4 ** The author disclaims copyright to this source code. In place of
5 ** a legal notice, here is a blessing:
7 ** May you do good and not evil.
8 ** May you find forgiveness for yourself and forgive others.
9 ** May you share freely, never taking more than you give.
11 *************************************************************************
13 ** This file contains the implementation of a write-ahead log (WAL) used in
14 ** "journal_mode=WAL" mode.
16 ** WRITE-AHEAD LOG (WAL) FILE FORMAT
18 ** A WAL file consists of a header followed by zero or more "frames".
19 ** Each frame records the revised content of a single page from the
20 ** database file. All changes to the database are recorded by writing
21 ** frames into the WAL. Transactions commit when a frame is written that
22 ** contains a commit marker. A single WAL can and usually does record
23 ** multiple transactions. Periodically, the content of the WAL is
24 ** transferred back into the database file in an operation called a
27 ** A single WAL file can be used multiple times. In other words, the
28 ** WAL can fill up with frames and then be checkpointed and then new
29 ** frames can overwrite the old ones. A WAL always grows from beginning
30 ** toward the end. Checksums and counters attached to each frame are
31 ** used to determine which frames within the WAL are valid and which
32 ** are leftovers from prior checkpoints.
34 ** The WAL header is 32 bytes in size and consists of the following eight
35 ** big-endian 32-bit unsigned integer values:
37 ** 0: Magic number. 0x377f0682 or 0x377f0683
38 ** 4: File format version. Currently 3007000
39 ** 8: Database page size. Example: 1024
40 ** 12: Checkpoint sequence number
41 ** 16: Salt-1, random integer incremented with each checkpoint
42 ** 20: Salt-2, a different random integer changing with each ckpt
43 ** 24: Checksum-1 (first part of checksum for first 24 bytes of header).
44 ** 28: Checksum-2 (second part of checksum for first 24 bytes of header).
46 ** Immediately following the wal-header are zero or more frames. Each
47 ** frame consists of a 24-byte frame-header followed by a <page-size> bytes
48 ** of page data. The frame-header is six big-endian 32-bit unsigned
49 ** integer values, as follows:
52 ** 4: For commit records, the size of the database image in pages
53 ** after the commit. For all other records, zero.
54 ** 8: Salt-1 (copied from the header)
55 ** 12: Salt-2 (copied from the header)
59 ** A frame is considered valid if and only if the following conditions are
62 ** (1) The salt-1 and salt-2 values in the frame-header match
63 ** salt values in the wal-header
65 ** (2) The checksum values in the final 8 bytes of the frame-header
66 ** exactly match the checksum computed consecutively on the
67 ** WAL header and the first 8 bytes and the content of all frames
68 ** up to and including the current frame.
70 ** The checksum is computed using 32-bit big-endian integers if the
71 ** magic number in the first 4 bytes of the WAL is 0x377f0683 and it
72 ** is computed using little-endian if the magic number is 0x377f0682.
73 ** The checksum values are always stored in the frame header in a
74 ** big-endian format regardless of which byte order is used to compute
75 ** the checksum. The checksum is computed by interpreting the input as
76 ** an even number of unsigned 32-bit integers: x[0] through x[N]. The
77 ** algorithm used for the checksum is as follows:
79 ** for i from 0 to n-1 step 2:
84 ** Note that s0 and s1 are both weighted checksums using fibonacci weights
85 ** in reverse order (the largest fibonacci weight occurs on the first element
86 ** of the sequence being summed.) The s1 value spans all 32-bit
87 ** terms of the sequence whereas s0 omits the final term.
89 ** On a checkpoint, the WAL is first VFS.xSync-ed, then valid content of the
90 ** WAL is transferred into the database, then the database is VFS.xSync-ed.
91 ** The VFS.xSync operations serve as write barriers - all writes launched
92 ** before the xSync must complete before any write that launches after the
95 ** After each checkpoint, the salt-1 value is incremented and the salt-2
96 ** value is randomized. This prevents old and new frames in the WAL from
97 ** being considered valid at the same time and being checkpointing together
102 ** To read a page from the database (call it page number P), a reader
103 ** first checks the WAL to see if it contains page P. If so, then the
104 ** last valid instance of page P that is a followed by a commit frame
105 ** or is a commit frame itself becomes the value read. If the WAL
106 ** contains no copies of page P that are valid and which are a commit
107 ** frame or are followed by a commit frame, then page P is read from
108 ** the database file.
110 ** To start a read transaction, the reader records the index of the last
111 ** valid frame in the WAL. The reader uses this recorded "mxFrame" value
112 ** for all subsequent read operations. New transactions can be appended
113 ** to the WAL, but as long as the reader uses its original mxFrame value
114 ** and ignores the newly appended content, it will see a consistent snapshot
115 ** of the database from a single point in time. This technique allows
116 ** multiple concurrent readers to view different versions of the database
117 ** content simultaneously.
119 ** The reader algorithm in the previous paragraphs works correctly, but
120 ** because frames for page P can appear anywhere within the WAL, the
121 ** reader has to scan the entire WAL looking for page P frames. If the
122 ** WAL is large (multiple megabytes is typical) that scan can be slow,
123 ** and read performance suffers. To overcome this problem, a separate
124 ** data structure called the wal-index is maintained to expedite the
125 ** search for frames of a particular page.
129 ** Conceptually, the wal-index is shared memory, though VFS implementations
130 ** might choose to implement the wal-index using a mmapped file. Because
131 ** the wal-index is shared memory, SQLite does not support journal_mode=WAL
132 ** on a network filesystem. All users of the database must be able to
135 ** The wal-index is transient. After a crash, the wal-index can (and should
136 ** be) reconstructed from the original WAL file. In fact, the VFS is required
137 ** to either truncate or zero the header of the wal-index when the last
138 ** connection to it closes. Because the wal-index is transient, it can
139 ** use an architecture-specific format; it does not have to be cross-platform.
140 ** Hence, unlike the database and WAL file formats which store all values
141 ** as big endian, the wal-index can store multi-byte values in the native
142 ** byte order of the host computer.
144 ** The purpose of the wal-index is to answer this question quickly: Given
145 ** a page number P and a maximum frame index M, return the index of the
146 ** last frame in the wal before frame M for page P in the WAL, or return
147 ** NULL if there are no frames for page P in the WAL prior to M.
149 ** The wal-index consists of a header region, followed by an one or
150 ** more index blocks.
152 ** The wal-index header contains the total number of frames within the WAL
153 ** in the mxFrame field.
155 ** Each index block except for the first contains information on
156 ** HASHTABLE_NPAGE frames. The first index block contains information on
157 ** HASHTABLE_NPAGE_ONE frames. The values of HASHTABLE_NPAGE_ONE and
158 ** HASHTABLE_NPAGE are selected so that together the wal-index header and
159 ** first index block are the same size as all other index blocks in the
162 ** Each index block contains two sections, a page-mapping that contains the
163 ** database page number associated with each wal frame, and a hash-table
164 ** that allows readers to query an index block for a specific page number.
165 ** The page-mapping is an array of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE
166 ** for the first index block) 32-bit page numbers. The first entry in the
167 ** first index-block contains the database page number corresponding to the
168 ** first frame in the WAL file. The first entry in the second index block
169 ** in the WAL file corresponds to the (HASHTABLE_NPAGE_ONE+1)th frame in
170 ** the log, and so on.
172 ** The last index block in a wal-index usually contains less than the full
173 ** complement of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE) page-numbers,
174 ** depending on the contents of the WAL file. This does not change the
175 ** allocated size of the page-mapping array - the page-mapping array merely
176 ** contains unused entries.
178 ** Even without using the hash table, the last frame for page P
179 ** can be found by scanning the page-mapping sections of each index block
180 ** starting with the last index block and moving toward the first, and
181 ** within each index block, starting at the end and moving toward the
182 ** beginning. The first entry that equals P corresponds to the frame
183 ** holding the content for that page.
185 ** The hash table consists of HASHTABLE_NSLOT 16-bit unsigned integers.
186 ** HASHTABLE_NSLOT = 2*HASHTABLE_NPAGE, and there is one entry in the
187 ** hash table for each page number in the mapping section, so the hash
188 ** table is never more than half full. The expected number of collisions
189 ** prior to finding a match is 1. Each entry of the hash table is an
190 ** 1-based index of an entry in the mapping section of the same
191 ** index block. Let K be the 1-based index of the largest entry in
192 ** the mapping section. (For index blocks other than the last, K will
193 ** always be exactly HASHTABLE_NPAGE (4096) and for the last index block
194 ** K will be (mxFrame%HASHTABLE_NPAGE).) Unused slots of the hash table
195 ** contain a value of 0.
197 ** To look for page P in the hash table, first compute a hash iKey on
200 ** iKey = (P * 383) % HASHTABLE_NSLOT
202 ** Then start scanning entries of the hash table, starting with iKey
203 ** (wrapping around to the beginning when the end of the hash table is
204 ** reached) until an unused hash slot is found. Let the first unused slot
205 ** be at index iUnused. (iUnused might be less than iKey if there was
206 ** wrap-around.) Because the hash table is never more than half full,
207 ** the search is guaranteed to eventually hit an unused entry. Let
208 ** iMax be the value between iKey and iUnused, closest to iUnused,
209 ** where aHash[iMax]==P. If there is no iMax entry (if there exists
210 ** no hash slot such that aHash[i]==p) then page P is not in the
211 ** current index block. Otherwise the iMax-th mapping entry of the
212 ** current index block corresponds to the last entry that references
215 ** A hash search begins with the last index block and moves toward the
216 ** first index block, looking for entries corresponding to page P. On
217 ** average, only two or three slots in each index block need to be
218 ** examined in order to either find the last entry for page P, or to
219 ** establish that no such entry exists in the block. Each index block
220 ** holds over 4000 entries. So two or three index blocks are sufficient
221 ** to cover a typical 10 megabyte WAL file, assuming 1K pages. 8 or 10
222 ** comparisons (on average) suffice to either locate a frame in the
223 ** WAL or to establish that the frame does not exist in the WAL. This
224 ** is much faster than scanning the entire 10MB WAL.
226 ** Note that entries are added in order of increasing K. Hence, one
227 ** reader might be using some value K0 and a second reader that started
228 ** at a later time (after additional transactions were added to the WAL
229 ** and to the wal-index) might be using a different value K1, where K1>K0.
230 ** Both readers can use the same hash table and mapping section to get
231 ** the correct result. There may be entries in the hash table with
232 ** K>K0 but to the first reader, those entries will appear to be unused
233 ** slots in the hash table and so the first reader will get an answer as
234 ** if no values greater than K0 had ever been inserted into the hash table
235 ** in the first place - which is what reader one wants. Meanwhile, the
236 ** second reader using K1 will see additional values that were inserted
237 ** later, which is exactly what reader two wants.
239 ** When a rollback occurs, the value of K is decreased. Hash table entries
240 ** that correspond to frames greater than the new K value are removed
241 ** from the hash table at this point.
243 #ifndef SQLITE_OMIT_WAL
248 ** Trace output macros
250 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
251 int sqlite3WalTrace
= 0;
252 # define WALTRACE(X) if(sqlite3WalTrace) sqlite3DebugPrintf X
258 ** The maximum (and only) versions of the wal and wal-index formats
259 ** that may be interpreted by this version of SQLite.
261 ** If a client begins recovering a WAL file and finds that (a) the checksum
262 ** values in the wal-header are correct and (b) the version field is not
263 ** WAL_MAX_VERSION, recovery fails and SQLite returns SQLITE_CANTOPEN.
265 ** Similarly, if a client successfully reads a wal-index header (i.e. the
266 ** checksum test is successful) and finds that the version field is not
267 ** WALINDEX_MAX_VERSION, then no read-transaction is opened and SQLite
268 ** returns SQLITE_CANTOPEN.
270 #define WAL_MAX_VERSION 3007000
271 #define WALINDEX_MAX_VERSION 3007000
274 ** Indices of various locking bytes. WAL_NREADER is the number
275 ** of available reader locks and should be at least 3.
277 #define WAL_WRITE_LOCK 0
278 #define WAL_ALL_BUT_WRITE 1
279 #define WAL_CKPT_LOCK 1
280 #define WAL_RECOVER_LOCK 2
281 #define WAL_READ_LOCK(I) (3+(I))
282 #define WAL_NREADER (SQLITE_SHM_NLOCK-3)
285 /* Object declarations */
286 typedef struct WalIndexHdr WalIndexHdr
;
287 typedef struct WalIterator WalIterator
;
288 typedef struct WalCkptInfo WalCkptInfo
;
292 ** The following object holds a copy of the wal-index header content.
294 ** The actual header in the wal-index consists of two copies of this
297 ** The szPage value can be any power of 2 between 512 and 32768, inclusive.
298 ** Or it can be 1 to represent a 65536-byte page. The latter case was
299 ** added in 3.7.1 when support for 64K pages was added.
302 u32 iVersion
; /* Wal-index version */
303 u32 unused
; /* Unused (padding) field */
304 u32 iChange
; /* Counter incremented each transaction */
305 u8 isInit
; /* 1 when initialized */
306 u8 bigEndCksum
; /* True if checksums in WAL are big-endian */
307 u16 szPage
; /* Database page size in bytes. 1==64K */
308 u32 mxFrame
; /* Index of last valid frame in the WAL */
309 u32 nPage
; /* Size of database in pages */
310 u32 aFrameCksum
[2]; /* Checksum of last frame in log */
311 u32 aSalt
[2]; /* Two salt values copied from WAL header */
312 u32 aCksum
[2]; /* Checksum over all prior fields */
316 ** A copy of the following object occurs in the wal-index immediately
317 ** following the second copy of the WalIndexHdr. This object stores
318 ** information used by checkpoint.
320 ** nBackfill is the number of frames in the WAL that have been written
321 ** back into the database. (We call the act of moving content from WAL to
322 ** database "backfilling".) The nBackfill number is never greater than
323 ** WalIndexHdr.mxFrame. nBackfill can only be increased by threads
324 ** holding the WAL_CKPT_LOCK lock (which includes a recovery thread).
325 ** However, a WAL_WRITE_LOCK thread can move the value of nBackfill from
326 ** mxFrame back to zero when the WAL is reset.
328 ** There is one entry in aReadMark[] for each reader lock. If a reader
329 ** holds read-lock K, then the value in aReadMark[K] is no greater than
330 ** the mxFrame for that reader. The value READMARK_NOT_USED (0xffffffff)
331 ** for any aReadMark[] means that entry is unused. aReadMark[0] is
332 ** a special case; its value is never used and it exists as a place-holder
333 ** to avoid having to offset aReadMark[] indexs by one. Readers holding
334 ** WAL_READ_LOCK(0) always ignore the entire WAL and read all content
335 ** directly from the database.
337 ** The value of aReadMark[K] may only be changed by a thread that
338 ** is holding an exclusive lock on WAL_READ_LOCK(K). Thus, the value of
339 ** aReadMark[K] cannot changed while there is a reader is using that mark
340 ** since the reader will be holding a shared lock on WAL_READ_LOCK(K).
342 ** The checkpointer may only transfer frames from WAL to database where
343 ** the frame numbers are less than or equal to every aReadMark[] that is
344 ** in use (that is, every aReadMark[j] for which there is a corresponding
345 ** WAL_READ_LOCK(j)). New readers (usually) pick the aReadMark[] with the
346 ** largest value and will increase an unused aReadMark[] to mxFrame if there
347 ** is not already an aReadMark[] equal to mxFrame. The exception to the
348 ** previous sentence is when nBackfill equals mxFrame (meaning that everything
349 ** in the WAL has been backfilled into the database) then new readers
350 ** will choose aReadMark[0] which has value 0 and hence such reader will
351 ** get all their all content directly from the database file and ignore
354 ** Writers normally append new frames to the end of the WAL. However,
355 ** if nBackfill equals mxFrame (meaning that all WAL content has been
356 ** written back into the database) and if no readers are using the WAL
357 ** (in other words, if there are no WAL_READ_LOCK(i) where i>0) then
358 ** the writer will first "reset" the WAL back to the beginning and start
359 ** writing new content beginning at frame 1.
361 ** We assume that 32-bit loads are atomic and so no locks are needed in
362 ** order to read from any aReadMark[] entries.
365 u32 nBackfill
; /* Number of WAL frames backfilled into DB */
366 u32 aReadMark
[WAL_NREADER
]; /* Reader marks */
368 #define READMARK_NOT_USED 0xffffffff
371 /* A block of WALINDEX_LOCK_RESERVED bytes beginning at
372 ** WALINDEX_LOCK_OFFSET is reserved for locks. Since some systems
373 ** only support mandatory file-locks, we do not read or write data
374 ** from the region of the file on which locks are applied.
376 #define WALINDEX_LOCK_OFFSET (sizeof(WalIndexHdr)*2 + sizeof(WalCkptInfo))
377 #define WALINDEX_LOCK_RESERVED 16
378 #define WALINDEX_HDR_SIZE (WALINDEX_LOCK_OFFSET+WALINDEX_LOCK_RESERVED)
380 /* Size of header before each frame in wal */
381 #define WAL_FRAME_HDRSIZE 24
383 /* Size of write ahead log header, including checksum. */
384 /* #define WAL_HDRSIZE 24 */
385 #define WAL_HDRSIZE 32
387 /* WAL magic value. Either this value, or the same value with the least
388 ** significant bit also set (WAL_MAGIC | 0x00000001) is stored in 32-bit
389 ** big-endian format in the first 4 bytes of a WAL file.
391 ** If the LSB is set, then the checksums for each frame within the WAL
392 ** file are calculated by treating all data as an array of 32-bit
393 ** big-endian words. Otherwise, they are calculated by interpreting
394 ** all data as 32-bit little-endian words.
396 #define WAL_MAGIC 0x377f0682
399 ** Return the offset of frame iFrame in the write-ahead log file,
400 ** assuming a database page size of szPage bytes. The offset returned
401 ** is to the start of the write-ahead log frame-header.
403 #define walFrameOffset(iFrame, szPage) ( \
404 WAL_HDRSIZE + ((iFrame)-1)*(i64)((szPage)+WAL_FRAME_HDRSIZE) \
408 ** An open write-ahead log file is represented by an instance of the
412 sqlite3_vfs
*pVfs
; /* The VFS used to create pDbFd */
413 sqlite3_file
*pDbFd
; /* File handle for the database file */
414 sqlite3_file
*pWalFd
; /* File handle for WAL file */
415 u32 iCallback
; /* Value to pass to log callback (or 0) */
416 i64 mxWalSize
; /* Truncate WAL to this size upon reset */
417 int nWiData
; /* Size of array apWiData */
418 int szFirstBlock
; /* Size of first block written to WAL file */
419 volatile u32
**apWiData
; /* Pointer to wal-index content in memory */
420 u32 szPage
; /* Database page size */
421 i16 readLock
; /* Which read lock is being held. -1 for none */
422 u8 syncFlags
; /* Flags to use to sync header writes */
423 u8 exclusiveMode
; /* Non-zero if connection is in exclusive mode */
424 u8 writeLock
; /* True if in a write transaction */
425 u8 ckptLock
; /* True if holding a checkpoint lock */
426 u8 readOnly
; /* WAL_RDWR, WAL_RDONLY, or WAL_SHM_RDONLY */
427 u8 truncateOnCommit
; /* True to truncate WAL file on commit */
428 u8 syncHeader
; /* Fsync the WAL header if true */
429 u8 padToSectorBoundary
; /* Pad transactions out to the next sector */
430 WalIndexHdr hdr
; /* Wal-index header for current transaction */
431 const char *zWalName
; /* Name of WAL file */
432 u32 nCkpt
; /* Checkpoint sequence counter in the wal-header */
434 u8 lockError
; /* True if a locking error has occurred */
439 ** Candidate values for Wal.exclusiveMode.
441 #define WAL_NORMAL_MODE 0
442 #define WAL_EXCLUSIVE_MODE 1
443 #define WAL_HEAPMEMORY_MODE 2
446 ** Possible values for WAL.readOnly
448 #define WAL_RDWR 0 /* Normal read/write connection */
449 #define WAL_RDONLY 1 /* The WAL file is readonly */
450 #define WAL_SHM_RDONLY 2 /* The SHM file is readonly */
453 ** Each page of the wal-index mapping contains a hash-table made up of
454 ** an array of HASHTABLE_NSLOT elements of the following type.
459 ** This structure is used to implement an iterator that loops through
460 ** all frames in the WAL in database page order. Where two or more frames
461 ** correspond to the same database page, the iterator visits only the
462 ** frame most recently written to the WAL (in other words, the frame with
463 ** the largest index).
465 ** The internals of this structure are only accessed by:
467 ** walIteratorInit() - Create a new iterator,
468 ** walIteratorNext() - Step an iterator,
469 ** walIteratorFree() - Free an iterator.
471 ** This functionality is used by the checkpoint code (see walCheckpoint()).
474 int iPrior
; /* Last result returned from the iterator */
475 int nSegment
; /* Number of entries in aSegment[] */
477 int iNext
; /* Next slot in aIndex[] not yet returned */
478 ht_slot
*aIndex
; /* i0, i1, i2... such that aPgno[iN] ascend */
479 u32
*aPgno
; /* Array of page numbers. */
480 int nEntry
; /* Nr. of entries in aPgno[] and aIndex[] */
481 int iZero
; /* Frame number associated with aPgno[0] */
482 } aSegment
[1]; /* One for every 32KB page in the wal-index */
486 ** Define the parameters of the hash tables in the wal-index file. There
487 ** is a hash-table following every HASHTABLE_NPAGE page numbers in the
490 ** Changing any of these constants will alter the wal-index format and
491 ** create incompatibilities.
493 #define HASHTABLE_NPAGE 4096 /* Must be power of 2 */
494 #define HASHTABLE_HASH_1 383 /* Should be prime */
495 #define HASHTABLE_NSLOT (HASHTABLE_NPAGE*2) /* Must be a power of 2 */
498 ** The block of page numbers associated with the first hash-table in a
499 ** wal-index is smaller than usual. This is so that there is a complete
500 ** hash-table on each aligned 32KB page of the wal-index.
502 #define HASHTABLE_NPAGE_ONE (HASHTABLE_NPAGE - (WALINDEX_HDR_SIZE/sizeof(u32)))
504 /* The wal-index is divided into pages of WALINDEX_PGSZ bytes each. */
505 #define WALINDEX_PGSZ ( \
506 sizeof(ht_slot)*HASHTABLE_NSLOT + HASHTABLE_NPAGE*sizeof(u32) \
510 ** Obtain a pointer to the iPage'th page of the wal-index. The wal-index
511 ** is broken into pages of WALINDEX_PGSZ bytes. Wal-index pages are
512 ** numbered from zero.
514 ** If this call is successful, *ppPage is set to point to the wal-index
515 ** page and SQLITE_OK is returned. If an error (an OOM or VFS error) occurs,
516 ** then an SQLite error code is returned and *ppPage is set to 0.
518 static int walIndexPage(Wal
*pWal
, int iPage
, volatile u32
**ppPage
){
521 /* Enlarge the pWal->apWiData[] array if required */
522 if( pWal
->nWiData
<=iPage
){
523 int nByte
= sizeof(u32
*)*(iPage
+1);
524 volatile u32
**apNew
;
525 apNew
= (volatile u32
**)sqlite3_realloc((void *)pWal
->apWiData
, nByte
);
530 memset((void*)&apNew
[pWal
->nWiData
], 0,
531 sizeof(u32
*)*(iPage
+1-pWal
->nWiData
));
532 pWal
->apWiData
= apNew
;
533 pWal
->nWiData
= iPage
+1;
536 /* Request a pointer to the required page from the VFS */
537 if( pWal
->apWiData
[iPage
]==0 ){
538 if( pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
){
539 pWal
->apWiData
[iPage
] = (u32
volatile *)sqlite3MallocZero(WALINDEX_PGSZ
);
540 if( !pWal
->apWiData
[iPage
] ) rc
= SQLITE_NOMEM
;
542 rc
= sqlite3OsShmMap(pWal
->pDbFd
, iPage
, WALINDEX_PGSZ
,
543 pWal
->writeLock
, (void volatile **)&pWal
->apWiData
[iPage
]
545 if( rc
==SQLITE_READONLY
){
546 pWal
->readOnly
|= WAL_SHM_RDONLY
;
552 *ppPage
= pWal
->apWiData
[iPage
];
553 assert( iPage
==0 || *ppPage
|| rc
!=SQLITE_OK
);
558 ** Return a pointer to the WalCkptInfo structure in the wal-index.
560 static volatile WalCkptInfo
*walCkptInfo(Wal
*pWal
){
561 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
562 return (volatile WalCkptInfo
*)&(pWal
->apWiData
[0][sizeof(WalIndexHdr
)/2]);
566 ** Return a pointer to the WalIndexHdr structure in the wal-index.
568 static volatile WalIndexHdr
*walIndexHdr(Wal
*pWal
){
569 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
570 return (volatile WalIndexHdr
*)pWal
->apWiData
[0];
574 ** The argument to this macro must be of type u32. On a little-endian
575 ** architecture, it returns the u32 value that results from interpreting
576 ** the 4 bytes as a big-endian value. On a big-endian architecture, it
577 ** returns the value that would be produced by intepreting the 4 bytes
578 ** of the input value as a little-endian integer.
580 #define BYTESWAP32(x) ( \
581 (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8) \
582 + (((x)&0x00FF0000)>>8) + (((x)&0xFF000000)>>24) \
586 ** Generate or extend an 8 byte checksum based on the data in
587 ** array aByte[] and the initial values of aIn[0] and aIn[1] (or
588 ** initial values of 0 and 0 if aIn==NULL).
590 ** The checksum is written back into aOut[] before returning.
592 ** nByte must be a positive multiple of 8.
594 static void walChecksumBytes(
595 int nativeCksum
, /* True for native byte-order, false for non-native */
596 u8
*a
, /* Content to be checksummed */
597 int nByte
, /* Bytes of content in a[]. Must be a multiple of 8. */
598 const u32
*aIn
, /* Initial checksum value input */
599 u32
*aOut
/* OUT: Final checksum value output */
602 u32
*aData
= (u32
*)a
;
603 u32
*aEnd
= (u32
*)&a
[nByte
];
613 assert( (nByte
&0x00000007)==0 );
619 }while( aData
<aEnd
);
622 s1
+= BYTESWAP32(aData
[0]) + s2
;
623 s2
+= BYTESWAP32(aData
[1]) + s1
;
625 }while( aData
<aEnd
);
632 static void walShmBarrier(Wal
*pWal
){
633 if( pWal
->exclusiveMode
!=WAL_HEAPMEMORY_MODE
){
634 sqlite3OsShmBarrier(pWal
->pDbFd
);
639 ** Write the header information in pWal->hdr into the wal-index.
641 ** The checksum on pWal->hdr is updated before it is written.
643 static void walIndexWriteHdr(Wal
*pWal
){
644 volatile WalIndexHdr
*aHdr
= walIndexHdr(pWal
);
645 const int nCksum
= offsetof(WalIndexHdr
, aCksum
);
647 assert( pWal
->writeLock
);
648 pWal
->hdr
.isInit
= 1;
649 pWal
->hdr
.iVersion
= WALINDEX_MAX_VERSION
;
650 walChecksumBytes(1, (u8
*)&pWal
->hdr
, nCksum
, 0, pWal
->hdr
.aCksum
);
651 memcpy((void *)&aHdr
[1], (void *)&pWal
->hdr
, sizeof(WalIndexHdr
));
653 memcpy((void *)&aHdr
[0], (void *)&pWal
->hdr
, sizeof(WalIndexHdr
));
657 ** This function encodes a single frame header and writes it to a buffer
658 ** supplied by the caller. A frame-header is made up of a series of
659 ** 4-byte big-endian integers, as follows:
662 ** 4: For commit records, the size of the database image in pages
663 ** after the commit. For all other records, zero.
664 ** 8: Salt-1 (copied from the wal-header)
665 ** 12: Salt-2 (copied from the wal-header)
669 static void walEncodeFrame(
670 Wal
*pWal
, /* The write-ahead log */
671 u32 iPage
, /* Database page number for frame */
672 u32 nTruncate
, /* New db size (or 0 for non-commit frames) */
673 u8
*aData
, /* Pointer to page data */
674 u8
*aFrame
/* OUT: Write encoded frame here */
676 int nativeCksum
; /* True for native byte-order checksums */
677 u32
*aCksum
= pWal
->hdr
.aFrameCksum
;
678 assert( WAL_FRAME_HDRSIZE
==24 );
679 sqlite3Put4byte(&aFrame
[0], iPage
);
680 sqlite3Put4byte(&aFrame
[4], nTruncate
);
681 memcpy(&aFrame
[8], pWal
->hdr
.aSalt
, 8);
683 nativeCksum
= (pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
);
684 walChecksumBytes(nativeCksum
, aFrame
, 8, aCksum
, aCksum
);
685 walChecksumBytes(nativeCksum
, aData
, pWal
->szPage
, aCksum
, aCksum
);
687 sqlite3Put4byte(&aFrame
[16], aCksum
[0]);
688 sqlite3Put4byte(&aFrame
[20], aCksum
[1]);
692 ** Check to see if the frame with header in aFrame[] and content
693 ** in aData[] is valid. If it is a valid frame, fill *piPage and
694 ** *pnTruncate and return true. Return if the frame is not valid.
696 static int walDecodeFrame(
697 Wal
*pWal
, /* The write-ahead log */
698 u32
*piPage
, /* OUT: Database page number for frame */
699 u32
*pnTruncate
, /* OUT: New db size (or 0 if not commit) */
700 u8
*aData
, /* Pointer to page data (for checksum) */
701 u8
*aFrame
/* Frame data */
703 int nativeCksum
; /* True for native byte-order checksums */
704 u32
*aCksum
= pWal
->hdr
.aFrameCksum
;
705 u32 pgno
; /* Page number of the frame */
706 assert( WAL_FRAME_HDRSIZE
==24 );
708 /* A frame is only valid if the salt values in the frame-header
709 ** match the salt values in the wal-header.
711 if( memcmp(&pWal
->hdr
.aSalt
, &aFrame
[8], 8)!=0 ){
715 /* A frame is only valid if the page number is creater than zero.
717 pgno
= sqlite3Get4byte(&aFrame
[0]);
722 /* A frame is only valid if a checksum of the WAL header,
723 ** all prior frams, the first 16 bytes of this frame-header,
724 ** and the frame-data matches the checksum in the last 8
725 ** bytes of this frame-header.
727 nativeCksum
= (pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
);
728 walChecksumBytes(nativeCksum
, aFrame
, 8, aCksum
, aCksum
);
729 walChecksumBytes(nativeCksum
, aData
, pWal
->szPage
, aCksum
, aCksum
);
730 if( aCksum
[0]!=sqlite3Get4byte(&aFrame
[16])
731 || aCksum
[1]!=sqlite3Get4byte(&aFrame
[20])
733 /* Checksum failed. */
737 /* If we reach this point, the frame is valid. Return the page number
738 ** and the new database size.
741 *pnTruncate
= sqlite3Get4byte(&aFrame
[4]);
746 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
748 ** Names of locks. This routine is used to provide debugging output and is not
749 ** a part of an ordinary build.
751 static const char *walLockName(int lockIdx
){
752 if( lockIdx
==WAL_WRITE_LOCK
){
754 }else if( lockIdx
==WAL_CKPT_LOCK
){
756 }else if( lockIdx
==WAL_RECOVER_LOCK
){
757 return "RECOVER-LOCK";
759 static char zName
[15];
760 sqlite3_snprintf(sizeof(zName
), zName
, "READ-LOCK[%d]",
761 lockIdx
-WAL_READ_LOCK(0));
765 #endif /*defined(SQLITE_TEST) || defined(SQLITE_DEBUG) */
769 ** Set or release locks on the WAL. Locks are either shared or exclusive.
770 ** A lock cannot be moved directly between shared and exclusive - it must go
771 ** through the unlocked state first.
773 ** In locking_mode=EXCLUSIVE, all of these routines become no-ops.
775 static int walLockShared(Wal
*pWal
, int lockIdx
){
777 if( pWal
->exclusiveMode
) return SQLITE_OK
;
778 rc
= sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, 1,
779 SQLITE_SHM_LOCK
| SQLITE_SHM_SHARED
);
780 WALTRACE(("WAL%p: acquire SHARED-%s %s\n", pWal
,
781 walLockName(lockIdx
), rc
? "failed" : "ok"));
782 VVA_ONLY( pWal
->lockError
= (u8
)(rc
!=SQLITE_OK
&& rc
!=SQLITE_BUSY
); )
785 static void walUnlockShared(Wal
*pWal
, int lockIdx
){
786 if( pWal
->exclusiveMode
) return;
787 (void)sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, 1,
788 SQLITE_SHM_UNLOCK
| SQLITE_SHM_SHARED
);
789 WALTRACE(("WAL%p: release SHARED-%s\n", pWal
, walLockName(lockIdx
)));
791 static int walLockExclusive(Wal
*pWal
, int lockIdx
, int n
){
793 if( pWal
->exclusiveMode
) return SQLITE_OK
;
794 rc
= sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, n
,
795 SQLITE_SHM_LOCK
| SQLITE_SHM_EXCLUSIVE
);
796 WALTRACE(("WAL%p: acquire EXCLUSIVE-%s cnt=%d %s\n", pWal
,
797 walLockName(lockIdx
), n
, rc
? "failed" : "ok"));
798 VVA_ONLY( pWal
->lockError
= (u8
)(rc
!=SQLITE_OK
&& rc
!=SQLITE_BUSY
); )
801 static void walUnlockExclusive(Wal
*pWal
, int lockIdx
, int n
){
802 if( pWal
->exclusiveMode
) return;
803 (void)sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, n
,
804 SQLITE_SHM_UNLOCK
| SQLITE_SHM_EXCLUSIVE
);
805 WALTRACE(("WAL%p: release EXCLUSIVE-%s cnt=%d\n", pWal
,
806 walLockName(lockIdx
), n
));
810 ** Compute a hash on a page number. The resulting hash value must land
811 ** between 0 and (HASHTABLE_NSLOT-1). The walHashNext() function advances
812 ** the hash to the next value in the event of a collision.
814 static int walHash(u32 iPage
){
816 assert( (HASHTABLE_NSLOT
& (HASHTABLE_NSLOT
-1))==0 );
817 return (iPage
*HASHTABLE_HASH_1
) & (HASHTABLE_NSLOT
-1);
819 static int walNextHash(int iPriorHash
){
820 return (iPriorHash
+1)&(HASHTABLE_NSLOT
-1);
824 ** Return pointers to the hash table and page number array stored on
825 ** page iHash of the wal-index. The wal-index is broken into 32KB pages
826 ** numbered starting from 0.
828 ** Set output variable *paHash to point to the start of the hash table
829 ** in the wal-index file. Set *piZero to one less than the frame
830 ** number of the first frame indexed by this hash table. If a
831 ** slot in the hash table is set to N, it refers to frame number
832 ** (*piZero+N) in the log.
834 ** Finally, set *paPgno so that *paPgno[1] is the page number of the
835 ** first frame indexed by the hash table, frame (*piZero+1).
837 static int walHashGet(
838 Wal
*pWal
, /* WAL handle */
839 int iHash
, /* Find the iHash'th table */
840 volatile ht_slot
**paHash
, /* OUT: Pointer to hash index */
841 volatile u32
**paPgno
, /* OUT: Pointer to page number array */
842 u32
*piZero
/* OUT: Frame associated with *paPgno[0] */
844 int rc
; /* Return code */
847 rc
= walIndexPage(pWal
, iHash
, &aPgno
);
848 assert( rc
==SQLITE_OK
|| iHash
>0 );
852 volatile ht_slot
*aHash
;
854 aHash
= (volatile ht_slot
*)&aPgno
[HASHTABLE_NPAGE
];
856 aPgno
= &aPgno
[WALINDEX_HDR_SIZE
/sizeof(u32
)];
859 iZero
= HASHTABLE_NPAGE_ONE
+ (iHash
-1)*HASHTABLE_NPAGE
;
862 *paPgno
= &aPgno
[-1];
870 ** Return the number of the wal-index page that contains the hash-table
871 ** and page-number array that contain entries corresponding to WAL frame
872 ** iFrame. The wal-index is broken up into 32KB pages. Wal-index pages
873 ** are numbered starting from 0.
875 static int walFramePage(u32 iFrame
){
876 int iHash
= (iFrame
+HASHTABLE_NPAGE
-HASHTABLE_NPAGE_ONE
-1) / HASHTABLE_NPAGE
;
877 assert( (iHash
==0 || iFrame
>HASHTABLE_NPAGE_ONE
)
878 && (iHash
>=1 || iFrame
<=HASHTABLE_NPAGE_ONE
)
879 && (iHash
<=1 || iFrame
>(HASHTABLE_NPAGE_ONE
+HASHTABLE_NPAGE
))
880 && (iHash
>=2 || iFrame
<=HASHTABLE_NPAGE_ONE
+HASHTABLE_NPAGE
)
881 && (iHash
<=2 || iFrame
>(HASHTABLE_NPAGE_ONE
+2*HASHTABLE_NPAGE
))
887 ** Return the page number associated with frame iFrame in this WAL.
889 static u32
walFramePgno(Wal
*pWal
, u32 iFrame
){
890 int iHash
= walFramePage(iFrame
);
892 return pWal
->apWiData
[0][WALINDEX_HDR_SIZE
/sizeof(u32
) + iFrame
- 1];
894 return pWal
->apWiData
[iHash
][(iFrame
-1-HASHTABLE_NPAGE_ONE
)%HASHTABLE_NPAGE
];
898 ** Remove entries from the hash table that point to WAL slots greater
899 ** than pWal->hdr.mxFrame.
901 ** This function is called whenever pWal->hdr.mxFrame is decreased due
902 ** to a rollback or savepoint.
904 ** At most only the hash table containing pWal->hdr.mxFrame needs to be
905 ** updated. Any later hash tables will be automatically cleared when
906 ** pWal->hdr.mxFrame advances to the point where those hash tables are
909 static void walCleanupHash(Wal
*pWal
){
910 volatile ht_slot
*aHash
= 0; /* Pointer to hash table to clear */
911 volatile u32
*aPgno
= 0; /* Page number array for hash table */
912 u32 iZero
= 0; /* frame == (aHash[x]+iZero) */
913 int iLimit
= 0; /* Zero values greater than this */
914 int nByte
; /* Number of bytes to zero in aPgno[] */
915 int i
; /* Used to iterate through aHash[] */
917 assert( pWal
->writeLock
);
918 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
-1 );
919 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
);
920 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
+1 );
922 if( pWal
->hdr
.mxFrame
==0 ) return;
924 /* Obtain pointers to the hash-table and page-number array containing
925 ** the entry that corresponds to frame pWal->hdr.mxFrame. It is guaranteed
926 ** that the page said hash-table and array reside on is already mapped.
928 assert( pWal
->nWiData
>walFramePage(pWal
->hdr
.mxFrame
) );
929 assert( pWal
->apWiData
[walFramePage(pWal
->hdr
.mxFrame
)] );
930 walHashGet(pWal
, walFramePage(pWal
->hdr
.mxFrame
), &aHash
, &aPgno
, &iZero
);
932 /* Zero all hash-table entries that correspond to frame numbers greater
933 ** than pWal->hdr.mxFrame.
935 iLimit
= pWal
->hdr
.mxFrame
- iZero
;
937 for(i
=0; i
<HASHTABLE_NSLOT
; i
++){
938 if( aHash
[i
]>iLimit
){
943 /* Zero the entries in the aPgno array that correspond to frames with
944 ** frame numbers greater than pWal->hdr.mxFrame.
946 nByte
= (int)((char *)aHash
- (char *)&aPgno
[iLimit
+1]);
947 memset((void *)&aPgno
[iLimit
+1], 0, nByte
);
949 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
950 /* Verify that the every entry in the mapping region is still reachable
951 ** via the hash table even after the cleanup.
954 int i
; /* Loop counter */
955 int iKey
; /* Hash key */
956 for(i
=1; i
<=iLimit
; i
++){
957 for(iKey
=walHash(aPgno
[i
]); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
958 if( aHash
[iKey
]==i
) break;
960 assert( aHash
[iKey
]==i
);
963 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
968 ** Set an entry in the wal-index that will map database page number
969 ** pPage into WAL frame iFrame.
971 static int walIndexAppend(Wal
*pWal
, u32 iFrame
, u32 iPage
){
972 int rc
; /* Return code */
973 u32 iZero
= 0; /* One less than frame number of aPgno[1] */
974 volatile u32
*aPgno
= 0; /* Page number array */
975 volatile ht_slot
*aHash
= 0; /* Hash table */
977 rc
= walHashGet(pWal
, walFramePage(iFrame
), &aHash
, &aPgno
, &iZero
);
979 /* Assuming the wal-index file was successfully mapped, populate the
980 ** page number array and hash table entry.
983 int iKey
; /* Hash table key */
984 int idx
; /* Value to write to hash-table slot */
985 int nCollide
; /* Number of hash collisions */
987 idx
= iFrame
- iZero
;
988 assert( idx
<= HASHTABLE_NSLOT
/2 + 1 );
990 /* If this is the first entry to be added to this hash-table, zero the
991 ** entire hash table and aPgno[] array before proceding.
994 int nByte
= (int)((u8
*)&aHash
[HASHTABLE_NSLOT
] - (u8
*)&aPgno
[1]);
995 memset((void*)&aPgno
[1], 0, nByte
);
998 /* If the entry in aPgno[] is already set, then the previous writer
999 ** must have exited unexpectedly in the middle of a transaction (after
1000 ** writing one or more dirty pages to the WAL to free up memory).
1001 ** Remove the remnants of that writers uncommitted transaction from
1002 ** the hash-table before writing any new entries.
1005 walCleanupHash(pWal
);
1006 assert( !aPgno
[idx
] );
1009 /* Write the aPgno[] array entry and the hash-table slot. */
1011 for(iKey
=walHash(iPage
); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
1012 if( (nCollide
--)==0 ) return SQLITE_CORRUPT_BKPT
;
1015 aHash
[iKey
] = (ht_slot
)idx
;
1017 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
1018 /* Verify that the number of entries in the hash table exactly equals
1019 ** the number of entries in the mapping region.
1022 int i
; /* Loop counter */
1023 int nEntry
= 0; /* Number of entries in the hash table */
1024 for(i
=0; i
<HASHTABLE_NSLOT
; i
++){ if( aHash
[i
] ) nEntry
++; }
1025 assert( nEntry
==idx
);
1028 /* Verify that the every entry in the mapping region is reachable
1029 ** via the hash table. This turns out to be a really, really expensive
1030 ** thing to check, so only do this occasionally - not on every
1033 if( (idx
&0x3ff)==0 ){
1034 int i
; /* Loop counter */
1035 for(i
=1; i
<=idx
; i
++){
1036 for(iKey
=walHash(aPgno
[i
]); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
1037 if( aHash
[iKey
]==i
) break;
1039 assert( aHash
[iKey
]==i
);
1042 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
1051 ** Recover the wal-index by reading the write-ahead log file.
1053 ** This routine first tries to establish an exclusive lock on the
1054 ** wal-index to prevent other threads/processes from doing anything
1055 ** with the WAL or wal-index while recovery is running. The
1056 ** WAL_RECOVER_LOCK is also held so that other threads will know
1057 ** that this thread is running recovery. If unable to establish
1058 ** the necessary locks, this routine returns SQLITE_BUSY.
1060 static int walIndexRecover(Wal
*pWal
){
1061 int rc
; /* Return Code */
1062 i64 nSize
; /* Size of log file */
1063 u32 aFrameCksum
[2] = {0, 0};
1064 int iLock
; /* Lock offset to lock for checkpoint */
1065 int nLock
; /* Number of locks to hold */
1067 /* Obtain an exclusive lock on all byte in the locking range not already
1068 ** locked by the caller. The caller is guaranteed to have locked the
1069 ** WAL_WRITE_LOCK byte, and may have also locked the WAL_CKPT_LOCK byte.
1070 ** If successful, the same bytes that are locked here are unlocked before
1071 ** this function returns.
1073 assert( pWal
->ckptLock
==1 || pWal
->ckptLock
==0 );
1074 assert( WAL_ALL_BUT_WRITE
==WAL_WRITE_LOCK
+1 );
1075 assert( WAL_CKPT_LOCK
==WAL_ALL_BUT_WRITE
);
1076 assert( pWal
->writeLock
);
1077 iLock
= WAL_ALL_BUT_WRITE
+ pWal
->ckptLock
;
1078 nLock
= SQLITE_SHM_NLOCK
- iLock
;
1079 rc
= walLockExclusive(pWal
, iLock
, nLock
);
1083 WALTRACE(("WAL%p: recovery begin...\n", pWal
));
1085 memset(&pWal
->hdr
, 0, sizeof(WalIndexHdr
));
1087 rc
= sqlite3OsFileSize(pWal
->pWalFd
, &nSize
);
1088 if( rc
!=SQLITE_OK
){
1089 goto recovery_error
;
1092 if( nSize
>WAL_HDRSIZE
){
1093 u8 aBuf
[WAL_HDRSIZE
]; /* Buffer to load WAL header into */
1094 u8
*aFrame
= 0; /* Malloc'd buffer to load entire frame */
1095 int szFrame
; /* Number of bytes in buffer aFrame[] */
1096 u8
*aData
; /* Pointer to data part of aFrame buffer */
1097 int iFrame
; /* Index of last frame read */
1098 i64 iOffset
; /* Next offset to read from log file */
1099 int szPage
; /* Page size according to the log */
1100 u32 magic
; /* Magic value read from WAL header */
1101 u32 version
; /* Magic value read from WAL header */
1102 int isValid
; /* True if this frame is valid */
1104 /* Read in the WAL header. */
1105 rc
= sqlite3OsRead(pWal
->pWalFd
, aBuf
, WAL_HDRSIZE
, 0);
1106 if( rc
!=SQLITE_OK
){
1107 goto recovery_error
;
1110 /* If the database page size is not a power of two, or is greater than
1111 ** SQLITE_MAX_PAGE_SIZE, conclude that the WAL file contains no valid
1112 ** data. Similarly, if the 'magic' value is invalid, ignore the whole
1115 magic
= sqlite3Get4byte(&aBuf
[0]);
1116 szPage
= sqlite3Get4byte(&aBuf
[8]);
1117 if( (magic
&0xFFFFFFFE)!=WAL_MAGIC
1118 || szPage
&(szPage
-1)
1119 || szPage
>SQLITE_MAX_PAGE_SIZE
1124 pWal
->hdr
.bigEndCksum
= (u8
)(magic
&0x00000001);
1125 pWal
->szPage
= szPage
;
1126 pWal
->nCkpt
= sqlite3Get4byte(&aBuf
[12]);
1127 memcpy(&pWal
->hdr
.aSalt
, &aBuf
[16], 8);
1129 /* Verify that the WAL header checksum is correct */
1130 walChecksumBytes(pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
,
1131 aBuf
, WAL_HDRSIZE
-2*4, 0, pWal
->hdr
.aFrameCksum
1133 if( pWal
->hdr
.aFrameCksum
[0]!=sqlite3Get4byte(&aBuf
[24])
1134 || pWal
->hdr
.aFrameCksum
[1]!=sqlite3Get4byte(&aBuf
[28])
1139 /* Verify that the version number on the WAL format is one that
1140 ** are able to understand */
1141 version
= sqlite3Get4byte(&aBuf
[4]);
1142 if( version
!=WAL_MAX_VERSION
){
1143 rc
= SQLITE_CANTOPEN_BKPT
;
1147 /* Malloc a buffer to read frames into. */
1148 szFrame
= szPage
+ WAL_FRAME_HDRSIZE
;
1149 aFrame
= (u8
*)sqlite3_malloc(szFrame
);
1152 goto recovery_error
;
1154 aData
= &aFrame
[WAL_FRAME_HDRSIZE
];
1156 /* Read all frames from the log file. */
1158 for(iOffset
=WAL_HDRSIZE
; (iOffset
+szFrame
)<=nSize
; iOffset
+=szFrame
){
1159 u32 pgno
; /* Database page number for frame */
1160 u32 nTruncate
; /* dbsize field from frame header */
1162 /* Read and decode the next log frame. */
1164 rc
= sqlite3OsRead(pWal
->pWalFd
, aFrame
, szFrame
, iOffset
);
1165 if( rc
!=SQLITE_OK
) break;
1166 isValid
= walDecodeFrame(pWal
, &pgno
, &nTruncate
, aData
, aFrame
);
1167 if( !isValid
) break;
1168 rc
= walIndexAppend(pWal
, iFrame
, pgno
);
1169 if( rc
!=SQLITE_OK
) break;
1171 /* If nTruncate is non-zero, this is a commit record. */
1173 pWal
->hdr
.mxFrame
= iFrame
;
1174 pWal
->hdr
.nPage
= nTruncate
;
1175 pWal
->hdr
.szPage
= (u16
)((szPage
&0xff00) | (szPage
>>16));
1176 testcase( szPage
<=32768 );
1177 testcase( szPage
>=65536 );
1178 aFrameCksum
[0] = pWal
->hdr
.aFrameCksum
[0];
1179 aFrameCksum
[1] = pWal
->hdr
.aFrameCksum
[1];
1183 sqlite3_free(aFrame
);
1187 if( rc
==SQLITE_OK
){
1188 volatile WalCkptInfo
*pInfo
;
1190 pWal
->hdr
.aFrameCksum
[0] = aFrameCksum
[0];
1191 pWal
->hdr
.aFrameCksum
[1] = aFrameCksum
[1];
1192 walIndexWriteHdr(pWal
);
1194 /* Reset the checkpoint-header. This is safe because this thread is
1195 ** currently holding locks that exclude all other readers, writers and
1198 pInfo
= walCkptInfo(pWal
);
1199 pInfo
->nBackfill
= 0;
1200 pInfo
->aReadMark
[0] = 0;
1201 for(i
=1; i
<WAL_NREADER
; i
++) pInfo
->aReadMark
[i
] = READMARK_NOT_USED
;
1202 if( pWal
->hdr
.mxFrame
) pInfo
->aReadMark
[1] = pWal
->hdr
.mxFrame
;
1204 /* If more than one frame was recovered from the log file, report an
1205 ** event via sqlite3_log(). This is to help with identifying performance
1206 ** problems caused by applications routinely shutting down without
1207 ** checkpointing the log file.
1209 if( pWal
->hdr
.nPage
){
1210 sqlite3_log(SQLITE_NOTICE_RECOVER_WAL
,
1211 "recovered %d frames from WAL file %s",
1212 pWal
->hdr
.mxFrame
, pWal
->zWalName
1218 WALTRACE(("WAL%p: recovery %s\n", pWal
, rc
? "failed" : "ok"));
1219 walUnlockExclusive(pWal
, iLock
, nLock
);
1224 ** Close an open wal-index.
1226 static void walIndexClose(Wal
*pWal
, int isDelete
){
1227 if( pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
){
1229 for(i
=0; i
<pWal
->nWiData
; i
++){
1230 sqlite3_free((void *)pWal
->apWiData
[i
]);
1231 pWal
->apWiData
[i
] = 0;
1234 sqlite3OsShmUnmap(pWal
->pDbFd
, isDelete
);
1239 ** Open a connection to the WAL file zWalName. The database file must
1240 ** already be opened on connection pDbFd. The buffer that zWalName points
1241 ** to must remain valid for the lifetime of the returned Wal* handle.
1243 ** A SHARED lock should be held on the database file when this function
1244 ** is called. The purpose of this SHARED lock is to prevent any other
1245 ** client from unlinking the WAL or wal-index file. If another process
1246 ** were to do this just after this client opened one of these files, the
1247 ** system would be badly broken.
1249 ** If the log file is successfully opened, SQLITE_OK is returned and
1250 ** *ppWal is set to point to a new WAL handle. If an error occurs,
1251 ** an SQLite error code is returned and *ppWal is left unmodified.
1254 sqlite3_vfs
*pVfs
, /* vfs module to open wal and wal-index */
1255 sqlite3_file
*pDbFd
, /* The open database file */
1256 const char *zWalName
, /* Name of the WAL file */
1257 int bNoShm
, /* True to run in heap-memory mode */
1258 i64 mxWalSize
, /* Truncate WAL to this size on reset */
1259 Wal
**ppWal
/* OUT: Allocated Wal handle */
1261 int rc
; /* Return Code */
1262 Wal
*pRet
; /* Object to allocate and return */
1263 int flags
; /* Flags passed to OsOpen() */
1265 assert( zWalName
&& zWalName
[0] );
1268 /* In the amalgamation, the os_unix.c and os_win.c source files come before
1269 ** this source file. Verify that the #defines of the locking byte offsets
1270 ** in os_unix.c and os_win.c agree with the WALINDEX_LOCK_OFFSET value.
1273 assert( WIN_SHM_BASE
==WALINDEX_LOCK_OFFSET
);
1275 #ifdef UNIX_SHM_BASE
1276 assert( UNIX_SHM_BASE
==WALINDEX_LOCK_OFFSET
);
1280 /* Allocate an instance of struct Wal to return. */
1282 pRet
= (Wal
*)sqlite3MallocZero(sizeof(Wal
) + pVfs
->szOsFile
);
1284 return SQLITE_NOMEM
;
1288 pRet
->pWalFd
= (sqlite3_file
*)&pRet
[1];
1289 pRet
->pDbFd
= pDbFd
;
1290 pRet
->readLock
= -1;
1291 pRet
->mxWalSize
= mxWalSize
;
1292 pRet
->zWalName
= zWalName
;
1293 pRet
->syncHeader
= 1;
1294 pRet
->padToSectorBoundary
= 1;
1295 pRet
->exclusiveMode
= (bNoShm
? WAL_HEAPMEMORY_MODE
: WAL_NORMAL_MODE
);
1297 /* Open file handle on the write-ahead log file. */
1298 flags
= (SQLITE_OPEN_READWRITE
|SQLITE_OPEN_CREATE
|SQLITE_OPEN_WAL
);
1299 rc
= sqlite3OsOpen(pVfs
, zWalName
, pRet
->pWalFd
, flags
, &flags
);
1300 if( rc
==SQLITE_OK
&& flags
&SQLITE_OPEN_READONLY
){
1301 pRet
->readOnly
= WAL_RDONLY
;
1304 if( rc
!=SQLITE_OK
){
1305 walIndexClose(pRet
, 0);
1306 sqlite3OsClose(pRet
->pWalFd
);
1309 int iDC
= sqlite3OsDeviceCharacteristics(pDbFd
);
1310 if( iDC
& SQLITE_IOCAP_SEQUENTIAL
){ pRet
->syncHeader
= 0; }
1311 if( iDC
& SQLITE_IOCAP_POWERSAFE_OVERWRITE
){
1312 pRet
->padToSectorBoundary
= 0;
1315 WALTRACE(("WAL%d: opened\n", pRet
));
1321 ** Change the size to which the WAL file is trucated on each reset.
1323 void sqlite3WalLimit(Wal
*pWal
, i64 iLimit
){
1324 if( pWal
) pWal
->mxWalSize
= iLimit
;
1328 ** Find the smallest page number out of all pages held in the WAL that
1329 ** has not been returned by any prior invocation of this method on the
1330 ** same WalIterator object. Write into *piFrame the frame index where
1331 ** that page was last written into the WAL. Write into *piPage the page
1334 ** Return 0 on success. If there are no pages in the WAL with a page
1335 ** number larger than *piPage, then return 1.
1337 static int walIteratorNext(
1338 WalIterator
*p
, /* Iterator */
1339 u32
*piPage
, /* OUT: The page number of the next page */
1340 u32
*piFrame
/* OUT: Wal frame index of next page */
1342 u32 iMin
; /* Result pgno must be greater than iMin */
1343 u32 iRet
= 0xFFFFFFFF; /* 0xffffffff is never a valid page number */
1344 int i
; /* For looping through segments */
1347 assert( iMin
<0xffffffff );
1348 for(i
=p
->nSegment
-1; i
>=0; i
--){
1349 struct WalSegment
*pSegment
= &p
->aSegment
[i
];
1350 while( pSegment
->iNext
<pSegment
->nEntry
){
1351 u32 iPg
= pSegment
->aPgno
[pSegment
->aIndex
[pSegment
->iNext
]];
1355 *piFrame
= pSegment
->iZero
+ pSegment
->aIndex
[pSegment
->iNext
];
1363 *piPage
= p
->iPrior
= iRet
;
1364 return (iRet
==0xFFFFFFFF);
1368 ** This function merges two sorted lists into a single sorted list.
1370 ** aLeft[] and aRight[] are arrays of indices. The sort key is
1371 ** aContent[aLeft[]] and aContent[aRight[]]. Upon entry, the following
1372 ** is guaranteed for all J<K:
1374 ** aContent[aLeft[J]] < aContent[aLeft[K]]
1375 ** aContent[aRight[J]] < aContent[aRight[K]]
1377 ** This routine overwrites aRight[] with a new (probably longer) sequence
1378 ** of indices such that the aRight[] contains every index that appears in
1379 ** either aLeft[] or the old aRight[] and such that the second condition
1380 ** above is still met.
1382 ** The aContent[aLeft[X]] values will be unique for all X. And the
1383 ** aContent[aRight[X]] values will be unique too. But there might be
1384 ** one or more combinations of X and Y such that
1386 ** aLeft[X]!=aRight[Y] && aContent[aLeft[X]] == aContent[aRight[Y]]
1388 ** When that happens, omit the aLeft[X] and use the aRight[Y] index.
1390 static void walMerge(
1391 const u32
*aContent
, /* Pages in wal - keys for the sort */
1392 ht_slot
*aLeft
, /* IN: Left hand input list */
1393 int nLeft
, /* IN: Elements in array *paLeft */
1394 ht_slot
**paRight
, /* IN/OUT: Right hand input list */
1395 int *pnRight
, /* IN/OUT: Elements in *paRight */
1396 ht_slot
*aTmp
/* Temporary buffer */
1398 int iLeft
= 0; /* Current index in aLeft */
1399 int iRight
= 0; /* Current index in aRight */
1400 int iOut
= 0; /* Current index in output buffer */
1401 int nRight
= *pnRight
;
1402 ht_slot
*aRight
= *paRight
;
1404 assert( nLeft
>0 && nRight
>0 );
1405 while( iRight
<nRight
|| iLeft
<nLeft
){
1410 && (iRight
>=nRight
|| aContent
[aLeft
[iLeft
]]<aContent
[aRight
[iRight
]])
1412 logpage
= aLeft
[iLeft
++];
1414 logpage
= aRight
[iRight
++];
1416 dbpage
= aContent
[logpage
];
1418 aTmp
[iOut
++] = logpage
;
1419 if( iLeft
<nLeft
&& aContent
[aLeft
[iLeft
]]==dbpage
) iLeft
++;
1421 assert( iLeft
>=nLeft
|| aContent
[aLeft
[iLeft
]]>dbpage
);
1422 assert( iRight
>=nRight
|| aContent
[aRight
[iRight
]]>dbpage
);
1427 memcpy(aLeft
, aTmp
, sizeof(aTmp
[0])*iOut
);
1431 ** Sort the elements in list aList using aContent[] as the sort key.
1432 ** Remove elements with duplicate keys, preferring to keep the
1433 ** larger aList[] values.
1435 ** The aList[] entries are indices into aContent[]. The values in
1436 ** aList[] are to be sorted so that for all J<K:
1438 ** aContent[aList[J]] < aContent[aList[K]]
1440 ** For any X and Y such that
1442 ** aContent[aList[X]] == aContent[aList[Y]]
1444 ** Keep the larger of the two values aList[X] and aList[Y] and discard
1447 static void walMergesort(
1448 const u32
*aContent
, /* Pages in wal */
1449 ht_slot
*aBuffer
, /* Buffer of at least *pnList items to use */
1450 ht_slot
*aList
, /* IN/OUT: List to sort */
1451 int *pnList
/* IN/OUT: Number of elements in aList[] */
1454 int nList
; /* Number of elements in aList */
1455 ht_slot
*aList
; /* Pointer to sub-list content */
1458 const int nList
= *pnList
; /* Size of input list */
1459 int nMerge
= 0; /* Number of elements in list aMerge */
1460 ht_slot
*aMerge
= 0; /* List to be merged */
1461 int iList
; /* Index into input list */
1462 int iSub
= 0; /* Index into aSub array */
1463 struct Sublist aSub
[13]; /* Array of sub-lists */
1465 memset(aSub
, 0, sizeof(aSub
));
1466 assert( nList
<=HASHTABLE_NPAGE
&& nList
>0 );
1467 assert( HASHTABLE_NPAGE
==(1<<(ArraySize(aSub
)-1)) );
1469 for(iList
=0; iList
<nList
; iList
++){
1471 aMerge
= &aList
[iList
];
1472 for(iSub
=0; iList
& (1<<iSub
); iSub
++){
1473 struct Sublist
*p
= &aSub
[iSub
];
1474 assert( p
->aList
&& p
->nList
<=(1<<iSub
) );
1475 assert( p
->aList
==&aList
[iList
&~((2<<iSub
)-1)] );
1476 walMerge(aContent
, p
->aList
, p
->nList
, &aMerge
, &nMerge
, aBuffer
);
1478 aSub
[iSub
].aList
= aMerge
;
1479 aSub
[iSub
].nList
= nMerge
;
1482 for(iSub
++; iSub
<ArraySize(aSub
); iSub
++){
1483 if( nList
& (1<<iSub
) ){
1484 struct Sublist
*p
= &aSub
[iSub
];
1485 assert( p
->nList
<=(1<<iSub
) );
1486 assert( p
->aList
==&aList
[nList
&~((2<<iSub
)-1)] );
1487 walMerge(aContent
, p
->aList
, p
->nList
, &aMerge
, &nMerge
, aBuffer
);
1490 assert( aMerge
==aList
);
1496 for(i
=1; i
<*pnList
; i
++){
1497 assert( aContent
[aList
[i
]] > aContent
[aList
[i
-1]] );
1504 ** Free an iterator allocated by walIteratorInit().
1506 static void walIteratorFree(WalIterator
*p
){
1507 sqlite3ScratchFree(p
);
1511 ** Construct a WalInterator object that can be used to loop over all
1512 ** pages in the WAL in ascending order. The caller must hold the checkpoint
1515 ** On success, make *pp point to the newly allocated WalInterator object
1516 ** return SQLITE_OK. Otherwise, return an error code. If this routine
1517 ** returns an error, the value of *pp is undefined.
1519 ** The calling routine should invoke walIteratorFree() to destroy the
1520 ** WalIterator object when it has finished with it.
1522 static int walIteratorInit(Wal
*pWal
, WalIterator
**pp
){
1523 WalIterator
*p
; /* Return value */
1524 int nSegment
; /* Number of segments to merge */
1525 u32 iLast
; /* Last frame in log */
1526 int nByte
; /* Number of bytes to allocate */
1527 int i
; /* Iterator variable */
1528 ht_slot
*aTmp
; /* Temp space used by merge-sort */
1529 int rc
= SQLITE_OK
; /* Return Code */
1531 /* This routine only runs while holding the checkpoint lock. And
1532 ** it only runs if there is actually content in the log (mxFrame>0).
1534 assert( pWal
->ckptLock
&& pWal
->hdr
.mxFrame
>0 );
1535 iLast
= pWal
->hdr
.mxFrame
;
1537 /* Allocate space for the WalIterator object. */
1538 nSegment
= walFramePage(iLast
) + 1;
1539 nByte
= sizeof(WalIterator
)
1540 + (nSegment
-1)*sizeof(struct WalSegment
)
1541 + iLast
*sizeof(ht_slot
);
1542 p
= (WalIterator
*)sqlite3ScratchMalloc(nByte
);
1544 return SQLITE_NOMEM
;
1546 memset(p
, 0, nByte
);
1547 p
->nSegment
= nSegment
;
1549 /* Allocate temporary space used by the merge-sort routine. This block
1550 ** of memory will be freed before this function returns.
1552 aTmp
= (ht_slot
*)sqlite3ScratchMalloc(
1553 sizeof(ht_slot
) * (iLast
>HASHTABLE_NPAGE
?HASHTABLE_NPAGE
:iLast
)
1559 for(i
=0; rc
==SQLITE_OK
&& i
<nSegment
; i
++){
1560 volatile ht_slot
*aHash
;
1562 volatile u32
*aPgno
;
1564 rc
= walHashGet(pWal
, i
, &aHash
, &aPgno
, &iZero
);
1565 if( rc
==SQLITE_OK
){
1566 int j
; /* Counter variable */
1567 int nEntry
; /* Number of entries in this segment */
1568 ht_slot
*aIndex
; /* Sorted index for this segment */
1571 if( (i
+1)==nSegment
){
1572 nEntry
= (int)(iLast
- iZero
);
1574 nEntry
= (int)((u32
*)aHash
- (u32
*)aPgno
);
1576 aIndex
= &((ht_slot
*)&p
->aSegment
[p
->nSegment
])[iZero
];
1579 for(j
=0; j
<nEntry
; j
++){
1580 aIndex
[j
] = (ht_slot
)j
;
1582 walMergesort((u32
*)aPgno
, aTmp
, aIndex
, &nEntry
);
1583 p
->aSegment
[i
].iZero
= iZero
;
1584 p
->aSegment
[i
].nEntry
= nEntry
;
1585 p
->aSegment
[i
].aIndex
= aIndex
;
1586 p
->aSegment
[i
].aPgno
= (u32
*)aPgno
;
1589 sqlite3ScratchFree(aTmp
);
1591 if( rc
!=SQLITE_OK
){
1599 ** Attempt to obtain the exclusive WAL lock defined by parameters lockIdx and
1600 ** n. If the attempt fails and parameter xBusy is not NULL, then it is a
1601 ** busy-handler function. Invoke it and retry the lock until either the
1602 ** lock is successfully obtained or the busy-handler returns 0.
1604 static int walBusyLock(
1605 Wal
*pWal
, /* WAL connection */
1606 int (*xBusy
)(void*), /* Function to call when busy */
1607 void *pBusyArg
, /* Context argument for xBusyHandler */
1608 int lockIdx
, /* Offset of first byte to lock */
1609 int n
/* Number of bytes to lock */
1613 rc
= walLockExclusive(pWal
, lockIdx
, n
);
1614 }while( xBusy
&& rc
==SQLITE_BUSY
&& xBusy(pBusyArg
) );
1619 ** The cache of the wal-index header must be valid to call this function.
1620 ** Return the page-size in bytes used by the database.
1622 static int walPagesize(Wal
*pWal
){
1623 return (pWal
->hdr
.szPage
&0xfe00) + ((pWal
->hdr
.szPage
&0x0001)<<16);
1627 ** Copy as much content as we can from the WAL back into the database file
1628 ** in response to an sqlite3_wal_checkpoint() request or the equivalent.
1630 ** The amount of information copies from WAL to database might be limited
1631 ** by active readers. This routine will never overwrite a database page
1632 ** that a concurrent reader might be using.
1634 ** All I/O barrier operations (a.k.a fsyncs) occur in this routine when
1635 ** SQLite is in WAL-mode in synchronous=NORMAL. That means that if
1636 ** checkpoints are always run by a background thread or background
1637 ** process, foreground threads will never block on a lengthy fsync call.
1639 ** Fsync is called on the WAL before writing content out of the WAL and
1640 ** into the database. This ensures that if the new content is persistent
1641 ** in the WAL and can be recovered following a power-loss or hard reset.
1643 ** Fsync is also called on the database file if (and only if) the entire
1644 ** WAL content is copied into the database file. This second fsync makes
1645 ** it safe to delete the WAL since the new content will persist in the
1648 ** This routine uses and updates the nBackfill field of the wal-index header.
1649 ** This is the only routine tha will increase the value of nBackfill.
1650 ** (A WAL reset or recovery will revert nBackfill to zero, but not increase
1653 ** The caller must be holding sufficient locks to ensure that no other
1654 ** checkpoint is running (in any other thread or process) at the same
1657 static int walCheckpoint(
1658 Wal
*pWal
, /* Wal connection */
1659 int eMode
, /* One of PASSIVE, FULL or RESTART */
1660 int (*xBusyCall
)(void*), /* Function to call when busy */
1661 void *pBusyArg
, /* Context argument for xBusyHandler */
1662 int sync_flags
, /* Flags for OsSync() (or 0) */
1663 u8
*zBuf
/* Temporary buffer to use */
1665 int rc
; /* Return code */
1666 int szPage
; /* Database page-size */
1667 WalIterator
*pIter
= 0; /* Wal iterator context */
1668 u32 iDbpage
= 0; /* Next database page to write */
1669 u32 iFrame
= 0; /* Wal frame containing data for iDbpage */
1670 u32 mxSafeFrame
; /* Max frame that can be backfilled */
1671 u32 mxPage
; /* Max database page to write */
1672 int i
; /* Loop counter */
1673 volatile WalCkptInfo
*pInfo
; /* The checkpoint status information */
1674 int (*xBusy
)(void*) = 0; /* Function to call when waiting for locks */
1676 szPage
= walPagesize(pWal
);
1677 testcase( szPage
<=32768 );
1678 testcase( szPage
>=65536 );
1679 pInfo
= walCkptInfo(pWal
);
1680 if( pInfo
->nBackfill
>=pWal
->hdr
.mxFrame
) return SQLITE_OK
;
1682 /* Allocate the iterator */
1683 rc
= walIteratorInit(pWal
, &pIter
);
1684 if( rc
!=SQLITE_OK
){
1689 if( eMode
!=SQLITE_CHECKPOINT_PASSIVE
) xBusy
= xBusyCall
;
1691 /* Compute in mxSafeFrame the index of the last frame of the WAL that is
1692 ** safe to write into the database. Frames beyond mxSafeFrame might
1693 ** overwrite database pages that are in use by active readers and thus
1694 ** cannot be backfilled from the WAL.
1696 mxSafeFrame
= pWal
->hdr
.mxFrame
;
1697 mxPage
= pWal
->hdr
.nPage
;
1698 for(i
=1; i
<WAL_NREADER
; i
++){
1699 u32 y
= pInfo
->aReadMark
[i
];
1700 if( mxSafeFrame
>y
){
1701 assert( y
<=pWal
->hdr
.mxFrame
);
1702 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(i
), 1);
1703 if( rc
==SQLITE_OK
){
1704 pInfo
->aReadMark
[i
] = (i
==1 ? mxSafeFrame
: READMARK_NOT_USED
);
1705 walUnlockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
1706 }else if( rc
==SQLITE_BUSY
){
1710 goto walcheckpoint_out
;
1715 if( pInfo
->nBackfill
<mxSafeFrame
1716 && (rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(0), 1))==SQLITE_OK
1718 i64 nSize
; /* Current size of database file */
1719 u32 nBackfill
= pInfo
->nBackfill
;
1721 /* Sync the WAL to disk */
1723 rc
= sqlite3OsSync(pWal
->pWalFd
, sync_flags
);
1726 /* If the database may grow as a result of this checkpoint, hint
1727 ** about the eventual size of the db file to the VFS layer.
1729 if( rc
==SQLITE_OK
){
1730 i64 nReq
= ((i64
)mxPage
* szPage
);
1731 rc
= sqlite3OsFileSize(pWal
->pDbFd
, &nSize
);
1732 if( rc
==SQLITE_OK
&& nSize
<nReq
){
1733 sqlite3OsFileControlHint(pWal
->pDbFd
, SQLITE_FCNTL_SIZE_HINT
, &nReq
);
1738 /* Iterate through the contents of the WAL, copying data to the db file. */
1739 while( rc
==SQLITE_OK
&& 0==walIteratorNext(pIter
, &iDbpage
, &iFrame
) ){
1741 assert( walFramePgno(pWal
, iFrame
)==iDbpage
);
1742 if( iFrame
<=nBackfill
|| iFrame
>mxSafeFrame
|| iDbpage
>mxPage
) continue;
1743 iOffset
= walFrameOffset(iFrame
, szPage
) + WAL_FRAME_HDRSIZE
;
1744 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL file */
1745 rc
= sqlite3OsRead(pWal
->pWalFd
, zBuf
, szPage
, iOffset
);
1746 if( rc
!=SQLITE_OK
) break;
1747 iOffset
= (iDbpage
-1)*(i64
)szPage
;
1748 testcase( IS_BIG_INT(iOffset
) );
1749 rc
= sqlite3OsWrite(pWal
->pDbFd
, zBuf
, szPage
, iOffset
);
1750 if( rc
!=SQLITE_OK
) break;
1753 /* If work was actually accomplished... */
1754 if( rc
==SQLITE_OK
){
1755 if( mxSafeFrame
==walIndexHdr(pWal
)->mxFrame
){
1756 i64 szDb
= pWal
->hdr
.nPage
*(i64
)szPage
;
1757 testcase( IS_BIG_INT(szDb
) );
1758 rc
= sqlite3OsTruncate(pWal
->pDbFd
, szDb
);
1759 if( rc
==SQLITE_OK
&& sync_flags
){
1760 rc
= sqlite3OsSync(pWal
->pDbFd
, sync_flags
);
1763 if( rc
==SQLITE_OK
){
1764 pInfo
->nBackfill
= mxSafeFrame
;
1768 /* Release the reader lock held while backfilling */
1769 walUnlockExclusive(pWal
, WAL_READ_LOCK(0), 1);
1772 if( rc
==SQLITE_BUSY
){
1773 /* Reset the return code so as not to report a checkpoint failure
1774 ** just because there are active readers. */
1778 /* If this is an SQLITE_CHECKPOINT_RESTART operation, and the entire wal
1779 ** file has been copied into the database file, then block until all
1780 ** readers have finished using the wal file. This ensures that the next
1781 ** process to write to the database restarts the wal file.
1783 if( rc
==SQLITE_OK
&& eMode
!=SQLITE_CHECKPOINT_PASSIVE
){
1784 assert( pWal
->writeLock
);
1785 if( pInfo
->nBackfill
<pWal
->hdr
.mxFrame
){
1787 }else if( eMode
==SQLITE_CHECKPOINT_RESTART
){
1788 assert( mxSafeFrame
==pWal
->hdr
.mxFrame
);
1789 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1790 if( rc
==SQLITE_OK
){
1791 walUnlockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1797 walIteratorFree(pIter
);
1802 ** If the WAL file is currently larger than nMax bytes in size, truncate
1803 ** it to exactly nMax bytes. If an error occurs while doing so, ignore it.
1805 static void walLimitSize(Wal
*pWal
, i64 nMax
){
1808 sqlite3BeginBenignMalloc();
1809 rx
= sqlite3OsFileSize(pWal
->pWalFd
, &sz
);
1810 if( rx
==SQLITE_OK
&& (sz
> nMax
) ){
1811 rx
= sqlite3OsTruncate(pWal
->pWalFd
, nMax
);
1813 sqlite3EndBenignMalloc();
1815 sqlite3_log(rx
, "cannot limit WAL size: %s", pWal
->zWalName
);
1820 ** Close a connection to a log file.
1822 int sqlite3WalClose(
1823 Wal
*pWal
, /* Wal to close */
1824 int sync_flags
, /* Flags to pass to OsSync() (or 0) */
1826 u8
*zBuf
/* Buffer of at least nBuf bytes */
1830 int isDelete
= 0; /* True to unlink wal and wal-index files */
1832 /* If an EXCLUSIVE lock can be obtained on the database file (using the
1833 ** ordinary, rollback-mode locking methods, this guarantees that the
1834 ** connection associated with this log file is the only connection to
1835 ** the database. In this case checkpoint the database and unlink both
1836 ** the wal and wal-index files.
1838 ** The EXCLUSIVE lock is not released before returning.
1840 rc
= sqlite3OsLock(pWal
->pDbFd
, SQLITE_LOCK_EXCLUSIVE
);
1841 if( rc
==SQLITE_OK
){
1842 if( pWal
->exclusiveMode
==WAL_NORMAL_MODE
){
1843 pWal
->exclusiveMode
= WAL_EXCLUSIVE_MODE
;
1845 rc
= sqlite3WalCheckpoint(
1846 pWal
, SQLITE_CHECKPOINT_PASSIVE
, 0, 0, sync_flags
, nBuf
, zBuf
, 0, 0
1848 if( rc
==SQLITE_OK
){
1850 sqlite3OsFileControlHint(
1851 pWal
->pDbFd
, SQLITE_FCNTL_PERSIST_WAL
, &bPersist
1854 /* Try to delete the WAL file if the checkpoint completed and
1855 ** fsyned (rc==SQLITE_OK) and if we are not in persistent-wal
1856 ** mode (!bPersist) */
1858 }else if( pWal
->mxWalSize
>=0 ){
1859 /* Try to truncate the WAL file to zero bytes if the checkpoint
1860 ** completed and fsynced (rc==SQLITE_OK) and we are in persistent
1861 ** WAL mode (bPersist) and if the PRAGMA journal_size_limit is a
1862 ** non-negative value (pWal->mxWalSize>=0). Note that we truncate
1863 ** to zero bytes as truncating to the journal_size_limit might
1864 ** leave a corrupt WAL file on disk. */
1865 walLimitSize(pWal
, 0);
1870 walIndexClose(pWal
, isDelete
);
1871 sqlite3OsClose(pWal
->pWalFd
);
1873 sqlite3BeginBenignMalloc();
1874 sqlite3OsDelete(pWal
->pVfs
, pWal
->zWalName
, 0);
1875 sqlite3EndBenignMalloc();
1877 WALTRACE(("WAL%p: closed\n", pWal
));
1878 sqlite3_free((void *)pWal
->apWiData
);
1885 ** Try to read the wal-index header. Return 0 on success and 1 if
1886 ** there is a problem.
1888 ** The wal-index is in shared memory. Another thread or process might
1889 ** be writing the header at the same time this procedure is trying to
1890 ** read it, which might result in inconsistency. A dirty read is detected
1891 ** by verifying that both copies of the header are the same and also by
1892 ** a checksum on the header.
1894 ** If and only if the read is consistent and the header is different from
1895 ** pWal->hdr, then pWal->hdr is updated to the content of the new header
1896 ** and *pChanged is set to 1.
1898 ** If the checksum cannot be verified return non-zero. If the header
1899 ** is read successfully and the checksum verified, return zero.
1901 static int walIndexTryHdr(Wal
*pWal
, int *pChanged
){
1902 u32 aCksum
[2]; /* Checksum on the header content */
1903 WalIndexHdr h1
, h2
; /* Two copies of the header content */
1904 WalIndexHdr
volatile *aHdr
; /* Header in shared memory */
1906 /* The first page of the wal-index must be mapped at this point. */
1907 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
1909 /* Read the header. This might happen concurrently with a write to the
1910 ** same area of shared memory on a different CPU in a SMP,
1911 ** meaning it is possible that an inconsistent snapshot is read
1912 ** from the file. If this happens, return non-zero.
1914 ** There are two copies of the header at the beginning of the wal-index.
1915 ** When reading, read [0] first then [1]. Writes are in the reverse order.
1916 ** Memory barriers are used to prevent the compiler or the hardware from
1917 ** reordering the reads and writes.
1919 aHdr
= walIndexHdr(pWal
);
1920 memcpy(&h1
, (void *)&aHdr
[0], sizeof(h1
));
1921 walShmBarrier(pWal
);
1922 memcpy(&h2
, (void *)&aHdr
[1], sizeof(h2
));
1924 if( memcmp(&h1
, &h2
, sizeof(h1
))!=0 ){
1925 return 1; /* Dirty read */
1928 return 1; /* Malformed header - probably all zeros */
1930 walChecksumBytes(1, (u8
*)&h1
, sizeof(h1
)-sizeof(h1
.aCksum
), 0, aCksum
);
1931 if( aCksum
[0]!=h1
.aCksum
[0] || aCksum
[1]!=h1
.aCksum
[1] ){
1932 return 1; /* Checksum does not match */
1935 if( memcmp(&pWal
->hdr
, &h1
, sizeof(WalIndexHdr
)) ){
1937 memcpy(&pWal
->hdr
, &h1
, sizeof(WalIndexHdr
));
1938 pWal
->szPage
= (pWal
->hdr
.szPage
&0xfe00) + ((pWal
->hdr
.szPage
&0x0001)<<16);
1939 testcase( pWal
->szPage
<=32768 );
1940 testcase( pWal
->szPage
>=65536 );
1943 /* The header was successfully read. Return zero. */
1948 ** Read the wal-index header from the wal-index and into pWal->hdr.
1949 ** If the wal-header appears to be corrupt, try to reconstruct the
1950 ** wal-index from the WAL before returning.
1952 ** Set *pChanged to 1 if the wal-index header value in pWal->hdr is
1953 ** changed by this opertion. If pWal->hdr is unchanged, set *pChanged
1956 ** If the wal-index header is successfully read, return SQLITE_OK.
1957 ** Otherwise an SQLite error code.
1959 static int walIndexReadHdr(Wal
*pWal
, int *pChanged
){
1960 int rc
; /* Return code */
1961 int badHdr
; /* True if a header read failed */
1962 volatile u32
*page0
; /* Chunk of wal-index containing header */
1964 /* Ensure that page 0 of the wal-index (the page that contains the
1965 ** wal-index header) is mapped. Return early if an error occurs here.
1968 rc
= walIndexPage(pWal
, 0, &page0
);
1969 if( rc
!=SQLITE_OK
){
1972 assert( page0
|| pWal
->writeLock
==0 );
1974 /* If the first page of the wal-index has been mapped, try to read the
1975 ** wal-index header immediately, without holding any lock. This usually
1976 ** works, but may fail if the wal-index header is corrupt or currently
1977 ** being modified by another thread or process.
1979 badHdr
= (page0
? walIndexTryHdr(pWal
, pChanged
) : 1);
1981 /* If the first attempt failed, it might have been due to a race
1982 ** with a writer. So get a WRITE lock and try again.
1984 assert( badHdr
==0 || pWal
->writeLock
==0 );
1986 if( pWal
->readOnly
& WAL_SHM_RDONLY
){
1987 if( SQLITE_OK
==(rc
= walLockShared(pWal
, WAL_WRITE_LOCK
)) ){
1988 walUnlockShared(pWal
, WAL_WRITE_LOCK
);
1989 rc
= SQLITE_READONLY_RECOVERY
;
1991 }else if( SQLITE_OK
==(rc
= walLockExclusive(pWal
, WAL_WRITE_LOCK
, 1)) ){
1992 pWal
->writeLock
= 1;
1993 if( SQLITE_OK
==(rc
= walIndexPage(pWal
, 0, &page0
)) ){
1994 badHdr
= walIndexTryHdr(pWal
, pChanged
);
1996 /* If the wal-index header is still malformed even while holding
1997 ** a WRITE lock, it can only mean that the header is corrupted and
1998 ** needs to be reconstructed. So run recovery to do exactly that.
2000 rc
= walIndexRecover(pWal
);
2004 pWal
->writeLock
= 0;
2005 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2009 /* If the header is read successfully, check the version number to make
2010 ** sure the wal-index was not constructed with some future format that
2011 ** this version of SQLite cannot understand.
2013 if( badHdr
==0 && pWal
->hdr
.iVersion
!=WALINDEX_MAX_VERSION
){
2014 rc
= SQLITE_CANTOPEN_BKPT
;
2021 ** This is the value that walTryBeginRead returns when it needs to
2024 #define WAL_RETRY (-1)
2027 ** Attempt to start a read transaction. This might fail due to a race or
2028 ** other transient condition. When that happens, it returns WAL_RETRY to
2029 ** indicate to the caller that it is safe to retry immediately.
2031 ** On success return SQLITE_OK. On a permanent failure (such an
2032 ** I/O error or an SQLITE_BUSY because another process is running
2033 ** recovery) return a positive error code.
2035 ** The useWal parameter is true to force the use of the WAL and disable
2036 ** the case where the WAL is bypassed because it has been completely
2037 ** checkpointed. If useWal==0 then this routine calls walIndexReadHdr()
2038 ** to make a copy of the wal-index header into pWal->hdr. If the
2039 ** wal-index header has changed, *pChanged is set to 1 (as an indication
2040 ** to the caller that the local paget cache is obsolete and needs to be
2041 ** flushed.) When useWal==1, the wal-index header is assumed to already
2042 ** be loaded and the pChanged parameter is unused.
2044 ** The caller must set the cnt parameter to the number of prior calls to
2045 ** this routine during the current read attempt that returned WAL_RETRY.
2046 ** This routine will start taking more aggressive measures to clear the
2047 ** race conditions after multiple WAL_RETRY returns, and after an excessive
2048 ** number of errors will ultimately return SQLITE_PROTOCOL. The
2049 ** SQLITE_PROTOCOL return indicates that some other process has gone rogue
2050 ** and is not honoring the locking protocol. There is a vanishingly small
2051 ** chance that SQLITE_PROTOCOL could be returned because of a run of really
2052 ** bad luck when there is lots of contention for the wal-index, but that
2053 ** possibility is so small that it can be safely neglected, we believe.
2055 ** On success, this routine obtains a read lock on
2056 ** WAL_READ_LOCK(pWal->readLock). The pWal->readLock integer is
2057 ** in the range 0 <= pWal->readLock < WAL_NREADER. If pWal->readLock==(-1)
2058 ** that means the Wal does not hold any read lock. The reader must not
2059 ** access any database page that is modified by a WAL frame up to and
2060 ** including frame number aReadMark[pWal->readLock]. The reader will
2061 ** use WAL frames up to and including pWal->hdr.mxFrame if pWal->readLock>0
2062 ** Or if pWal->readLock==0, then the reader will ignore the WAL
2063 ** completely and get all content directly from the database file.
2064 ** If the useWal parameter is 1 then the WAL will never be ignored and
2065 ** this routine will always set pWal->readLock>0 on success.
2066 ** When the read transaction is completed, the caller must release the
2067 ** lock on WAL_READ_LOCK(pWal->readLock) and set pWal->readLock to -1.
2069 ** This routine uses the nBackfill and aReadMark[] fields of the header
2070 ** to select a particular WAL_READ_LOCK() that strives to let the
2071 ** checkpoint process do as much work as possible. This routine might
2072 ** update values of the aReadMark[] array in the header, but if it does
2073 ** so it takes care to hold an exclusive lock on the corresponding
2074 ** WAL_READ_LOCK() while changing values.
2076 static int walTryBeginRead(Wal
*pWal
, int *pChanged
, int useWal
, int cnt
){
2077 volatile WalCkptInfo
*pInfo
; /* Checkpoint information in wal-index */
2078 u32 mxReadMark
; /* Largest aReadMark[] value */
2079 int mxI
; /* Index of largest aReadMark[] value */
2080 int i
; /* Loop counter */
2081 int rc
= SQLITE_OK
; /* Return code */
2083 assert( pWal
->readLock
<0 ); /* Not currently locked */
2085 /* Take steps to avoid spinning forever if there is a protocol error.
2087 ** Circumstances that cause a RETRY should only last for the briefest
2088 ** instances of time. No I/O or other system calls are done while the
2089 ** locks are held, so the locks should not be held for very long. But
2090 ** if we are unlucky, another process that is holding a lock might get
2091 ** paged out or take a page-fault that is time-consuming to resolve,
2092 ** during the few nanoseconds that it is holding the lock. In that case,
2093 ** it might take longer than normal for the lock to free.
2095 ** After 5 RETRYs, we begin calling sqlite3OsSleep(). The first few
2096 ** calls to sqlite3OsSleep() have a delay of 1 microsecond. Really this
2097 ** is more of a scheduler yield than an actual delay. But on the 10th
2098 ** an subsequent retries, the delays start becoming longer and longer,
2099 ** so that on the 100th (and last) RETRY we delay for 21 milliseconds.
2100 ** The total delay time before giving up is less than 1 second.
2103 int nDelay
= 1; /* Pause time in microseconds */
2105 VVA_ONLY( pWal
->lockError
= 1; )
2106 return SQLITE_PROTOCOL
;
2108 if( cnt
>=10 ) nDelay
= (cnt
-9)*238; /* Max delay 21ms. Total delay 996ms */
2109 sqlite3OsSleep(pWal
->pVfs
, nDelay
);
2113 rc
= walIndexReadHdr(pWal
, pChanged
);
2114 if( rc
==SQLITE_BUSY
){
2115 /* If there is not a recovery running in another thread or process
2116 ** then convert BUSY errors to WAL_RETRY. If recovery is known to
2117 ** be running, convert BUSY to BUSY_RECOVERY. There is a race here
2118 ** which might cause WAL_RETRY to be returned even if BUSY_RECOVERY
2119 ** would be technically correct. But the race is benign since with
2120 ** WAL_RETRY this routine will be called again and will probably be
2121 ** right on the second iteration.
2123 if( pWal
->apWiData
[0]==0 ){
2124 /* This branch is taken when the xShmMap() method returns SQLITE_BUSY.
2125 ** We assume this is a transient condition, so return WAL_RETRY. The
2126 ** xShmMap() implementation used by the default unix and win32 VFS
2127 ** modules may return SQLITE_BUSY due to a race condition in the
2128 ** code that determines whether or not the shared-memory region
2129 ** must be zeroed before the requested page is returned.
2132 }else if( SQLITE_OK
==(rc
= walLockShared(pWal
, WAL_RECOVER_LOCK
)) ){
2133 walUnlockShared(pWal
, WAL_RECOVER_LOCK
);
2135 }else if( rc
==SQLITE_BUSY
){
2136 rc
= SQLITE_BUSY_RECOVERY
;
2139 if( rc
!=SQLITE_OK
){
2144 pInfo
= walCkptInfo(pWal
);
2145 if( !useWal
&& pInfo
->nBackfill
==pWal
->hdr
.mxFrame
){
2146 /* The WAL has been completely backfilled (or it is empty).
2147 ** and can be safely ignored.
2149 rc
= walLockShared(pWal
, WAL_READ_LOCK(0));
2150 walShmBarrier(pWal
);
2151 if( rc
==SQLITE_OK
){
2152 if( memcmp((void *)walIndexHdr(pWal
), &pWal
->hdr
, sizeof(WalIndexHdr
)) ){
2153 /* It is not safe to allow the reader to continue here if frames
2154 ** may have been appended to the log before READ_LOCK(0) was obtained.
2155 ** When holding READ_LOCK(0), the reader ignores the entire log file,
2156 ** which implies that the database file contains a trustworthy
2157 ** snapshoT. Since holding READ_LOCK(0) prevents a checkpoint from
2158 ** happening, this is usually correct.
2160 ** However, if frames have been appended to the log (or if the log
2161 ** is wrapped and written for that matter) before the READ_LOCK(0)
2162 ** is obtained, that is not necessarily true. A checkpointer may
2163 ** have started to backfill the appended frames but crashed before
2164 ** it finished. Leaving a corrupt image in the database file.
2166 walUnlockShared(pWal
, WAL_READ_LOCK(0));
2171 }else if( rc
!=SQLITE_BUSY
){
2176 /* If we get this far, it means that the reader will want to use
2177 ** the WAL to get at content from recent commits. The job now is
2178 ** to select one of the aReadMark[] entries that is closest to
2179 ** but not exceeding pWal->hdr.mxFrame and lock that entry.
2183 for(i
=1; i
<WAL_NREADER
; i
++){
2184 u32 thisMark
= pInfo
->aReadMark
[i
];
2185 if( mxReadMark
<=thisMark
&& thisMark
<=pWal
->hdr
.mxFrame
){
2186 assert( thisMark
!=READMARK_NOT_USED
);
2187 mxReadMark
= thisMark
;
2191 /* There was once an "if" here. The extra "{" is to preserve indentation. */
2193 if( (pWal
->readOnly
& WAL_SHM_RDONLY
)==0
2194 && (mxReadMark
<pWal
->hdr
.mxFrame
|| mxI
==0)
2196 for(i
=1; i
<WAL_NREADER
; i
++){
2197 rc
= walLockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
2198 if( rc
==SQLITE_OK
){
2199 mxReadMark
= pInfo
->aReadMark
[i
] = pWal
->hdr
.mxFrame
;
2201 walUnlockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
2203 }else if( rc
!=SQLITE_BUSY
){
2209 assert( rc
==SQLITE_BUSY
|| (pWal
->readOnly
& WAL_SHM_RDONLY
)!=0 );
2210 return rc
==SQLITE_BUSY
? WAL_RETRY
: SQLITE_READONLY_CANTLOCK
;
2213 rc
= walLockShared(pWal
, WAL_READ_LOCK(mxI
));
2215 return rc
==SQLITE_BUSY
? WAL_RETRY
: rc
;
2217 /* Now that the read-lock has been obtained, check that neither the
2218 ** value in the aReadMark[] array or the contents of the wal-index
2219 ** header have changed.
2221 ** It is necessary to check that the wal-index header did not change
2222 ** between the time it was read and when the shared-lock was obtained
2223 ** on WAL_READ_LOCK(mxI) was obtained to account for the possibility
2224 ** that the log file may have been wrapped by a writer, or that frames
2225 ** that occur later in the log than pWal->hdr.mxFrame may have been
2226 ** copied into the database by a checkpointer. If either of these things
2227 ** happened, then reading the database with the current value of
2228 ** pWal->hdr.mxFrame risks reading a corrupted snapshot. So, retry
2231 ** This does not guarantee that the copy of the wal-index header is up to
2232 ** date before proceeding. That would not be possible without somehow
2233 ** blocking writers. It only guarantees that a dangerous checkpoint or
2234 ** log-wrap (either of which would require an exclusive lock on
2235 ** WAL_READ_LOCK(mxI)) has not occurred since the snapshot was valid.
2237 walShmBarrier(pWal
);
2238 if( pInfo
->aReadMark
[mxI
]!=mxReadMark
2239 || memcmp((void *)walIndexHdr(pWal
), &pWal
->hdr
, sizeof(WalIndexHdr
))
2241 walUnlockShared(pWal
, WAL_READ_LOCK(mxI
));
2244 assert( mxReadMark
<=pWal
->hdr
.mxFrame
);
2245 pWal
->readLock
= (i16
)mxI
;
2252 ** Begin a read transaction on the database.
2254 ** This routine used to be called sqlite3OpenSnapshot() and with good reason:
2255 ** it takes a snapshot of the state of the WAL and wal-index for the current
2256 ** instant in time. The current thread will continue to use this snapshot.
2257 ** Other threads might append new content to the WAL and wal-index but
2258 ** that extra content is ignored by the current thread.
2260 ** If the database contents have changes since the previous read
2261 ** transaction, then *pChanged is set to 1 before returning. The
2262 ** Pager layer will use this to know that is cache is stale and
2263 ** needs to be flushed.
2265 int sqlite3WalBeginReadTransaction(Wal
*pWal
, int *pChanged
){
2266 int rc
; /* Return code */
2267 int cnt
= 0; /* Number of TryBeginRead attempts */
2270 rc
= walTryBeginRead(pWal
, pChanged
, 0, ++cnt
);
2271 }while( rc
==WAL_RETRY
);
2272 testcase( (rc
&0xff)==SQLITE_BUSY
);
2273 testcase( (rc
&0xff)==SQLITE_IOERR
);
2274 testcase( rc
==SQLITE_PROTOCOL
);
2275 testcase( rc
==SQLITE_OK
);
2280 ** Finish with a read transaction. All this does is release the
2283 void sqlite3WalEndReadTransaction(Wal
*pWal
){
2284 sqlite3WalEndWriteTransaction(pWal
);
2285 if( pWal
->readLock
>=0 ){
2286 walUnlockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
));
2287 pWal
->readLock
= -1;
2292 ** Search the wal file for page pgno. If found, set *piRead to the frame that
2293 ** contains the page. Otherwise, if pgno is not in the wal file, set *piRead
2296 ** Return SQLITE_OK if successful, or an error code if an error occurs. If an
2297 ** error does occur, the final value of *piRead is undefined.
2299 int sqlite3WalFindFrame(
2300 Wal
*pWal
, /* WAL handle */
2301 Pgno pgno
, /* Database page number to read data for */
2302 u32
*piRead
/* OUT: Frame number (or zero) */
2304 u32 iRead
= 0; /* If !=0, WAL frame to return data from */
2305 u32 iLast
= pWal
->hdr
.mxFrame
; /* Last page in WAL for this reader */
2306 int iHash
; /* Used to loop through N hash tables */
2308 /* This routine is only be called from within a read transaction. */
2309 assert( pWal
->readLock
>=0 || pWal
->lockError
);
2311 /* If the "last page" field of the wal-index header snapshot is 0, then
2312 ** no data will be read from the wal under any circumstances. Return early
2313 ** in this case as an optimization. Likewise, if pWal->readLock==0,
2314 ** then the WAL is ignored by the reader so return early, as if the
2317 if( iLast
==0 || pWal
->readLock
==0 ){
2322 /* Search the hash table or tables for an entry matching page number
2323 ** pgno. Each iteration of the following for() loop searches one
2324 ** hash table (each hash table indexes up to HASHTABLE_NPAGE frames).
2326 ** This code might run concurrently to the code in walIndexAppend()
2327 ** that adds entries to the wal-index (and possibly to this hash
2328 ** table). This means the value just read from the hash
2329 ** slot (aHash[iKey]) may have been added before or after the
2330 ** current read transaction was opened. Values added after the
2331 ** read transaction was opened may have been written incorrectly -
2332 ** i.e. these slots may contain garbage data. However, we assume
2333 ** that any slots written before the current read transaction was
2334 ** opened remain unmodified.
2336 ** For the reasons above, the if(...) condition featured in the inner
2337 ** loop of the following block is more stringent that would be required
2338 ** if we had exclusive access to the hash-table:
2340 ** (aPgno[iFrame]==pgno):
2341 ** This condition filters out normal hash-table collisions.
2344 ** This condition filters out entries that were added to the hash
2345 ** table after the current read-transaction had started.
2347 for(iHash
=walFramePage(iLast
); iHash
>=0 && iRead
==0; iHash
--){
2348 volatile ht_slot
*aHash
; /* Pointer to hash table */
2349 volatile u32
*aPgno
; /* Pointer to array of page numbers */
2350 u32 iZero
; /* Frame number corresponding to aPgno[0] */
2351 int iKey
; /* Hash slot index */
2352 int nCollide
; /* Number of hash collisions remaining */
2353 int rc
; /* Error code */
2355 rc
= walHashGet(pWal
, iHash
, &aHash
, &aPgno
, &iZero
);
2356 if( rc
!=SQLITE_OK
){
2359 nCollide
= HASHTABLE_NSLOT
;
2360 for(iKey
=walHash(pgno
); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
2361 u32 iFrame
= aHash
[iKey
] + iZero
;
2362 if( iFrame
<=iLast
&& aPgno
[aHash
[iKey
]]==pgno
){
2363 /* assert( iFrame>iRead ); -- not true if there is corruption */
2366 if( (nCollide
--)==0 ){
2367 return SQLITE_CORRUPT_BKPT
;
2372 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
2373 /* If expensive assert() statements are available, do a linear search
2374 ** of the wal-index file content. Make sure the results agree with the
2375 ** result obtained using the hash indexes above. */
2379 for(iTest
=iLast
; iTest
>0; iTest
--){
2380 if( walFramePgno(pWal
, iTest
)==pgno
){
2385 assert( iRead
==iRead2
);
2394 ** Read the contents of frame iRead from the wal file into buffer pOut
2395 ** (which is nOut bytes in size). Return SQLITE_OK if successful, or an
2396 ** error code otherwise.
2398 int sqlite3WalReadFrame(
2399 Wal
*pWal
, /* WAL handle */
2400 u32 iRead
, /* Frame to read */
2401 int nOut
, /* Size of buffer pOut in bytes */
2402 u8
*pOut
/* Buffer to write page data to */
2406 sz
= pWal
->hdr
.szPage
;
2407 sz
= (sz
&0xfe00) + ((sz
&0x0001)<<16);
2408 testcase( sz
<=32768 );
2409 testcase( sz
>=65536 );
2410 iOffset
= walFrameOffset(iRead
, sz
) + WAL_FRAME_HDRSIZE
;
2411 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL */
2412 return sqlite3OsRead(pWal
->pWalFd
, pOut
, (nOut
>sz
? sz
: nOut
), iOffset
);
2416 ** Return the size of the database in pages (or zero, if unknown).
2418 Pgno
sqlite3WalDbsize(Wal
*pWal
){
2419 if( pWal
&& ALWAYS(pWal
->readLock
>=0) ){
2420 return pWal
->hdr
.nPage
;
2427 ** This function starts a write transaction on the WAL.
2429 ** A read transaction must have already been started by a prior call
2430 ** to sqlite3WalBeginReadTransaction().
2432 ** If another thread or process has written into the database since
2433 ** the read transaction was started, then it is not possible for this
2434 ** thread to write as doing so would cause a fork. So this routine
2435 ** returns SQLITE_BUSY in that case and no write transaction is started.
2437 ** There can only be a single writer active at a time.
2439 int sqlite3WalBeginWriteTransaction(Wal
*pWal
){
2442 /* Cannot start a write transaction without first holding a read
2444 assert( pWal
->readLock
>=0 );
2446 if( pWal
->readOnly
){
2447 return SQLITE_READONLY
;
2450 /* Only one writer allowed at a time. Get the write lock. Return
2451 ** SQLITE_BUSY if unable.
2453 rc
= walLockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2457 pWal
->writeLock
= 1;
2459 /* If another connection has written to the database file since the
2460 ** time the read transaction on this connection was started, then
2461 ** the write is disallowed.
2463 if( memcmp(&pWal
->hdr
, (void *)walIndexHdr(pWal
), sizeof(WalIndexHdr
))!=0 ){
2464 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2465 pWal
->writeLock
= 0;
2466 rc
= SQLITE_BUSY_SNAPSHOT
;
2473 ** End a write transaction. The commit has already been done. This
2474 ** routine merely releases the lock.
2476 int sqlite3WalEndWriteTransaction(Wal
*pWal
){
2477 if( pWal
->writeLock
){
2478 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2479 pWal
->writeLock
= 0;
2480 pWal
->truncateOnCommit
= 0;
2486 ** If any data has been written (but not committed) to the log file, this
2487 ** function moves the write-pointer back to the start of the transaction.
2489 ** Additionally, the callback function is invoked for each frame written
2490 ** to the WAL since the start of the transaction. If the callback returns
2491 ** other than SQLITE_OK, it is not invoked again and the error code is
2492 ** returned to the caller.
2494 ** Otherwise, if the callback function does not return an error, this
2495 ** function returns SQLITE_OK.
2497 int sqlite3WalUndo(Wal
*pWal
, int (*xUndo
)(void *, Pgno
), void *pUndoCtx
){
2499 if( ALWAYS(pWal
->writeLock
) ){
2500 Pgno iMax
= pWal
->hdr
.mxFrame
;
2503 /* Restore the clients cache of the wal-index header to the state it
2504 ** was in before the client began writing to the database.
2506 memcpy(&pWal
->hdr
, (void *)walIndexHdr(pWal
), sizeof(WalIndexHdr
));
2508 for(iFrame
=pWal
->hdr
.mxFrame
+1;
2509 ALWAYS(rc
==SQLITE_OK
) && iFrame
<=iMax
;
2512 /* This call cannot fail. Unless the page for which the page number
2513 ** is passed as the second argument is (a) in the cache and
2514 ** (b) has an outstanding reference, then xUndo is either a no-op
2515 ** (if (a) is false) or simply expels the page from the cache (if (b)
2518 ** If the upper layer is doing a rollback, it is guaranteed that there
2519 ** are no outstanding references to any page other than page 1. And
2520 ** page 1 is never written to the log until the transaction is
2521 ** committed. As a result, the call to xUndo may not fail.
2523 assert( walFramePgno(pWal
, iFrame
)!=1 );
2524 rc
= xUndo(pUndoCtx
, walFramePgno(pWal
, iFrame
));
2526 if( iMax
!=pWal
->hdr
.mxFrame
) walCleanupHash(pWal
);
2528 assert( rc
==SQLITE_OK
);
2533 ** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32
2534 ** values. This function populates the array with values required to
2535 ** "rollback" the write position of the WAL handle back to the current
2536 ** point in the event of a savepoint rollback (via WalSavepointUndo()).
2538 void sqlite3WalSavepoint(Wal
*pWal
, u32
*aWalData
){
2539 assert( pWal
->writeLock
);
2540 aWalData
[0] = pWal
->hdr
.mxFrame
;
2541 aWalData
[1] = pWal
->hdr
.aFrameCksum
[0];
2542 aWalData
[2] = pWal
->hdr
.aFrameCksum
[1];
2543 aWalData
[3] = pWal
->nCkpt
;
2547 ** Move the write position of the WAL back to the point identified by
2548 ** the values in the aWalData[] array. aWalData must point to an array
2549 ** of WAL_SAVEPOINT_NDATA u32 values that has been previously populated
2550 ** by a call to WalSavepoint().
2552 int sqlite3WalSavepointUndo(Wal
*pWal
, u32
*aWalData
){
2555 assert( pWal
->writeLock
);
2556 assert( aWalData
[3]!=pWal
->nCkpt
|| aWalData
[0]<=pWal
->hdr
.mxFrame
);
2558 if( aWalData
[3]!=pWal
->nCkpt
){
2559 /* This savepoint was opened immediately after the write-transaction
2560 ** was started. Right after that, the writer decided to wrap around
2561 ** to the start of the log. Update the savepoint values to match.
2564 aWalData
[3] = pWal
->nCkpt
;
2567 if( aWalData
[0]<pWal
->hdr
.mxFrame
){
2568 pWal
->hdr
.mxFrame
= aWalData
[0];
2569 pWal
->hdr
.aFrameCksum
[0] = aWalData
[1];
2570 pWal
->hdr
.aFrameCksum
[1] = aWalData
[2];
2571 walCleanupHash(pWal
);
2579 ** This function is called just before writing a set of frames to the log
2580 ** file (see sqlite3WalFrames()). It checks to see if, instead of appending
2581 ** to the current log file, it is possible to overwrite the start of the
2582 ** existing log file with the new frames (i.e. "reset" the log). If so,
2583 ** it sets pWal->hdr.mxFrame to 0. Otherwise, pWal->hdr.mxFrame is left
2586 ** SQLITE_OK is returned if no error is encountered (regardless of whether
2587 ** or not pWal->hdr.mxFrame is modified). An SQLite error code is returned
2588 ** if an error occurs.
2590 static int walRestartLog(Wal
*pWal
){
2594 if( pWal
->readLock
==0 ){
2595 volatile WalCkptInfo
*pInfo
= walCkptInfo(pWal
);
2596 assert( pInfo
->nBackfill
==pWal
->hdr
.mxFrame
);
2597 if( pInfo
->nBackfill
>0 ){
2599 sqlite3_randomness(4, &salt1
);
2600 rc
= walLockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
2601 if( rc
==SQLITE_OK
){
2602 /* If all readers are using WAL_READ_LOCK(0) (in other words if no
2603 ** readers are currently using the WAL), then the transactions
2604 ** frames will overwrite the start of the existing log. Update the
2605 ** wal-index header to reflect this.
2607 ** In theory it would be Ok to update the cache of the header only
2608 ** at this point. But updating the actual wal-index header is also
2609 ** safe and means there is no special case for sqlite3WalUndo()
2610 ** to handle if this transaction is rolled back.
2612 int i
; /* Loop counter */
2613 u32
*aSalt
= pWal
->hdr
.aSalt
; /* Big-endian salt values */
2616 pWal
->hdr
.mxFrame
= 0;
2617 sqlite3Put4byte((u8
*)&aSalt
[0], 1 + sqlite3Get4byte((u8
*)&aSalt
[0]));
2619 walIndexWriteHdr(pWal
);
2620 pInfo
->nBackfill
= 0;
2621 pInfo
->aReadMark
[1] = 0;
2622 for(i
=2; i
<WAL_NREADER
; i
++) pInfo
->aReadMark
[i
] = READMARK_NOT_USED
;
2623 assert( pInfo
->aReadMark
[0]==0 );
2624 walUnlockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
2625 }else if( rc
!=SQLITE_BUSY
){
2629 walUnlockShared(pWal
, WAL_READ_LOCK(0));
2630 pWal
->readLock
= -1;
2634 rc
= walTryBeginRead(pWal
, ¬Used
, 1, ++cnt
);
2635 }while( rc
==WAL_RETRY
);
2636 assert( (rc
&0xff)!=SQLITE_BUSY
); /* BUSY not possible when useWal==1 */
2637 testcase( (rc
&0xff)==SQLITE_IOERR
);
2638 testcase( rc
==SQLITE_PROTOCOL
);
2639 testcase( rc
==SQLITE_OK
);
2645 ** Information about the current state of the WAL file and where
2646 ** the next fsync should occur - passed from sqlite3WalFrames() into
2649 typedef struct WalWriter
{
2650 Wal
*pWal
; /* The complete WAL information */
2651 sqlite3_file
*pFd
; /* The WAL file to which we write */
2652 sqlite3_int64 iSyncPoint
; /* Fsync at this offset */
2653 int syncFlags
; /* Flags for the fsync */
2654 int szPage
; /* Size of one page */
2658 ** Write iAmt bytes of content into the WAL file beginning at iOffset.
2659 ** Do a sync when crossing the p->iSyncPoint boundary.
2661 ** In other words, if iSyncPoint is in between iOffset and iOffset+iAmt,
2662 ** first write the part before iSyncPoint, then sync, then write the
2665 static int walWriteToLog(
2666 WalWriter
*p
, /* WAL to write to */
2667 void *pContent
, /* Content to be written */
2668 int iAmt
, /* Number of bytes to write */
2669 sqlite3_int64 iOffset
/* Start writing at this offset */
2672 if( iOffset
<p
->iSyncPoint
&& iOffset
+iAmt
>=p
->iSyncPoint
){
2673 int iFirstAmt
= (int)(p
->iSyncPoint
- iOffset
);
2674 rc
= sqlite3OsWrite(p
->pFd
, pContent
, iFirstAmt
, iOffset
);
2676 iOffset
+= iFirstAmt
;
2678 pContent
= (void*)(iFirstAmt
+ (char*)pContent
);
2679 assert( p
->syncFlags
& (SQLITE_SYNC_NORMAL
|SQLITE_SYNC_FULL
) );
2680 rc
= sqlite3OsSync(p
->pFd
, p
->syncFlags
& SQLITE_SYNC_MASK
);
2681 if( iAmt
==0 || rc
) return rc
;
2683 rc
= sqlite3OsWrite(p
->pFd
, pContent
, iAmt
, iOffset
);
2688 ** Write out a single frame of the WAL
2690 static int walWriteOneFrame(
2691 WalWriter
*p
, /* Where to write the frame */
2692 PgHdr
*pPage
, /* The page of the frame to be written */
2693 int nTruncate
, /* The commit flag. Usually 0. >0 for commit */
2694 sqlite3_int64 iOffset
/* Byte offset at which to write */
2696 int rc
; /* Result code from subfunctions */
2697 void *pData
; /* Data actually written */
2698 u8 aFrame
[WAL_FRAME_HDRSIZE
]; /* Buffer to assemble frame-header in */
2699 #if defined(SQLITE_HAS_CODEC)
2700 if( (pData
= sqlite3PagerCodec(pPage
))==0 ) return SQLITE_NOMEM
;
2702 pData
= pPage
->pData
;
2704 walEncodeFrame(p
->pWal
, pPage
->pgno
, nTruncate
, pData
, aFrame
);
2705 rc
= walWriteToLog(p
, aFrame
, sizeof(aFrame
), iOffset
);
2707 /* Write the page data */
2708 rc
= walWriteToLog(p
, pData
, p
->szPage
, iOffset
+sizeof(aFrame
));
2713 ** Write a set of frames to the log. The caller must hold the write-lock
2714 ** on the log file (obtained using sqlite3WalBeginWriteTransaction()).
2716 int sqlite3WalFrames(
2717 Wal
*pWal
, /* Wal handle to write to */
2718 int szPage
, /* Database page-size in bytes */
2719 PgHdr
*pList
, /* List of dirty pages to write */
2720 Pgno nTruncate
, /* Database size after this commit */
2721 int isCommit
, /* True if this is a commit */
2722 int sync_flags
/* Flags to pass to OsSync() (or 0) */
2724 int rc
; /* Used to catch return codes */
2725 u32 iFrame
; /* Next frame address */
2726 PgHdr
*p
; /* Iterator to run through pList with. */
2727 PgHdr
*pLast
= 0; /* Last frame in list */
2728 int nExtra
= 0; /* Number of extra copies of last page */
2729 int szFrame
; /* The size of a single frame */
2730 i64 iOffset
; /* Next byte to write in WAL file */
2731 WalWriter w
; /* The writer */
2734 assert( pWal
->writeLock
);
2736 /* If this frame set completes a transaction, then nTruncate>0. If
2737 ** nTruncate==0 then this frame set does not complete the transaction. */
2738 assert( (isCommit
!=0)==(nTruncate
!=0) );
2740 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
2741 { int cnt
; for(cnt
=0, p
=pList
; p
; p
=p
->pDirty
, cnt
++){}
2742 WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n",
2743 pWal
, cnt
, pWal
->hdr
.mxFrame
, isCommit
? "Commit" : "Spill"));
2747 /* See if it is possible to write these frames into the start of the
2748 ** log file, instead of appending to it at pWal->hdr.mxFrame.
2750 if( SQLITE_OK
!=(rc
= walRestartLog(pWal
)) ){
2754 /* If this is the first frame written into the log, write the WAL
2755 ** header to the start of the WAL file. See comments at the top of
2756 ** this source file for a description of the WAL header format.
2758 iFrame
= pWal
->hdr
.mxFrame
;
2760 u8 aWalHdr
[WAL_HDRSIZE
]; /* Buffer to assemble wal-header in */
2761 u32 aCksum
[2]; /* Checksum for wal-header */
2763 sqlite3Put4byte(&aWalHdr
[0], (WAL_MAGIC
| SQLITE_BIGENDIAN
));
2764 sqlite3Put4byte(&aWalHdr
[4], WAL_MAX_VERSION
);
2765 sqlite3Put4byte(&aWalHdr
[8], szPage
);
2766 sqlite3Put4byte(&aWalHdr
[12], pWal
->nCkpt
);
2767 if( pWal
->nCkpt
==0 ) sqlite3_randomness(8, pWal
->hdr
.aSalt
);
2768 memcpy(&aWalHdr
[16], pWal
->hdr
.aSalt
, 8);
2769 walChecksumBytes(1, aWalHdr
, WAL_HDRSIZE
-2*4, 0, aCksum
);
2770 sqlite3Put4byte(&aWalHdr
[24], aCksum
[0]);
2771 sqlite3Put4byte(&aWalHdr
[28], aCksum
[1]);
2773 pWal
->szPage
= szPage
;
2774 pWal
->hdr
.bigEndCksum
= SQLITE_BIGENDIAN
;
2775 pWal
->hdr
.aFrameCksum
[0] = aCksum
[0];
2776 pWal
->hdr
.aFrameCksum
[1] = aCksum
[1];
2777 pWal
->truncateOnCommit
= 1;
2779 rc
= sqlite3OsWrite(pWal
->pWalFd
, aWalHdr
, sizeof(aWalHdr
), 0);
2780 WALTRACE(("WAL%p: wal-header write %s\n", pWal
, rc
? "failed" : "ok"));
2781 if( rc
!=SQLITE_OK
){
2785 /* Sync the header (unless SQLITE_IOCAP_SEQUENTIAL is true or unless
2786 ** all syncing is turned off by PRAGMA synchronous=OFF). Otherwise
2787 ** an out-of-order write following a WAL restart could result in
2788 ** database corruption. See the ticket:
2790 ** http://localhost:591/sqlite/info/ff5be73dee
2792 if( pWal
->syncHeader
&& sync_flags
){
2793 rc
= sqlite3OsSync(pWal
->pWalFd
, sync_flags
& SQLITE_SYNC_MASK
);
2797 assert( (int)pWal
->szPage
==szPage
);
2799 /* Setup information needed to write frames into the WAL */
2801 w
.pFd
= pWal
->pWalFd
;
2803 w
.syncFlags
= sync_flags
;
2805 iOffset
= walFrameOffset(iFrame
+1, szPage
);
2806 szFrame
= szPage
+ WAL_FRAME_HDRSIZE
;
2808 /* Write all frames into the log file exactly once */
2809 for(p
=pList
; p
; p
=p
->pDirty
){
2810 int nDbSize
; /* 0 normally. Positive == commit flag */
2812 assert( iOffset
==walFrameOffset(iFrame
, szPage
) );
2813 nDbSize
= (isCommit
&& p
->pDirty
==0) ? nTruncate
: 0;
2814 rc
= walWriteOneFrame(&w
, p
, nDbSize
, iOffset
);
2820 /* If this is the end of a transaction, then we might need to pad
2821 ** the transaction and/or sync the WAL file.
2823 ** Padding and syncing only occur if this set of frames complete a
2824 ** transaction and if PRAGMA synchronous=FULL. If synchronous==NORMAL
2825 ** or synchonous==OFF, then no padding or syncing are needed.
2827 ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not
2828 ** needed and only the sync is done. If padding is needed, then the
2829 ** final frame is repeated (with its commit mark) until the next sector
2830 ** boundary is crossed. Only the part of the WAL prior to the last
2831 ** sector boundary is synced; the part of the last frame that extends
2832 ** past the sector boundary is written after the sync.
2834 if( isCommit
&& (sync_flags
& WAL_SYNC_TRANSACTIONS
)!=0 ){
2835 if( pWal
->padToSectorBoundary
){
2836 int sectorSize
= sqlite3SectorSize(pWal
->pWalFd
);
2837 w
.iSyncPoint
= ((iOffset
+sectorSize
-1)/sectorSize
)*sectorSize
;
2838 while( iOffset
<w
.iSyncPoint
){
2839 rc
= walWriteOneFrame(&w
, pLast
, nTruncate
, iOffset
);
2845 rc
= sqlite3OsSync(w
.pFd
, sync_flags
& SQLITE_SYNC_MASK
);
2849 /* If this frame set completes the first transaction in the WAL and
2850 ** if PRAGMA journal_size_limit is set, then truncate the WAL to the
2851 ** journal size limit, if possible.
2853 if( isCommit
&& pWal
->truncateOnCommit
&& pWal
->mxWalSize
>=0 ){
2854 i64 sz
= pWal
->mxWalSize
;
2855 if( walFrameOffset(iFrame
+nExtra
+1, szPage
)>pWal
->mxWalSize
){
2856 sz
= walFrameOffset(iFrame
+nExtra
+1, szPage
);
2858 walLimitSize(pWal
, sz
);
2859 pWal
->truncateOnCommit
= 0;
2862 /* Append data to the wal-index. It is not necessary to lock the
2863 ** wal-index to do this as the SQLITE_SHM_WRITE lock held on the wal-index
2864 ** guarantees that there are no other writers, and no data that may
2865 ** be in use by existing readers is being overwritten.
2867 iFrame
= pWal
->hdr
.mxFrame
;
2868 for(p
=pList
; p
&& rc
==SQLITE_OK
; p
=p
->pDirty
){
2870 rc
= walIndexAppend(pWal
, iFrame
, p
->pgno
);
2872 while( rc
==SQLITE_OK
&& nExtra
>0 ){
2875 rc
= walIndexAppend(pWal
, iFrame
, pLast
->pgno
);
2878 if( rc
==SQLITE_OK
){
2879 /* Update the private copy of the header. */
2880 pWal
->hdr
.szPage
= (u16
)((szPage
&0xff00) | (szPage
>>16));
2881 testcase( szPage
<=32768 );
2882 testcase( szPage
>=65536 );
2883 pWal
->hdr
.mxFrame
= iFrame
;
2885 pWal
->hdr
.iChange
++;
2886 pWal
->hdr
.nPage
= nTruncate
;
2888 /* If this is a commit, update the wal-index header too. */
2890 walIndexWriteHdr(pWal
);
2891 pWal
->iCallback
= iFrame
;
2895 WALTRACE(("WAL%p: frame write %s\n", pWal
, rc
? "failed" : "ok"));
2900 ** This routine is called to implement sqlite3_wal_checkpoint() and
2901 ** related interfaces.
2903 ** Obtain a CHECKPOINT lock and then backfill as much information as
2904 ** we can from WAL into the database.
2906 ** If parameter xBusy is not NULL, it is a pointer to a busy-handler
2907 ** callback. In this case this function runs a blocking checkpoint.
2909 int sqlite3WalCheckpoint(
2910 Wal
*pWal
, /* Wal connection */
2911 int eMode
, /* PASSIVE, FULL or RESTART */
2912 int (*xBusy
)(void*), /* Function to call when busy */
2913 void *pBusyArg
, /* Context argument for xBusyHandler */
2914 int sync_flags
, /* Flags to sync db file with (or 0) */
2915 int nBuf
, /* Size of temporary buffer */
2916 u8
*zBuf
, /* Temporary buffer to use */
2917 int *pnLog
, /* OUT: Number of frames in WAL */
2918 int *pnCkpt
/* OUT: Number of backfilled frames in WAL */
2920 int rc
; /* Return code */
2921 int isChanged
= 0; /* True if a new wal-index header is loaded */
2922 int eMode2
= eMode
; /* Mode to pass to walCheckpoint() */
2924 assert( pWal
->ckptLock
==0 );
2925 assert( pWal
->writeLock
==0 );
2927 if( pWal
->readOnly
) return SQLITE_READONLY
;
2928 WALTRACE(("WAL%p: checkpoint begins\n", pWal
));
2929 rc
= walLockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
2931 /* Usually this is SQLITE_BUSY meaning that another thread or process
2932 ** is already running a checkpoint, or maybe a recovery. But it might
2933 ** also be SQLITE_IOERR. */
2938 /* If this is a blocking-checkpoint, then obtain the write-lock as well
2939 ** to prevent any writers from running while the checkpoint is underway.
2940 ** This has to be done before the call to walIndexReadHdr() below.
2942 ** If the writer lock cannot be obtained, then a passive checkpoint is
2943 ** run instead. Since the checkpointer is not holding the writer lock,
2944 ** there is no point in blocking waiting for any readers. Assuming no
2945 ** other error occurs, this function will return SQLITE_BUSY to the caller.
2947 if( eMode
!=SQLITE_CHECKPOINT_PASSIVE
){
2948 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_WRITE_LOCK
, 1);
2949 if( rc
==SQLITE_OK
){
2950 pWal
->writeLock
= 1;
2951 }else if( rc
==SQLITE_BUSY
){
2952 eMode2
= SQLITE_CHECKPOINT_PASSIVE
;
2957 /* Read the wal-index header. */
2958 if( rc
==SQLITE_OK
){
2959 rc
= walIndexReadHdr(pWal
, &isChanged
);
2960 if( isChanged
&& pWal
->pDbFd
->pMethods
->iVersion
>=3 ){
2961 sqlite3OsUnfetch(pWal
->pDbFd
, 0, 0);
2965 /* Copy data from the log to the database file. */
2966 if( rc
==SQLITE_OK
){
2967 if( pWal
->hdr
.mxFrame
&& walPagesize(pWal
)!=nBuf
){
2968 rc
= SQLITE_CORRUPT_BKPT
;
2970 rc
= walCheckpoint(pWal
, eMode2
, xBusy
, pBusyArg
, sync_flags
, zBuf
);
2973 /* If no error occurred, set the output variables. */
2974 if( rc
==SQLITE_OK
|| rc
==SQLITE_BUSY
){
2975 if( pnLog
) *pnLog
= (int)pWal
->hdr
.mxFrame
;
2976 if( pnCkpt
) *pnCkpt
= (int)(walCkptInfo(pWal
)->nBackfill
);
2981 /* If a new wal-index header was loaded before the checkpoint was
2982 ** performed, then the pager-cache associated with pWal is now
2983 ** out of date. So zero the cached wal-index header to ensure that
2984 ** next time the pager opens a snapshot on this database it knows that
2985 ** the cache needs to be reset.
2987 memset(&pWal
->hdr
, 0, sizeof(WalIndexHdr
));
2990 /* Release the locks. */
2991 sqlite3WalEndWriteTransaction(pWal
);
2992 walUnlockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
2994 WALTRACE(("WAL%p: checkpoint %s\n", pWal
, rc
? "failed" : "ok"));
2995 return (rc
==SQLITE_OK
&& eMode
!=eMode2
? SQLITE_BUSY
: rc
);
2998 /* Return the value to pass to a sqlite3_wal_hook callback, the
2999 ** number of frames in the WAL at the point of the last commit since
3000 ** sqlite3WalCallback() was called. If no commits have occurred since
3001 ** the last call, then return 0.
3003 int sqlite3WalCallback(Wal
*pWal
){
3006 ret
= pWal
->iCallback
;
3007 pWal
->iCallback
= 0;
3013 ** This function is called to change the WAL subsystem into or out
3014 ** of locking_mode=EXCLUSIVE.
3016 ** If op is zero, then attempt to change from locking_mode=EXCLUSIVE
3017 ** into locking_mode=NORMAL. This means that we must acquire a lock
3018 ** on the pWal->readLock byte. If the WAL is already in locking_mode=NORMAL
3019 ** or if the acquisition of the lock fails, then return 0. If the
3020 ** transition out of exclusive-mode is successful, return 1. This
3021 ** operation must occur while the pager is still holding the exclusive
3022 ** lock on the main database file.
3024 ** If op is one, then change from locking_mode=NORMAL into
3025 ** locking_mode=EXCLUSIVE. This means that the pWal->readLock must
3026 ** be released. Return 1 if the transition is made and 0 if the
3027 ** WAL is already in exclusive-locking mode - meaning that this
3028 ** routine is a no-op. The pager must already hold the exclusive lock
3029 ** on the main database file before invoking this operation.
3031 ** If op is negative, then do a dry-run of the op==1 case but do
3032 ** not actually change anything. The pager uses this to see if it
3033 ** should acquire the database exclusive lock prior to invoking
3036 int sqlite3WalExclusiveMode(Wal
*pWal
, int op
){
3038 assert( pWal
->writeLock
==0 );
3039 assert( pWal
->exclusiveMode
!=WAL_HEAPMEMORY_MODE
|| op
==-1 );
3041 /* pWal->readLock is usually set, but might be -1 if there was a
3042 ** prior error while attempting to acquire are read-lock. This cannot
3043 ** happen if the connection is actually in exclusive mode (as no xShmLock
3044 ** locks are taken in this case). Nor should the pager attempt to
3045 ** upgrade to exclusive-mode following such an error.
3047 assert( pWal
->readLock
>=0 || pWal
->lockError
);
3048 assert( pWal
->readLock
>=0 || (op
<=0 && pWal
->exclusiveMode
==0) );
3051 if( pWal
->exclusiveMode
){
3052 pWal
->exclusiveMode
= 0;
3053 if( walLockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
))!=SQLITE_OK
){
3054 pWal
->exclusiveMode
= 1;
3056 rc
= pWal
->exclusiveMode
==0;
3058 /* Already in locking_mode=NORMAL */
3062 assert( pWal
->exclusiveMode
==0 );
3063 assert( pWal
->readLock
>=0 );
3064 walUnlockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
));
3065 pWal
->exclusiveMode
= 1;
3068 rc
= pWal
->exclusiveMode
==0;
3074 ** Return true if the argument is non-NULL and the WAL module is using
3075 ** heap-memory for the wal-index. Otherwise, if the argument is NULL or the
3076 ** WAL module is using shared-memory, return false.
3078 int sqlite3WalHeapMemory(Wal
*pWal
){
3079 return (pWal
&& pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
);
3082 #ifdef SQLITE_ENABLE_ZIPVFS
3084 ** If the argument is not NULL, it points to a Wal object that holds a
3085 ** read-lock. This function returns the database page-size if it is known,
3086 ** or zero if it is not (or if pWal is NULL).
3088 int sqlite3WalFramesize(Wal
*pWal
){
3089 assert( pWal
==0 || pWal
->readLock
>=0 );
3090 return (pWal
? pWal
->szPage
: 0);
3094 #endif /* #ifndef SQLITE_OMIT_WAL */