition to playing back the statement journal, also |
|
1984 ** playback all pages of the transaction journal beginning |
|
1985 ** at offset pPager->stmtJSize. |
|
1986 */ |
|
1987 static int pager_stmt_playback(Pager *pPager){ |
|
1988 i64 szJ; /* Size of the full journal */ |
|
1989 i64 hdrOff; |
|
1990 int nRec; /* Number of Records */ |
|
1991 int i; /* Loop counter */ |
|
1992 int rc; |
|
1993 |
|
1994 szJ = pPager->journalOff; |
|
1995 |
|
1996 /* Set hdrOff to be the offset just after the end of the last journal |
|
1997 ** page written before the first journal-header for this statement |
|
1998 ** transaction was written, or the end of the file if no journal |
|
1999 ** header was written. |
|
2000 */ |
|
2001 hdrOff = pPager->stmtHdrOff; |
|
2002 assert( pPager->fullSync || !hdrOff ); |
|
2003 if( !hdrOff ){ |
|
2004 hdrOff = szJ; |
|
2005 } |
|
2006 |
|
2007 /* Truncate the database back to its original size. |
|
2008 */ |
|
2009 rc = pager_truncate(pPager, pPager->stmtSize); |
|
2010 assert( pPager->state>=PAGER_SHARED ); |
|
2011 |
|
2012 /* Figure out how many records are in the statement journal. |
|
2013 */ |
|
2014 assert( pPager->stmtInUse && pPager->journalOpen ); |
|
2015 nRec = pPager->stmtNRec; |
|
2016 |
|
2017 /* Copy original pages out of the statement journal and back into the |
|
2018 ** database file. Note that the statement journal omits checksums from |
|
2019 ** each record since power-failure recovery is not important to statement |
|
2020 ** journals. |
|
2021 */ |
|
2022 for(i=0; i<nRec; i++){ |
|
2023 i64 offset = i*(4+pPager->pageSize); |
|
2024 rc = pager_playback_one_page(pPager, pPager->stfd, offset, 0); |
|
2025 assert( rc!=SQLITE_DONE ); |
|
2026 if( rc!=SQLITE_OK ) goto end_stmt_playback; |
|
2027 } |
|
2028 |
|
2029 /* Now roll some pages back from the transaction journal. Pager.stmtJSize |
|
2030 ** was the size of the journal file when this statement was started, so |
|
2031 ** everything after that needs to be rolled back, either into the |
|
2032 ** database, the memory cache, or both. |
|
2033 ** |
|
2034 ** If it is not zero, then Pager.stmtHdrOff is the offset to the start |
|
2035 ** of the first journal header written during this statement transaction. |
|
2036 */ |
|
2037 pPager->journalOff = pPager->stmtJSize; |
|
2038 pPager->cksumInit = pPager->stmtCksum; |
|
2039 while( pPager->journalOff < hdrOff ){ |
|
2040 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1); |
|
2041 assert( rc!=SQLITE_DONE ); |
|
2042 if( rc!=SQLITE_OK ) goto end_stmt_playback; |
|
2043 } |
|
2044 |
|
2045 while( pPager->journalOff < szJ ){ |
|
2046 u32 nJRec; /* Number of Journal Records */ |
|
2047 u32 dummy; |
|
2048 rc = readJournalHdr(pPager, szJ, &nJRec, &dummy); |
|
2049 if( rc!=SQLITE_OK ){ |
|
2050 assert( rc!=SQLITE_DONE ); |
|
2051 goto end_stmt_playback; |
|
2052 } |
|
2053 if( nJRec==0 ){ |
|
2054 nJRec = (szJ - pPager->journalOff) / (pPager->pageSize+8); |
|
2055 } |
|
2056 for(i=nJRec-1; i>=0 && pPager->journalOff < szJ; i--){ |
|
2057 rc = pager_playback_one_page(pPager, pPager->jfd, pPager->journalOff, 1); |
|
2058 assert( rc!=SQLITE_DONE ); |
|
2059 if( rc!=SQLITE_OK ) goto end_stmt_playback; |
|
2060 } |
|
2061 } |
|
2062 |
|
2063 pPager->journalOff = szJ; |
|
2064 |
|
2065 end_stmt_playback: |
|
2066 if( rc==SQLITE_OK) { |
|
2067 pPager->journalOff = szJ; |
|
2068 /* pager_reload_cache(pPager); */ |
|
2069 } |
|
2070 return rc; |
|
2071 } |
|
2072 |
|
2073 /* |
|
2074 ** Change the maximum number of in-memory pages that are allowed. |
|
2075 */ |
|
2076 void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){ |
|
2077 if( mxPage>10 ){ |
|
2078 pPager->mxPage = mxPage; |
|
2079 }else{ |
|
2080 pPager->mxPage = 10; |
|
2081 } |
|
2082 } |
|
2083 |
|
2084 /* |
|
2085 ** Adjust the robustness of the database to damage due to OS crashes |
|
2086 ** or power failures by changing the number of syncs()s when writing |
|
2087 ** the rollback journal. There are three levels: |
|
2088 ** |
|
2089 ** OFF sqlite3OsSync() is never called. This is the default |
|
2090 ** for temporary and transient files. |
|
2091 ** |
|
2092 ** NORMAL The journal is synced once before writes begin on the |
|
2093 ** database. This is normally adequate protection, but |
|
2094 ** it is theoretically possible, though very unlikely, |
|
2095 ** that an inopertune power failure could leave the journal |
|
2096 ** in a state which would cause damage to the database |
|
2097 ** when it is rolled back. |
|
2098 ** |
|
2099 ** FULL The journal is synced twice before writes begin on the |
|
2100 ** database (with some additional information - the nRec field |
|
2101 ** of the journal header - being written in between the two |
|
2102 ** syncs). If we assume that writing a |
|
2103 ** single disk sector is atomic, then this mode provides |
|
2104 ** assurance that the journal will not be corrupted to the |
|
2105 ** point of causing damage to the database during rollback. |
|
2106 ** |
|
2107 ** Numeric values associated with these states are OFF==1, NORMAL=2, |
|
2108 ** and FULL=3. |
|
2109 */ |
|
2110 #ifndef SQLITE_OMIT_PAGER_PRAGMAS |
|
2111 void sqlite3PagerSetSafetyLevel(Pager *pPager, int level, int full_fsync){ |
|
2112 pPager->noSync = level==1 || pPager->tempFile; |
|
2113 pPager->fullSync = level==3 && !pPager->tempFile; |
|
2114 pPager->sync_flags = (full_fsync?SQLITE_SYNC_FULL:SQLITE_SYNC_NORMAL); |
|
2115 if( pPager->noSync ) pPager->needSync = 0; |
|
2116 } |
|
2117 #endif |
|
2118 |
|
2119 /* |
|
2120 ** The following global variable is incremented whenever the library |
|
2121 ** attempts to open a temporary file. This information is used for |
|
2122 ** testing and analysis only. |
|
2123 */ |
|
2124 #ifdef SQLITE_TEST |
|
2125 int sqlite3_opentemp_count = 0; |
|
2126 #endif |
|
2127 |
|
2128 /* |
|
2129 ** Open a temporary file. |
|
2130 ** |
|
2131 ** Write the file descriptor into *fd. Return SQLITE_OK on success or some |
|
2132 ** other error code if we fail. The OS will automatically delete the temporary |
|
2133 ** file when it is closed. |
|
2134 */ |
|
2135 static int sqlite3PagerOpentemp( |
|
2136 Pager *pPager, /* The pager object */ |
|
2137 sqlite3_file *pFile, /* Write the file descriptor here */ |
|
2138 int vfsFlags /* Flags passed through to the VFS */ |
|
2139 ){ |
|
2140 int rc; |
|
2141 |
|
2142 #ifdef SQLITE_TEST |
|
2143 sqlite3_opentemp_count++; /* Used for testing and analysis only */ |
|
2144 #endif |
|
2145 |
|
2146 vfsFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE | |
|
2147 SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE; |
|
2148 rc = sqlite3OsOpen(pPager->pVfs, 0, pFile, vfsFlags, 0); |
|
2149 assert( rc!=SQLITE_OK || pFile->pMethods ); |
|
2150 return rc; |
|
2151 } |
|
2152 |
|
2153 /* |
|
2154 ** Create a new page cache and put a pointer to the page cache in *ppPager. |
|
2155 ** The file to be cached need not exist. The file is not locked until |
|
2156 ** the first call to sqlite3PagerGet() and is only held open until the |
|
2157 ** last page is released using sqlite3PagerUnref(). |
|
2158 ** |
|
2159 ** If zFilename is NULL then a randomly-named temporary file is created |
|
2160 ** and used as the file to be cached. The file will be deleted |
|
2161 ** automatically when it is closed. |
|
2162 ** |
|
2163 ** If zFilename is ":memory:" then all information is held in cache. |
|
2164 ** It is never written to disk. This can be used to implement an |
|
2165 ** in-memory database. |
|
2166 */ |
|
2167 int sqlite3PagerOpen( |
|
2168 sqlite3_vfs *pVfs, /* The virtual file system to use */ |
|
2169 Pager **ppPager, /* Return the Pager structure here */ |
|
2170 const char *zFilename, /* Name of the database file to open */ |
|
2171 int nExtra, /* Extra bytes append to each in-memory page */ |
|
2172 int flags, /* flags controlling this file */ |
|
2173 int vfsFlags /* flags passed through to sqlite3_vfs.xOpen() */ |
|
2174 ){ |
|
2175 u8 *pPtr; |
|
2176 Pager *pPager = 0; |
|
2177 int rc = SQLITE_OK; |
|
2178 int i; |
|
2179 int tempFile = 0; |
|
2180 int memDb = 0; |
|
2181 int readOnly = 0; |
|
2182 int useJournal = (flags & PAGER_OMIT_JOURNAL)==0; |
|
2183 int noReadlock = (flags & PAGER_NO_READLOCK)!=0; |
|
2184 int journalFileSize = sqlite3JournalSize(pVfs); |
|
2185 int szPageDflt = SQLITE_DEFAULT_PAGE_SIZE; |
|
2186 char *zPathname = 0; |
|
2187 int nPathname = 0; |
|
2188 |
|
2189 /* The default return is a NULL pointer */ |
|
2190 *ppPager = 0; |
|
2191 |
|
2192 /* Compute and store the full pathname in an allocated buffer pointed |
|
2193 ** to by zPathname, length nPathname. Or, if this is a temporary file, |
|
2194 ** leave both nPathname and zPathname set to 0. |
|
2195 */ |
|
2196 if( zFilename && zFilename[0] ){ |
|
2197 nPathname = pVfs->mxPathname+1; |
|
2198 zPathname = sqlite3Malloc(nPathname*2); |
|
2199 if( zPathname==0 ){ |
|
2200 return SQLITE_NOMEM; |
|
2201 } |
|
2202 #ifndef SQLITE_OMIT_MEMORYDB |
|
2203 if( strcmp(zFilename,":memory:")==0 ){ |
|
2204 memDb = 1; |
|
2205 zPathname[0] = 0; |
|
2206 }else |
|
2207 #endif |
|
2208 { |
|
2209 rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname); |
|
2210 } |
|
2211 if( rc!=SQLITE_OK ){ |
|
2212 sqlite3_free(zPathname); |
|
2213 return rc; |
|
2214 } |
|
2215 nPathname = strlen(zPathname); |
|
2216 } |
|
2217 |
|
2218 /* Allocate memory for the pager structure */ |
|
2219 pPager = sqlite3MallocZero( |
|
2220 sizeof(*pPager) + /* Pager structure */ |
|
2221 journalFileSize + /* The journal file structure */ |
|
2222 pVfs->szOsFile * 3 + /* The main db and two journal files */ |
|
2223 3*nPathname + 40 /* zFilename, zDirectory, zJournal */ |
|
2224 ); |
|
2225 if( !pPager ){ |
|
2226 sqlite3_free(zPathname); |
|
2227 return SQLITE_NOMEM; |
|
2228 } |
|
2229 pPtr = (u8 *)&pPager[1]; |
|
2230 pPager->vfsFlags = vfsFlags; |
|
2231 pPager->fd = (sqlite3_file*)&pPtr[pVfs->szOsFile*0]; |
|
2232 pPager->stfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*1]; |
|
2233 pPager->jfd = (sqlite3_file*)&pPtr[pVfs->szOsFile*2]; |
|
2234 pPager->zFilename = (char*)&pPtr[pVfs->szOsFile*2+journalFileSize]; |
|
2235 pPager->zDirectory = &pPager->zFilename[nPathname+1]; |
|
2236 pPager->zJournal = &pPager->zDirectory[nPathname+1]; |
|
2237 pPager->pVfs = pVfs; |
|
2238 if( zPathname ){ |
|
2239 memcpy(pPager->zFilename, zPathname, nPathname+1); |
|
2240 sqlite3_free(zPathname); |
|
2241 } |
|
2242 |
|
2243 /* Open the pager file. |
|
2244 */ |
|
2245 if( zFilename && zFilename[0] && !memDb ){ |
|
2246 if( nPathname>(pVfs->mxPathname - sizeof("-journal")) ){ |
|
2247 rc = SQLITE_CANTOPEN; |
|
2248 }else{ |
|
2249 int fout = 0; |
|
2250 rc = sqlite3OsOpen(pVfs, pPager->zFilename, pPager->fd, |
|
2251 pPager->vfsFlags, &fout); |
|
2252 readOnly = (fout&SQLITE_OPEN_READONLY); |
|
2253 |
|
2254 /* If the file was successfully opened for read/write access, |
|
2255 ** choose a default page size in case we have to create the |
|
2256 ** database file. The default page size is the maximum of: |
|
2257 ** |
|
2258 ** + SQLITE_DEFAULT_PAGE_SIZE, |
|
2259 ** + The value returned by sqlite3OsSectorSize() |
|
2260 ** + The largest page size that can be written atomically. |
|
2261 */ |
|
2262 if( rc==SQLITE_OK && !readOnly ){ |
|
2263 int iSectorSize = sqlite3OsSectorSize(pPager->fd); |
|
2264 if( szPageDflt<iSectorSize ){ |
|
2265 szPageDflt = iSectorSize; |
|
2266 } |
|
2267 #ifdef SQLITE_ENABLE_ATOMIC_WRITE |
|
2268 { |
|
2269 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); |
|
2270 int ii; |
|
2271 assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); |
|
2272 assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); |
|
2273 assert(SQLITE_MAX_DEFAULT_PAGE_SIZE<=65536); |
|
2274 for(ii=szPageDflt; ii<=SQLITE_MAX_DEFAULT_PAGE_SIZE; ii=ii*2){ |
|
2275 if( iDc&(SQLITE_IOCAP_ATOMIC|(ii>>8)) ) szPageDflt = ii; |
|
2276 } |
|
2277 } |
|
2278 #endif |
|
2279 if( szPageDflt>SQLITE_MAX_DEFAULT_PAGE_SIZE ){ |
|
2280 szPageDflt = SQLITE_MAX_DEFAULT_PAGE_SIZE; |
|
2281 } |
|
2282 } |
|
2283 } |
|
2284 }else if( !memDb ){ |
|
2285 /* If a temporary file is requested, it is not opened immediately. |
|
2286 ** In this case we accept the default page size and delay actually |
|
2287 ** opening the file until the first call to OsWrite(). |
|
2288 */ |
|
2289 tempFile = 1; |
|
2290 pPager->state = PAGER_EXCLUSIVE; |
|
2291 } |
|
2292 |
|
2293 if( pPager && rc==SQLITE_OK ){ |
|
2294 pPager->pTmpSpace = sqlite3PageMalloc(szPageDflt); |
|
2295 } |
|
2296 |
|
2297 /* If an error occured in either of the blocks above. |
|
2298 ** Free the Pager structure and close the file. |
|
2299 ** Since the pager is not allocated there is no need to set |
|
2300 ** any Pager.errMask variables. |
|
2301 */ |
|
2302 if( !pPager || !pPager->pTmpSpace ){ |
|
2303 sqlite3OsClose(pPager->fd); |
|
2304 sqlite3_free(pPager); |
|
2305 return ((rc==SQLITE_OK)?SQLITE_NOMEM:rc); |
|
2306 } |
|
2307 |
|
2308 PAGERTRACE3("OPEN %d %s\n", FILEHANDLEID(pPager->fd), pPager->zFilename); |
|
2309 IOTRACE(("OPEN %p %s\n", pPager, pPager->zFilename)) |
|
2310 |
|
2311 /* Fill in Pager.zDirectory[] */ |
|
2312 memcpy(pPager->zDirectory, pPager->zFilename, nPathname+1); |
|
2313 for(i=strlen(pPager->zDirectory); i>0 && pPager->zDirectory[i-1]!='/'; i--){} |
|
2314 if( i>0 ) pPager->zDirectory[i-1] = 0; |
|
2315 |
|
2316 /* Fill in Pager.zJournal[] */ |
|
2317 if( zPathname ){ |
|
2318 memcpy(pPager->zJournal, pPager->zFilename, nPathname); |
|
2319 memcpy(&pPager->zJournal[nPathname], "-journal", 9); |
|
2320 }else{ |
|
2321 pPager->zJournal = 0; |
|
2322 } |
|
2323 |
|
2324 /* pPager->journalOpen = 0; */ |
|
2325 pPager->useJournal = useJournal && !memDb; |
|
2326 pPager->noReadlock = noReadlock && readOnly; |
|
2327 /* pPager->stmtOpen = 0; */ |
|
2328 /* pPager->stmtInUse = 0; */ |
|
2329 /* pPager->nRef = 0; */ |
|
2330 pPager->dbSize = memDb-1; |
|
2331 pPager->pageSize = szPageDflt; |
|
2332 /* pPager->stmtSize = 0; */ |
|
2333 /* pPager->stmtJSize = 0; */ |
|
2334 /* pPager->nPage = 0; */ |
|
2335 pPager->mxPage = 100; |
|
2336 pPager->mxPgno = SQLITE_MAX_PAGE_COUNT; |
|
2337 /* pPager->state = PAGER_UNLOCK; */ |
|
2338 assert( pPager->state == (tempFile ? PAGER_EXCLUSIVE : PAGER_UNLOCK) ); |
|
2339 /* pPager->errMask = 0; */ |
|
2340 pPager->tempFile = tempFile; |
|
2341 assert( tempFile==PAGER_LOCKINGMODE_NORMAL |
|
2342 || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE ); |
|
2343 assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 ); |
|
2344 pPager->exclusiveMode = tempFile; |
|
2345 pPager->memDb = memDb; |
|
2346 pPager->readOnly = readOnly; |
|
2347 /* pPager->needSync = 0; */ |
|
2348 pPager->noSync = pPager->tempFile || !useJournal; |
|
2349 pPager->fullSync = (pPager->noSync?0:1); |
|
2350 pPager->sync_flags = SQLITE_SYNC_NORMAL; |
|
2351 /* pPager->pFirst = 0; */ |
|
2352 /* pPager->pFirstSynced = 0; */ |
|
2353 /* pPager->pLast = 0; */ |
|
2354 pPager->nExtra = FORCE_ALIGNMENT(nExtra); |
|
2355 pPager->journalSizeLimit = SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT; |
|
2356 assert(pPager->fd->pMethods||memDb||tempFile); |
|
2357 if( !memDb ){ |
|
2358 setSectorSize(pPager); |
|
2359 } |
|
2360 /* pPager->pBusyHandler = 0; */ |
|
2361 /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */ |
|
2362 *ppPager = pPager; |
|
2363 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
2364 pPager->iInUseMM = 0; |
|
2365 pPager->iInUseDB = 0; |
|
2366 if( !memDb ){ |
|
2367 #ifndef SQLITE_MUTEX_NOOP |
|
2368 sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM2); |
|
2369 #endif |
|
2370 sqlite3_mutex_enter(mutex); |
|
2371 pPager->pNext = sqlite3PagerList; |
|
2372 if( sqlite3PagerList ){ |
|
2373 assert( sqlite3PagerList->pPrev==0 ); |
|
2374 sqlite3PagerList->pPrev = pPager; |
|
2375 } |
|
2376 pPager->pPrev = 0; |
|
2377 sqlite3PagerList = pPager; |
|
2378 sqlite3_mutex_leave(mutex); |
|
2379 } |
|
2380 #endif |
|
2381 return SQLITE_OK; |
|
2382 } |
|
2383 |
|
2384 /* |
|
2385 ** Set the busy handler function. |
|
2386 */ |
|
2387 void sqlite3PagerSetBusyhandler(Pager *pPager, BusyHandler *pBusyHandler){ |
|
2388 pPager->pBusyHandler = pBusyHandler; |
|
2389 } |
|
2390 |
|
2391 /* |
|
2392 ** Set the destructor for this pager. If not NULL, the destructor is called |
|
2393 ** when the reference count on each page reaches zero. The destructor can |
|
2394 ** be used to clean up information in the extra segment appended to each page. |
|
2395 ** |
|
2396 ** The destructor is not called as a result sqlite3PagerClose(). |
|
2397 ** Destructors are only called by sqlite3PagerUnref(). |
|
2398 */ |
|
2399 void sqlite3PagerSetDestructor(Pager *pPager, void (*xDesc)(DbPage*,int)){ |
|
2400 pPager->xDestructor = xDesc; |
|
2401 } |
|
2402 |
|
2403 /* |
|
2404 ** Set the reinitializer for this pager. If not NULL, the reinitializer |
|
2405 ** is called when the content of a page in cache is restored to its original |
|
2406 ** value as a result of a rollback. The callback gives higher-level code |
|
2407 ** an opportunity to restore the EXTRA section to agree with the restored |
|
2408 ** page data. |
|
2409 */ |
|
2410 void sqlite3PagerSetReiniter(Pager *pPager, void (*xReinit)(DbPage*,int)){ |
|
2411 pPager->xReiniter = xReinit; |
|
2412 } |
|
2413 |
|
2414 /* |
|
2415 ** Set the page size to *pPageSize. If the suggest new page size is |
|
2416 ** inappropriate, then an alternative page size is set to that |
|
2417 ** value before returning. |
|
2418 */ |
|
2419 int sqlite3PagerSetPagesize(Pager *pPager, u16 *pPageSize){ |
|
2420 int rc = SQLITE_OK; |
|
2421 u16 pageSize = *pPageSize; |
|
2422 assert( pageSize==0 || (pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE) ); |
|
2423 if( pageSize && pageSize!=pPager->pageSize |
|
2424 && !pPager->memDb && pPager->nRef==0 |
|
2425 ){ |
|
2426 char *pNew = (char *)sqlite3PageMalloc(pageSize); |
|
2427 if( !pNew ){ |
|
2428 rc = SQLITE_NOMEM; |
|
2429 }else{ |
|
2430 pagerEnter(pPager); |
|
2431 pager_reset(pPager); |
|
2432 pPager->pageSize = pageSize; |
|
2433 setSectorSize(pPager); |
|
2434 sqlite3PageFree(pPager->pTmpSpace); |
|
2435 pPager->pTmpSpace = pNew; |
|
2436 pagerLeave(pPager); |
|
2437 } |
|
2438 } |
|
2439 *pPageSize = pPager->pageSize; |
|
2440 return rc; |
|
2441 } |
|
2442 |
|
2443 /* |
|
2444 ** Return a pointer to the "temporary page" buffer held internally |
|
2445 ** by the pager. This is a buffer that is big enough to hold the |
|
2446 ** entire content of a database page. This buffer is used internally |
|
2447 ** during rollback and will be overwritten whenever a rollback |
|
2448 ** occurs. But other modules are free to use it too, as long as |
|
2449 ** no rollbacks are happening. |
|
2450 */ |
|
2451 void *sqlite3PagerTempSpace(Pager *pPager){ |
|
2452 return pPager->pTmpSpace; |
|
2453 } |
|
2454 |
|
2455 /* |
|
2456 ** Attempt to set the maximum database page count if mxPage is positive. |
|
2457 ** Make no changes if mxPage is zero or negative. And never reduce the |
|
2458 ** maximum page count below the current size of the database. |
|
2459 ** |
|
2460 ** Regardless of mxPage, return the current maximum page count. |
|
2461 */ |
|
2462 int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){ |
|
2463 if( mxPage>0 ){ |
|
2464 pPager->mxPgno = mxPage; |
|
2465 } |
|
2466 sqlite3PagerPagecount(pPager, 0); |
|
2467 return pPager->mxPgno; |
|
2468 } |
|
2469 |
|
2470 /* |
|
2471 ** The following set of routines are used to disable the simulated |
|
2472 ** I/O error mechanism. These routines are used to avoid simulated |
|
2473 ** errors in places where we do not care about errors. |
|
2474 ** |
|
2475 ** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops |
|
2476 ** and generate no code. |
|
2477 */ |
|
2478 #ifdef SQLITE_TEST |
|
2479 extern int sqlite3_io_error_pending; |
|
2480 extern int sqlite3_io_error_hit; |
|
2481 static int saved_cnt; |
|
2482 void disable_simulated_io_errors(void){ |
|
2483 saved_cnt = sqlite3_io_error_pending; |
|
2484 sqlite3_io_error_pending = -1; |
|
2485 } |
|
2486 void enable_simulated_io_errors(void){ |
|
2487 sqlite3_io_error_pending = saved_cnt; |
|
2488 } |
|
2489 #else |
|
2490 # define disable_simulated_io_errors() |
|
2491 # define enable_simulated_io_errors() |
|
2492 #endif |
|
2493 |
|
2494 /* |
|
2495 ** Read the first N bytes from the beginning of the file into memory |
|
2496 ** that pDest points to. |
|
2497 ** |
|
2498 ** No error checking is done. The rational for this is that this function |
|
2499 ** may be called even if the file does not exist or contain a header. In |
|
2500 ** these cases sqlite3OsRead() will return an error, to which the correct |
|
2501 ** response is to zero the memory at pDest and continue. A real IO error |
|
2502 ** will presumably recur and be picked up later (Todo: Think about this). |
|
2503 */ |
|
2504 int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){ |
|
2505 int rc = SQLITE_OK; |
|
2506 memset(pDest, 0, N); |
|
2507 assert(MEMDB||pPager->fd->pMethods||pPager->tempFile); |
|
2508 if( pPager->fd->pMethods ){ |
|
2509 IOTRACE(("DBHDR %p 0 %d\n", pPager, N)) |
|
2510 rc = sqlite3OsRead(pPager->fd, pDest, N, 0); |
|
2511 if( rc==SQLITE_IOERR_SHORT_READ ){ |
|
2512 rc = SQLITE_OK; |
|
2513 } |
|
2514 } |
|
2515 return rc; |
|
2516 } |
|
2517 |
|
2518 /* |
|
2519 ** Return the total number of pages in the disk file associated with |
|
2520 ** pPager. |
|
2521 ** |
|
2522 ** If the PENDING_BYTE lies on the page directly after the end of the |
|
2523 ** file, then consider this page part of the file too. For example, if |
|
2524 ** PENDING_BYTE is byte 4096 (the first byte of page 5) and the size of the |
|
2525 ** file is 4096 bytes, 5 is returned instead of 4. |
|
2526 */ |
|
2527 int sqlite3PagerPagecount(Pager *pPager, int *pnPage){ |
|
2528 i64 n = 0; |
|
2529 int rc; |
|
2530 assert( pPager!=0 ); |
|
2531 if( pPager->errCode ){ |
|
2532 return pPager->errCode; |
|
2533 } |
|
2534 if( pPager->dbSize>=0 ){ |
|
2535 n = pPager->dbSize; |
|
2536 } else { |
|
2537 assert(pPager->fd->pMethods||pPager->tempFile); |
|
2538 if( (pPager->fd->pMethods) |
|
2539 && (rc = sqlite3OsFileSize(pPager->fd, &n))!=SQLITE_OK ){ |
|
2540 pPager->nRef++; |
|
2541 pager_error(pPager, rc); |
|
2542 pPager->nRef--; |
|
2543 return rc; |
|
2544 } |
|
2545 if( n>0 && n<pPager->pageSize ){ |
|
2546 n = 1; |
|
2547 }else{ |
|
2548 n /= pPager->pageSize; |
|
2549 } |
|
2550 if( pPager->state!=PAGER_UNLOCK ){ |
|
2551 pPager->dbSize = n; |
|
2552 } |
|
2553 } |
|
2554 if( n==(PENDING_BYTE/pPager->pageSize) ){ |
|
2555 n++; |
|
2556 } |
|
2557 if( n>pPager->mxPgno ){ |
|
2558 pPager->mxPgno = n; |
|
2559 } |
|
2560 if( pnPage ){ |
|
2561 *pnPage = n; |
|
2562 } |
|
2563 return SQLITE_OK; |
|
2564 } |
|
2565 |
|
2566 |
|
2567 #ifndef SQLITE_OMIT_MEMORYDB |
|
2568 /* |
|
2569 ** Clear a PgHistory block |
|
2570 */ |
|
2571 static void clearHistory(PgHistory *pHist){ |
|
2572 sqlite3PageFree(pHist->pOrig); |
|
2573 sqlite3PageFree(pHist->pStmt); |
|
2574 pHist->pOrig = 0; |
|
2575 pHist->pStmt = 0; |
|
2576 } |
|
2577 #else |
|
2578 #define clearHistory(x) |
|
2579 #endif |
|
2580 |
|
2581 /* |
|
2582 ** Forward declaration |
|
2583 */ |
|
2584 static int syncJournal(Pager*); |
|
2585 |
|
2586 /* |
|
2587 ** Unlink pPg from its hash chain. Also set the page number to 0 to indicate |
|
2588 ** that the page is not part of any hash chain. This is required because the |
|
2589 ** sqlite3PagerMovepage() routine can leave a page in the |
|
2590 ** pNextFree/pPrevFree list that is not a part of any hash-chain. |
|
2591 */ |
|
2592 static void unlinkHashChain(Pager *pPager, PgHdr *pPg){ |
|
2593 if( pPg->pgno==0 ){ |
|
2594 assert( pPg->pNextHash==0 && pPg->pPrevHash==0 ); |
|
2595 return; |
|
2596 } |
|
2597 if( pPg->pNextHash ){ |
|
2598 pPg->pNextHash->pPrevHash = pPg->pPrevHash; |
|
2599 } |
|
2600 if( pPg->pPrevHash ){ |
|
2601 assert( pPager->aHash[pPg->pgno & (pPager->nHash-1)]!=pPg ); |
|
2602 pPg->pPrevHash->pNextHash = pPg->pNextHash; |
|
2603 }else{ |
|
2604 int h = pPg->pgno & (pPager->nHash-1); |
|
2605 pPager->aHash[h] = pPg->pNextHash; |
|
2606 } |
|
2607 if( MEMDB ){ |
|
2608 clearHistory(PGHDR_TO_HIST(pPg, pPager)); |
|
2609 } |
|
2610 pPg->pgno = 0; |
|
2611 pPg->pNextHash = pPg->pPrevHash = 0; |
|
2612 } |
|
2613 |
|
2614 /* |
|
2615 ** Unlink a page from the free list (the list of all pages where nRef==0) |
|
2616 ** and from its hash collision chain. |
|
2617 */ |
|
2618 static void unlinkPage(PgHdr *pPg){ |
|
2619 Pager *pPager = pPg->pPager; |
|
2620 |
|
2621 /* Unlink from free page list */ |
|
2622 lruListRemove(pPg); |
|
2623 |
|
2624 /* Unlink from the pgno hash table */ |
|
2625 unlinkHashChain(pPager, pPg); |
|
2626 } |
|
2627 |
|
2628 /* |
|
2629 ** This routine is used to truncate the cache when a database |
|
2630 ** is truncated. Drop from the cache all pages whose pgno is |
|
2631 ** larger than pPager->dbSize and is unreferenced. |
|
2632 ** |
|
2633 ** Referenced pages larger than pPager->dbSize are zeroed. |
|
2634 ** |
|
2635 ** Actually, at the point this routine is called, it would be |
|
2636 ** an error to have a referenced page. But rather than delete |
|
2637 ** that page and guarantee a subsequent segfault, it seems better |
|
2638 ** to zero it and hope that we error out sanely. |
|
2639 */ |
|
2640 static void pager_truncate_cache(Pager *pPager){ |
|
2641 PgHdr *pPg; |
|
2642 PgHdr **ppPg; |
|
2643 int dbSize = pPager->dbSize; |
|
2644 |
|
2645 ppPg = &pPager->pAll; |
|
2646 while( (pPg = *ppPg)!=0 ){ |
|
2647 if( pPg->pgno<=dbSize ){ |
|
2648 ppPg = &pPg->pNextAll; |
|
2649 }else if( pPg->nRef>0 ){ |
|
2650 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize); |
|
2651 ppPg = &pPg->pNextAll; |
|
2652 }else{ |
|
2653 *ppPg = pPg->pNextAll; |
|
2654 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
2655 if( *ppPg ){ |
|
2656 (*ppPg)->pPrevAll = pPg->pPrevAll; |
|
2657 } |
|
2658 #endif |
|
2659 IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno)); |
|
2660 PAGER_INCR(sqlite3_pager_pgfree_count); |
|
2661 unlinkPage(pPg); |
|
2662 makeClean(pPg); |
|
2663 sqlite3PageFree(pPg->pData); |
|
2664 sqlite3_free(pPg); |
|
2665 pPager->nPage--; |
|
2666 } |
|
2667 } |
|
2668 } |
|
2669 |
|
2670 /* |
|
2671 ** Try to obtain a lock on a file. Invoke the busy callback if the lock |
|
2672 ** is currently not available. Repeat until the busy callback returns |
|
2673 ** false or until the lock succeeds. |
|
2674 ** |
|
2675 ** Return SQLITE_OK on success and an error code if we cannot obtain |
|
2676 ** the lock. |
|
2677 */ |
|
2678 static int pager_wait_on_lock(Pager *pPager, int locktype){ |
|
2679 int rc; |
|
2680 |
|
2681 /* The OS lock values must be the same as the Pager lock values */ |
|
2682 assert( PAGER_SHARED==SHARED_LOCK ); |
|
2683 assert( PAGER_RESERVED==RESERVED_LOCK ); |
|
2684 assert( PAGER_EXCLUSIVE==EXCLUSIVE_LOCK ); |
|
2685 |
|
2686 /* If the file is currently unlocked then the size must be unknown */ |
|
2687 assert( pPager->state>=PAGER_SHARED || pPager->dbSize<0 || MEMDB ); |
|
2688 |
|
2689 if( pPager->state>=locktype ){ |
|
2690 rc = SQLITE_OK; |
|
2691 }else{ |
|
2692 if( pPager->pBusyHandler ) pPager->pBusyHandler->nBusy = 0; |
|
2693 do { |
|
2694 rc = sqlite3OsLock(pPager->fd, locktype); |
|
2695 }while( rc==SQLITE_BUSY && sqlite3InvokeBusyHandler(pPager->pBusyHandler) ); |
|
2696 if( rc==SQLITE_OK ){ |
|
2697 pPager->state = locktype; |
|
2698 IOTRACE(("LOCK %p %d\n", pPager, locktype)) |
|
2699 } |
|
2700 } |
|
2701 return rc; |
|
2702 } |
|
2703 |
|
2704 /* |
|
2705 ** Truncate the file to the number of pages specified. |
|
2706 */ |
|
2707 int sqlite3PagerTruncate(Pager *pPager, Pgno nPage){ |
|
2708 int rc; |
|
2709 assert( pPager->state>=PAGER_SHARED || MEMDB ); |
|
2710 sqlite3PagerPagecount(pPager, 0); |
|
2711 if( pPager->errCode ){ |
|
2712 rc = pPager->errCode; |
|
2713 return rc; |
|
2714 } |
|
2715 if( nPage>=(unsigned)pPager->dbSize ){ |
|
2716 return SQLITE_OK; |
|
2717 } |
|
2718 if( MEMDB ){ |
|
2719 pPager->dbSize = nPage; |
|
2720 pager_truncate_cache(pPager); |
|
2721 return SQLITE_OK; |
|
2722 } |
|
2723 pagerEnter(pPager); |
|
2724 rc = syncJournal(pPager); |
|
2725 pagerLeave(pPager); |
|
2726 if( rc!=SQLITE_OK ){ |
|
2727 return rc; |
|
2728 } |
|
2729 |
|
2730 /* Get an exclusive lock on the database before truncating. */ |
|
2731 pagerEnter(pPager); |
|
2732 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); |
|
2733 pagerLeave(pPager); |
|
2734 if( rc!=SQLITE_OK ){ |
|
2735 return rc; |
|
2736 } |
|
2737 |
|
2738 rc = pager_truncate(pPager, nPage); |
|
2739 return rc; |
|
2740 } |
|
2741 |
|
2742 /* |
|
2743 ** Shutdown the page cache. Free all memory and close all files. |
|
2744 ** |
|
2745 ** If a transaction was in progress when this routine is called, that |
|
2746 ** transaction is rolled back. All outstanding pages are invalidated |
|
2747 ** and their memory is freed. Any attempt to use a page associated |
|
2748 ** with this page cache after this function returns will likely |
|
2749 ** result in a coredump. |
|
2750 ** |
|
2751 ** This function always succeeds. If a transaction is active an attempt |
|
2752 ** is made to roll it back. If an error occurs during the rollback |
|
2753 ** a hot journal may be left in the filesystem but no error is returned |
|
2754 ** to the caller. |
|
2755 */ |
|
2756 int sqlite3PagerClose(Pager *pPager){ |
|
2757 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
2758 if( !MEMDB ){ |
|
2759 #ifndef SQLITE_MUTEX_NOOP |
|
2760 sqlite3_mutex *mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM2); |
|
2761 #endif |
|
2762 sqlite3_mutex_enter(mutex); |
|
2763 if( pPager->pPrev ){ |
|
2764 pPager->pPrev->pNext = pPager->pNext; |
|
2765 }else{ |
|
2766 sqlite3PagerList = pPager->pNext; |
|
2767 } |
|
2768 if( pPager->pNext ){ |
|
2769 pPager->pNext->pPrev = pPager->pPrev; |
|
2770 } |
|
2771 sqlite3_mutex_leave(mutex); |
|
2772 } |
|
2773 #endif |
|
2774 |
|
2775 disable_simulated_io_errors(); |
|
2776 sqlite3BeginBenignMalloc(); |
|
2777 pPager->errCode = 0; |
|
2778 pPager->exclusiveMode = 0; |
|
2779 pager_reset(pPager); |
|
2780 pagerUnlockAndRollback(pPager); |
|
2781 enable_simulated_io_errors(); |
|
2782 sqlite3EndBenignMalloc(); |
|
2783 PAGERTRACE2("CLOSE %d\n", PAGERID(pPager)); |
|
2784 IOTRACE(("CLOSE %p\n", pPager)) |
|
2785 if( pPager->journalOpen ){ |
|
2786 sqlite3OsClose(pPager->jfd); |
|
2787 } |
|
2788 sqlite3BitvecDestroy(pPager->pInJournal); |
|
2789 if( pPager->stmtOpen ){ |
|
2790 sqlite3OsClose(pPager->stfd); |
|
2791 } |
|
2792 sqlite3OsClose(pPager->fd); |
|
2793 /* Temp files are automatically deleted by the OS |
|
2794 ** if( pPager->tempFile ){ |
|
2795 ** sqlite3OsDelete(pPager->zFilename); |
|
2796 ** } |
|
2797 */ |
|
2798 |
|
2799 sqlite3_free(pPager->aHash); |
|
2800 sqlite3PageFree(pPager->pTmpSpace); |
|
2801 sqlite3_free(pPager); |
|
2802 return SQLITE_OK; |
|
2803 } |
|
2804 |
|
2805 #if !defined(NDEBUG) || defined(SQLITE_TEST) |
|
2806 /* |
|
2807 ** Return the page number for the given page data. |
|
2808 */ |
|
2809 Pgno sqlite3PagerPagenumber(DbPage *p){ |
|
2810 return p->pgno; |
|
2811 } |
|
2812 #endif |
|
2813 |
|
2814 /* |
|
2815 ** The page_ref() function increments the reference count for a page. |
|
2816 ** If the page is currently on the freelist (the reference count is zero) then |
|
2817 ** remove it from the freelist. |
|
2818 ** |
|
2819 ** For non-test systems, page_ref() is a macro that calls _page_ref() |
|
2820 ** online of the reference count is zero. For test systems, page_ref() |
|
2821 ** is a real function so that we can set breakpoints and trace it. |
|
2822 */ |
|
2823 static void _page_ref(PgHdr *pPg){ |
|
2824 if( pPg->nRef==0 ){ |
|
2825 /* The page is currently on the freelist. Remove it. */ |
|
2826 lruListRemove(pPg); |
|
2827 pPg->pPager->nRef++; |
|
2828 } |
|
2829 pPg->nRef++; |
|
2830 } |
|
2831 #ifdef SQLITE_DEBUG |
|
2832 static void page_ref(PgHdr *pPg){ |
|
2833 if( pPg->nRef==0 ){ |
|
2834 _page_ref(pPg); |
|
2835 }else{ |
|
2836 pPg->nRef++; |
|
2837 } |
|
2838 } |
|
2839 #else |
|
2840 # define page_ref(P) ((P)->nRef==0?_page_ref(P):(void)(P)->nRef++) |
|
2841 #endif |
|
2842 |
|
2843 /* |
|
2844 ** Increment the reference count for a page. The input pointer is |
|
2845 ** a reference to the page data. |
|
2846 */ |
|
2847 int sqlite3PagerRef(DbPage *pPg){ |
|
2848 pagerEnter(pPg->pPager); |
|
2849 page_ref(pPg); |
|
2850 pagerLeave(pPg->pPager); |
|
2851 return SQLITE_OK; |
|
2852 } |
|
2853 |
|
2854 /* |
|
2855 ** Sync the journal. In other words, make sure all the pages that have |
|
2856 ** been written to the journal have actually reached the surface of the |
|
2857 ** disk. It is not safe to modify the original database file until after |
|
2858 ** the journal has been synced. If the original database is modified before |
|
2859 ** the journal is synced and a power failure occurs, the unsynced journal |
|
2860 ** data would be lost and we would be unable to completely rollback the |
|
2861 ** database changes. Database corruption would occur. |
|
2862 ** |
|
2863 ** This routine also updates the nRec field in the header of the journal. |
|
2864 ** (See comments on the pager_playback() routine for additional information.) |
|
2865 ** If the sync mode is FULL, two syncs will occur. First the whole journal |
|
2866 ** is synced, then the nRec field is updated, then a second sync occurs. |
|
2867 ** |
|
2868 ** For temporary databases, we do not care if we are able to rollback |
|
2869 ** after a power failure, so no sync occurs. |
|
2870 ** |
|
2871 ** If the IOCAP_SEQUENTIAL flag is set for the persistent media on which |
|
2872 ** the database is stored, then OsSync() is never called on the journal |
|
2873 ** file. In this case all that is required is to update the nRec field in |
|
2874 ** the journal header. |
|
2875 ** |
|
2876 ** This routine clears the needSync field of every page current held in |
|
2877 ** memory. |
|
2878 */ |
|
2879 static int syncJournal(Pager *pPager){ |
|
2880 PgHdr *pPg; |
|
2881 int rc = SQLITE_OK; |
|
2882 |
|
2883 /* Sync the journal before modifying the main database |
|
2884 ** (assuming there is a journal and it needs to be synced.) |
|
2885 */ |
|
2886 if( pPager->needSync ){ |
|
2887 if( !pPager->tempFile ){ |
|
2888 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); |
|
2889 assert( pPager->journalOpen ); |
|
2890 |
|
2891 if( 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ |
|
2892 /* Write the nRec value into the journal file header. If in |
|
2893 ** full-synchronous mode, sync the journal first. This ensures that |
|
2894 ** all data has really hit the disk before nRec is updated to mark |
|
2895 ** it as a candidate for rollback. |
|
2896 ** |
|
2897 ** This is not required if the persistent media supports the |
|
2898 ** SAFE_APPEND property. Because in this case it is not possible |
|
2899 ** for garbage data to be appended to the file, the nRec field |
|
2900 ** is populated with 0xFFFFFFFF when the journal header is written |
|
2901 ** and never needs to be updated. |
|
2902 */ |
|
2903 i64 jrnlOff; |
|
2904 if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ |
|
2905 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager)); |
|
2906 IOTRACE(("JSYNC %p\n", pPager)) |
|
2907 rc = sqlite3OsSync(pPager->jfd, pPager->sync_flags); |
|
2908 if( rc!=0 ) return rc; |
|
2909 } |
|
2910 |
|
2911 jrnlOff = pPager->journalHdr + sizeof(aJournalMagic); |
|
2912 IOTRACE(("JHDR %p %lld %d\n", pPager, jrnlOff, 4)); |
|
2913 rc = write32bits(pPager->jfd, jrnlOff, pPager->nRec); |
|
2914 if( rc ) return rc; |
|
2915 } |
|
2916 if( 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ |
|
2917 PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager)); |
|
2918 IOTRACE(("JSYNC %p\n", pPager)) |
|
2919 rc = sqlite3OsSync(pPager->jfd, pPager->sync_flags| |
|
2920 (pPager->sync_flags==SQLITE_SYNC_FULL?SQLITE_SYNC_DATAONLY:0) |
|
2921 ); |
|
2922 if( rc!=0 ) return rc; |
|
2923 } |
|
2924 pPager->journalStarted = 1; |
|
2925 } |
|
2926 pPager->needSync = 0; |
|
2927 |
|
2928 /* Erase the needSync flag from every page. |
|
2929 */ |
|
2930 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ |
|
2931 pPg->needSync = 0; |
|
2932 } |
|
2933 lruListSetFirstSynced(pPager); |
|
2934 } |
|
2935 |
|
2936 #ifndef NDEBUG |
|
2937 /* If the Pager.needSync flag is clear then the PgHdr.needSync |
|
2938 ** flag must also be clear for all pages. Verify that this |
|
2939 ** invariant is true. |
|
2940 */ |
|
2941 else{ |
|
2942 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ |
|
2943 assert( pPg->needSync==0 ); |
|
2944 } |
|
2945 assert( pPager->lru.pFirstSynced==pPager->lru.pFirst ); |
|
2946 } |
|
2947 #endif |
|
2948 |
|
2949 return rc; |
|
2950 } |
|
2951 |
|
2952 /* |
|
2953 ** Merge two lists of pages connected by pDirty and in pgno order. |
|
2954 ** Do not both fixing the pPrevDirty pointers. |
|
2955 */ |
|
2956 static PgHdr *merge_pagelist(PgHdr *pA, PgHdr *pB){ |
|
2957 PgHdr result, *pTail; |
|
2958 pTail = &result; |
|
2959 while( pA && pB ){ |
|
2960 if( pA->pgno<pB->pgno ){ |
|
2961 pTail->pDirty = pA; |
|
2962 pTail = pA; |
|
2963 pA = pA->pDirty; |
|
2964 }else{ |
|
2965 pTail->pDirty = pB; |
|
2966 pTail = pB; |
|
2967 pB = pB->pDirty; |
|
2968 } |
|
2969 } |
|
2970 if( pA ){ |
|
2971 pTail->pDirty = pA; |
|
2972 }else if( pB ){ |
|
2973 pTail->pDirty = pB; |
|
2974 }else{ |
|
2975 pTail->pDirty = 0; |
|
2976 } |
|
2977 return result.pDirty; |
|
2978 } |
|
2979 |
|
2980 /* |
|
2981 ** Sort the list of pages in accending order by pgno. Pages are |
|
2982 ** connected by pDirty pointers. The pPrevDirty pointers are |
|
2983 ** corrupted by this sort. |
|
2984 */ |
|
2985 #define N_SORT_BUCKET_ALLOC 25 |
|
2986 #define N_SORT_BUCKET 25 |
|
2987 #ifdef SQLITE_TEST |
|
2988 int sqlite3_pager_n_sort_bucket = 0; |
|
2989 #undef N_SORT_BUCKET |
|
2990 #define N_SORT_BUCKET \ |
|
2991 (sqlite3_pager_n_sort_bucket?sqlite3_pager_n_sort_bucket:N_SORT_BUCKET_ALLOC) |
|
2992 #endif |
|
2993 static PgHdr *sort_pagelist(PgHdr *pIn){ |
|
2994 PgHdr *a[N_SORT_BUCKET_ALLOC], *p; |
|
2995 int i; |
|
2996 memset(a, 0, sizeof(a)); |
|
2997 while( pIn ){ |
|
2998 p = pIn; |
|
2999 pIn = p->pDirty; |
|
3000 p->pDirty = 0; |
|
3001 for(i=0; i<N_SORT_BUCKET-1; i++){ |
|
3002 if( a[i]==0 ){ |
|
3003 a[i] = p; |
|
3004 break; |
|
3005 }else{ |
|
3006 p = merge_pagelist(a[i], p); |
|
3007 a[i] = 0; |
|
3008 } |
|
3009 } |
|
3010 if( i==N_SORT_BUCKET-1 ){ |
|
3011 /* Coverage: To get here, there need to be 2^(N_SORT_BUCKET) |
|
3012 ** elements in the input list. This is possible, but impractical. |
|
3013 ** Testing this line is the point of global variable |
|
3014 ** sqlite3_pager_n_sort_bucket. |
|
3015 */ |
|
3016 a[i] = merge_pagelist(a[i], p); |
|
3017 } |
|
3018 } |
|
3019 p = a[0]; |
|
3020 for(i=1; i<N_SORT_BUCKET; i++){ |
|
3021 p = merge_pagelist(p, a[i]); |
|
3022 } |
|
3023 return p; |
|
3024 } |
|
3025 |
|
3026 /* |
|
3027 ** Given a list of pages (connected by the PgHdr.pDirty pointer) write |
|
3028 ** every one of those pages out to the database file and mark them all |
|
3029 ** as clean. |
|
3030 */ |
|
3031 static int pager_write_pagelist(PgHdr *pList){ |
|
3032 Pager *pPager; |
|
3033 PgHdr *p; |
|
3034 int rc; |
|
3035 |
|
3036 if( pList==0 ) return SQLITE_OK; |
|
3037 pPager = pList->pPager; |
|
3038 |
|
3039 /* At this point there may be either a RESERVED or EXCLUSIVE lock on the |
|
3040 ** database file. If there is already an EXCLUSIVE lock, the following |
|
3041 ** calls to sqlite3OsLock() are no-ops. |
|
3042 ** |
|
3043 ** Moving the lock from RESERVED to EXCLUSIVE actually involves going |
|
3044 ** through an intermediate state PENDING. A PENDING lock prevents new |
|
3045 ** readers from attaching to the database but is unsufficient for us to |
|
3046 ** write. The idea of a PENDING lock is to prevent new readers from |
|
3047 ** coming in while we wait for existing readers to clear. |
|
3048 ** |
|
3049 ** While the pager is in the RESERVED state, the original database file |
|
3050 ** is unchanged and we can rollback without having to playback the |
|
3051 ** journal into the original database file. Once we transition to |
|
3052 ** EXCLUSIVE, it means the database file has been changed and any rollback |
|
3053 ** will require a journal playback. |
|
3054 */ |
|
3055 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); |
|
3056 if( rc!=SQLITE_OK ){ |
|
3057 return rc; |
|
3058 } |
|
3059 |
|
3060 pList = sort_pagelist(pList); |
|
3061 for(p=pList; p; p=p->pDirty){ |
|
3062 assert( p->dirty ); |
|
3063 p->dirty = 0; |
|
3064 } |
|
3065 while( pList ){ |
|
3066 |
|
3067 /* If the file has not yet been opened, open it now. */ |
|
3068 if( !pPager->fd->pMethods ){ |
|
3069 assert(pPager->tempFile); |
|
3070 rc = sqlite3PagerOpentemp(pPager, pPager->fd, pPager->vfsFlags); |
|
3071 if( rc ) return rc; |
|
3072 } |
|
3073 |
|
3074 /* If there are dirty pages in the page cache with page numbers greater |
|
3075 ** than Pager.dbSize, this means sqlite3PagerTruncate() was called to |
|
3076 ** make the file smaller (presumably by auto-vacuum code). Do not write |
|
3077 ** any such pages to the file. |
|
3078 */ |
|
3079 if( pList->pgno<=pPager->dbSize ){ |
|
3080 i64 offset = (pList->pgno-1)*(i64)pPager->pageSize; |
|
3081 char *pData = CODEC2(pPager, PGHDR_TO_DATA(pList), pList->pgno, 6); |
|
3082 PAGERTRACE4("STORE %d page %d hash(%08x)\n", |
|
3083 PAGERID(pPager), pList->pgno, pager_pagehash(pList)); |
|
3084 IOTRACE(("PGOUT %p %d\n", pPager, pList->pgno)); |
|
3085 rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize, offset); |
|
3086 PAGER_INCR(sqlite3_pager_writedb_count); |
|
3087 PAGER_INCR(pPager->nWrite); |
|
3088 if( pList->pgno==1 ){ |
|
3089 memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers)); |
|
3090 } |
|
3091 } |
|
3092 #ifndef NDEBUG |
|
3093 else{ |
|
3094 PAGERTRACE3("NOSTORE %d page %d\n", PAGERID(pPager), pList->pgno); |
|
3095 } |
|
3096 #endif |
|
3097 if( rc ) return rc; |
|
3098 #ifdef SQLITE_CHECK_PAGES |
|
3099 pList->pageHash = pager_pagehash(pList); |
|
3100 #endif |
|
3101 pList = pList->pDirty; |
|
3102 } |
|
3103 return SQLITE_OK; |
|
3104 } |
|
3105 |
|
3106 /* |
|
3107 ** Collect every dirty page into a dirty list and |
|
3108 ** return a pointer to the head of that list. All pages are |
|
3109 ** collected even if they are still in use. |
|
3110 */ |
|
3111 static PgHdr *pager_get_all_dirty_pages(Pager *pPager){ |
|
3112 |
|
3113 #ifndef NDEBUG |
|
3114 /* Verify the sanity of the dirty list when we are running |
|
3115 ** in debugging mode. This is expensive, so do not |
|
3116 ** do this on a normal build. */ |
|
3117 int n1 = 0; |
|
3118 int n2 = 0; |
|
3119 PgHdr *p; |
|
3120 for(p=pPager->pAll; p; p=p->pNextAll){ if( p->dirty ) n1++; } |
|
3121 for(p=pPager->pDirty; p; p=p->pDirty){ n2++; } |
|
3122 assert( n1==n2 ); |
|
3123 #endif |
|
3124 |
|
3125 return pPager->pDirty; |
|
3126 } |
|
3127 |
|
3128 /* |
|
3129 ** Return 1 if there is a hot journal on the given pager. |
|
3130 ** A hot journal is one that needs to be played back. |
|
3131 ** |
|
3132 ** If the current size of the database file is 0 but a journal file |
|
3133 ** exists, that is probably an old journal left over from a prior |
|
3134 ** database with the same name. Just delete the journal. |
|
3135 ** |
|
3136 ** Return negative if unable to determine the status of the journal. |
|
3137 ** |
|
3138 ** This routine does not open the journal file to examine its |
|
3139 ** content. Hence, the journal might contain the name of a master |
|
3140 ** journal file that has been deleted, and hence not be hot. Or |
|
3141 ** the header of the journal might be zeroed out. This routine |
|
3142 ** does not discover these cases of a non-hot journal - if the |
|
3143 ** journal file exists and is not empty this routine assumes it |
|
3144 ** is hot. The pager_playback() routine will discover that the |
|
3145 ** journal file is not really hot and will no-op. |
|
3146 */ |
|
3147 static int hasHotJournal(Pager *pPager, int *pExists){ |
|
3148 sqlite3_vfs *pVfs = pPager->pVfs; |
|
3149 int rc = SQLITE_OK; |
|
3150 *pExists = 0; |
|
3151 if( pPager->useJournal && pPager->fd->pMethods ){ |
|
3152 int exists; |
|
3153 int locked; |
|
3154 |
|
3155 rc = sqlite3OsAccess(pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS, &exists); |
|
3156 if( rc==SQLITE_OK && exists ){ |
|
3157 rc = sqlite3OsCheckReservedLock(pPager->fd, &locked); |
|
3158 } |
|
3159 |
|
3160 if( rc==SQLITE_OK && exists && !locked ){ |
|
3161 int nPage; |
|
3162 rc = sqlite3PagerPagecount(pPager, &nPage); |
|
3163 if( rc==SQLITE_OK ){ |
|
3164 if( nPage==0 ){ |
|
3165 sqlite3OsDelete(pVfs, pPager->zJournal, 0); |
|
3166 }else{ |
|
3167 *pExists = 1; |
|
3168 } |
|
3169 } |
|
3170 } |
|
3171 } |
|
3172 |
|
3173 return rc; |
|
3174 } |
|
3175 |
|
3176 /* |
|
3177 ** Try to find a page in the cache that can be recycled. |
|
3178 ** |
|
3179 ** This routine may return SQLITE_IOERR, SQLITE_FULL or SQLITE_OK. It |
|
3180 ** does not set the pPager->errCode variable. |
|
3181 */ |
|
3182 static int pager_recycle(Pager *pPager, PgHdr **ppPg){ |
|
3183 PgHdr *pPg; |
|
3184 *ppPg = 0; |
|
3185 |
|
3186 /* It is illegal to call this function unless the pager object |
|
3187 ** pointed to by pPager has at least one free page (page with nRef==0). |
|
3188 */ |
|
3189 assert(!MEMDB); |
|
3190 assert(pPager->lru.pFirst); |
|
3191 |
|
3192 /* Find a page to recycle. Try to locate a page that does not |
|
3193 ** require us to do an fsync() on the journal. |
|
3194 */ |
|
3195 pPg = pPager->lru.pFirstSynced; |
|
3196 |
|
3197 /* If we could not find a page that does not require an fsync() |
|
3198 ** on the journal file then fsync the journal file. This is a |
|
3199 ** very slow operation, so we work hard to avoid it. But sometimes |
|
3200 ** it can't be helped. |
|
3201 */ |
|
3202 if( pPg==0 && pPager->lru.pFirst ){ |
|
3203 if( !pPager->errCode ){ |
|
3204 int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); |
|
3205 int rc = syncJournal(pPager); |
|
3206 if( rc!=0 ){ |
|
3207 return rc; |
|
3208 } |
|
3209 if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ |
|
3210 /* If in full-sync mode, write a new journal header into the |
|
3211 ** journal file. This is done to avoid ever modifying a journal |
|
3212 ** header that is involved in the rollback of pages that have |
|
3213 ** already been written to the database (in case the header is |
|
3214 ** trashed when the nRec field is updated). |
|
3215 */ |
|
3216 pPager->nRec = 0; |
|
3217 assert( pPager->journalOff > 0 ); |
|
3218 assert( pPager->doNotSync==0 ); |
|
3219 rc = writeJournalHdr(pPager); |
|
3220 if( rc!=0 ){ |
|
3221 return rc; |
|
3222 } |
|
3223 } |
|
3224 } |
|
3225 pPg = pPager->lru.pFirst; |
|
3226 } |
|
3227 |
|
3228 assert( pPg->nRef==0 ); |
|
3229 |
|
3230 /* Write the page to the database file if it is dirty. |
|
3231 */ |
|
3232 if( pPg->dirty && !pPager->errCode ){ |
|
3233 int rc; |
|
3234 assert( pPg->needSync==0 ); |
|
3235 makeClean(pPg); |
|
3236 pPg->dirty = 1; |
|
3237 pPg->pDirty = 0; |
|
3238 rc = pager_write_pagelist( pPg ); |
|
3239 pPg->dirty = 0; |
|
3240 if( rc!=SQLITE_OK ){ |
|
3241 return rc; |
|
3242 } |
|
3243 } |
|
3244 assert( pPg->dirty==0 || pPager->errCode ); |
|
3245 |
|
3246 /* If the page we are recycling is marked as alwaysRollback, then |
|
3247 ** set the global alwaysRollback flag, thus disabling the |
|
3248 ** sqlite3PagerDontRollback() optimization for the rest of this transaction. |
|
3249 ** It is necessary to do this because the page marked alwaysRollback |
|
3250 ** might be reloaded at a later time but at that point we won't remember |
|
3251 ** that is was marked alwaysRollback. This means that all pages must |
|
3252 ** be marked as alwaysRollback from here on out. |
|
3253 */ |
|
3254 if( pPg->alwaysRollback ){ |
|
3255 IOTRACE(("ALWAYS_ROLLBACK %p\n", pPager)) |
|
3256 pPager->alwaysRollback = 1; |
|
3257 } |
|
3258 |
|
3259 /* Unlink the old page from the free list and the hash table |
|
3260 */ |
|
3261 unlinkPage(pPg); |
|
3262 assert( pPg->pgno==0 ); |
|
3263 |
|
3264 *ppPg = pPg; |
|
3265 return SQLITE_OK; |
|
3266 } |
|
3267 |
|
3268 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
3269 /* |
|
3270 ** This function is called to free superfluous dynamically allocated memory |
|
3271 ** held by the pager system. Memory in use by any SQLite pager allocated |
|
3272 ** by the current thread may be sqlite3_free()ed. |
|
3273 ** |
|
3274 ** nReq is the number of bytes of memory required. Once this much has |
|
3275 ** been released, the function returns. The return value is the total number |
|
3276 ** of bytes of memory released. |
|
3277 */ |
|
3278 int sqlite3PagerReleaseMemory(int nReq){ |
|
3279 int nReleased = 0; /* Bytes of memory released so far */ |
|
3280 Pager *pPager; /* For looping over pagers */ |
|
3281 BusyHandler *savedBusy; /* Saved copy of the busy handler */ |
|
3282 int rc = SQLITE_OK; |
|
3283 |
|
3284 /* Acquire the memory-management mutex |
|
3285 */ |
|
3286 #ifndef SQLITE_MUTEX_NOOP |
|
3287 sqlite3_mutex *mutex; /* The MEM2 mutex */ |
|
3288 mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MEM2); |
|
3289 #endif |
|
3290 sqlite3_mutex_enter(mutex); |
|
3291 |
|
3292 /* Signal all database connections that memory management wants |
|
3293 ** to have access to the pagers. |
|
3294 */ |
|
3295 for(pPager=sqlite3PagerList; pPager; pPager=pPager->pNext){ |
|
3296 pPager->iInUseMM = 1; |
|
3297 } |
|
3298 |
|
3299 while( rc==SQLITE_OK && (nReq<0 || nReleased<nReq) ){ |
|
3300 PgHdr *pPg; |
|
3301 PgHdr *pRecycled; |
|
3302 |
|
3303 /* Try to find a page to recycle that does not require a sync(). If |
|
3304 ** this is not possible, find one that does require a sync(). |
|
3305 */ |
|
3306 sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
3307 pPg = sqlite3LruPageList.pFirstSynced; |
|
3308 while( pPg && (pPg->needSync || pPg->pPager->iInUseDB) ){ |
|
3309 pPg = pPg->gfree.pNext; |
|
3310 } |
|
3311 if( !pPg ){ |
|
3312 pPg = sqlite3LruPageList.pFirst; |
|
3313 while( pPg && pPg->pPager->iInUseDB ){ |
|
3314 pPg = pPg->gfree.pNext; |
|
3315 } |
|
3316 } |
|
3317 sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU)); |
|
3318 |
|
3319 /* If pPg==0, then the block above has failed to find a page to |
|
3320 ** recycle. In this case return early - no further memory will |
|
3321 ** be released. |
|
3322 */ |
|
3323 if( !pPg ) break; |
|
3324 |
|
3325 pPager = pPg->pPager; |
|
3326 assert(!pPg->needSync || pPg==pPager->lru.pFirst); |
|
3327 assert(pPg->needSync || pPg==pPager->lru.pFirstSynced); |
|
3328 |
|
3329 savedBusy = pPager->pBusyHandler; |
|
3330 pPager->pBusyHandler = 0; |
|
3331 rc = pager_recycle(pPager, &pRecycled); |
|
3332 pPager->pBusyHandler = savedBusy; |
|
3333 assert(pRecycled==pPg || rc!=SQLITE_OK); |
|
3334 if( rc==SQLITE_OK ){ |
|
3335 /* We've found a page to free. At this point the page has been |
|
3336 ** removed from the page hash-table, free-list and synced-list |
|
3337 ** (pFirstSynced). It is still in the all pages (pAll) list. |
|
3338 ** Remove it from this list before freeing. |
|
3339 ** |
|
3340 ** Todo: Check the Pager.pStmt list to make sure this is Ok. It |
|
3341 ** probably is though. |
|
3342 */ |
|
3343 PgHdr *pTmp; |
|
3344 assert( pPg ); |
|
3345 if( pPg==pPager->pAll ){ |
|
3346 assert(pPg->pPrevAll==0); |
|
3347 assert(pPg->pNextAll==0 || pPg->pNextAll->pPrevAll==pPg); |
|
3348 pPager->pAll = pPg->pNextAll; |
|
3349 if( pPager->pAll ){ |
|
3350 pPager->pAll->pPrevAll = 0; |
|
3351 } |
|
3352 }else{ |
|
3353 assert(pPg->pPrevAll); |
|
3354 assert(pPg->pPrevAll->pNextAll==pPg); |
|
3355 pTmp = pPg->pPrevAll; |
|
3356 pTmp->pNextAll = pPg->pNextAll; |
|
3357 if( pTmp->pNextAll ){ |
|
3358 pTmp->pNextAll->pPrevAll = pTmp; |
|
3359 } |
|
3360 } |
|
3361 nReleased += ( |
|
3362 sizeof(*pPg) + pPager->pageSize |
|
3363 + sizeof(u32) + pPager->nExtra |
|
3364 + MEMDB*sizeof(PgHistory) |
|
3365 ); |
|
3366 IOTRACE(("PGFREE %p %d *\n", pPager, pPg->pgno)); |
|
3367 PAGER_INCR(sqlite3_pager_pgfree_count); |
|
3368 sqlite3PageFree(pPg->pData); |
|
3369 sqlite3_free(pPg); |
|
3370 pPager->nPage--; |
|
3371 }else{ |
|
3372 /* An error occured whilst writing to the database file or |
|
3373 ** journal in pager_recycle(). The error is not returned to the |
|
3374 ** caller of this function. Instead, set the Pager.errCode variable. |
|
3375 ** The error will be returned to the user (or users, in the case |
|
3376 ** of a shared pager cache) of the pager for which the error occured. |
|
3377 */ |
|
3378 assert( |
|
3379 (rc&0xff)==SQLITE_IOERR || |
|
3380 rc==SQLITE_FULL || |
|
3381 rc==SQLITE_BUSY |
|
3382 ); |
|
3383 assert( pPager->state>=PAGER_RESERVED ); |
|
3384 pager_error(pPager, rc); |
|
3385 } |
|
3386 } |
|
3387 |
|
3388 /* Clear the memory management flags and release the mutex |
|
3389 */ |
|
3390 for(pPager=sqlite3PagerList; pPager; pPager=pPager->pNext){ |
|
3391 pPager->iInUseMM = 0; |
|
3392 } |
|
3393 sqlite3_mutex_leave(mutex); |
|
3394 |
|
3395 /* Return the number of bytes released |
|
3396 */ |
|
3397 return nReleased; |
|
3398 } |
|
3399 #endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT */ |
|
3400 |
|
3401 /* |
|
3402 ** Read the content of page pPg out of the database file. |
|
3403 */ |
|
3404 static int readDbPage(Pager *pPager, PgHdr *pPg, Pgno pgno){ |
|
3405 int rc; |
|
3406 i64 offset; |
|
3407 assert( MEMDB==0 ); |
|
3408 assert(pPager->fd->pMethods||pPager->tempFile); |
|
3409 if( !pPager->fd->pMethods ){ |
|
3410 return SQLITE_IOERR_SHORT_READ; |
|
3411 } |
|
3412 offset = (pgno-1)*(i64)pPager->pageSize; |
|
3413 rc = sqlite3OsRead(pPager->fd, PGHDR_TO_DATA(pPg), pPager->pageSize, offset); |
|
3414 PAGER_INCR(sqlite3_pager_readdb_count); |
|
3415 PAGER_INCR(pPager->nRead); |
|
3416 IOTRACE(("PGIN %p %d\n", pPager, pgno)); |
|
3417 if( pgno==1 ){ |
|
3418 memcpy(&pPager->dbFileVers, &((u8*)PGHDR_TO_DATA(pPg))[24], |
|
3419 sizeof(pPager->dbFileVers)); |
|
3420 } |
|
3421 CODEC1(pPager, PGHDR_TO_DATA(pPg), pPg->pgno, 3); |
|
3422 PAGERTRACE4("FETCH %d page %d hash(%08x)\n", |
|
3423 PAGERID(pPager), pPg->pgno, pager_pagehash(pPg)); |
|
3424 return rc; |
|
3425 } |
|
3426 |
|
3427 |
|
3428 /* |
|
3429 ** This function is called to obtain the shared lock required before |
|
3430 ** data may be read from the pager cache. If the shared lock has already |
|
3431 ** been obtained, this function is a no-op. |
|
3432 ** |
|
3433 ** Immediately after obtaining the shared lock (if required), this function |
|
3434 ** checks for a hot-journal file. If one is found, an emergency rollback |
|
3435 ** is performed immediately. |
|
3436 */ |
|
3437 static int pagerSharedLock(Pager *pPager){ |
|
3438 int rc = SQLITE_OK; |
|
3439 int isErrorReset = 0; |
|
3440 |
|
3441 /* If this database is opened for exclusive access, has no outstanding |
|
3442 ** page references and is in an error-state, now is the chance to clear |
|
3443 ** the error. Discard the contents of the pager-cache and treat any |
|
3444 ** open journal file as a hot-journal. |
|
3445 */ |
|
3446 if( !MEMDB && pPager->exclusiveMode && pPager->nRef==0 && pPager->errCode ){ |
|
3447 if( pPager->journalOpen ){ |
|
3448 isErrorReset = 1; |
|
3449 } |
|
3450 pPager->errCode = SQLITE_OK; |
|
3451 pager_reset(pPager); |
|
3452 } |
|
3453 |
|
3454 /* If the pager is still in an error state, do not proceed. The error |
|
3455 ** state will be cleared at some point in the future when all page |
|
3456 ** references are dropped and the cache can be discarded. |
|
3457 */ |
|
3458 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){ |
|
3459 return pPager->errCode; |
|
3460 } |
|
3461 |
|
3462 if( pPager->state==PAGER_UNLOCK || isErrorReset ){ |
|
3463 sqlite3_vfs *pVfs = pPager->pVfs; |
|
3464 if( !MEMDB ){ |
|
3465 int isHotJournal; |
|
3466 assert( pPager->nRef==0 ); |
|
3467 if( !pPager->noReadlock ){ |
|
3468 rc = pager_wait_on_lock(pPager, SHARED_LOCK); |
|
3469 if( rc!=SQLITE_OK ){ |
|
3470 assert( pPager->state==PAGER_UNLOCK ); |
|
3471 return pager_error(pPager, rc); |
|
3472 } |
|
3473 assert( pPager->state>=SHARED_LOCK ); |
|
3474 } |
|
3475 |
|
3476 /* If a journal file exists, and there is no RESERVED lock on the |
|
3477 ** database file, then it either needs to be played back or deleted. |
|
3478 */ |
|
3479 if( !isErrorReset ){ |
|
3480 rc = hasHotJournal(pPager, &isHotJournal); |
|
3481 if( rc!=SQLITE_OK ){ |
|
3482 goto failed; |
|
3483 } |
|
3484 } |
|
3485 if( isErrorReset || isHotJournal ){ |
|
3486 /* Get an EXCLUSIVE lock on the database file. At this point it is |
|
3487 ** important that a RESERVED lock is not obtained on the way to the |
|
3488 ** EXCLUSIVE lock. If it were, another process might open the |
|
3489 ** database file, detect the RESERVED lock, and conclude that the |
|
3490 ** database is safe to read while this process is still rolling it |
|
3491 ** back. |
|
3492 ** |
|
3493 ** Because the intermediate RESERVED lock is not requested, the |
|
3494 ** second process will get to this point in the code and fail to |
|
3495 ** obtain its own EXCLUSIVE lock on the database file. |
|
3496 */ |
|
3497 if( pPager->state<EXCLUSIVE_LOCK ){ |
|
3498 rc = sqlite3OsLock(pPager->fd, EXCLUSIVE_LOCK); |
|
3499 if( rc!=SQLITE_OK ){ |
|
3500 rc = pager_error(pPager, rc); |
|
3501 goto failed; |
|
3502 } |
|
3503 pPager->state = PAGER_EXCLUSIVE; |
|
3504 } |
|
3505 |
|
3506 /* Open the journal for read/write access. This is because in |
|
3507 ** exclusive-access mode the file descriptor will be kept open and |
|
3508 ** possibly used for a transaction later on. On some systems, the |
|
3509 ** OsTruncate() call used in exclusive-access mode also requires |
|
3510 ** a read/write file handle. |
|
3511 */ |
|
3512 if( !isErrorReset && pPager->journalOpen==0 ){ |
|
3513 int res; |
|
3514 rc = sqlite3OsAccess(pVfs,pPager->zJournal,SQLITE_ACCESS_EXISTS,&res); |
|
3515 if( rc==SQLITE_OK ){ |
|
3516 if( res ){ |
|
3517 int fout = 0; |
|
3518 int f = SQLITE_OPEN_READWRITE|SQLITE_OPEN_MAIN_JOURNAL; |
|
3519 assert( !pPager->tempFile ); |
|
3520 rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &fout); |
|
3521 assert( rc!=SQLITE_OK || pPager->jfd->pMethods ); |
|
3522 if( fout&SQLITE_OPEN_READONLY ){ |
|
3523 rc = SQLITE_BUSY; |
|
3524 sqlite3OsClose(pPager->jfd); |
|
3525 } |
|
3526 }else{ |
|
3527 /* If the journal does not exist, that means some other process |
|
3528 ** has already rolled it back */ |
|
3529 rc = SQLITE_BUSY; |
|
3530 } |
|
3531 } |
|
3532 } |
|
3533 if( rc!=SQLITE_OK ){ |
|
3534 if( rc!=SQLITE_NOMEM && rc!=SQLITE_IOERR_UNLOCK |
|
3535 && rc!=SQLITE_IOERR_NOMEM |
|
3536 ){ |
|
3537 rc = SQLITE_BUSY; |
|
3538 } |
|
3539 goto failed; |
|
3540 } |
|
3541 pPager->journalOpen = 1; |
|
3542 pPager->journalStarted = 0; |
|
3543 pPager->journalOff = 0; |
|
3544 pPager->setMaster = 0; |
|
3545 pPager->journalHdr = 0; |
|
3546 |
|
3547 /* Playback and delete the journal. Drop the database write |
|
3548 ** lock and reacquire the read lock. |
|
3549 */ |
|
3550 rc = pager_playback(pPager, 1); |
|
3551 if( rc!=SQLITE_OK ){ |
|
3552 rc = pager_error(pPager, rc); |
|
3553 goto failed; |
|
3554 } |
|
3555 assert(pPager->state==PAGER_SHARED || |
|
3556 (pPager->exclusiveMode && pPager->state>PAGER_SHARED) |
|
3557 ); |
|
3558 } |
|
3559 |
|
3560 if( pPager->pAll ){ |
|
3561 /* The shared-lock has just been acquired on the database file |
|
3562 ** and there are already pages in the cache (from a previous |
|
3563 ** read or write transaction). Check to see if the database |
|
3564 ** has been modified. If the database has changed, flush the |
|
3565 ** cache. |
|
3566 ** |
|
3567 ** Database changes is detected by looking at 15 bytes beginning |
|
3568 ** at offset 24 into the file. The first 4 of these 16 bytes are |
|
3569 ** a 32-bit counter that is incremented with each change. The |
|
3570 ** other bytes change randomly with each file change when |
|
3571 ** a codec is in use. |
|
3572 ** |
|
3573 ** There is a vanishingly small chance that a change will not be |
|
3574 ** detected. The chance of an undetected change is so small that |
|
3575 ** it can be neglected. |
|
3576 */ |
|
3577 char dbFileVers[sizeof(pPager->dbFileVers)]; |
|
3578 sqlite3PagerPagecount(pPager, 0); |
|
3579 |
|
3580 if( pPager->errCode ){ |
|
3581 rc = pPager->errCode; |
|
3582 goto failed; |
|
3583 } |
|
3584 |
|
3585 if( pPager->dbSize>0 ){ |
|
3586 IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers))); |
|
3587 rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers), 24); |
|
3588 if( rc!=SQLITE_OK ){ |
|
3589 goto failed; |
|
3590 } |
|
3591 }else{ |
|
3592 memset(dbFileVers, 0, sizeof(dbFileVers)); |
|
3593 } |
|
3594 |
|
3595 if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){ |
|
3596 pager_reset(pPager); |
|
3597 } |
|
3598 } |
|
3599 } |
|
3600 assert( pPager->exclusiveMode || pPager->state<=PAGER_SHARED ); |
|
3601 if( pPager->state==PAGER_UNLOCK ){ |
|
3602 pPager->state = PAGER_SHARED; |
|
3603 } |
|
3604 } |
|
3605 |
|
3606 failed: |
|
3607 if( rc!=SQLITE_OK ){ |
|
3608 /* pager_unlock() is a no-op for exclusive mode and in-memory databases. */ |
|
3609 pager_unlock(pPager); |
|
3610 } |
|
3611 return rc; |
|
3612 } |
|
3613 |
|
3614 /* |
|
3615 ** Allocate a PgHdr object. Either create a new one or reuse |
|
3616 ** an existing one that is not otherwise in use. |
|
3617 ** |
|
3618 ** A new PgHdr structure is created if any of the following are |
|
3619 ** true: |
|
3620 ** |
|
3621 ** (1) We have not exceeded our maximum allocated cache size |
|
3622 ** as set by the "PRAGMA cache_size" command. |
|
3623 ** |
|
3624 ** (2) There are no unused PgHdr objects available at this time. |
|
3625 ** |
|
3626 ** (3) This is an in-memory database. |
|
3627 ** |
|
3628 ** (4) There are no PgHdr objects that do not require a journal |
|
3629 ** file sync and a sync of the journal file is currently |
|
3630 ** prohibited. |
|
3631 ** |
|
3632 ** Otherwise, reuse an existing PgHdr. In other words, reuse an |
|
3633 ** existing PgHdr if all of the following are true: |
|
3634 ** |
|
3635 ** (1) We have reached or exceeded the maximum cache size |
|
3636 ** allowed by "PRAGMA cache_size". |
|
3637 ** |
|
3638 ** (2) There is a PgHdr available with PgHdr->nRef==0 |
|
3639 ** |
|
3640 ** (3) We are not in an in-memory database |
|
3641 ** |
|
3642 ** (4) Either there is an available PgHdr that does not need |
|
3643 ** to be synced to disk or else disk syncing is currently |
|
3644 ** allowed. |
|
3645 */ |
|
3646 static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){ |
|
3647 int rc = SQLITE_OK; |
|
3648 PgHdr *pPg; |
|
3649 int nByteHdr; |
|
3650 |
|
3651 /* Create a new PgHdr if any of the four conditions defined |
|
3652 ** above are met: */ |
|
3653 if( pPager->nPage<pPager->mxPage |
|
3654 || pPager->lru.pFirst==0 |
|
3655 || MEMDB |
|
3656 || (pPager->lru.pFirstSynced==0 && pPager->doNotSync) |
|
3657 ){ |
|
3658 void *pData; |
|
3659 if( pPager->nPage>=pPager->nHash ){ |
|
3660 pager_resize_hash_table(pPager, |
|
3661 pPager->nHash<256 ? 256 : pPager->nHash*2); |
|
3662 if( pPager->nHash==0 ){ |
|
3663 rc = SQLITE_NOMEM; |
|
3664 goto pager_allocate_out; |
|
3665 } |
|
3666 } |
|
3667 pagerLeave(pPager); |
|
3668 nByteHdr = sizeof(*pPg) + sizeof(u32) + pPager->nExtra |
|
3669 + MEMDB*sizeof(PgHistory); |
|
3670 pPg = sqlite3Malloc( nByteHdr ); |
|
3671 if( pPg ){ |
|
3672 pData = sqlite3PageMalloc( pPager->pageSize ); |
|
3673 if( pData==0 ){ |
|
3674 sqlite3_free(pPg); |
|
3675 pPg = 0; |
|
3676 } |
|
3677 } |
|
3678 pagerEnter(pPager); |
|
3679 if( pPg==0 ){ |
|
3680 rc = SQLITE_NOMEM; |
|
3681 goto pager_allocate_out; |
|
3682 } |
|
3683 memset(pPg, 0, nByteHdr); |
|
3684 pPg->pData = pData; |
|
3685 pPg->pPager = pPager; |
|
3686 pPg->pNextAll = pPager->pAll; |
|
3687 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT |
|
3688 if( pPg->pNextAll ){ |
|
3689 pPg->pNextAll->pPrevAll = pPg; |
|
3690 } |
|
3691 #endif |
|
3692 pPager->pAll = pPg; |
|
3693 pPager->nPage++; |
|
3694 }else{ |
|
3695 /* Recycle an existing page with a zero ref-count. */ |
|
3696 rc = pager_recycle(pPager, &pPg); |
|
3697 if( rc==SQLITE_BUSY ){ |
|
3698 rc = SQLITE_IOERR_BLOCKED; |
|
3699 } |
|
3700 if( rc!=SQLITE_OK ){ |
|
3701 goto pager_allocate_out; |
|
3702 } |
|
3703 assert( pPager->state>=SHARED_LOCK ); |
|
3704 assert(pPg); |
|
3705 } |
|
3706 *ppPg = pPg; |
|
3707 |
|
3708 pager_allocate_out: |
|
3709 return rc; |
|
3710 } |
|
3711 |
|
3712 /* |
|
3713 ** Make sure we have the content for a page. If the page was |
|
3714 ** previously acquired with noContent==1, then the content was |
|
3715 ** just initialized to zeros instead of being read from disk. |
|
3716 ** But now we need the real data off of disk. So make sure we |
|
3717 ** have it. Read it in if we do not have it already. |
|
3718 */ |
|
3719 static int pager_get_content(PgHdr *pPg){ |
|
3720 if( pPg->needRead ){ |
|
3721 int rc = readDbPage(pPg->pPager, pPg, pPg->pgno); |
|
3722 if( rc==SQLITE_OK ){ |
|
3723 pPg->needRead = 0; |
|
3724 }else{ |
|
3725 return rc; |
|
3726 } |
|
3727 } |
|
3728 return SQLITE_OK; |
|
3729 } |
|
3730 |
|
3731 /* |
|
3732 ** Acquire a page. |
|
3733 ** |
|
3734 ** A read lock on the disk file is obtained when the first page is acquired. |
|
3735 ** This read lock is dropped when the last page is released. |
|
3736 ** |
|
3737 ** This routine works for any page number greater than 0. If the database |
|
3738 ** file is smaller than the requested page, then no actual disk |
|
3739 ** read occurs and the memory image of the page is initialized to |
|
3740 ** all zeros. The extra data appended to a page is always initialized |
|
3741 ** to zeros the first time a page is loaded into memory. |
|
3742 ** |
|
3743 ** The acquisition might fail for several reasons. In all cases, |
|
3744 ** an appropriate error code is returned and *ppPage is set to NULL. |
|
3745 ** |
|
3746 ** See also sqlite3PagerLookup(). Both this routine and Lookup() attempt |
|
3747 ** to find a page in the in-memory cache first. If the page is not already |
|
3748 ** in memory, this routine goes to disk to read it in whereas Lookup() |
|
3749 ** just returns 0. This routine acquires a read-lock the first time it |
|
3750 ** has to go to disk, and could also playback an old journal if necessary. |
|
3751 ** Since Lookup() never goes to disk, it never has to deal with locks |
|
3752 ** or journal files. |
|
3753 ** |
|
3754 ** If noContent is false, the page contents are actually read from disk. |
|
3755 ** If noContent is true, it means that we do not care about the contents |
|
3756 ** of the page at this time, so do not do a disk read. Just fill in the |
|
3757 ** page content with zeros. But mark the fact that we have not read the |
|
3758 ** content by setting the PgHdr.needRead flag. Later on, if |
|
3759 ** sqlite3PagerWrite() is called on this page or if this routine is |
|
3760 ** called again with noContent==0, that means that the content is needed |
|
3761 ** and the disk read should occur at that point. |
|
3762 */ |
|
3763 static int pagerAcquire( |
|
3764 Pager *pPager, /* The pager open on the database file */ |
|
3765 Pgno pgno, /* Page number to fetch */ |
|
3766 DbPage **ppPage, /* Write a pointer to the page here */ |
|
3767 int noContent /* Do not bother reading content from disk if true */ |
|
3768 ){ |
|
3769 PgHdr *pPg; |
|
3770 int rc; |
|
3771 |
|
3772 assert( pPager->state==PAGER_UNLOCK || pPager->nRef>0 || pgno==1 ); |
|
3773 |
|
3774 /* The maximum page number is 2^31. Return SQLITE_CORRUPT if a page |
|
3775 ** number greater than this, or zero, is requested. |
|
3776 */ |
|
3777 if( pgno>PAGER_MAX_PGNO || pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){ |
|
3778 return SQLITE_CORRUPT_BKPT; |
|
3779 } |
|
3780 |
|
3781 /* Make sure we have not hit any critical errors. |
|
3782 */ |
|
3783 assert( pPager!=0 ); |
|
3784 *ppPage = 0; |
|
3785 |
|
3786 /* If this is the first page accessed, then get a SHARED lock |
|
3787 ** on the database file. pagerSharedLock() is a no-op if |
|
3788 ** a database lock is already held. |
|
3789 */ |
|
3790 rc = pagerSharedLock(pPager); |
|
3791 if( rc!=SQLITE_OK ){ |
|
3792 return rc; |
|
3793 } |
|
3794 assert( pPager->state!=PAGER_UNLOCK ); |
|
3795 |
|
3796 pPg = pager_lookup(pPager, pgno); |
|
3797 if( pPg==0 ){ |
|
3798 /* The requested page is not in the page cache. */ |
|
3799 int nMax; |
|
3800 int h; |
|
3801 PAGER_INCR(pPager->nMiss); |
|
3802 rc = pagerAllocatePage(pPager, &pPg); |
|
3803 if( rc!=SQLITE_OK ){ |
|
3804 return rc; |
|
3805 } |
|
3806 |
|
3807 pPg->pgno = pgno; |
|
3808 assert( !MEMDB || pgno>pPager->stmtSize ); |
|
3809 pPg->inJournal = sqlite3BitvecTest(pPager->pInJournal, pgno); |
|
3810 pPg->needSync = 0; |
|
3811 |
|
3812 makeClean(pPg); |
|
3813 pPg->nRef = 1; |
|
3814 |
|
3815 pPager->nRef++; |
|
3816 if( pPager->nExtra>0 ){ |
|
3817 memset(PGHDR_TO_EXTRA(pPg, pPager), 0, pPager->nExtra); |
|
3818 } |
|
3819 rc = sqlite3PagerPagecount(pPager, &nMax); |
|
3820 if( rc!=SQLITE_OK ){ |
|
3821 sqlite3PagerUnref(pPg); |
|
3822 return rc; |
|
3823 } |
|
3824 |
|
3825 /* Populate the page with data, either by reading from the database |
|
3826 ** file, or by setting the entire page to zero. |
|
3827 */ |
|
3828 if( nMax<(int)pgno || MEMDB || (noContent && !pPager->alwaysRollback) ){ |
|
3829 if( pgno>pPager->mxPgno ){ |
|
3830 sqlite3PagerUnref(pPg); |
|
3831 return SQLITE_FULL; |
|
3832 } |
|
3833 memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize); |
|
3834 pPg->needRead = noContent && !pPager->alwaysRollback; |
|
3835 IOTRACE(("ZERO %p %d\n", pPager, pgno)); |
|
3836 }else{ |
|
3837 rc = readDbPage(pPager, pPg, pgno); |
|
3838 if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){ |
|
3839 pPg->pgno = 0; |
|
3840 sqlite3PagerUnref(pPg); |
|
3841 return rc; |
|
3842 } |
|
3843 pPg->needRead = 0; |
|
3844 } |
|
3845 |
|
3846 /* Link the page into the page hash table */ |
|
3847 h = pgno & (pPager->nHash-1); |
|
3848 assert( pgno!=0 ); |
|
3849 pPg->pNextHash = pPager->aHash[h]; |
|
3850 pPager->aHash[h] = pPg; |
|
3851 if( pPg->pNextHash ){ |
|
3852 assert( pPg->pNextHash->pPrevHash==0 ); |
|
3853 pPg->pNextHash->pPrevHash = pPg; |
|
3854 } |
|
3855 |
|
3856 #ifdef SQLITE_CHECK_PAGES |
|
3857 pPg->pageHash = pager_pagehash(pPg); |
|
3858 #endif |
|
3859 }else{ |
|
3860 /* The requested page is in the page cache. */ |
|
3861 assert(pPager->nRef>0 || pgno==1); |
|
3862 PAGER_INCR(pPager->nHit); |
|
3863 if( !noContent ){ |
|
3864 rc = pager_get_content(pPg); |
|
3865 if( rc ){ |
|
3866 return rc; |
|
3867 } |
|
3868 } |
|
3869 page_ref(pPg); |
|
3870 } |
|
3871 *ppPage = pPg; |
|
3872 return SQLITE_OK; |
|
3873 } |
|
3874 int sqlite3PagerAcquire( |
|
3875 Pager *pPager, /* The pager open on the database file */ |
|
3876 Pgno pgno, /* Page number to fetch */ |
|
3877 DbPage **ppPage, /* Write a pointer to the page here */ |
|
3878 int noContent /* Do not bother reading content from disk if true */ |
|
3879 ){ |
|
3880 int rc; |
|
3881 pagerEnter(pPager); |
|
3882 rc = pagerAcquire(pPager, pgno, ppPage, noContent); |
|
3883 pagerLeave(pPager); |
|
3884 return rc; |
|
3885 } |
|
3886 |
|
3887 |
|
3888 /* |
|
3889 ** Acquire a page if it is already in the in-memory cache. Do |
|
3890 ** not read the page from disk. Return a pointer to the page, |
|
3891 ** or 0 if the page is not in cache. |
|
3892 ** |
|
3893 ** See also sqlite3PagerGet(). The difference between this routine |
|
3894 ** and sqlite3PagerGet() is that _get() will go to the disk and read |
|
3895 ** in the page if the page is not already in cache. This routine |
|
3896 ** returns NULL if the page is not in cache or if a disk I/O error |
|
3897 ** has ever happened. |
|
3898 */ |
|
3899 DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){ |
|
3900 PgHdr *pPg = 0; |
|
3901 |
|
3902 assert( pPager!=0 ); |
|
3903 assert( pgno!=0 ); |
|
3904 |
|
3905 pagerEnter(pPager); |
|
3906 if( pPager->state==PAGER_UNLOCK ){ |
|
3907 assert( !pPager->pAll || pPager->exclusiveMode ); |
|
3908 }else if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){ |
|
3909 /* Do nothing */ |
|
3910 }else if( (pPg = pager_lookup(pPager, pgno))!=0 ){ |
|
3911 page_ref(pPg); |
|
3912 } |
|
3913 pagerLeave(pPager); |
|
3914 return pPg; |
|
3915 } |
|
3916 |
|
3917 /* |
|
3918 ** Release a page. |
|
3919 ** |
|
3920 ** If the number of references to the page drop to zero, then the |
|
3921 ** page is added to the LRU list. When all references to all pages |
|
3922 ** are released, a rollback occurs and the lock on the database is |
|
3923 ** removed. |
|
3924 */ |
|
3925 int sqlite3PagerUnref(DbPage *pPg){ |
|
3926 Pager *pPager; |
|
3927 |
|
3928 if( pPg==0 ) return SQLITE_OK; |
|
3929 pPager = pPg->pPager; |
|
3930 |
|
3931 /* Decrement the reference count for this page |
|
3932 */ |
|
3933 assert( pPg->nRef>0 ); |
|
3934 pagerEnter(pPg->pPager); |
|
3935 pPg->nRef--; |
|
3936 |
|
3937 CHECK_PAGE(pPg); |
|
3938 |
|
3939 /* When the number of references to a page reach 0, call the |
|
3940 ** destructor and add the page to the freelist. |
|
3941 */ |
|
3942 if( pPg->nRef==0 ){ |
|
3943 |
|
3944 lruListAdd(pPg); |
|
3945 if( pPager->xDestructor ){ |
|
3946 pPager->xDestructor(pPg, pPager->pageSize); |
|
3947 } |
|
3948 |
|
3949 /* When all pages reach the freelist, drop the read lock from |
|
3950 ** the database file. |
|
3951 */ |
|
3952 pPager->nRef--; |
|
3953 assert( pPager->nRef>=0 ); |
|
3954 if( pPager->nRef==0 && (!pPager->exclusiveMode || pPager->journalOff>0) ){ |
|
3955 pagerUnlockAndRollback(pPager); |
|
3956 } |
|
3957 } |
|
3958 pagerLeave(pPager); |
|
3959 return SQLITE_OK; |
|
3960 } |
|
3961 |
|
3962 /* |
|
3963 ** Create a journal file for pPager. There should already be a RESERVED |
|
3964 ** or EXCLUSIVE lock on the database file when this routine is called. |
|
3965 ** |
|
3966 ** Return SQLITE_OK if everything. Return an error code and release the |
|
3967 ** write lock if anything goes wrong. |
|
3968 */ |
|
3969 static int pager_open_journal(Pager *pPager){ |
|
3970 sqlite3_vfs *pVfs = pPager->pVfs; |
|
3971 int flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_EXCLUSIVE|SQLITE_OPEN_CREATE); |
|
3972 |
|
3973 int rc; |
|
3974 assert( !MEMDB ); |
|
3975 assert( pPager->state>=PAGER_RESERVED ); |
|
3976 assert( pPager->useJournal ); |
|
3977 assert( pPager->pInJournal==0 ); |
|
3978 sqlite3PagerPagecount(pPager, 0); |
|
3979 pagerLeave(pPager); |
|
3980 pPager->pInJournal = sqlite3BitvecCreate(pPager->dbSize); |
|
3981 pagerEnter(pPager); |
|
3982 if( pPager->pInJournal==0 ){ |
|
3983 rc = SQLITE_NOMEM; |
|
3984 goto failed_to_open_journal; |
|
3985 } |
|
3986 |
|
3987 if( pPager->journalOpen==0 ){ |
|
3988 if( pPager->tempFile ){ |
|
3989 flags |= (SQLITE_OPEN_DELETEONCLOSE|SQLITE_OPEN_TEMP_JOURNAL); |
|
3990 }else{ |
|
3991 flags |= (SQLITE_OPEN_MAIN_JOURNAL); |
|
3992 } |
|
3993 #ifdef SQLITE_ENABLE_ATOMIC_WRITE |
|
3994 rc = sqlite3JournalOpen( |
|
3995 pVfs, pPager->zJournal, pPager->jfd, flags, jrnlBufferSize(pPager) |
|
3996 ); |
|
3997 #else |
|
3998 rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, flags, 0); |
|
3999 #endif |
|
4000 assert( rc!=SQLITE_OK || pPager->jfd->pMethods ); |
|
4001 pPager->journalOff = 0; |
|
4002 pPager->setMaster = 0; |
|
4003 pPager->journalHdr = 0; |
|
4004 if( rc!=SQLITE_OK ){ |
|
4005 if( rc==SQLITE_NOMEM ){ |
|
4006 sqlite3OsDelete(pVfs, pPager->zJournal, 0); |
|
4007 } |
|
4008 goto failed_to_open_journal; |
|
4009 } |
|
4010 } |
|
4011 pPager->journalOpen = 1; |
|
4012 pPager->journalStarted = 0; |
|
4013 pPager->needSync = 0; |
|
4014 pPager->alwaysRollback = 0; |
|
4015 pPager->nRec = 0; |
|
4016 if( pPager->errCode ){ |
|
4017 rc = pPager->errCode; |
|
4018 goto failed_to_open_journal; |
|
4019 } |
|
4020 pPager->origDbSize = pPager->dbSize; |
|
4021 |
|
4022 rc = writeJournalHdr(pPager); |
|
4023 |
|
4024 if( pPager->stmtAutoopen && rc==SQLITE_OK ){ |
|
4025 rc = sqlite3PagerStmtBegin(pPager); |
|
4026 } |
|
4027 if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM && rc!=SQLITE_IOERR_NOMEM ){ |
|
4028 rc = pager_end_transaction(pPager, 0); |
|
4029 if( rc==SQLITE_OK ){ |
|
4030 rc = SQLITE_FULL; |
|
4031 } |
|
4032 } |
|
4033 return rc; |
|
4034 |
|
4035 failed_to_open_journal: |
|
4036 sqlite3BitvecDestroy(pPager->pInJournal); |
|
4037 pPager->pInJournal = 0; |
|
4038 return rc; |
|
4039 } |
|
4040 |
|
4041 /* |
|
4042 ** Acquire a write-lock on the database. The lock is removed when |
|
4043 ** the any of the following happen: |
|
4044 ** |
|
4045 ** * sqlite3PagerCommitPhaseTwo() is called. |
|
4046 ** * sqlite3PagerRollback() is called. |
|
4047 ** * sqlite3PagerClose() is called. |
|
4048 ** * sqlite3PagerUnref() is called to on every outstanding page. |
|
4049 ** |
|
4050 ** The first parameter to this routine is a pointer to any open page of the |
|
4051 ** database file. Nothing changes about the page - it is used merely to |
|
4052 ** acquire a pointer to the Pager structure and as proof that there is |
|
4053 ** already a read-lock on the database. |
|
4054 ** |
|
4055 ** The second parameter indicates how much space in bytes to reserve for a |
|
4056 ** master journal file-name at the start of the journal when it is created. |
|
4057 ** |
|
4058 ** A journal file is opened if this is not a temporary file. For temporary |
|
4059 ** files, the opening of the journal file is deferred until there is an |
|
4060 ** actual need to write to the journal. |
|
4061 ** |
|
4062 ** If the database is already reserved for writing, this routine is a no-op. |
|
4063 ** |
|
4064 ** If exFlag is true, go ahead and get an EXCLUSIVE lock on the file |
|
4065 ** immediately instead of waiting until we try to flush the cache. The |
|
4066 ** exFlag is ignored if a transaction is already active. |
|
4067 */ |
|
4068 int sqlite3PagerBegin(DbPage *pPg, int exFlag){ |
|
4069 Pager *pPager = pPg->pPager; |
|
4070 int rc = SQLITE_OK; |
|
4071 pagerEnter(pPager); |
|
4072 assert( pPg->nRef>0 ); |
|
4073 assert( pPager->state!=PAGER_UNLOCK ); |
|
4074 if( pPager->state==PAGER_SHARED ){ |
|
4075 assert( pPager->pInJournal==0 ); |
|
4076 if( MEMDB ){ |
|
4077 pPager->state = PAGER_EXCLUSIVE; |
|
4078 pPager->origDbSize = pPager->dbSize; |
|
4079 }else{ |
|
4080 rc = sqlite3OsLock(pPager->fd, RESERVED_LOCK); |
|
4081 if( rc==SQLITE_OK ){ |
|
4082 pPager->state = PAGER_RESERVED; |
|
4083 if( exFlag ){ |
|
4084 rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); |
|
4085 } |
|
4086 } |
|
4087 if( rc!=SQLITE_OK ){ |
|
4088 pagerLeave(pPager); |
|
4089 return rc; |
|
4090 } |
|
4091 pPager->dirtyCache = 0; |
|
4092 PAGERTRACE2("TRANSACTION %d\n", PAGERID(pPager)); |
|
4093 if( pPager->useJournal && !pPager->tempFile |
|
4094 && pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ |
|
4095 rc = pager_open_journal(pPager); |
|
4096 } |
|
4097 } |
|
4098 }else if( pPager->journalOpen && pPager->journalOff==0 ){ |
|
4099 /* This happens when the pager was in exclusive-access mode the last |
|
4100 ** time a (read or write) transaction was successfully concluded |
|
4101 ** by this connection. Instead of deleting the journal file it was |
|
4102 ** kept open and either was truncated to 0 bytes or its header was |
|
4103 ** overwritten with zeros. |
|
4104 */ |
|
4105 assert( pPager->nRec==0 ); |
|
4106 assert( pPager->origDbSize==0 ); |
|
4107 assert( pPager->pInJournal==0 ); |
|
4108 sqlite3PagerPagecount(pPager, 0); |
|
4109 pagerLeave(pPager); |
|
4110 pPager->pInJournal = sqlite3BitvecCreate( pPager->dbSize ); |
|
4111 pagerEnter(pPager); |
|
4112 if( !pPager->pInJournal ){ |
|
4113 rc = SQLITE_NOMEM; |
|
4114 }else{ |
|
4115 pPager->origDbSize = pPager->dbSize; |
|
4116 rc = writeJournalHdr(pPager); |
|
4117 } |
|
4118 } |
|
4119 assert( !pPager->journalOpen || pPager->journalOff>0 || rc!=SQLITE_OK ); |
|
4120 pagerLeave(pPager); |
|
4121 return rc; |
|
4122 } |
|
4123 |
|
4124 /* |
|
4125 ** Make a page dirty. Set its dirty flag and add it to the dirty |
|
4126 ** page list. |
|
4127 */ |
|
4128 static void makeDirty(PgHdr *pPg){ |
|
4129 if( pPg->dirty==0 ){ |
|
4130 Pager *pPager = pPg->pPager; |
|
4131 pPg->dirty = 1; |
|
4132 pPg->pDirty = pPager->pDirty; |
|
4133 if( pPager->pDirty ){ |
|
4134 pPager->pDirty->pPrevDirty = pPg; |
|
4135 } |
|
4136 pPg->pPrevDirty = 0; |
|
4137 pPager->pDirty = pPg; |
|
4138 } |
|
4139 } |
|
4140 |
|
4141 /* |
|
4142 ** Make a page clean. Clear its dirty bit and remove it from the |
|
4143 ** dirty page list. |
|
4144 */ |
|
4145 static void makeClean(PgHdr *pPg){ |
|
4146 if( pPg->dirty ){ |
|
4147 pPg->dirty = 0; |
|
4148 if( pPg->pDirty ){ |
|
4149 assert( pPg->pDirty->pPrevDirty==pPg ); |
|
4150 pPg->pDirty->pPrevDirty = pPg->pPrevDirty; |
|
4151 } |
|
4152 if( pPg->pPrevDirty ){ |
|
4153 assert( pPg->pPrevDirty->pDirty==pPg ); |
|
4154 pPg->pPrevDirty->pDirty = pPg->pDirty; |
|
4155 }else{ |
|
4156 assert( pPg->pPager->pDirty==pPg ); |
|
4157 pPg->pPager->pDirty = pPg->pDirty; |
|
4158 } |
|
4159 } |
|
4160 } |
|
4161 |
|
4162 |
|
4163 /* |
|
4164 ** Mark a data page as writeable. The page is written into the journal |
|
4165 ** if it is not there already. This routine must be called before making |
|
4166 ** changes to a page. |
|
4167 ** |
|
4168 ** The first time this routine is called, the pager creates a new |
|
4169 ** journal and acquires a RESERVED lock on the database. If the RESERVED |
|
4170 ** lock could not be acquired, this routine returns SQLITE_BUSY. The |
|
4171 ** calling routine must check for that return value and be careful not to |
|
4172 ** change any page data until this routine returns SQLITE_OK. |
|
4173 ** |
|
4174 ** If the journal file could not be written because the disk is full, |
|
4175 ** then this routine returns SQLITE_FULL and does an immediate rollback. |
|
4176 ** All subsequent write attempts also return SQLITE_FULL until there |
|
4177 ** is a call to sqlite3PagerCommit() or sqlite3PagerRollback() to |
|
4178 ** reset. |
|
4179 */ |
|
4180 static int pager_write(PgHdr *pPg){ |
|
4181 void *pData = PGHDR_TO_DATA(pPg); |
|
4182 Pager *pPager = pPg->pPager; |
|
4183 int rc = SQLITE_OK; |
|
4184 |
|
4185 /* Check for errors |
|
4186 */ |
|
4187 if( pPager->errCode ){ |
|
4188 return pPager->errCode; |
|
4189 } |
|
4190 if( pPager->readOnly ){ |
|
4191 return SQLITE_PERM; |
|
4192 } |
|
4193 |
|
4194 assert( !pPager->setMaster ); |
|
4195 |
|
4196 CHECK_PAGE(pPg); |
|
4197 |
|
4198 /* If this page was previously acquired with noContent==1, that means |
|
4199 ** we didn't really read in the content of the page. This can happen |
|
4200 ** (for example) when the page is being moved to the freelist. But |
|
4201 ** now we are (perhaps) moving the page off of the freelist for |
|
4202 ** reuse and we need to know its original content so that content |
|
4203 ** can be stored in the rollback journal. So do the read at this |
|
4204 ** time. |
|
4205 */ |
|
4206 rc = pager_get_content(pPg); |
|
4207 if( rc ){ |
|
4208 return rc; |
|
4209 } |
|
4210 |
|
4211 /* Mark the page as dirty. If the page has already been written |
|
4212 ** to the journal then we can return right away. |
|
4213 */ |
|
4214 makeDirty(pPg); |
|
4215 if( pPg->inJournal && (pageInStatement(pPg) || pPager->stmtInUse==0) ){ |
|
4216 pPager->dirtyCache = 1; |
|
4217 pPager->dbModified = 1; |
|
4218 }else{ |
|
4219 |
|
4220 /* If we get this far, it means that the page needs to be |
|
4221 ** written to the transaction journal or the ckeckpoint journal |
|
4222 ** or both. |
|
4223 ** |
|
4224 ** First check to see that the transaction journal exists and |
|
4225 ** create it if it does not. |
|
4226 */ |
|
4227 assert( pPager->state!=PAGER_UNLOCK ); |
|
4228 rc = sqlite3PagerBegin(pPg, 0); |
|
4229 if( rc!=SQLITE_OK ){ |
|
4230 return rc; |
|
4231 } |
|
4232 assert( pPager->state>=PAGER_RESERVED ); |
|
4233 if( !pPager->journalOpen && pPager->useJournal |
|
4234 && pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ |
|
4235 rc = pager_open_journal(pPager); |
|
4236 if( rc!=SQLITE_OK ) return rc; |
|
4237 } |
|
4238 pPager->dirtyCache = 1; |
|
4239 pPager->dbModified = 1; |
|
4240 |
|
4241 /* The transaction journal now exists and we have a RESERVED or an |
|
4242 ** EXCLUSIVE lock on the main database file. Write the current page to |
|
4243 ** the transaction journal if it is not there already. |
|
4244 */ |
|
4245 if( !pPg->inJournal && (pPager->journalOpen || MEMDB) ){ |
|
4246 if( (int)pPg->pgno <= pPager->origDbSize ){ |
|
4247 if( MEMDB ){ |
|
4248 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
4249 PAGERTRACE3("JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno); |
|
4250 assert( pHist->pOrig==0 ); |
|
4251 pHist->pOrig = sqlite3PageMalloc( pPager->pageSize ); |
|
4252 if( !pHist->pOrig ){ |
|
4253 return SQLITE_NOMEM; |
|
4254 } |
|
4255 memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize); |
|
4256 }else{ |
|
4257 u32 cksum; |
|
4258 char *pData2; |
|
4259 |
|
4260 /* We should never write to the journal file the page that |
|
4261 ** contains the database locks. The following assert verifies |
|
4262 ** that we do not. */ |
|
4263 assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) ); |
|
4264 pData2 = CODEC2(pPager, pData, pPg->pgno, 7); |
|
4265 cksum = pager_cksum(pPager, (u8*)pData2); |
|
4266 rc = write32bits(pPager->jfd, pPager->journalOff, pPg->pgno); |
|
4267 if( rc==SQLITE_OK ){ |
|
4268 rc = sqlite3OsWrite(pPager->jfd, pData2, pPager->pageSize, |
|
4269 pPager->journalOff + 4); |
|
4270 pPager->journalOff += pPager->pageSize+4; |
|
4271 } |
|
4272 if( rc==SQLITE_OK ){ |
|
4273 rc = write32bits(pPager->jfd, pPager->journalOff, cksum); |
|
4274 pPager->journalOff += 4; |
|
4275 } |
|
4276 IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno, |
|
4277 pPager->journalOff, pPager->pageSize)); |
|
4278 PAGER_INCR(sqlite3_pager_writej_count); |
|
4279 PAGERTRACE5("JOURNAL %d page %d needSync=%d hash(%08x)\n", |
|
4280 PAGERID(pPager), pPg->pgno, pPg->needSync, pager_pagehash(pPg)); |
|
4281 |
|
4282 /* An error has occured writing to the journal file. The |
|
4283 ** transaction will be rolled back by the layer above. |
|
4284 */ |
|
4285 if( rc!=SQLITE_OK ){ |
|
4286 return rc; |
|
4287 } |
|
4288 |
|
4289 pPager->nRec++; |
|
4290 assert( pPager->pInJournal!=0 ); |
|
4291 sqlite3BitvecSet(pPager->pInJournal, pPg->pgno); |
|
4292 pPg->needSync = !pPager->noSync; |
|
4293 if( pPager->stmtInUse ){ |
|
4294 sqlite3BitvecSet(pPager->pInStmt, pPg->pgno); |
|
4295 } |
|
4296 } |
|
4297 }else{ |
|
4298 pPg->needSync = !pPager->journalStarted && !pPager->noSync; |
|
4299 PAGERTRACE4("APPEND %d page %d needSync=%d\n", |
|
4300 PAGERID(pPager), pPg->pgno, pPg->needSync); |
|
4301 } |
|
4302 if( pPg->needSync ){ |
|
4303 pPager->needSync = 1; |
|
4304 } |
|
4305 pPg->inJournal = 1; |
|
4306 } |
|
4307 |
|
4308 /* If the statement journal is open and the page is not in it, |
|
4309 ** then write the current page to the statement journal. Note that |
|
4310 ** the statement journal format differs from the standard journal format |
|
4311 ** in that it omits the checksums and the header. |
|
4312 */ |
|
4313 if( pPager->stmtInUse |
|
4314 && !pageInStatement(pPg) |
|
4315 && (int)pPg->pgno<=pPager->stmtSize |
|
4316 ){ |
|
4317 assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize ); |
|
4318 if( MEMDB ){ |
|
4319 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
4320 assert( pHist->pStmt==0 ); |
|
4321 pHist->pStmt = sqlite3PageMalloc( pPager->pageSize ); |
|
4322 if( pHist->pStmt ){ |
|
4323 memcpy(pHist->pStmt, PGHDR_TO_DATA(pPg), pPager->pageSize); |
|
4324 } |
|
4325 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno); |
|
4326 page_add_to_stmt_list(pPg); |
|
4327 }else{ |
|
4328 i64 offset = pPager->stmtNRec*(4+pPager->pageSize); |
|
4329 char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7); |
|
4330 rc = write32bits(pPager->stfd, offset, pPg->pgno); |
|
4331 if( rc==SQLITE_OK ){ |
|
4332 rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize, offset+4); |
|
4333 } |
|
4334 PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno); |
|
4335 if( rc!=SQLITE_OK ){ |
|
4336 return rc; |
|
4337 } |
|
4338 pPager->stmtNRec++; |
|
4339 assert( pPager->pInStmt!=0 ); |
|
4340 sqlite3BitvecSet(pPager->pInStmt, pPg->pgno); |
|
4341 } |
|
4342 } |
|
4343 } |
|
4344 |
|
4345 /* Update the database size and return. |
|
4346 */ |
|
4347 assert( pPager->state>=PAGER_SHARED ); |
|
4348 if( pPager->dbSize<(int)pPg->pgno ){ |
|
4349 pPager->dbSize = pPg->pgno; |
|
4350 if( !MEMDB && pPager->dbSize==PENDING_BYTE/pPager->pageSize ){ |
|
4351 pPager->dbSize++; |
|
4352 } |
|
4353 } |
|
4354 return rc; |
|
4355 } |
|
4356 |
|
4357 /* |
|
4358 ** This function is used to mark a data-page as writable. It uses |
|
4359 ** pager_write() to open a journal file (if it is not already open) |
|
4360 ** and write the page *pData to the journal. |
|
4361 ** |
|
4362 ** The difference between this function and pager_write() is that this |
|
4363 ** function also deals with the special case where 2 or more pages |
|
4364 ** fit on a single disk sector. In this case all co-resident pages |
|
4365 ** must have been written to the journal file before returning. |
|
4366 */ |
|
4367 int sqlite3PagerWrite(DbPage *pDbPage){ |
|
4368 int rc = SQLITE_OK; |
|
4369 |
|
4370 PgHdr *pPg = pDbPage; |
|
4371 Pager *pPager = pPg->pPager; |
|
4372 Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize); |
|
4373 |
|
4374 pagerEnter(pPager); |
|
4375 if( !MEMDB && nPagePerSector>1 ){ |
|
4376 Pgno nPageCount; /* Total number of pages in database file */ |
|
4377 Pgno pg1; /* First page of the sector pPg is located on. */ |
|
4378 int nPage; /* Number of pages starting at pg1 to journal */ |
|
4379 int ii; |
|
4380 int needSync = 0; |
|
4381 |
|
4382 /* Set the doNotSync flag to 1. This is because we cannot allow a journal |
|
4383 ** header to be written between the pages journaled by this function. |
|
4384 */ |
|
4385 assert( pPager->doNotSync==0 ); |
|
4386 pPager->doNotSync = 1; |
|
4387 |
|
4388 /* This trick assumes that both the page-size and sector-size are |
|
4389 ** an integer power of 2. It sets variable pg1 to the identifier |
|
4390 ** of the first page of the sector pPg is located on. |
|
4391 */ |
|
4392 pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1; |
|
4393 |
|
4394 sqlite3PagerPagecount(pPager, (int *)&nPageCount); |
|
4395 if( pPg->pgno>nPageCount ){ |
|
4396 nPage = (pPg->pgno - pg1)+1; |
|
4397 }else if( (pg1+nPagePerSector-1)>nPageCount ){ |
|
4398 nPage = nPageCount+1-pg1; |
|
4399 }else{ |
|
4400 nPage = nPagePerSector; |
|
4401 } |
|
4402 assert(nPage>0); |
|
4403 assert(pg1<=pPg->pgno); |
|
4404 assert((pg1+nPage)>pPg->pgno); |
|
4405 |
|
4406 for(ii=0; ii<nPage && rc==SQLITE_OK; ii++){ |
|
4407 Pgno pg = pg1+ii; |
|
4408 PgHdr *pPage; |
|
4409 if( pg==pPg->pgno || !sqlite3BitvecTest(pPager->pInJournal, pg) ){ |
|
4410 if( pg!=PAGER_MJ_PGNO(pPager) ){ |
|
4411 rc = sqlite3PagerGet(pPager, pg, &pPage); |
|
4412 if( rc==SQLITE_OK ){ |
|
4413 rc = pager_write(pPage); |
|
4414 if( pPage->needSync ){ |
|
4415 needSync = 1; |
|
4416 } |
|
4417 sqlite3PagerUnref(pPage); |
|
4418 } |
|
4419 } |
|
4420 }else if( (pPage = pager_lookup(pPager, pg))!=0 ){ |
|
4421 if( pPage->needSync ){ |
|
4422 needSync = 1; |
|
4423 } |
|
4424 } |
|
4425 } |
|
4426 |
|
4427 /* If the PgHdr.needSync flag is set for any of the nPage pages |
|
4428 ** starting at pg1, then it needs to be set for all of them. Because |
|
4429 ** writing to any of these nPage pages may damage the others, the |
|
4430 ** journal file must contain sync()ed copies of all of them |
|
4431 ** before any of them can be written out to the database file. |
|
4432 */ |
|
4433 if( needSync ){ |
|
4434 for(ii=0; ii<nPage && needSync; ii++){ |
|
4435 PgHdr *pPage = pager_lookup(pPager, pg1+ii); |
|
4436 if( pPage ) pPage->needSync = 1; |
|
4437 } |
|
4438 assert(pPager->needSync); |
|
4439 } |
|
4440 |
|
4441 assert( pPager->doNotSync==1 ); |
|
4442 pPager->doNotSync = 0; |
|
4443 }else{ |
|
4444 rc = pager_write(pDbPage); |
|
4445 } |
|
4446 pagerLeave(pPager); |
|
4447 return rc; |
|
4448 } |
|
4449 |
|
4450 /* |
|
4451 ** Return TRUE if the page given in the argument was previously passed |
|
4452 ** to sqlite3PagerWrite(). In other words, return TRUE if it is ok |
|
4453 ** to change the content of the page. |
|
4454 */ |
|
4455 #ifndef NDEBUG |
|
4456 int sqlite3PagerIswriteable(DbPage *pPg){ |
|
4457 return pPg->dirty; |
|
4458 } |
|
4459 #endif |
|
4460 |
|
4461 /* |
|
4462 ** A call to this routine tells the pager that it is not necessary to |
|
4463 ** write the information on page pPg back to the disk, even though |
|
4464 ** that page might be marked as dirty. |
|
4465 ** |
|
4466 ** The overlying software layer calls this routine when all of the data |
|
4467 ** on the given page is unused. The pager marks the page as clean so |
|
4468 ** that it does not get written to disk. |
|
4469 ** |
|
4470 ** Tests show that this optimization, together with the |
|
4471 ** sqlite3PagerDontRollback() below, more than double the speed |
|
4472 ** of large INSERT operations and quadruple the speed of large DELETEs. |
|
4473 ** |
|
4474 ** When this routine is called, set the alwaysRollback flag to true. |
|
4475 ** Subsequent calls to sqlite3PagerDontRollback() for the same page |
|
4476 ** will thereafter be ignored. This is necessary to avoid a problem |
|
4477 ** where a page with data is added to the freelist during one part of |
|
4478 ** a transaction then removed from the freelist during a later part |
|
4479 ** of the same transaction and reused for some other purpose. When it |
|
4480 ** is first added to the freelist, this routine is called. When reused, |
|
4481 ** the sqlite3PagerDontRollback() routine is called. But because the |
|
4482 ** page contains critical data, we still need to be sure it gets |
|
4483 ** rolled back in spite of the sqlite3PagerDontRollback() call. |
|
4484 */ |
|
4485 void sqlite3PagerDontWrite(DbPage *pDbPage){ |
|
4486 PgHdr *pPg = pDbPage; |
|
4487 Pager *pPager = pPg->pPager; |
|
4488 |
|
4489 if( MEMDB ) return; |
|
4490 pagerEnter(pPager); |
|
4491 pPg->alwaysRollback = 1; |
|
4492 if( pPg->dirty && !pPager->stmtInUse ){ |
|
4493 assert( pPager->state>=PAGER_SHARED ); |
|
4494 if( pPager->dbSize==(int)pPg->pgno && pPager->origDbSize<pPager->dbSize ){ |
|
4495 /* If this pages is the last page in the file and the file has grown |
|
4496 ** during the current transaction, then do NOT mark the page as clean. |
|
4497 ** When the database file grows, we must make sure that the last page |
|
4498 ** gets written at least once so that the disk file will be the correct |
|
4499 ** size. If you do not write this page and the size of the file |
|
4500 ** on the disk ends up being too small, that can lead to database |
|
4501 ** corruption during the next transaction. |
|
4502 */ |
|
4503 }else{ |
|
4504 PAGERTRACE3("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager)); |
|
4505 IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno)) |
|
4506 makeClean(pPg); |
|
4507 #ifdef SQLITE_CHECK_PAGES |
|
4508 pPg->pageHash = pager_pagehash(pPg); |
|
4509 #endif |
|
4510 } |
|
4511 } |
|
4512 pagerLeave(pPager); |
|
4513 } |
|
4514 |
|
4515 /* |
|
4516 ** A call to this routine tells the pager that if a rollback occurs, |
|
4517 ** it is not necessary to restore the data on the given page. This |
|
4518 ** means that the pager does not have to record the given page in the |
|
4519 ** rollback journal. |
|
4520 ** |
|
4521 ** If we have not yet actually read the content of this page (if |
|
4522 ** the PgHdr.needRead flag is set) then this routine acts as a promise |
|
4523 ** that we will never need to read the page content in the future. |
|
4524 ** so the needRead flag can be cleared at this point. |
|
4525 */ |
|
4526 void sqlite3PagerDontRollback(DbPage *pPg){ |
|
4527 Pager *pPager = pPg->pPager; |
|
4528 |
|
4529 pagerEnter(pPager); |
|
4530 assert( pPager->state>=PAGER_RESERVED ); |
|
4531 |
|
4532 /* If the journal file is not open, or DontWrite() has been called on |
|
4533 ** this page (DontWrite() sets the alwaysRollback flag), then this |
|
4534 ** function is a no-op. |
|
4535 */ |
|
4536 if( pPager->journalOpen==0 || pPg->alwaysRollback || pPager->alwaysRollback ){ |
|
4537 pagerLeave(pPager); |
|
4538 return; |
|
4539 } |
|
4540 assert( !MEMDB ); /* For a memdb, pPager->journalOpen is always 0 */ |
|
4541 |
|
4542 #ifdef SQLITE_SECURE_DELETE |
|
4543 if( pPg->inJournal || (int)pPg->pgno > pPager->origDbSize ){ |
|
4544 return; |
|
4545 } |
|
4546 #endif |
|
4547 |
|
4548 /* If SECURE_DELETE is disabled, then there is no way that this |
|
4549 ** routine can be called on a page for which sqlite3PagerDontWrite() |
|
4550 ** has not been previously called during the same transaction. |
|
4551 ** And if DontWrite() has previously been called, the following |
|
4552 ** conditions must be met. |
|
4553 ** |
|
4554 ** (Later:) Not true. If the database is corrupted by having duplicate |
|
4555 ** pages on the freelist (ex: corrupt9.test) then the following is not |
|
4556 ** necessarily true: |
|
4557 */ |
|
4558 /* assert( !pPg->inJournal && (int)pPg->pgno <= pPager->origDbSize ); */ |
|
4559 |
|
4560 assert( pPager->pInJournal!=0 ); |
|
4561 sqlite3BitvecSet(pPager->pInJournal, pPg->pgno); |
|
4562 pPg->inJournal = 1; |
|
4563 pPg->needRead = 0; |
|
4564 if( pPager->stmtInUse ){ |
|
4565 assert( pPager->stmtSize >= pPager->origDbSize ); |
|
4566 sqlite3BitvecSet(pPager->pInStmt, pPg->pgno); |
|
4567 } |
|
4568 PAGERTRACE3("DONT_ROLLBACK page %d of %d\n", pPg->pgno, PAGERID(pPager)); |
|
4569 IOTRACE(("GARBAGE %p %d\n", pPager, pPg->pgno)) |
|
4570 pagerLeave(pPager); |
|
4571 } |
|
4572 |
|
4573 |
|
4574 /* |
|
4575 ** This routine is called to increment the database file change-counter, |
|
4576 ** stored at byte 24 of the pager file. |
|
4577 */ |
|
4578 static int pager_incr_changecounter(Pager *pPager, int isDirect){ |
|
4579 PgHdr *pPgHdr; |
|
4580 u32 change_counter; |
|
4581 int rc = SQLITE_OK; |
|
4582 |
|
4583 #ifndef SQLITE_ENABLE_ATOMIC_WRITE |
|
4584 assert( isDirect==0 ); /* isDirect is only true for atomic writes */ |
|
4585 #endif |
|
4586 if( !pPager->changeCountDone ){ |
|
4587 /* Open page 1 of the file for writing. */ |
|
4588 rc = sqlite3PagerGet(pPager, 1, &pPgHdr); |
|
4589 if( rc!=SQLITE_OK ) return rc; |
|
4590 |
|
4591 if( !isDirect ){ |
|
4592 rc = sqlite3PagerWrite(pPgHdr); |
|
4593 if( rc!=SQLITE_OK ){ |
|
4594 sqlite3PagerUnref(pPgHdr); |
|
4595 return rc; |
|
4596 } |
|
4597 } |
|
4598 |
|
4599 /* Increment the value just read and write it back to byte 24. */ |
|
4600 change_counter = sqlite3Get4byte((u8*)pPager->dbFileVers); |
|
4601 change_counter++; |
|
4602 put32bits(((char*)PGHDR_TO_DATA(pPgHdr))+24, change_counter); |
|
4603 |
|
4604 #ifdef SQLITE_ENABLE_ATOMIC_WRITE |
|
4605 if( isDirect && pPager->fd->pMethods ){ |
|
4606 const void *zBuf = PGHDR_TO_DATA(pPgHdr); |
|
4607 rc = sqlite3OsWrite(pPager->fd, zBuf, pPager->pageSize, 0); |
|
4608 } |
|
4609 #endif |
|
4610 |
|
4611 /* Release the page reference. */ |
|
4612 sqlite3PagerUnref(pPgHdr); |
|
4613 pPager->changeCountDone = 1; |
|
4614 } |
|
4615 return rc; |
|
4616 } |
|
4617 |
|
4618 /* |
|
4619 ** Sync the pager file to disk. |
|
4620 */ |
|
4621 int sqlite3PagerSync(Pager *pPager){ |
|
4622 int rc; |
|
4623 pagerEnter(pPager); |
|
4624 rc = sqlite3OsSync(pPager->fd, pPager->sync_flags); |
|
4625 pagerLeave(pPager); |
|
4626 return rc; |
|
4627 } |
|
4628 |
|
4629 /* |
|
4630 ** Sync the database file for the pager pPager. zMaster points to the name |
|
4631 ** of a master journal file that should be written into the individual |
|
4632 ** journal file. zMaster may be NULL, which is interpreted as no master |
|
4633 ** journal (a single database transaction). |
|
4634 ** |
|
4635 ** This routine ensures that the journal is synced, all dirty pages written |
|
4636 ** to the database file and the database file synced. The only thing that |
|
4637 ** remains to commit the transaction is to delete the journal file (or |
|
4638 ** master journal file if specified). |
|
4639 ** |
|
4640 ** Note that if zMaster==NULL, this does not overwrite a previous value |
|
4641 ** passed to an sqlite3PagerCommitPhaseOne() call. |
|
4642 ** |
|
4643 ** If parameter nTrunc is non-zero, then the pager file is truncated to |
|
4644 ** nTrunc pages (this is used by auto-vacuum databases). |
|
4645 ** |
|
4646 ** If the final parameter - noSync - is true, then the database file itself |
|
4647 ** is not synced. The caller must call sqlite3PagerSync() directly to |
|
4648 ** sync the database file before calling CommitPhaseTwo() to delete the |
|
4649 ** journal file in this case. |
|
4650 */ |
|
4651 int sqlite3PagerCommitPhaseOne( |
|
4652 Pager *pPager, |
|
4653 const char *zMaster, |
|
4654 Pgno nTrunc, |
|
4655 int noSync |
|
4656 ){ |
|
4657 int rc = SQLITE_OK; |
|
4658 |
|
4659 if( pPager->errCode ){ |
|
4660 return pPager->errCode; |
|
4661 } |
|
4662 |
|
4663 /* If no changes have been made, we can leave the transaction early. |
|
4664 */ |
|
4665 if( pPager->dbModified==0 && |
|
4666 (pPager->journalMode!=PAGER_JOURNALMODE_DELETE || |
|
4667 pPager->exclusiveMode!=0) ){ |
|
4668 assert( pPager->dirtyCache==0 || pPager->journalOpen==0 ); |
|
4669 return SQLITE_OK; |
|
4670 } |
|
4671 |
|
4672 PAGERTRACE4("DATABASE SYNC: File=%s zMaster=%s nTrunc=%d\n", |
|
4673 pPager->zFilename, zMaster, nTrunc); |
|
4674 pagerEnter(pPager); |
|
4675 |
|
4676 /* If this is an in-memory db, or no pages have been written to, or this |
|
4677 ** function has already been called, it is a no-op. |
|
4678 */ |
|
4679 if( pPager->state!=PAGER_SYNCED && !MEMDB && pPager->dirtyCache ){ |
|
4680 PgHdr *pPg; |
|
4681 |
|
4682 #ifdef SQLITE_ENABLE_ATOMIC_WRITE |
|
4683 /* The atomic-write optimization can be used if all of the |
|
4684 ** following are true: |
|
4685 ** |
|
4686 ** + The file-system supports the atomic-write property for |
|
4687 ** blocks of size page-size, and |
|
4688 ** + This commit is not part of a multi-file transaction, and |
|
4689 ** + Exactly one page has been modified and store in the journal file. |
|
4690 ** |
|
4691 ** If the optimization can be used, then the journal file will never |
|
4692 ** be created for this transaction. |
|
4693 */ |
|
4694 int useAtomicWrite = ( |
|
4695 !zMaster && |
|
4696 pPager->journalOpen && |
|
4697 pPager->journalOff==jrnlBufferSize(pPager) && |
|
4698 nTrunc==0 && |
|
4699 (0==pPager->pDirty || 0==pPager->pDirty->pDirty) |
|
4700 ); |
|
4701 assert( pPager->journalOpen || pPager->journalMode==PAGER_JOURNALMODE_OFF ); |
|
4702 if( useAtomicWrite ){ |
|
4703 /* Update the nRec field in the journal file. */ |
|
4704 int offset = pPager->journalHdr + sizeof(aJournalMagic); |
|
4705 assert(pPager->nRec==1); |
|
4706 rc = write32bits(pPager->jfd, offset, pPager->nRec); |
|
4707 |
|
4708 /* Update the db file change counter. The following call will modify |
|
4709 ** the in-memory representation of page 1 to include the updated |
|
4710 ** change counter and then write page 1 directly to the database |
|
4711 ** file. Because of the atomic-write property of the host file-system, |
|
4712 ** this is safe. |
|
4713 */ |
|
4714 if( rc==SQLITE_OK ){ |
|
4715 rc = pager_incr_changecounter(pPager, 1); |
|
4716 } |
|
4717 }else{ |
|
4718 rc = sqlite3JournalCreate(pPager->jfd); |
|
4719 } |
|
4720 |
|
4721 if( !useAtomicWrite && rc==SQLITE_OK ) |
|
4722 #endif |
|
4723 |
|
4724 /* If a master journal file name has already been written to the |
|
4725 ** journal file, then no sync is required. This happens when it is |
|
4726 ** written, then the process fails to upgrade from a RESERVED to an |
|
4727 ** EXCLUSIVE lock. The next time the process tries to commit the |
|
4728 ** transaction the m-j name will have already been written. |
|
4729 */ |
|
4730 if( !pPager->setMaster ){ |
|
4731 rc = pager_incr_changecounter(pPager, 0); |
|
4732 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4733 if( pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ |
|
4734 #ifndef SQLITE_OMIT_AUTOVACUUM |
|
4735 if( nTrunc!=0 ){ |
|
4736 /* If this transaction has made the database smaller, then all pages |
|
4737 ** being discarded by the truncation must be written to the journal |
|
4738 ** file. |
|
4739 */ |
|
4740 Pgno i; |
|
4741 int iSkip = PAGER_MJ_PGNO(pPager); |
|
4742 for( i=nTrunc+1; i<=pPager->origDbSize; i++ ){ |
|
4743 if( !sqlite3BitvecTest(pPager->pInJournal, i) && i!=iSkip ){ |
|
4744 rc = sqlite3PagerGet(pPager, i, &pPg); |
|
4745 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4746 rc = sqlite3PagerWrite(pPg); |
|
4747 sqlite3PagerUnref(pPg); |
|
4748 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4749 } |
|
4750 } |
|
4751 } |
|
4752 #endif |
|
4753 rc = writeMasterJournal(pPager, zMaster); |
|
4754 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4755 rc = syncJournal(pPager); |
|
4756 } |
|
4757 } |
|
4758 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4759 |
|
4760 #ifndef SQLITE_OMIT_AUTOVACUUM |
|
4761 if( nTrunc!=0 ){ |
|
4762 rc = sqlite3PagerTruncate(pPager, nTrunc); |
|
4763 if( rc!=SQLITE_OK ) goto sync_exit; |
|
4764 } |
|
4765 #endif |
|
4766 |
|
4767 /* Write all dirty pages to the database file */ |
|
4768 pPg = pager_get_all_dirty_pages(pPager); |
|
4769 rc = pager_write_pagelist(pPg); |
|
4770 if( rc!=SQLITE_OK ){ |
|
4771 assert( rc!=SQLITE_IOERR_BLOCKED ); |
|
4772 /* The error might have left the dirty list all fouled up here, |
|
4773 ** but that does not matter because if the if the dirty list did |
|
4774 ** get corrupted, then the transaction will roll back and |
|
4775 ** discard the dirty list. There is an assert in |
|
4776 ** pager_get_all_dirty_pages() that verifies that no attempt |
|
4777 ** is made to use an invalid dirty list. |
|
4778 */ |
|
4779 goto sync_exit; |
|
4780 } |
|
4781 pPager->pDirty = 0; |
|
4782 |
|
4783 /* Sync the database file. */ |
|
4784 if( !pPager->noSync && !noSync ){ |
|
4785 rc = sqlite3OsSync(pPager->fd, pPager->sync_flags); |
|
4786 } |
|
4787 IOTRACE(("DBSYNC %p\n", pPager)) |
|
4788 |
|
4789 pPager->state = PAGER_SYNCED; |
|
4790 }else if( MEMDB && nTrunc!=0 ){ |
|
4791 rc = sqlite3PagerTruncate(pPager, nTrunc); |
|
4792 } |
|
4793 |
|
4794 sync_exit: |
|
4795 if( rc==SQLITE_IOERR_BLOCKED ){ |
|
4796 /* pager_incr_changecounter() may attempt to obtain an exclusive |
|
4797 * lock to spill the cache and return IOERR_BLOCKED. But since |
|
4798 * there is no chance the cache is inconsistent, it is |
|
4799 * better to return SQLITE_BUSY. |
|
4800 */ |
|
4801 rc = SQLITE_BUSY; |
|
4802 } |
|
4803 pagerLeave(pPager); |
|
4804 return rc; |
|
4805 } |
|
4806 |
|
4807 |
|
4808 /* |
|
4809 ** Commit all changes to the database and release the write lock. |
|
4810 ** |
|
4811 ** If the commit fails for any reason, a rollback attempt is made |
|
4812 ** and an error code is returned. If the commit worked, SQLITE_OK |
|
4813 ** is returned. |
|
4814 */ |
|
4815 int sqlite3PagerCommitPhaseTwo(Pager *pPager){ |
|
4816 int rc; |
|
4817 PgHdr *pPg; |
|
4818 |
|
4819 if( pPager->errCode ){ |
|
4820 return pPager->errCode; |
|
4821 } |
|
4822 if( pPager->state<PAGER_RESERVED ){ |
|
4823 return SQLITE_ERROR; |
|
4824 } |
|
4825 if( pPager->dbModified==0 && |
|
4826 (pPager->journalMode!=PAGER_JOURNALMODE_DELETE || |
|
4827 pPager->exclusiveMode!=0) ){ |
|
4828 assert( pPager->dirtyCache==0 || pPager->journalOpen==0 ); |
|
4829 return SQLITE_OK; |
|
4830 } |
|
4831 pagerEnter(pPager); |
|
4832 PAGERTRACE2("COMMIT %d\n", PAGERID(pPager)); |
|
4833 if( MEMDB ){ |
|
4834 pPg = pager_get_all_dirty_pages(pPager); |
|
4835 while( pPg ){ |
|
4836 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
4837 clearHistory(pHist); |
|
4838 pPg->dirty = 0; |
|
4839 pPg->inJournal = 0; |
|
4840 pHist->inStmt = 0; |
|
4841 pPg->needSync = 0; |
|
4842 pHist->pPrevStmt = pHist->pNextStmt = 0; |
|
4843 pPg = pPg->pDirty; |
|
4844 } |
|
4845 pPager->pDirty = 0; |
|
4846 #ifndef NDEBUG |
|
4847 for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){ |
|
4848 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
4849 assert( !pPg->alwaysRollback ); |
|
4850 assert( !pHist->pOrig ); |
|
4851 assert( !pHist->pStmt ); |
|
4852 } |
|
4853 #endif |
|
4854 pPager->pStmt = 0; |
|
4855 pPager->state = PAGER_SHARED; |
|
4856 pagerLeave(pPager); |
|
4857 return SQLITE_OK; |
|
4858 } |
|
4859 assert( pPager->state==PAGER_SYNCED || !pPager->dirtyCache ); |
|
4860 rc = pager_end_transaction(pPager, pPager->setMaster); |
|
4861 rc = pager_error(pPager, rc); |
|
4862 pagerLeave(pPager); |
|
4863 return rc; |
|
4864 } |
|
4865 |
|
4866 /* |
|
4867 ** Rollback all changes. The database falls back to PAGER_SHARED mode. |
|
4868 ** All in-memory cache pages revert to their original data contents. |
|
4869 ** The journal is deleted. |
|
4870 ** |
|
4871 ** This routine cannot fail unless some other process is not following |
|
4872 ** the correct locking protocol or unless some other |
|
4873 ** process is writing trash into the journal file (SQLITE_CORRUPT) or |
|
4874 ** unless a prior malloc() failed (SQLITE_NOMEM). Appropriate error |
|
4875 ** codes are returned for all these occasions. Otherwise, |
|
4876 ** SQLITE_OK is returned. |
|
4877 */ |
|
4878 int sqlite3PagerRollback(Pager *pPager){ |
|
4879 int rc; |
|
4880 PAGERTRACE2("ROLLBACK %d\n", PAGERID(pPager)); |
|
4881 if( MEMDB ){ |
|
4882 PgHdr *p; |
|
4883 for(p=pPager->pAll; p; p=p->pNextAll){ |
|
4884 PgHistory *pHist; |
|
4885 assert( !p->alwaysRollback ); |
|
4886 if( !p->dirty ){ |
|
4887 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pOrig ); |
|
4888 assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pStmt ); |
|
4889 continue; |
|
4890 } |
|
4891 |
|
4892 pHist = PGHDR_TO_HIST(p, pPager); |
|
4893 if( pHist->pOrig ){ |
|
4894 memcpy(PGHDR_TO_DATA(p), pHist->pOrig, pPager->pageSize); |
|
4895 PAGERTRACE3("ROLLBACK-PAGE %d of %d\n", p->pgno, PAGERID(pPager)); |
|
4896 }else{ |
|
4897 PAGERTRACE3("PAGE %d is clean on %d\n", p->pgno, PAGERID(pPager)); |
|
4898 } |
|
4899 clearHistory(pHist); |
|
4900 p->dirty = 0; |
|
4901 p->inJournal = 0; |
|
4902 pHist->inStmt = 0; |
|
4903 pHist->pPrevStmt = pHist->pNextStmt = 0; |
|
4904 if( pPager->xReiniter ){ |
|
4905 pPager->xReiniter(p, pPager->pageSize); |
|
4906 } |
|
4907 } |
|
4908 pPager->pDirty = 0; |
|
4909 pPager->pStmt = 0; |
|
4910 pPager->dbSize = pPager->origDbSize; |
|
4911 pager_truncate_cache(pPager); |
|
4912 pPager->stmtInUse = 0; |
|
4913 pPager->state = PAGER_SHARED; |
|
4914 return SQLITE_OK; |
|
4915 } |
|
4916 |
|
4917 pagerEnter(pPager); |
|
4918 if( !pPager->dirtyCache || !pPager->journalOpen ){ |
|
4919 rc = pager_end_transaction(pPager, pPager->setMaster); |
|
4920 pagerLeave(pPager); |
|
4921 return rc; |
|
4922 } |
|
4923 |
|
4924 if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){ |
|
4925 if( pPager->state>=PAGER_EXCLUSIVE ){ |
|
4926 pager_playback(pPager, 0); |
|
4927 } |
|
4928 pagerLeave(pPager); |
|
4929 return pPager->errCode; |
|
4930 } |
|
4931 if( pPager->state==PAGER_RESERVED ){ |
|
4932 int rc2; |
|
4933 rc = pager_playback(pPager, 0); |
|
4934 rc2 = pager_end_transaction(pPager, pPager->setMaster); |
|
4935 if( rc==SQLITE_OK ){ |
|
4936 rc = rc2; |
|
4937 } |
|
4938 }else{ |
|
4939 rc = pager_playback(pPager, 0); |
|
4940 } |
|
4941 /* pager_reset(pPager); */ |
|
4942 pPager->dbSize = -1; |
|
4943 |
|
4944 /* If an error occurs during a ROLLBACK, we can no longer trust the pager |
|
4945 ** cache. So call pager_error() on the way out to make any error |
|
4946 ** persistent. |
|
4947 */ |
|
4948 rc = pager_error(pPager, rc); |
|
4949 pagerLeave(pPager); |
|
4950 return rc; |
|
4951 } |
|
4952 |
|
4953 /* |
|
4954 ** Return TRUE if the database file is opened read-only. Return FALSE |
|
4955 ** if the database is (in theory) writable. |
|
4956 */ |
|
4957 int sqlite3PagerIsreadonly(Pager *pPager){ |
|
4958 return pPager->readOnly; |
|
4959 } |
|
4960 |
|
4961 /* |
|
4962 ** Return the number of references to the pager. |
|
4963 */ |
|
4964 int sqlite3PagerRefcount(Pager *pPager){ |
|
4965 return pPager->nRef; |
|
4966 } |
|
4967 |
|
4968 #ifdef SQLITE_TEST |
|
4969 /* |
|
4970 ** This routine is used for testing and analysis only. |
|
4971 */ |
|
4972 int *sqlite3PagerStats(Pager *pPager){ |
|
4973 static int a[11]; |
|
4974 a[0] = pPager->nRef; |
|
4975 a[1] = pPager->nPage; |
|
4976 a[2] = pPager->mxPage; |
|
4977 a[3] = pPager->dbSize; |
|
4978 a[4] = pPager->state; |
|
4979 a[5] = pPager->errCode; |
|
4980 a[6] = pPager->nHit; |
|
4981 a[7] = pPager->nMiss; |
|
4982 a[8] = 0; /* Used to be pPager->nOvfl */ |
|
4983 a[9] = pPager->nRead; |
|
4984 a[10] = pPager->nWrite; |
|
4985 return a; |
|
4986 } |
|
4987 int sqlite3PagerIsMemdb(Pager *pPager){ |
|
4988 return MEMDB; |
|
4989 } |
|
4990 #endif |
|
4991 |
|
4992 /* |
|
4993 ** Set the statement rollback point. |
|
4994 ** |
|
4995 ** This routine should be called with the transaction journal already |
|
4996 ** open. A new statement journal is created that can be used to rollback |
|
4997 ** changes of a single SQL command within a larger transaction. |
|
4998 */ |
|
4999 static int pagerStmtBegin(Pager *pPager){ |
|
5000 int rc; |
|
5001 assert( !pPager->stmtInUse ); |
|
5002 assert( pPager->state>=PAGER_SHARED ); |
|
5003 assert( pPager->dbSize>=0 ); |
|
5004 PAGERTRACE2("STMT-BEGIN %d\n", PAGERID(pPager)); |
|
5005 if( MEMDB ){ |
|
5006 pPager->stmtInUse = 1; |
|
5007 pPager->stmtSize = pPager->dbSize; |
|
5008 return SQLITE_OK; |
|
5009 } |
|
5010 if( !pPager->journalOpen ){ |
|
5011 pPager->stmtAutoopen = 1; |
|
5012 return SQLITE_OK; |
|
5013 } |
|
5014 assert( pPager->journalOpen ); |
|
5015 pagerLeave(pPager); |
|
5016 assert( pPager->pInStmt==0 ); |
|
5017 pPager->pInStmt = sqlite3BitvecCreate(pPager->dbSize); |
|
5018 pagerEnter(pPager); |
|
5019 if( pPager->pInStmt==0 ){ |
|
5020 /* sqlite3OsLock(pPager->fd, SHARED_LOCK); */ |
|
5021 return SQLITE_NOMEM; |
|
5022 } |
|
5023 pPager->stmtJSize = pPager->journalOff; |
|
5024 pPager->stmtSize = pPager->dbSize; |
|
5025 pPager->stmtHdrOff = 0; |
|
5026 pPager->stmtCksum = pPager->cksumInit; |
|
5027 if( !pPager->stmtOpen ){ |
|
5028 rc = sqlite3PagerOpentemp(pPager, pPager->stfd, SQLITE_OPEN_SUBJOURNAL); |
|
5029 if( rc ){ |
|
5030 goto stmt_begin_failed; |
|
5031 } |
|
5032 pPager->stmtOpen = 1; |
|
5033 pPager->stmtNRec = 0; |
|
5034 } |
|
5035 pPager->stmtInUse = 1; |
|
5036 return SQLITE_OK; |
|
5037 |
|
5038 stmt_begin_failed: |
|
5039 if( pPager->pInStmt ){ |
|
5040 sqlite3BitvecDestroy(pPager->pInStmt); |
|
5041 pPager->pInStmt = 0; |
|
5042 } |
|
5043 return rc; |
|
5044 } |
|
5045 int sqlite3PagerStmtBegin(Pager *pPager){ |
|
5046 int rc; |
|
5047 pagerEnter(pPager); |
|
5048 rc = pagerStmtBegin(pPager); |
|
5049 pagerLeave(pPager); |
|
5050 return rc; |
|
5051 } |
|
5052 |
|
5053 /* |
|
5054 ** Commit a statement. |
|
5055 */ |
|
5056 int sqlite3PagerStmtCommit(Pager *pPager){ |
|
5057 pagerEnter(pPager); |
|
5058 if( pPager->stmtInUse ){ |
|
5059 PgHdr *pPg, *pNext; |
|
5060 PAGERTRACE2("STMT-COMMIT %d\n", PAGERID(pPager)); |
|
5061 if( !MEMDB ){ |
|
5062 /* sqlite3OsTruncate(pPager->stfd, 0); */ |
|
5063 sqlite3BitvecDestroy(pPager->pInStmt); |
|
5064 pPager->pInStmt = 0; |
|
5065 }else{ |
|
5066 for(pPg=pPager->pStmt; pPg; pPg=pNext){ |
|
5067 PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager); |
|
5068 pNext = pHist->pNextStmt; |
|
5069 assert( pHist->inStmt ); |
|
5070 pHist->inStmt = 0; |
|
5071 pHist->pPrevStmt = pHist->pNextStmt = 0; |
|
5072 sqlite3PageFree(pHist->pStmt); |
|
5073 pHist->pStmt = 0; |
|
5074 } |
|
5075 } |
|
5076 pPager->stmtNRec = 0; |
|
5077 pPager->stmtInUse = 0; |
|
5078 pPager->pStmt = 0; |
|
5079 } |
|
5080 pPager->stmtAutoopen = 0; |
|
5081 pagerLeave(pPager); |
|
5082 return SQLITE_OK; |
|
5083 } |
|
5084 |
|
5085 /* |
|
5086 ** Rollback a statement. |
|
5087 */ |
|
5088 int sqlite3PagerStmtRollback(Pager *pPager){ |
|
5089 int rc; |
|
5090 pagerEnter(pPager); |
|
5091 if( pPager->stmtInUse ){ |
|
5092 PAGERTRACE2("STMT-ROLLBACK %d\n", PAGERID(pPager)); |
|
5093 if( MEMDB ){ |
|
5094 PgHdr *pPg; |
|
5095 PgHistory *pHist; |
|
5096 for(pPg=pPager->pStmt; pPg; pPg=pHist->pNextStmt){ |
|
5097 pHist = PGHDR_TO_HIST(pPg, pPager); |
|
5098 if( pHist->pStmt ){ |
|
5099 memcpy(PGHDR_TO_DATA(pPg), pHist->pStmt, pPager->pageSize); |
|
5100 sqlite3PageFree(pHist->pStmt); |
|
5101 pHist->pStmt = 0; |
|
5102 } |
|
5103 } |
|
5104 pPager->dbSize = pPager->stmtSize; |
|
5105 pager_truncate_cache(pPager); |
|
5106 rc = SQLITE_OK; |
|
5107 }else{ |
|
5108 rc = pager_stmt_playback(pPager); |
|
5109 } |
|
5110 sqlite3PagerStmtCommit(pPager); |
|
5111 }else{ |
|
5112 rc = SQLITE_OK; |
|
5113 } |
|
5114 pPager->stmtAutoopen = 0; |
|
5115 pagerLeave(pPager); |
|
5116 return rc; |
|
5117 } |
|
5118 |
|
5119 /* |
|
5120 ** Return the full pathname of the database file. |
|
5121 */ |
|
5122 const char *sqlite3PagerFilename(Pager *pPager){ |
|
5123 return pPager->zFilename; |
|
5124 } |
|
5125 |
|
5126 /* |
|
5127 ** Return the VFS structure for the pager. |
|
5128 */ |
|
5129 const sqlite3_vfs *sqlite3PagerVfs(Pager *pPager){ |
|
5130 return pPager->pVfs; |
|
5131 } |
|
5132 |
|
5133 /* |
|
5134 ** Return the file handle for the database file associated |
|
5135 ** with the pager. This might return NULL if the file has |
|
5136 ** not yet been opened. |
|
5137 */ |
|
5138 sqlite3_file *sqlite3PagerFile(Pager *pPager){ |
|
5139 return pPager->fd; |
|
5140 } |
|
5141 |
|
5142 /* |
|
5143 ** Return the directory of the database file. |
|
5144 */ |
|
5145 const char *sqlite3PagerDirname(Pager *pPager){ |
|
5146 return pPager->zDirectory; |
|
5147 } |
|
5148 |
|
5149 /* |
|
5150 ** Return the full pathname of the journal file. |
|
5151 */ |
|
5152 const char *sqlite3PagerJournalname(Pager *pPager){ |
|
5153 return pPager->zJournal; |
|
5154 } |
|
5155 |
|
5156 /* |
|
5157 ** Return true if fsync() calls are disabled for this pager. Return FALSE |
|
5158 ** if fsync()s are executed normally. |
|
5159 */ |
|
5160 int sqlite3PagerNosync(Pager *pPager){ |
|
5161 return pPager->noSync; |
|
5162 } |
|
5163 |
|
5164 #ifdef SQLITE_HAS_CODEC |
|
5165 /* |
|
5166 ** Set the codec for this pager |
|
5167 */ |
|
5168 void sqlite3PagerSetCodec( |
|
5169 Pager *pPager, |
|
5170 void *(*xCodec)(void*,void*,Pgno,int), |
|
5171 void *pCodecArg |
|
5172 ){ |
|
5173 pPager->xCodec = xCodec; |
|
5174 pPager->pCodecArg = pCodecArg; |
|
5175 } |
|
5176 #endif |
|
5177 |
|
5178 #ifndef SQLITE_OMIT_AUTOVACUUM |
|
5179 /* |
|
5180 ** Move the page pPg to location pgno in the file. |
|
5181 ** |
|
5182 ** There must be no references to the page previously located at |
|
5183 ** pgno (which we call pPgOld) though that page is allowed to be |
|
5184 ** in cache. If the page previous located at pgno is not already |
|
5185 ** in the rollback journal, it is not put there by by this routine. |
|
5186 ** |
|
5187 ** References to the page pPg remain valid. Updating any |
|
5188 ** meta-data associated with pPg (i.e. data stored in the nExtra bytes |
|
5189 ** allocated along with the page) is the responsibility of the caller. |
|
5190 ** |
|
5191 ** A transaction must be active when this routine is called. It used to be |
|
5192 ** required that a statement transaction was not active, but this restriction |
|
5193 ** has been removed (CREATE INDEX needs to move a page when a statement |
|
5194 ** transaction is active). |
|
5195 ** |
|
5196 ** If the fourth argument, isCommit, is non-zero, then this page is being |
|
5197 ** moved as part of a database reorganization just before the transaction |
|
5198 ** is being committed. In this case, it is guaranteed that the database page |
|
5199 ** pPg refers to will not be written to again within this transaction. |
|
5200 */ |
|
5201 int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno, int isCommit){ |
|
5202 PgHdr *pPgOld; /* The page being overwritten. */ |
|
5203 int h; |
|
5204 Pgno needSyncPgno = 0; |
|
5205 |
|
5206 pagerEnter(pPager); |
|
5207 assert( pPg->nRef>0 ); |
|
5208 |
|
5209 PAGERTRACE5("MOVE %d page %d (needSync=%d) moves to %d\n", |
|
5210 PAGERID(pPager), pPg->pgno, pPg->needSync, pgno); |
|
5211 IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno)) |
|
5212 |
|
5213 pager_get_content(pPg); |
|
5214 |
|
5215 /* If the journal needs to be sync()ed before page pPg->pgno can |
|
5216 ** be written to, store pPg->pgno in local variable needSyncPgno. |
|
5217 ** |
|
5218 ** If the isCommit flag is set, there is no need to remember that |
|
5219 ** the journal needs to be sync()ed before database page pPg->pgno |
|
5220 ** can be written to. The caller has already promised not to write to it. |
|
5221 */ |
|
5222 if( pPg->needSync && !isCommit ){ |
|
5223 needSyncPgno = pPg->pgno; |
|
5224 assert( pPg->inJournal || (int)pgno>pPager->origDbSize ); |
|
5225 assert( pPg->dirty ); |
|
5226 assert( pPager->needSync ); |
|
5227 } |
|
5228 |
|
5229 /* Unlink pPg from its hash-chain */ |
|
5230 unlinkHashChain(pPager, pPg); |
|
5231 |
|
5232 /* If the cache contains a page with page-number pgno, remove it |
|
5233 ** from its hash chain. Also, if the PgHdr.needSync was set for |
|
5234 ** page pgno before the 'move' operation, it needs to be retained |
|
5235 ** for the page moved there. |
|
5236 */ |
|
5237 pPg->needSync = 0; |
|
5238 pPgOld = pager_lookup(pPager, pgno); |
|
5239 if( pPgOld ){ |
|
5240 assert( pPgOld->nRef==0 ); |
|
5241 unlinkHashChain(pPager, pPgOld); |
|
5242 makeClean(pPgOld); |
|
5243 pPg->needSync = pPgOld->needSync; |
|
5244 }else{ |
|
5245 pPg->needSync = 0; |
|
5246 } |
|
5247 pPg->inJournal = sqlite3BitvecTest(pPager->pInJournal, pgno); |
|
5248 |
|
5249 /* Change the page number for pPg and insert it into the new hash-chain. */ |
|
5250 assert( pgno!=0 ); |
|
5251 pPg->pgno = pgno; |
|
5252 h = pgno & (pPager->nHash-1); |
|
5253 if( pPager->aHash[h] ){ |
|
5254 assert( pPager->aHash[h]->pPrevHash==0 ); |
|
5255 pPager->aHash[h]->pPrevHash = pPg; |
|
5256 } |
|
5257 pPg->pNextHash = pPager->aHash[h]; |
|
5258 pPager->aHash[h] = pPg; |
|
5259 pPg->pPrevHash = 0; |
|
5260 |
|
5261 makeDirty(pPg); |
|
5262 pPager->dirtyCache = 1; |
|
5263 pPager->dbModified = 1; |
|
5264 |
|
5265 if( needSyncPgno ){ |
|
5266 /* If needSyncPgno is non-zero, then the journal file needs to be |
|
5267 ** sync()ed before any data is written to database file page needSyncPgno. |
|
5268 ** Currently, no such page exists in the page-cache and the |
|
5269 ** "is journaled" bitvec flag has been set. This needs to be remedied by |
|
5270 ** loading the page into the pager-cache and setting the PgHdr.needSync |
|
5271 ** flag. |
|
5272 ** |
|
5273 ** If the attempt to load the page into the page-cache fails, (due |
|
5274 ** to a malloc() or IO failure), clear the bit in the pInJournal[] |
|
5275 ** array. Otherwise, if the page is loaded and written again in |
|
5276 ** this transaction, it may be written to the database file before |
|
5277 ** it is synced into the journal file. This way, it may end up in |
|
5278 ** the journal file twice, but that is not a problem. |
|
5279 ** |
|
5280 ** The sqlite3PagerGet() call may cause the journal to sync. So make |
|
5281 ** sure the Pager.needSync flag is set too. |
|
5282 */ |
|
5283 int rc; |
|
5284 PgHdr *pPgHdr; |
|
5285 assert( pPager->needSync ); |
|
5286 rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr); |
|
5287 if( rc!=SQLITE_OK ){ |
|
5288 if( pPager->pInJournal && (int)needSyncPgno<=pPager->origDbSize ){ |
|
5289 sqlite3BitvecClear(pPager->pInJournal, needSyncPgno); |
|
5290 } |
|
5291 pagerLeave(pPager); |
|
5292 return rc; |
|
5293 } |
|
5294 pPager->needSync = 1; |
|
5295 pPgHdr->needSync = 1; |
|
5296 pPgHdr->inJournal = 1; |
|
5297 makeDirty(pPgHdr); |
|
5298 sqlite3PagerUnref(pPgHdr); |
|
5299 } |
|
5300 |
|
5301 pagerLeave(pPager); |
|
5302 return SQLITE_OK; |
|
5303 } |
|
5304 #endif |
|
5305 |
|
5306 /* |
|
5307 ** Return a pointer to the data for the specified page. |
|
5308 */ |
|
5309 void *sqlite3PagerGetData(DbPage *pPg){ |
|
5310 return PGHDR_TO_DATA(pPg); |
|
5311 } |
|
5312 |
|
5313 /* |
|
5314 ** Return a pointer to the Pager.nExtra bytes of "extra" space |
|
5315 ** allocated along with the specified page. |
|
5316 */ |
|
5317 void *sqlite3PagerGetExtra(DbPage *pPg){ |
|
5318 Pager *pPager = pPg->pPager; |
|
5319 return (pPager?PGHDR_TO_EXTRA(pPg, pPager):0); |
|
5320 } |
|
5321 |
|
5322 /* |
|
5323 ** Get/set the locking-mode for this pager. Parameter eMode must be one |
|
5324 ** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or |
|
5325 ** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then |
|
5326 ** the locking-mode is set to the value specified. |
|
5327 ** |
|
5328 ** The returned value is either PAGER_LOCKINGMODE_NORMAL or |
|
5329 ** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated) |
|
5330 ** locking-mode. |
|
5331 */ |
|
5332 int sqlite3PagerLockingMode(Pager *pPager, int eMode){ |
|
5333 assert( eMode==PAGER_LOCKINGMODE_QUERY |
|
5334 || eMode==PAGER_LOCKINGMODE_NORMAL |
|
5335 || eMode==PAGER_LOCKINGMODE_EXCLUSIVE ); |
|
5336 assert( PAGER_LOCKINGMODE_QUERY<0 ); |
|
5337 assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 ); |
|
5338 if( eMode>=0 && !pPager->tempFile ){ |
|
5339 pPager->exclusiveMode = eMode; |
|
5340 } |
|
5341 return (int)pPager->exclusiveMode; |
|
5342 } |
|
5343 |
|
5344 /* |
|
5345 ** Get/set the journal-mode for this pager. Parameter eMode must be one |
|
5346 ** of PAGER_JOURNALMODE_QUERY, PAGER_JOURNALMODE_DELETE or |
|
5347 ** PAGER_JOURNALMODE_PERSIST. If the parameter is not _QUERY, then |
|
5348 ** the journal-mode is set to the value specified. |
|
5349 ** |
|
5350 ** The returned value is either PAGER_JOURNALMODE_DELETE or |
|
5351 ** PAGER_JOURNALMODE_PERSIST, indicating the current (possibly updated) |
|
5352 ** journal-mode. |
|
5353 */ |
|
5354 int sqlite3PagerJournalMode(Pager *pPager, int eMode){ |
|
5355 assert( eMode==PAGER_JOURNALMODE_QUERY |
|
5356 || eMode==PAGER_JOURNALMODE_DELETE |
|
5357 || eMode==PAGER_JOURNALMODE_PERSIST |
|
5358 || eMode==PAGER_JOURNALMODE_OFF ); |
|
5359 assert( PAGER_JOURNALMODE_QUERY<0 ); |
|
5360 assert( PAGER_JOURNALMODE_DELETE>=0 && PAGER_JOURNALMODE_PERSIST>=0 ); |
|
5361 if( eMode>=0 ){ |
|
5362 pPager->journalMode = eMode; |
|
5363 } |
|
5364 return (int)pPager->journalMode; |
|
5365 } |
|
5366 |
|
5367 /* |
|
5368 ** Get/set the size-limit used for persistent journal files. |
|
5369 */ |
|
5370 i64 sqlite3PagerJournalSizeLimit(Pager *pPager, i64 iLimit){ |
|
5371 if( iLimit>=-1 ){ |
|
5372 pPager->journalSizeLimit = iLimit; |
|
5373 } |
|
5374 return pPager->journalSizeLimit; |
|
5375 } |
|
5376 |
|
5377 #endif /* SQLITE_OMIT_DISKIO */ |