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 ** In the default unix and windows implementation, the wal-index is a mmapped
136 ** file whose name is the database name with a "-shm" suffix added. For that
137 ** reason, the wal-index is sometimes called the "shm" file.
139 ** The wal-index is transient. After a crash, the wal-index can (and should
140 ** be) reconstructed from the original WAL file. In fact, the VFS is required
141 ** to either truncate or zero the header of the wal-index when the last
142 ** connection to it closes. Because the wal-index is transient, it can
143 ** use an architecture-specific format; it does not have to be cross-platform.
144 ** Hence, unlike the database and WAL file formats which store all values
145 ** as big endian, the wal-index can store multi-byte values in the native
146 ** byte order of the host computer.
148 ** The purpose of the wal-index is to answer this question quickly: Given
149 ** a page number P and a maximum frame index M, return the index of the
150 ** last frame in the wal before frame M for page P in the WAL, or return
151 ** NULL if there are no frames for page P in the WAL prior to M.
153 ** The wal-index consists of a header region, followed by an one or
154 ** more index blocks.
156 ** The wal-index header contains the total number of frames within the WAL
157 ** in the mxFrame field.
159 ** Each index block except for the first contains information on
160 ** HASHTABLE_NPAGE frames. The first index block contains information on
161 ** HASHTABLE_NPAGE_ONE frames. The values of HASHTABLE_NPAGE_ONE and
162 ** HASHTABLE_NPAGE are selected so that together the wal-index header and
163 ** first index block are the same size as all other index blocks in the
166 ** Each index block contains two sections, a page-mapping that contains the
167 ** database page number associated with each wal frame, and a hash-table
168 ** that allows readers to query an index block for a specific page number.
169 ** The page-mapping is an array of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE
170 ** for the first index block) 32-bit page numbers. The first entry in the
171 ** first index-block contains the database page number corresponding to the
172 ** first frame in the WAL file. The first entry in the second index block
173 ** in the WAL file corresponds to the (HASHTABLE_NPAGE_ONE+1)th frame in
174 ** the log, and so on.
176 ** The last index block in a wal-index usually contains less than the full
177 ** complement of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE) page-numbers,
178 ** depending on the contents of the WAL file. This does not change the
179 ** allocated size of the page-mapping array - the page-mapping array merely
180 ** contains unused entries.
182 ** Even without using the hash table, the last frame for page P
183 ** can be found by scanning the page-mapping sections of each index block
184 ** starting with the last index block and moving toward the first, and
185 ** within each index block, starting at the end and moving toward the
186 ** beginning. The first entry that equals P corresponds to the frame
187 ** holding the content for that page.
189 ** The hash table consists of HASHTABLE_NSLOT 16-bit unsigned integers.
190 ** HASHTABLE_NSLOT = 2*HASHTABLE_NPAGE, and there is one entry in the
191 ** hash table for each page number in the mapping section, so the hash
192 ** table is never more than half full. The expected number of collisions
193 ** prior to finding a match is 1. Each entry of the hash table is an
194 ** 1-based index of an entry in the mapping section of the same
195 ** index block. Let K be the 1-based index of the largest entry in
196 ** the mapping section. (For index blocks other than the last, K will
197 ** always be exactly HASHTABLE_NPAGE (4096) and for the last index block
198 ** K will be (mxFrame%HASHTABLE_NPAGE).) Unused slots of the hash table
199 ** contain a value of 0.
201 ** To look for page P in the hash table, first compute a hash iKey on
204 ** iKey = (P * 383) % HASHTABLE_NSLOT
206 ** Then start scanning entries of the hash table, starting with iKey
207 ** (wrapping around to the beginning when the end of the hash table is
208 ** reached) until an unused hash slot is found. Let the first unused slot
209 ** be at index iUnused. (iUnused might be less than iKey if there was
210 ** wrap-around.) Because the hash table is never more than half full,
211 ** the search is guaranteed to eventually hit an unused entry. Let
212 ** iMax be the value between iKey and iUnused, closest to iUnused,
213 ** where aHash[iMax]==P. If there is no iMax entry (if there exists
214 ** no hash slot such that aHash[i]==p) then page P is not in the
215 ** current index block. Otherwise the iMax-th mapping entry of the
216 ** current index block corresponds to the last entry that references
219 ** A hash search begins with the last index block and moves toward the
220 ** first index block, looking for entries corresponding to page P. On
221 ** average, only two or three slots in each index block need to be
222 ** examined in order to either find the last entry for page P, or to
223 ** establish that no such entry exists in the block. Each index block
224 ** holds over 4000 entries. So two or three index blocks are sufficient
225 ** to cover a typical 10 megabyte WAL file, assuming 1K pages. 8 or 10
226 ** comparisons (on average) suffice to either locate a frame in the
227 ** WAL or to establish that the frame does not exist in the WAL. This
228 ** is much faster than scanning the entire 10MB WAL.
230 ** Note that entries are added in order of increasing K. Hence, one
231 ** reader might be using some value K0 and a second reader that started
232 ** at a later time (after additional transactions were added to the WAL
233 ** and to the wal-index) might be using a different value K1, where K1>K0.
234 ** Both readers can use the same hash table and mapping section to get
235 ** the correct result. There may be entries in the hash table with
236 ** K>K0 but to the first reader, those entries will appear to be unused
237 ** slots in the hash table and so the first reader will get an answer as
238 ** if no values greater than K0 had ever been inserted into the hash table
239 ** in the first place - which is what reader one wants. Meanwhile, the
240 ** second reader using K1 will see additional values that were inserted
241 ** later, which is exactly what reader two wants.
243 ** When a rollback occurs, the value of K is decreased. Hash table entries
244 ** that correspond to frames greater than the new K value are removed
245 ** from the hash table at this point.
247 #ifndef SQLITE_OMIT_WAL
252 ** Trace output macros
254 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
255 int sqlite3WalTrace
= 0;
256 # define WALTRACE(X) if(sqlite3WalTrace) sqlite3DebugPrintf X
262 ** The maximum (and only) versions of the wal and wal-index formats
263 ** that may be interpreted by this version of SQLite.
265 ** If a client begins recovering a WAL file and finds that (a) the checksum
266 ** values in the wal-header are correct and (b) the version field is not
267 ** WAL_MAX_VERSION, recovery fails and SQLite returns SQLITE_CANTOPEN.
269 ** Similarly, if a client successfully reads a wal-index header (i.e. the
270 ** checksum test is successful) and finds that the version field is not
271 ** WALINDEX_MAX_VERSION, then no read-transaction is opened and SQLite
272 ** returns SQLITE_CANTOPEN.
274 #define WAL_MAX_VERSION 3007000
275 #define WALINDEX_MAX_VERSION 3007000
278 ** Index numbers for various locking bytes. WAL_NREADER is the number
279 ** of available reader locks and should be at least 3. The default
280 ** is SQLITE_SHM_NLOCK==8 and WAL_NREADER==5.
282 ** Technically, the various VFSes are free to implement these locks however
283 ** they see fit. However, compatibility is encouraged so that VFSes can
284 ** interoperate. The standard implemention used on both unix and windows
285 ** is for the index number to indicate a byte offset into the
286 ** WalCkptInfo.aLock[] array in the wal-index header. In other words, all
287 ** locks are on the shm file. The WALINDEX_LOCK_OFFSET constant (which
288 ** should be 120) is the location in the shm file for the first locking
291 #define WAL_WRITE_LOCK 0
292 #define WAL_ALL_BUT_WRITE 1
293 #define WAL_CKPT_LOCK 1
294 #define WAL_RECOVER_LOCK 2
295 #define WAL_READ_LOCK(I) (3+(I))
296 #define WAL_NREADER (SQLITE_SHM_NLOCK-3)
299 /* Object declarations */
300 typedef struct WalIndexHdr WalIndexHdr
;
301 typedef struct WalIterator WalIterator
;
302 typedef struct WalCkptInfo WalCkptInfo
;
306 ** The following object holds a copy of the wal-index header content.
308 ** The actual header in the wal-index consists of two copies of this
309 ** object followed by one instance of the WalCkptInfo object.
310 ** For all versions of SQLite through 3.10.0 and probably beyond,
311 ** the locking bytes (WalCkptInfo.aLock) start at offset 120 and
312 ** the total header size is 136 bytes.
314 ** The szPage value can be any power of 2 between 512 and 32768, inclusive.
315 ** Or it can be 1 to represent a 65536-byte page. The latter case was
316 ** added in 3.7.1 when support for 64K pages was added.
319 u32 iVersion
; /* Wal-index version */
320 u32 unused
; /* Unused (padding) field */
321 u32 iChange
; /* Counter incremented each transaction */
322 u8 isInit
; /* 1 when initialized */
323 u8 bigEndCksum
; /* True if checksums in WAL are big-endian */
324 u16 szPage
; /* Database page size in bytes. 1==64K */
325 u32 mxFrame
; /* Index of last valid frame in the WAL */
326 u32 nPage
; /* Size of database in pages */
327 u32 aFrameCksum
[2]; /* Checksum of last frame in log */
328 u32 aSalt
[2]; /* Two salt values copied from WAL header */
329 u32 aCksum
[2]; /* Checksum over all prior fields */
333 ** A copy of the following object occurs in the wal-index immediately
334 ** following the second copy of the WalIndexHdr. This object stores
335 ** information used by checkpoint.
337 ** nBackfill is the number of frames in the WAL that have been written
338 ** back into the database. (We call the act of moving content from WAL to
339 ** database "backfilling".) The nBackfill number is never greater than
340 ** WalIndexHdr.mxFrame. nBackfill can only be increased by threads
341 ** holding the WAL_CKPT_LOCK lock (which includes a recovery thread).
342 ** However, a WAL_WRITE_LOCK thread can move the value of nBackfill from
343 ** mxFrame back to zero when the WAL is reset.
345 ** nBackfillAttempted is the largest value of nBackfill that a checkpoint
346 ** has attempted to achieve. Normally nBackfill==nBackfillAtempted, however
347 ** the nBackfillAttempted is set before any backfilling is done and the
348 ** nBackfill is only set after all backfilling completes. So if a checkpoint
349 ** crashes, nBackfillAttempted might be larger than nBackfill. The
350 ** WalIndexHdr.mxFrame must never be less than nBackfillAttempted.
352 ** The aLock[] field is a set of bytes used for locking. These bytes should
353 ** never be read or written.
355 ** There is one entry in aReadMark[] for each reader lock. If a reader
356 ** holds read-lock K, then the value in aReadMark[K] is no greater than
357 ** the mxFrame for that reader. The value READMARK_NOT_USED (0xffffffff)
358 ** for any aReadMark[] means that entry is unused. aReadMark[0] is
359 ** a special case; its value is never used and it exists as a place-holder
360 ** to avoid having to offset aReadMark[] indexs by one. Readers holding
361 ** WAL_READ_LOCK(0) always ignore the entire WAL and read all content
362 ** directly from the database.
364 ** The value of aReadMark[K] may only be changed by a thread that
365 ** is holding an exclusive lock on WAL_READ_LOCK(K). Thus, the value of
366 ** aReadMark[K] cannot changed while there is a reader is using that mark
367 ** since the reader will be holding a shared lock on WAL_READ_LOCK(K).
369 ** The checkpointer may only transfer frames from WAL to database where
370 ** the frame numbers are less than or equal to every aReadMark[] that is
371 ** in use (that is, every aReadMark[j] for which there is a corresponding
372 ** WAL_READ_LOCK(j)). New readers (usually) pick the aReadMark[] with the
373 ** largest value and will increase an unused aReadMark[] to mxFrame if there
374 ** is not already an aReadMark[] equal to mxFrame. The exception to the
375 ** previous sentence is when nBackfill equals mxFrame (meaning that everything
376 ** in the WAL has been backfilled into the database) then new readers
377 ** will choose aReadMark[0] which has value 0 and hence such reader will
378 ** get all their all content directly from the database file and ignore
381 ** Writers normally append new frames to the end of the WAL. However,
382 ** if nBackfill equals mxFrame (meaning that all WAL content has been
383 ** written back into the database) and if no readers are using the WAL
384 ** (in other words, if there are no WAL_READ_LOCK(i) where i>0) then
385 ** the writer will first "reset" the WAL back to the beginning and start
386 ** writing new content beginning at frame 1.
388 ** We assume that 32-bit loads are atomic and so no locks are needed in
389 ** order to read from any aReadMark[] entries.
392 u32 nBackfill
; /* Number of WAL frames backfilled into DB */
393 u32 aReadMark
[WAL_NREADER
]; /* Reader marks */
394 u8 aLock
[SQLITE_SHM_NLOCK
]; /* Reserved space for locks */
395 u32 nBackfillAttempted
; /* WAL frames perhaps written, or maybe not */
396 u32 notUsed0
; /* Available for future enhancements */
398 #define READMARK_NOT_USED 0xffffffff
401 /* A block of WALINDEX_LOCK_RESERVED bytes beginning at
402 ** WALINDEX_LOCK_OFFSET is reserved for locks. Since some systems
403 ** only support mandatory file-locks, we do not read or write data
404 ** from the region of the file on which locks are applied.
406 #define WALINDEX_LOCK_OFFSET (sizeof(WalIndexHdr)*2+offsetof(WalCkptInfo,aLock))
407 #define WALINDEX_HDR_SIZE (sizeof(WalIndexHdr)*2+sizeof(WalCkptInfo))
409 /* Size of header before each frame in wal */
410 #define WAL_FRAME_HDRSIZE 24
412 /* Size of write ahead log header, including checksum. */
413 #define WAL_HDRSIZE 32
415 /* WAL magic value. Either this value, or the same value with the least
416 ** significant bit also set (WAL_MAGIC | 0x00000001) is stored in 32-bit
417 ** big-endian format in the first 4 bytes of a WAL file.
419 ** If the LSB is set, then the checksums for each frame within the WAL
420 ** file are calculated by treating all data as an array of 32-bit
421 ** big-endian words. Otherwise, they are calculated by interpreting
422 ** all data as 32-bit little-endian words.
424 #define WAL_MAGIC 0x377f0682
427 ** Return the offset of frame iFrame in the write-ahead log file,
428 ** assuming a database page size of szPage bytes. The offset returned
429 ** is to the start of the write-ahead log frame-header.
431 #define walFrameOffset(iFrame, szPage) ( \
432 WAL_HDRSIZE + ((iFrame)-1)*(i64)((szPage)+WAL_FRAME_HDRSIZE) \
436 ** An open write-ahead log file is represented by an instance of the
440 sqlite3_vfs
*pVfs
; /* The VFS used to create pDbFd */
441 sqlite3_file
*pDbFd
; /* File handle for the database file */
442 sqlite3_file
*pWalFd
; /* File handle for WAL file */
443 u32 iCallback
; /* Value to pass to log callback (or 0) */
444 i64 mxWalSize
; /* Truncate WAL to this size upon reset */
445 int nWiData
; /* Size of array apWiData */
446 int szFirstBlock
; /* Size of first block written to WAL file */
447 volatile u32
**apWiData
; /* Pointer to wal-index content in memory */
448 u32 szPage
; /* Database page size */
449 i16 readLock
; /* Which read lock is being held. -1 for none */
450 u8 syncFlags
; /* Flags to use to sync header writes */
451 u8 exclusiveMode
; /* Non-zero if connection is in exclusive mode */
452 u8 writeLock
; /* True if in a write transaction */
453 u8 ckptLock
; /* True if holding a checkpoint lock */
454 u8 readOnly
; /* WAL_RDWR, WAL_RDONLY, or WAL_SHM_RDONLY */
455 u8 truncateOnCommit
; /* True to truncate WAL file on commit */
456 u8 syncHeader
; /* Fsync the WAL header if true */
457 u8 padToSectorBoundary
; /* Pad transactions out to the next sector */
459 WalIndexHdr hdr
; /* Wal-index header for current transaction */
460 u32 minFrame
; /* Ignore wal frames before this one */
461 u32 iReCksum
; /* On commit, recalculate checksums from here */
462 const char *zWalName
; /* Name of WAL file */
463 u32 nCkpt
; /* Checkpoint sequence counter in the wal-header */
465 u8 lockError
; /* True if a locking error has occurred */
467 #ifdef SQLITE_ENABLE_SNAPSHOT
468 WalIndexHdr
*pSnapshot
; /* Start transaction here if not NULL */
473 ** Candidate values for Wal.exclusiveMode.
475 #define WAL_NORMAL_MODE 0
476 #define WAL_EXCLUSIVE_MODE 1
477 #define WAL_HEAPMEMORY_MODE 2
480 ** Possible values for WAL.readOnly
482 #define WAL_RDWR 0 /* Normal read/write connection */
483 #define WAL_RDONLY 1 /* The WAL file is readonly */
484 #define WAL_SHM_RDONLY 2 /* The SHM file is readonly */
487 ** Each page of the wal-index mapping contains a hash-table made up of
488 ** an array of HASHTABLE_NSLOT elements of the following type.
493 ** This structure is used to implement an iterator that loops through
494 ** all frames in the WAL in database page order. Where two or more frames
495 ** correspond to the same database page, the iterator visits only the
496 ** frame most recently written to the WAL (in other words, the frame with
497 ** the largest index).
499 ** The internals of this structure are only accessed by:
501 ** walIteratorInit() - Create a new iterator,
502 ** walIteratorNext() - Step an iterator,
503 ** walIteratorFree() - Free an iterator.
505 ** This functionality is used by the checkpoint code (see walCheckpoint()).
508 int iPrior
; /* Last result returned from the iterator */
509 int nSegment
; /* Number of entries in aSegment[] */
511 int iNext
; /* Next slot in aIndex[] not yet returned */
512 ht_slot
*aIndex
; /* i0, i1, i2... such that aPgno[iN] ascend */
513 u32
*aPgno
; /* Array of page numbers. */
514 int nEntry
; /* Nr. of entries in aPgno[] and aIndex[] */
515 int iZero
; /* Frame number associated with aPgno[0] */
516 } aSegment
[1]; /* One for every 32KB page in the wal-index */
520 ** Define the parameters of the hash tables in the wal-index file. There
521 ** is a hash-table following every HASHTABLE_NPAGE page numbers in the
524 ** Changing any of these constants will alter the wal-index format and
525 ** create incompatibilities.
527 #define HASHTABLE_NPAGE 4096 /* Must be power of 2 */
528 #define HASHTABLE_HASH_1 383 /* Should be prime */
529 #define HASHTABLE_NSLOT (HASHTABLE_NPAGE*2) /* Must be a power of 2 */
532 ** The block of page numbers associated with the first hash-table in a
533 ** wal-index is smaller than usual. This is so that there is a complete
534 ** hash-table on each aligned 32KB page of the wal-index.
536 #define HASHTABLE_NPAGE_ONE (HASHTABLE_NPAGE - (WALINDEX_HDR_SIZE/sizeof(u32)))
538 /* The wal-index is divided into pages of WALINDEX_PGSZ bytes each. */
539 #define WALINDEX_PGSZ ( \
540 sizeof(ht_slot)*HASHTABLE_NSLOT + HASHTABLE_NPAGE*sizeof(u32) \
544 ** Obtain a pointer to the iPage'th page of the wal-index. The wal-index
545 ** is broken into pages of WALINDEX_PGSZ bytes. Wal-index pages are
546 ** numbered from zero.
548 ** If this call is successful, *ppPage is set to point to the wal-index
549 ** page and SQLITE_OK is returned. If an error (an OOM or VFS error) occurs,
550 ** then an SQLite error code is returned and *ppPage is set to 0.
552 static int walIndexPage(Wal
*pWal
, int iPage
, volatile u32
**ppPage
){
555 /* Enlarge the pWal->apWiData[] array if required */
556 if( pWal
->nWiData
<=iPage
){
557 int nByte
= sizeof(u32
*)*(iPage
+1);
558 volatile u32
**apNew
;
559 apNew
= (volatile u32
**)sqlite3_realloc64((void *)pWal
->apWiData
, nByte
);
562 return SQLITE_NOMEM_BKPT
;
564 memset((void*)&apNew
[pWal
->nWiData
], 0,
565 sizeof(u32
*)*(iPage
+1-pWal
->nWiData
));
566 pWal
->apWiData
= apNew
;
567 pWal
->nWiData
= iPage
+1;
570 /* Request a pointer to the required page from the VFS */
571 if( pWal
->apWiData
[iPage
]==0 ){
572 if( pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
){
573 pWal
->apWiData
[iPage
] = (u32
volatile *)sqlite3MallocZero(WALINDEX_PGSZ
);
574 if( !pWal
->apWiData
[iPage
] ) rc
= SQLITE_NOMEM_BKPT
;
576 rc
= sqlite3OsShmMap(pWal
->pDbFd
, iPage
, WALINDEX_PGSZ
,
577 pWal
->writeLock
, (void volatile **)&pWal
->apWiData
[iPage
]
579 if( (rc
&0xff)==SQLITE_READONLY
){
580 pWal
->readOnly
|= WAL_SHM_RDONLY
;
581 if( rc
==SQLITE_READONLY
){
588 *ppPage
= pWal
->apWiData
[iPage
];
589 assert( iPage
==0 || *ppPage
|| rc
!=SQLITE_OK
);
594 ** Return a pointer to the WalCkptInfo structure in the wal-index.
596 static volatile WalCkptInfo
*walCkptInfo(Wal
*pWal
){
597 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
598 return (volatile WalCkptInfo
*)&(pWal
->apWiData
[0][sizeof(WalIndexHdr
)/2]);
602 ** Return a pointer to the WalIndexHdr structure in the wal-index.
604 static volatile WalIndexHdr
*walIndexHdr(Wal
*pWal
){
605 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
606 return (volatile WalIndexHdr
*)pWal
->apWiData
[0];
610 ** The argument to this macro must be of type u32. On a little-endian
611 ** architecture, it returns the u32 value that results from interpreting
612 ** the 4 bytes as a big-endian value. On a big-endian architecture, it
613 ** returns the value that would be produced by interpreting the 4 bytes
614 ** of the input value as a little-endian integer.
616 #define BYTESWAP32(x) ( \
617 (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8) \
618 + (((x)&0x00FF0000)>>8) + (((x)&0xFF000000)>>24) \
622 ** Generate or extend an 8 byte checksum based on the data in
623 ** array aByte[] and the initial values of aIn[0] and aIn[1] (or
624 ** initial values of 0 and 0 if aIn==NULL).
626 ** The checksum is written back into aOut[] before returning.
628 ** nByte must be a positive multiple of 8.
630 static void walChecksumBytes(
631 int nativeCksum
, /* True for native byte-order, false for non-native */
632 u8
*a
, /* Content to be checksummed */
633 int nByte
, /* Bytes of content in a[]. Must be a multiple of 8. */
634 const u32
*aIn
, /* Initial checksum value input */
635 u32
*aOut
/* OUT: Final checksum value output */
638 u32
*aData
= (u32
*)a
;
639 u32
*aEnd
= (u32
*)&a
[nByte
];
649 assert( (nByte
&0x00000007)==0 );
655 }while( aData
<aEnd
);
658 s1
+= BYTESWAP32(aData
[0]) + s2
;
659 s2
+= BYTESWAP32(aData
[1]) + s1
;
661 }while( aData
<aEnd
);
668 static void walShmBarrier(Wal
*pWal
){
669 if( pWal
->exclusiveMode
!=WAL_HEAPMEMORY_MODE
){
670 sqlite3OsShmBarrier(pWal
->pDbFd
);
675 ** Write the header information in pWal->hdr into the wal-index.
677 ** The checksum on pWal->hdr is updated before it is written.
679 static void walIndexWriteHdr(Wal
*pWal
){
680 volatile WalIndexHdr
*aHdr
= walIndexHdr(pWal
);
681 const int nCksum
= offsetof(WalIndexHdr
, aCksum
);
683 assert( pWal
->writeLock
);
684 pWal
->hdr
.isInit
= 1;
685 pWal
->hdr
.iVersion
= WALINDEX_MAX_VERSION
;
686 walChecksumBytes(1, (u8
*)&pWal
->hdr
, nCksum
, 0, pWal
->hdr
.aCksum
);
687 memcpy((void*)&aHdr
[1], (const void*)&pWal
->hdr
, sizeof(WalIndexHdr
));
689 memcpy((void*)&aHdr
[0], (const void*)&pWal
->hdr
, sizeof(WalIndexHdr
));
693 ** This function encodes a single frame header and writes it to a buffer
694 ** supplied by the caller. A frame-header is made up of a series of
695 ** 4-byte big-endian integers, as follows:
698 ** 4: For commit records, the size of the database image in pages
699 ** after the commit. For all other records, zero.
700 ** 8: Salt-1 (copied from the wal-header)
701 ** 12: Salt-2 (copied from the wal-header)
705 static void walEncodeFrame(
706 Wal
*pWal
, /* The write-ahead log */
707 u32 iPage
, /* Database page number for frame */
708 u32 nTruncate
, /* New db size (or 0 for non-commit frames) */
709 u8
*aData
, /* Pointer to page data */
710 u8
*aFrame
/* OUT: Write encoded frame here */
712 int nativeCksum
; /* True for native byte-order checksums */
713 u32
*aCksum
= pWal
->hdr
.aFrameCksum
;
714 assert( WAL_FRAME_HDRSIZE
==24 );
715 sqlite3Put4byte(&aFrame
[0], iPage
);
716 sqlite3Put4byte(&aFrame
[4], nTruncate
);
717 if( pWal
->iReCksum
==0 ){
718 memcpy(&aFrame
[8], pWal
->hdr
.aSalt
, 8);
720 nativeCksum
= (pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
);
721 walChecksumBytes(nativeCksum
, aFrame
, 8, aCksum
, aCksum
);
722 walChecksumBytes(nativeCksum
, aData
, pWal
->szPage
, aCksum
, aCksum
);
724 sqlite3Put4byte(&aFrame
[16], aCksum
[0]);
725 sqlite3Put4byte(&aFrame
[20], aCksum
[1]);
727 memset(&aFrame
[8], 0, 16);
732 ** Check to see if the frame with header in aFrame[] and content
733 ** in aData[] is valid. If it is a valid frame, fill *piPage and
734 ** *pnTruncate and return true. Return if the frame is not valid.
736 static int walDecodeFrame(
737 Wal
*pWal
, /* The write-ahead log */
738 u32
*piPage
, /* OUT: Database page number for frame */
739 u32
*pnTruncate
, /* OUT: New db size (or 0 if not commit) */
740 u8
*aData
, /* Pointer to page data (for checksum) */
741 u8
*aFrame
/* Frame data */
743 int nativeCksum
; /* True for native byte-order checksums */
744 u32
*aCksum
= pWal
->hdr
.aFrameCksum
;
745 u32 pgno
; /* Page number of the frame */
746 assert( WAL_FRAME_HDRSIZE
==24 );
748 /* A frame is only valid if the salt values in the frame-header
749 ** match the salt values in the wal-header.
751 if( memcmp(&pWal
->hdr
.aSalt
, &aFrame
[8], 8)!=0 ){
755 /* A frame is only valid if the page number is creater than zero.
757 pgno
= sqlite3Get4byte(&aFrame
[0]);
762 /* A frame is only valid if a checksum of the WAL header,
763 ** all prior frams, the first 16 bytes of this frame-header,
764 ** and the frame-data matches the checksum in the last 8
765 ** bytes of this frame-header.
767 nativeCksum
= (pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
);
768 walChecksumBytes(nativeCksum
, aFrame
, 8, aCksum
, aCksum
);
769 walChecksumBytes(nativeCksum
, aData
, pWal
->szPage
, aCksum
, aCksum
);
770 if( aCksum
[0]!=sqlite3Get4byte(&aFrame
[16])
771 || aCksum
[1]!=sqlite3Get4byte(&aFrame
[20])
773 /* Checksum failed. */
777 /* If we reach this point, the frame is valid. Return the page number
778 ** and the new database size.
781 *pnTruncate
= sqlite3Get4byte(&aFrame
[4]);
786 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
788 ** Names of locks. This routine is used to provide debugging output and is not
789 ** a part of an ordinary build.
791 static const char *walLockName(int lockIdx
){
792 if( lockIdx
==WAL_WRITE_LOCK
){
794 }else if( lockIdx
==WAL_CKPT_LOCK
){
796 }else if( lockIdx
==WAL_RECOVER_LOCK
){
797 return "RECOVER-LOCK";
799 static char zName
[15];
800 sqlite3_snprintf(sizeof(zName
), zName
, "READ-LOCK[%d]",
801 lockIdx
-WAL_READ_LOCK(0));
805 #endif /*defined(SQLITE_TEST) || defined(SQLITE_DEBUG) */
809 ** Set or release locks on the WAL. Locks are either shared or exclusive.
810 ** A lock cannot be moved directly between shared and exclusive - it must go
811 ** through the unlocked state first.
813 ** In locking_mode=EXCLUSIVE, all of these routines become no-ops.
815 static int walLockShared(Wal
*pWal
, int lockIdx
){
817 if( pWal
->exclusiveMode
) return SQLITE_OK
;
818 rc
= sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, 1,
819 SQLITE_SHM_LOCK
| SQLITE_SHM_SHARED
);
820 WALTRACE(("WAL%p: acquire SHARED-%s %s\n", pWal
,
821 walLockName(lockIdx
), rc
? "failed" : "ok"));
822 VVA_ONLY( pWal
->lockError
= (u8
)(rc
!=SQLITE_OK
&& rc
!=SQLITE_BUSY
); )
825 static void walUnlockShared(Wal
*pWal
, int lockIdx
){
826 if( pWal
->exclusiveMode
) return;
827 (void)sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, 1,
828 SQLITE_SHM_UNLOCK
| SQLITE_SHM_SHARED
);
829 WALTRACE(("WAL%p: release SHARED-%s\n", pWal
, walLockName(lockIdx
)));
831 static int walLockExclusive(Wal
*pWal
, int lockIdx
, int n
){
833 if( pWal
->exclusiveMode
) return SQLITE_OK
;
834 rc
= sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, n
,
835 SQLITE_SHM_LOCK
| SQLITE_SHM_EXCLUSIVE
);
836 WALTRACE(("WAL%p: acquire EXCLUSIVE-%s cnt=%d %s\n", pWal
,
837 walLockName(lockIdx
), n
, rc
? "failed" : "ok"));
838 VVA_ONLY( pWal
->lockError
= (u8
)(rc
!=SQLITE_OK
&& rc
!=SQLITE_BUSY
); )
841 static void walUnlockExclusive(Wal
*pWal
, int lockIdx
, int n
){
842 if( pWal
->exclusiveMode
) return;
843 (void)sqlite3OsShmLock(pWal
->pDbFd
, lockIdx
, n
,
844 SQLITE_SHM_UNLOCK
| SQLITE_SHM_EXCLUSIVE
);
845 WALTRACE(("WAL%p: release EXCLUSIVE-%s cnt=%d\n", pWal
,
846 walLockName(lockIdx
), n
));
850 ** Compute a hash on a page number. The resulting hash value must land
851 ** between 0 and (HASHTABLE_NSLOT-1). The walHashNext() function advances
852 ** the hash to the next value in the event of a collision.
854 static int walHash(u32 iPage
){
856 assert( (HASHTABLE_NSLOT
& (HASHTABLE_NSLOT
-1))==0 );
857 return (iPage
*HASHTABLE_HASH_1
) & (HASHTABLE_NSLOT
-1);
859 static int walNextHash(int iPriorHash
){
860 return (iPriorHash
+1)&(HASHTABLE_NSLOT
-1);
864 ** Return pointers to the hash table and page number array stored on
865 ** page iHash of the wal-index. The wal-index is broken into 32KB pages
866 ** numbered starting from 0.
868 ** Set output variable *paHash to point to the start of the hash table
869 ** in the wal-index file. Set *piZero to one less than the frame
870 ** number of the first frame indexed by this hash table. If a
871 ** slot in the hash table is set to N, it refers to frame number
872 ** (*piZero+N) in the log.
874 ** Finally, set *paPgno so that *paPgno[1] is the page number of the
875 ** first frame indexed by the hash table, frame (*piZero+1).
877 static int walHashGet(
878 Wal
*pWal
, /* WAL handle */
879 int iHash
, /* Find the iHash'th table */
880 volatile ht_slot
**paHash
, /* OUT: Pointer to hash index */
881 volatile u32
**paPgno
, /* OUT: Pointer to page number array */
882 u32
*piZero
/* OUT: Frame associated with *paPgno[0] */
884 int rc
; /* Return code */
887 rc
= walIndexPage(pWal
, iHash
, &aPgno
);
888 assert( rc
==SQLITE_OK
|| iHash
>0 );
892 volatile ht_slot
*aHash
;
894 aHash
= (volatile ht_slot
*)&aPgno
[HASHTABLE_NPAGE
];
896 aPgno
= &aPgno
[WALINDEX_HDR_SIZE
/sizeof(u32
)];
899 iZero
= HASHTABLE_NPAGE_ONE
+ (iHash
-1)*HASHTABLE_NPAGE
;
902 *paPgno
= &aPgno
[-1];
910 ** Return the number of the wal-index page that contains the hash-table
911 ** and page-number array that contain entries corresponding to WAL frame
912 ** iFrame. The wal-index is broken up into 32KB pages. Wal-index pages
913 ** are numbered starting from 0.
915 static int walFramePage(u32 iFrame
){
916 int iHash
= (iFrame
+HASHTABLE_NPAGE
-HASHTABLE_NPAGE_ONE
-1) / HASHTABLE_NPAGE
;
917 assert( (iHash
==0 || iFrame
>HASHTABLE_NPAGE_ONE
)
918 && (iHash
>=1 || iFrame
<=HASHTABLE_NPAGE_ONE
)
919 && (iHash
<=1 || iFrame
>(HASHTABLE_NPAGE_ONE
+HASHTABLE_NPAGE
))
920 && (iHash
>=2 || iFrame
<=HASHTABLE_NPAGE_ONE
+HASHTABLE_NPAGE
)
921 && (iHash
<=2 || iFrame
>(HASHTABLE_NPAGE_ONE
+2*HASHTABLE_NPAGE
))
927 ** Return the page number associated with frame iFrame in this WAL.
929 static u32
walFramePgno(Wal
*pWal
, u32 iFrame
){
930 int iHash
= walFramePage(iFrame
);
932 return pWal
->apWiData
[0][WALINDEX_HDR_SIZE
/sizeof(u32
) + iFrame
- 1];
934 return pWal
->apWiData
[iHash
][(iFrame
-1-HASHTABLE_NPAGE_ONE
)%HASHTABLE_NPAGE
];
938 ** Remove entries from the hash table that point to WAL slots greater
939 ** than pWal->hdr.mxFrame.
941 ** This function is called whenever pWal->hdr.mxFrame is decreased due
942 ** to a rollback or savepoint.
944 ** At most only the hash table containing pWal->hdr.mxFrame needs to be
945 ** updated. Any later hash tables will be automatically cleared when
946 ** pWal->hdr.mxFrame advances to the point where those hash tables are
949 static void walCleanupHash(Wal
*pWal
){
950 volatile ht_slot
*aHash
= 0; /* Pointer to hash table to clear */
951 volatile u32
*aPgno
= 0; /* Page number array for hash table */
952 u32 iZero
= 0; /* frame == (aHash[x]+iZero) */
953 int iLimit
= 0; /* Zero values greater than this */
954 int nByte
; /* Number of bytes to zero in aPgno[] */
955 int i
; /* Used to iterate through aHash[] */
957 assert( pWal
->writeLock
);
958 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
-1 );
959 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
);
960 testcase( pWal
->hdr
.mxFrame
==HASHTABLE_NPAGE_ONE
+1 );
962 if( pWal
->hdr
.mxFrame
==0 ) return;
964 /* Obtain pointers to the hash-table and page-number array containing
965 ** the entry that corresponds to frame pWal->hdr.mxFrame. It is guaranteed
966 ** that the page said hash-table and array reside on is already mapped.
968 assert( pWal
->nWiData
>walFramePage(pWal
->hdr
.mxFrame
) );
969 assert( pWal
->apWiData
[walFramePage(pWal
->hdr
.mxFrame
)] );
970 walHashGet(pWal
, walFramePage(pWal
->hdr
.mxFrame
), &aHash
, &aPgno
, &iZero
);
972 /* Zero all hash-table entries that correspond to frame numbers greater
973 ** than pWal->hdr.mxFrame.
975 iLimit
= pWal
->hdr
.mxFrame
- iZero
;
977 for(i
=0; i
<HASHTABLE_NSLOT
; i
++){
978 if( aHash
[i
]>iLimit
){
983 /* Zero the entries in the aPgno array that correspond to frames with
984 ** frame numbers greater than pWal->hdr.mxFrame.
986 nByte
= (int)((char *)aHash
- (char *)&aPgno
[iLimit
+1]);
987 memset((void *)&aPgno
[iLimit
+1], 0, nByte
);
989 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
990 /* Verify that the every entry in the mapping region is still reachable
991 ** via the hash table even after the cleanup.
994 int j
; /* Loop counter */
995 int iKey
; /* Hash key */
996 for(j
=1; j
<=iLimit
; j
++){
997 for(iKey
=walHash(aPgno
[j
]); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
998 if( aHash
[iKey
]==j
) break;
1000 assert( aHash
[iKey
]==j
);
1003 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
1008 ** Set an entry in the wal-index that will map database page number
1009 ** pPage into WAL frame iFrame.
1011 static int walIndexAppend(Wal
*pWal
, u32 iFrame
, u32 iPage
){
1012 int rc
; /* Return code */
1013 u32 iZero
= 0; /* One less than frame number of aPgno[1] */
1014 volatile u32
*aPgno
= 0; /* Page number array */
1015 volatile ht_slot
*aHash
= 0; /* Hash table */
1017 rc
= walHashGet(pWal
, walFramePage(iFrame
), &aHash
, &aPgno
, &iZero
);
1019 /* Assuming the wal-index file was successfully mapped, populate the
1020 ** page number array and hash table entry.
1022 if( rc
==SQLITE_OK
){
1023 int iKey
; /* Hash table key */
1024 int idx
; /* Value to write to hash-table slot */
1025 int nCollide
; /* Number of hash collisions */
1027 idx
= iFrame
- iZero
;
1028 assert( idx
<= HASHTABLE_NSLOT
/2 + 1 );
1030 /* If this is the first entry to be added to this hash-table, zero the
1031 ** entire hash table and aPgno[] array before proceeding.
1034 int nByte
= (int)((u8
*)&aHash
[HASHTABLE_NSLOT
] - (u8
*)&aPgno
[1]);
1035 memset((void*)&aPgno
[1], 0, nByte
);
1038 /* If the entry in aPgno[] is already set, then the previous writer
1039 ** must have exited unexpectedly in the middle of a transaction (after
1040 ** writing one or more dirty pages to the WAL to free up memory).
1041 ** Remove the remnants of that writers uncommitted transaction from
1042 ** the hash-table before writing any new entries.
1045 walCleanupHash(pWal
);
1046 assert( !aPgno
[idx
] );
1049 /* Write the aPgno[] array entry and the hash-table slot. */
1051 for(iKey
=walHash(iPage
); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
1052 if( (nCollide
--)==0 ) return SQLITE_CORRUPT_BKPT
;
1055 aHash
[iKey
] = (ht_slot
)idx
;
1057 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
1058 /* Verify that the number of entries in the hash table exactly equals
1059 ** the number of entries in the mapping region.
1062 int i
; /* Loop counter */
1063 int nEntry
= 0; /* Number of entries in the hash table */
1064 for(i
=0; i
<HASHTABLE_NSLOT
; i
++){ if( aHash
[i
] ) nEntry
++; }
1065 assert( nEntry
==idx
);
1068 /* Verify that the every entry in the mapping region is reachable
1069 ** via the hash table. This turns out to be a really, really expensive
1070 ** thing to check, so only do this occasionally - not on every
1073 if( (idx
&0x3ff)==0 ){
1074 int i
; /* Loop counter */
1075 for(i
=1; i
<=idx
; i
++){
1076 for(iKey
=walHash(aPgno
[i
]); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
1077 if( aHash
[iKey
]==i
) break;
1079 assert( aHash
[iKey
]==i
);
1082 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
1091 ** Recover the wal-index by reading the write-ahead log file.
1093 ** This routine first tries to establish an exclusive lock on the
1094 ** wal-index to prevent other threads/processes from doing anything
1095 ** with the WAL or wal-index while recovery is running. The
1096 ** WAL_RECOVER_LOCK is also held so that other threads will know
1097 ** that this thread is running recovery. If unable to establish
1098 ** the necessary locks, this routine returns SQLITE_BUSY.
1100 static int walIndexRecover(Wal
*pWal
){
1101 int rc
; /* Return Code */
1102 i64 nSize
; /* Size of log file */
1103 u32 aFrameCksum
[2] = {0, 0};
1104 int iLock
; /* Lock offset to lock for checkpoint */
1106 /* Obtain an exclusive lock on all byte in the locking range not already
1107 ** locked by the caller. The caller is guaranteed to have locked the
1108 ** WAL_WRITE_LOCK byte, and may have also locked the WAL_CKPT_LOCK byte.
1109 ** If successful, the same bytes that are locked here are unlocked before
1110 ** this function returns.
1112 assert( pWal
->ckptLock
==1 || pWal
->ckptLock
==0 );
1113 assert( WAL_ALL_BUT_WRITE
==WAL_WRITE_LOCK
+1 );
1114 assert( WAL_CKPT_LOCK
==WAL_ALL_BUT_WRITE
);
1115 assert( pWal
->writeLock
);
1116 iLock
= WAL_ALL_BUT_WRITE
+ pWal
->ckptLock
;
1117 rc
= walLockExclusive(pWal
, iLock
, WAL_READ_LOCK(0)-iLock
);
1118 if( rc
==SQLITE_OK
){
1119 rc
= walLockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1120 if( rc
!=SQLITE_OK
){
1121 walUnlockExclusive(pWal
, iLock
, WAL_READ_LOCK(0)-iLock
);
1128 WALTRACE(("WAL%p: recovery begin...\n", pWal
));
1130 memset(&pWal
->hdr
, 0, sizeof(WalIndexHdr
));
1132 rc
= sqlite3OsFileSize(pWal
->pWalFd
, &nSize
);
1133 if( rc
!=SQLITE_OK
){
1134 goto recovery_error
;
1137 if( nSize
>WAL_HDRSIZE
){
1138 u8 aBuf
[WAL_HDRSIZE
]; /* Buffer to load WAL header into */
1139 u8
*aFrame
= 0; /* Malloc'd buffer to load entire frame */
1140 int szFrame
; /* Number of bytes in buffer aFrame[] */
1141 u8
*aData
; /* Pointer to data part of aFrame buffer */
1142 int iFrame
; /* Index of last frame read */
1143 i64 iOffset
; /* Next offset to read from log file */
1144 int szPage
; /* Page size according to the log */
1145 u32 magic
; /* Magic value read from WAL header */
1146 u32 version
; /* Magic value read from WAL header */
1147 int isValid
; /* True if this frame is valid */
1149 /* Read in the WAL header. */
1150 rc
= sqlite3OsRead(pWal
->pWalFd
, aBuf
, WAL_HDRSIZE
, 0);
1151 if( rc
!=SQLITE_OK
){
1152 goto recovery_error
;
1155 /* If the database page size is not a power of two, or is greater than
1156 ** SQLITE_MAX_PAGE_SIZE, conclude that the WAL file contains no valid
1157 ** data. Similarly, if the 'magic' value is invalid, ignore the whole
1160 magic
= sqlite3Get4byte(&aBuf
[0]);
1161 szPage
= sqlite3Get4byte(&aBuf
[8]);
1162 if( (magic
&0xFFFFFFFE)!=WAL_MAGIC
1163 || szPage
&(szPage
-1)
1164 || szPage
>SQLITE_MAX_PAGE_SIZE
1169 pWal
->hdr
.bigEndCksum
= (u8
)(magic
&0x00000001);
1170 pWal
->szPage
= szPage
;
1171 pWal
->nCkpt
= sqlite3Get4byte(&aBuf
[12]);
1172 memcpy(&pWal
->hdr
.aSalt
, &aBuf
[16], 8);
1174 /* Verify that the WAL header checksum is correct */
1175 walChecksumBytes(pWal
->hdr
.bigEndCksum
==SQLITE_BIGENDIAN
,
1176 aBuf
, WAL_HDRSIZE
-2*4, 0, pWal
->hdr
.aFrameCksum
1178 if( pWal
->hdr
.aFrameCksum
[0]!=sqlite3Get4byte(&aBuf
[24])
1179 || pWal
->hdr
.aFrameCksum
[1]!=sqlite3Get4byte(&aBuf
[28])
1184 /* Verify that the version number on the WAL format is one that
1185 ** are able to understand */
1186 version
= sqlite3Get4byte(&aBuf
[4]);
1187 if( version
!=WAL_MAX_VERSION
){
1188 rc
= SQLITE_CANTOPEN_BKPT
;
1192 /* Malloc a buffer to read frames into. */
1193 szFrame
= szPage
+ WAL_FRAME_HDRSIZE
;
1194 aFrame
= (u8
*)sqlite3_malloc64(szFrame
);
1196 rc
= SQLITE_NOMEM_BKPT
;
1197 goto recovery_error
;
1199 aData
= &aFrame
[WAL_FRAME_HDRSIZE
];
1201 /* Read all frames from the log file. */
1203 for(iOffset
=WAL_HDRSIZE
; (iOffset
+szFrame
)<=nSize
; iOffset
+=szFrame
){
1204 u32 pgno
; /* Database page number for frame */
1205 u32 nTruncate
; /* dbsize field from frame header */
1207 /* Read and decode the next log frame. */
1209 rc
= sqlite3OsRead(pWal
->pWalFd
, aFrame
, szFrame
, iOffset
);
1210 if( rc
!=SQLITE_OK
) break;
1211 isValid
= walDecodeFrame(pWal
, &pgno
, &nTruncate
, aData
, aFrame
);
1212 if( !isValid
) break;
1213 rc
= walIndexAppend(pWal
, iFrame
, pgno
);
1214 if( rc
!=SQLITE_OK
) break;
1216 /* If nTruncate is non-zero, this is a commit record. */
1218 pWal
->hdr
.mxFrame
= iFrame
;
1219 pWal
->hdr
.nPage
= nTruncate
;
1220 pWal
->hdr
.szPage
= (u16
)((szPage
&0xff00) | (szPage
>>16));
1221 testcase( szPage
<=32768 );
1222 testcase( szPage
>=65536 );
1223 aFrameCksum
[0] = pWal
->hdr
.aFrameCksum
[0];
1224 aFrameCksum
[1] = pWal
->hdr
.aFrameCksum
[1];
1228 sqlite3_free(aFrame
);
1232 if( rc
==SQLITE_OK
){
1233 volatile WalCkptInfo
*pInfo
;
1235 pWal
->hdr
.aFrameCksum
[0] = aFrameCksum
[0];
1236 pWal
->hdr
.aFrameCksum
[1] = aFrameCksum
[1];
1237 walIndexWriteHdr(pWal
);
1239 /* Reset the checkpoint-header. This is safe because this thread is
1240 ** currently holding locks that exclude all other readers, writers and
1243 pInfo
= walCkptInfo(pWal
);
1244 pInfo
->nBackfill
= 0;
1245 pInfo
->nBackfillAttempted
= pWal
->hdr
.mxFrame
;
1246 pInfo
->aReadMark
[0] = 0;
1247 for(i
=1; i
<WAL_NREADER
; i
++) pInfo
->aReadMark
[i
] = READMARK_NOT_USED
;
1248 if( pWal
->hdr
.mxFrame
) pInfo
->aReadMark
[1] = pWal
->hdr
.mxFrame
;
1250 /* If more than one frame was recovered from the log file, report an
1251 ** event via sqlite3_log(). This is to help with identifying performance
1252 ** problems caused by applications routinely shutting down without
1253 ** checkpointing the log file.
1255 if( pWal
->hdr
.nPage
){
1256 sqlite3_log(SQLITE_NOTICE_RECOVER_WAL
,
1257 "recovered %d frames from WAL file %s",
1258 pWal
->hdr
.mxFrame
, pWal
->zWalName
1264 WALTRACE(("WAL%p: recovery %s\n", pWal
, rc
? "failed" : "ok"));
1265 walUnlockExclusive(pWal
, iLock
, WAL_READ_LOCK(0)-iLock
);
1266 walUnlockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1271 ** Close an open wal-index.
1273 static void walIndexClose(Wal
*pWal
, int isDelete
){
1274 if( pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
|| pWal
->bUnlocked
){
1276 for(i
=0; i
<pWal
->nWiData
; i
++){
1277 sqlite3_free((void *)pWal
->apWiData
[i
]);
1278 pWal
->apWiData
[i
] = 0;
1281 if( pWal
->exclusiveMode
!=WAL_HEAPMEMORY_MODE
){
1282 sqlite3OsShmUnmap(pWal
->pDbFd
, isDelete
);
1287 ** Open a connection to the WAL file zWalName. The database file must
1288 ** already be opened on connection pDbFd. The buffer that zWalName points
1289 ** to must remain valid for the lifetime of the returned Wal* handle.
1291 ** A SHARED lock should be held on the database file when this function
1292 ** is called. The purpose of this SHARED lock is to prevent any other
1293 ** client from unlinking the WAL or wal-index file. If another process
1294 ** were to do this just after this client opened one of these files, the
1295 ** system would be badly broken.
1297 ** If the log file is successfully opened, SQLITE_OK is returned and
1298 ** *ppWal is set to point to a new WAL handle. If an error occurs,
1299 ** an SQLite error code is returned and *ppWal is left unmodified.
1302 sqlite3_vfs
*pVfs
, /* vfs module to open wal and wal-index */
1303 sqlite3_file
*pDbFd
, /* The open database file */
1304 const char *zWalName
, /* Name of the WAL file */
1305 int bNoShm
, /* True to run in heap-memory mode */
1306 i64 mxWalSize
, /* Truncate WAL to this size on reset */
1307 Wal
**ppWal
/* OUT: Allocated Wal handle */
1309 int rc
; /* Return Code */
1310 Wal
*pRet
; /* Object to allocate and return */
1311 int flags
; /* Flags passed to OsOpen() */
1313 assert( zWalName
&& zWalName
[0] );
1316 /* In the amalgamation, the os_unix.c and os_win.c source files come before
1317 ** this source file. Verify that the #defines of the locking byte offsets
1318 ** in os_unix.c and os_win.c agree with the WALINDEX_LOCK_OFFSET value.
1319 ** For that matter, if the lock offset ever changes from its initial design
1320 ** value of 120, we need to know that so there is an assert() to check it.
1322 assert( 120==WALINDEX_LOCK_OFFSET
);
1323 assert( 136==WALINDEX_HDR_SIZE
);
1325 assert( WIN_SHM_BASE
==WALINDEX_LOCK_OFFSET
);
1327 #ifdef UNIX_SHM_BASE
1328 assert( UNIX_SHM_BASE
==WALINDEX_LOCK_OFFSET
);
1332 /* Allocate an instance of struct Wal to return. */
1334 pRet
= (Wal
*)sqlite3MallocZero(sizeof(Wal
) + pVfs
->szOsFile
);
1336 return SQLITE_NOMEM_BKPT
;
1340 pRet
->pWalFd
= (sqlite3_file
*)&pRet
[1];
1341 pRet
->pDbFd
= pDbFd
;
1342 pRet
->readLock
= -1;
1343 pRet
->mxWalSize
= mxWalSize
;
1344 pRet
->zWalName
= zWalName
;
1345 pRet
->syncHeader
= 1;
1346 pRet
->padToSectorBoundary
= 1;
1347 pRet
->exclusiveMode
= (bNoShm
? WAL_HEAPMEMORY_MODE
: WAL_NORMAL_MODE
);
1349 /* Open file handle on the write-ahead log file. */
1350 flags
= (SQLITE_OPEN_READWRITE
|SQLITE_OPEN_CREATE
|SQLITE_OPEN_WAL
);
1351 rc
= sqlite3OsOpen(pVfs
, zWalName
, pRet
->pWalFd
, flags
, &flags
);
1352 if( rc
==SQLITE_OK
&& flags
&SQLITE_OPEN_READONLY
){
1353 pRet
->readOnly
= WAL_RDONLY
;
1356 if( rc
!=SQLITE_OK
){
1357 walIndexClose(pRet
, 0);
1358 sqlite3OsClose(pRet
->pWalFd
);
1361 int iDC
= sqlite3OsDeviceCharacteristics(pDbFd
);
1362 if( iDC
& SQLITE_IOCAP_SEQUENTIAL
){ pRet
->syncHeader
= 0; }
1363 if( iDC
& SQLITE_IOCAP_POWERSAFE_OVERWRITE
){
1364 pRet
->padToSectorBoundary
= 0;
1367 WALTRACE(("WAL%d: opened\n", pRet
));
1373 ** Change the size to which the WAL file is trucated on each reset.
1375 void sqlite3WalLimit(Wal
*pWal
, i64 iLimit
){
1376 if( pWal
) pWal
->mxWalSize
= iLimit
;
1380 ** Find the smallest page number out of all pages held in the WAL that
1381 ** has not been returned by any prior invocation of this method on the
1382 ** same WalIterator object. Write into *piFrame the frame index where
1383 ** that page was last written into the WAL. Write into *piPage the page
1386 ** Return 0 on success. If there are no pages in the WAL with a page
1387 ** number larger than *piPage, then return 1.
1389 static int walIteratorNext(
1390 WalIterator
*p
, /* Iterator */
1391 u32
*piPage
, /* OUT: The page number of the next page */
1392 u32
*piFrame
/* OUT: Wal frame index of next page */
1394 u32 iMin
; /* Result pgno must be greater than iMin */
1395 u32 iRet
= 0xFFFFFFFF; /* 0xffffffff is never a valid page number */
1396 int i
; /* For looping through segments */
1399 assert( iMin
<0xffffffff );
1400 for(i
=p
->nSegment
-1; i
>=0; i
--){
1401 struct WalSegment
*pSegment
= &p
->aSegment
[i
];
1402 while( pSegment
->iNext
<pSegment
->nEntry
){
1403 u32 iPg
= pSegment
->aPgno
[pSegment
->aIndex
[pSegment
->iNext
]];
1407 *piFrame
= pSegment
->iZero
+ pSegment
->aIndex
[pSegment
->iNext
];
1415 *piPage
= p
->iPrior
= iRet
;
1416 return (iRet
==0xFFFFFFFF);
1420 ** This function merges two sorted lists into a single sorted list.
1422 ** aLeft[] and aRight[] are arrays of indices. The sort key is
1423 ** aContent[aLeft[]] and aContent[aRight[]]. Upon entry, the following
1424 ** is guaranteed for all J<K:
1426 ** aContent[aLeft[J]] < aContent[aLeft[K]]
1427 ** aContent[aRight[J]] < aContent[aRight[K]]
1429 ** This routine overwrites aRight[] with a new (probably longer) sequence
1430 ** of indices such that the aRight[] contains every index that appears in
1431 ** either aLeft[] or the old aRight[] and such that the second condition
1432 ** above is still met.
1434 ** The aContent[aLeft[X]] values will be unique for all X. And the
1435 ** aContent[aRight[X]] values will be unique too. But there might be
1436 ** one or more combinations of X and Y such that
1438 ** aLeft[X]!=aRight[Y] && aContent[aLeft[X]] == aContent[aRight[Y]]
1440 ** When that happens, omit the aLeft[X] and use the aRight[Y] index.
1442 static void walMerge(
1443 const u32
*aContent
, /* Pages in wal - keys for the sort */
1444 ht_slot
*aLeft
, /* IN: Left hand input list */
1445 int nLeft
, /* IN: Elements in array *paLeft */
1446 ht_slot
**paRight
, /* IN/OUT: Right hand input list */
1447 int *pnRight
, /* IN/OUT: Elements in *paRight */
1448 ht_slot
*aTmp
/* Temporary buffer */
1450 int iLeft
= 0; /* Current index in aLeft */
1451 int iRight
= 0; /* Current index in aRight */
1452 int iOut
= 0; /* Current index in output buffer */
1453 int nRight
= *pnRight
;
1454 ht_slot
*aRight
= *paRight
;
1456 assert( nLeft
>0 && nRight
>0 );
1457 while( iRight
<nRight
|| iLeft
<nLeft
){
1462 && (iRight
>=nRight
|| aContent
[aLeft
[iLeft
]]<aContent
[aRight
[iRight
]])
1464 logpage
= aLeft
[iLeft
++];
1466 logpage
= aRight
[iRight
++];
1468 dbpage
= aContent
[logpage
];
1470 aTmp
[iOut
++] = logpage
;
1471 if( iLeft
<nLeft
&& aContent
[aLeft
[iLeft
]]==dbpage
) iLeft
++;
1473 assert( iLeft
>=nLeft
|| aContent
[aLeft
[iLeft
]]>dbpage
);
1474 assert( iRight
>=nRight
|| aContent
[aRight
[iRight
]]>dbpage
);
1479 memcpy(aLeft
, aTmp
, sizeof(aTmp
[0])*iOut
);
1483 ** Sort the elements in list aList using aContent[] as the sort key.
1484 ** Remove elements with duplicate keys, preferring to keep the
1485 ** larger aList[] values.
1487 ** The aList[] entries are indices into aContent[]. The values in
1488 ** aList[] are to be sorted so that for all J<K:
1490 ** aContent[aList[J]] < aContent[aList[K]]
1492 ** For any X and Y such that
1494 ** aContent[aList[X]] == aContent[aList[Y]]
1496 ** Keep the larger of the two values aList[X] and aList[Y] and discard
1499 static void walMergesort(
1500 const u32
*aContent
, /* Pages in wal */
1501 ht_slot
*aBuffer
, /* Buffer of at least *pnList items to use */
1502 ht_slot
*aList
, /* IN/OUT: List to sort */
1503 int *pnList
/* IN/OUT: Number of elements in aList[] */
1506 int nList
; /* Number of elements in aList */
1507 ht_slot
*aList
; /* Pointer to sub-list content */
1510 const int nList
= *pnList
; /* Size of input list */
1511 int nMerge
= 0; /* Number of elements in list aMerge */
1512 ht_slot
*aMerge
= 0; /* List to be merged */
1513 int iList
; /* Index into input list */
1514 u32 iSub
= 0; /* Index into aSub array */
1515 struct Sublist aSub
[13]; /* Array of sub-lists */
1517 memset(aSub
, 0, sizeof(aSub
));
1518 assert( nList
<=HASHTABLE_NPAGE
&& nList
>0 );
1519 assert( HASHTABLE_NPAGE
==(1<<(ArraySize(aSub
)-1)) );
1521 for(iList
=0; iList
<nList
; iList
++){
1523 aMerge
= &aList
[iList
];
1524 for(iSub
=0; iList
& (1<<iSub
); iSub
++){
1526 assert( iSub
<ArraySize(aSub
) );
1528 assert( p
->aList
&& p
->nList
<=(1<<iSub
) );
1529 assert( p
->aList
==&aList
[iList
&~((2<<iSub
)-1)] );
1530 walMerge(aContent
, p
->aList
, p
->nList
, &aMerge
, &nMerge
, aBuffer
);
1532 aSub
[iSub
].aList
= aMerge
;
1533 aSub
[iSub
].nList
= nMerge
;
1536 for(iSub
++; iSub
<ArraySize(aSub
); iSub
++){
1537 if( nList
& (1<<iSub
) ){
1539 assert( iSub
<ArraySize(aSub
) );
1541 assert( p
->nList
<=(1<<iSub
) );
1542 assert( p
->aList
==&aList
[nList
&~((2<<iSub
)-1)] );
1543 walMerge(aContent
, p
->aList
, p
->nList
, &aMerge
, &nMerge
, aBuffer
);
1546 assert( aMerge
==aList
);
1552 for(i
=1; i
<*pnList
; i
++){
1553 assert( aContent
[aList
[i
]] > aContent
[aList
[i
-1]] );
1560 ** Free an iterator allocated by walIteratorInit().
1562 static void walIteratorFree(WalIterator
*p
){
1567 ** Construct a WalInterator object that can be used to loop over all
1568 ** pages in the WAL in ascending order. The caller must hold the checkpoint
1571 ** On success, make *pp point to the newly allocated WalInterator object
1572 ** return SQLITE_OK. Otherwise, return an error code. If this routine
1573 ** returns an error, the value of *pp is undefined.
1575 ** The calling routine should invoke walIteratorFree() to destroy the
1576 ** WalIterator object when it has finished with it.
1578 static int walIteratorInit(Wal
*pWal
, WalIterator
**pp
){
1579 WalIterator
*p
; /* Return value */
1580 int nSegment
; /* Number of segments to merge */
1581 u32 iLast
; /* Last frame in log */
1582 int nByte
; /* Number of bytes to allocate */
1583 int i
; /* Iterator variable */
1584 ht_slot
*aTmp
; /* Temp space used by merge-sort */
1585 int rc
= SQLITE_OK
; /* Return Code */
1587 /* This routine only runs while holding the checkpoint lock. And
1588 ** it only runs if there is actually content in the log (mxFrame>0).
1590 assert( pWal
->ckptLock
&& pWal
->hdr
.mxFrame
>0 );
1591 iLast
= pWal
->hdr
.mxFrame
;
1593 /* Allocate space for the WalIterator object. */
1594 nSegment
= walFramePage(iLast
) + 1;
1595 nByte
= sizeof(WalIterator
)
1596 + (nSegment
-1)*sizeof(struct WalSegment
)
1597 + iLast
*sizeof(ht_slot
);
1598 p
= (WalIterator
*)sqlite3_malloc64(nByte
);
1600 return SQLITE_NOMEM_BKPT
;
1602 memset(p
, 0, nByte
);
1603 p
->nSegment
= nSegment
;
1605 /* Allocate temporary space used by the merge-sort routine. This block
1606 ** of memory will be freed before this function returns.
1608 aTmp
= (ht_slot
*)sqlite3_malloc64(
1609 sizeof(ht_slot
) * (iLast
>HASHTABLE_NPAGE
?HASHTABLE_NPAGE
:iLast
)
1612 rc
= SQLITE_NOMEM_BKPT
;
1615 for(i
=0; rc
==SQLITE_OK
&& i
<nSegment
; i
++){
1616 volatile ht_slot
*aHash
;
1618 volatile u32
*aPgno
;
1620 rc
= walHashGet(pWal
, i
, &aHash
, &aPgno
, &iZero
);
1621 if( rc
==SQLITE_OK
){
1622 int j
; /* Counter variable */
1623 int nEntry
; /* Number of entries in this segment */
1624 ht_slot
*aIndex
; /* Sorted index for this segment */
1627 if( (i
+1)==nSegment
){
1628 nEntry
= (int)(iLast
- iZero
);
1630 nEntry
= (int)((u32
*)aHash
- (u32
*)aPgno
);
1632 aIndex
= &((ht_slot
*)&p
->aSegment
[p
->nSegment
])[iZero
];
1635 for(j
=0; j
<nEntry
; j
++){
1636 aIndex
[j
] = (ht_slot
)j
;
1638 walMergesort((u32
*)aPgno
, aTmp
, aIndex
, &nEntry
);
1639 p
->aSegment
[i
].iZero
= iZero
;
1640 p
->aSegment
[i
].nEntry
= nEntry
;
1641 p
->aSegment
[i
].aIndex
= aIndex
;
1642 p
->aSegment
[i
].aPgno
= (u32
*)aPgno
;
1647 if( rc
!=SQLITE_OK
){
1655 ** Attempt to obtain the exclusive WAL lock defined by parameters lockIdx and
1656 ** n. If the attempt fails and parameter xBusy is not NULL, then it is a
1657 ** busy-handler function. Invoke it and retry the lock until either the
1658 ** lock is successfully obtained or the busy-handler returns 0.
1660 static int walBusyLock(
1661 Wal
*pWal
, /* WAL connection */
1662 int (*xBusy
)(void*), /* Function to call when busy */
1663 void *pBusyArg
, /* Context argument for xBusyHandler */
1664 int lockIdx
, /* Offset of first byte to lock */
1665 int n
/* Number of bytes to lock */
1669 rc
= walLockExclusive(pWal
, lockIdx
, n
);
1670 }while( xBusy
&& rc
==SQLITE_BUSY
&& xBusy(pBusyArg
) );
1675 ** The cache of the wal-index header must be valid to call this function.
1676 ** Return the page-size in bytes used by the database.
1678 static int walPagesize(Wal
*pWal
){
1679 return (pWal
->hdr
.szPage
&0xfe00) + ((pWal
->hdr
.szPage
&0x0001)<<16);
1683 ** The following is guaranteed when this function is called:
1685 ** a) the WRITER lock is held,
1686 ** b) the entire log file has been checkpointed, and
1687 ** c) any existing readers are reading exclusively from the database
1688 ** file - there are no readers that may attempt to read a frame from
1691 ** This function updates the shared-memory structures so that the next
1692 ** client to write to the database (which may be this one) does so by
1693 ** writing frames into the start of the log file.
1695 ** The value of parameter salt1 is used as the aSalt[1] value in the
1696 ** new wal-index header. It should be passed a pseudo-random value (i.e.
1697 ** one obtained from sqlite3_randomness()).
1699 static void walRestartHdr(Wal
*pWal
, u32 salt1
){
1700 volatile WalCkptInfo
*pInfo
= walCkptInfo(pWal
);
1701 int i
; /* Loop counter */
1702 u32
*aSalt
= pWal
->hdr
.aSalt
; /* Big-endian salt values */
1704 pWal
->hdr
.mxFrame
= 0;
1705 sqlite3Put4byte((u8
*)&aSalt
[0], 1 + sqlite3Get4byte((u8
*)&aSalt
[0]));
1706 memcpy(&pWal
->hdr
.aSalt
[1], &salt1
, 4);
1707 walIndexWriteHdr(pWal
);
1708 pInfo
->nBackfill
= 0;
1709 pInfo
->nBackfillAttempted
= 0;
1710 pInfo
->aReadMark
[1] = 0;
1711 for(i
=2; i
<WAL_NREADER
; i
++) pInfo
->aReadMark
[i
] = READMARK_NOT_USED
;
1712 assert( pInfo
->aReadMark
[0]==0 );
1716 ** Copy as much content as we can from the WAL back into the database file
1717 ** in response to an sqlite3_wal_checkpoint() request or the equivalent.
1719 ** The amount of information copies from WAL to database might be limited
1720 ** by active readers. This routine will never overwrite a database page
1721 ** that a concurrent reader might be using.
1723 ** All I/O barrier operations (a.k.a fsyncs) occur in this routine when
1724 ** SQLite is in WAL-mode in synchronous=NORMAL. That means that if
1725 ** checkpoints are always run by a background thread or background
1726 ** process, foreground threads will never block on a lengthy fsync call.
1728 ** Fsync is called on the WAL before writing content out of the WAL and
1729 ** into the database. This ensures that if the new content is persistent
1730 ** in the WAL and can be recovered following a power-loss or hard reset.
1732 ** Fsync is also called on the database file if (and only if) the entire
1733 ** WAL content is copied into the database file. This second fsync makes
1734 ** it safe to delete the WAL since the new content will persist in the
1737 ** This routine uses and updates the nBackfill field of the wal-index header.
1738 ** This is the only routine that will increase the value of nBackfill.
1739 ** (A WAL reset or recovery will revert nBackfill to zero, but not increase
1742 ** The caller must be holding sufficient locks to ensure that no other
1743 ** checkpoint is running (in any other thread or process) at the same
1746 static int walCheckpoint(
1747 Wal
*pWal
, /* Wal connection */
1748 sqlite3
*db
, /* Check for interrupts on this handle */
1749 int eMode
, /* One of PASSIVE, FULL or RESTART */
1750 int (*xBusy
)(void*), /* Function to call when busy */
1751 void *pBusyArg
, /* Context argument for xBusyHandler */
1752 int sync_flags
, /* Flags for OsSync() (or 0) */
1753 u8
*zBuf
/* Temporary buffer to use */
1755 int rc
= SQLITE_OK
; /* Return code */
1756 int szPage
; /* Database page-size */
1757 WalIterator
*pIter
= 0; /* Wal iterator context */
1758 u32 iDbpage
= 0; /* Next database page to write */
1759 u32 iFrame
= 0; /* Wal frame containing data for iDbpage */
1760 u32 mxSafeFrame
; /* Max frame that can be backfilled */
1761 u32 mxPage
; /* Max database page to write */
1762 int i
; /* Loop counter */
1763 volatile WalCkptInfo
*pInfo
; /* The checkpoint status information */
1765 szPage
= walPagesize(pWal
);
1766 testcase( szPage
<=32768 );
1767 testcase( szPage
>=65536 );
1768 pInfo
= walCkptInfo(pWal
);
1769 if( pInfo
->nBackfill
<pWal
->hdr
.mxFrame
){
1771 /* Allocate the iterator */
1772 rc
= walIteratorInit(pWal
, &pIter
);
1773 if( rc
!=SQLITE_OK
){
1778 /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked
1779 ** in the SQLITE_CHECKPOINT_PASSIVE mode. */
1780 assert( eMode
!=SQLITE_CHECKPOINT_PASSIVE
|| xBusy
==0 );
1782 /* Compute in mxSafeFrame the index of the last frame of the WAL that is
1783 ** safe to write into the database. Frames beyond mxSafeFrame might
1784 ** overwrite database pages that are in use by active readers and thus
1785 ** cannot be backfilled from the WAL.
1787 mxSafeFrame
= pWal
->hdr
.mxFrame
;
1788 mxPage
= pWal
->hdr
.nPage
;
1789 for(i
=1; i
<WAL_NREADER
; i
++){
1790 /* Thread-sanitizer reports that the following is an unsafe read,
1791 ** as some other thread may be in the process of updating the value
1792 ** of the aReadMark[] slot. The assumption here is that if that is
1793 ** happening, the other client may only be increasing the value,
1794 ** not decreasing it. So assuming either that either the "old" or
1795 ** "new" version of the value is read, and not some arbitrary value
1796 ** that would never be written by a real client, things are still
1798 u32 y
= pInfo
->aReadMark
[i
];
1799 if( mxSafeFrame
>y
){
1800 assert( y
<=pWal
->hdr
.mxFrame
);
1801 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(i
), 1);
1802 if( rc
==SQLITE_OK
){
1803 pInfo
->aReadMark
[i
] = (i
==1 ? mxSafeFrame
: READMARK_NOT_USED
);
1804 walUnlockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
1805 }else if( rc
==SQLITE_BUSY
){
1809 goto walcheckpoint_out
;
1814 if( pInfo
->nBackfill
<mxSafeFrame
1815 && (rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(0),1))==SQLITE_OK
1817 i64 nSize
; /* Current size of database file */
1818 u32 nBackfill
= pInfo
->nBackfill
;
1820 pInfo
->nBackfillAttempted
= mxSafeFrame
;
1822 /* Sync the WAL to disk */
1823 rc
= sqlite3OsSync(pWal
->pWalFd
, CKPT_SYNC_FLAGS(sync_flags
));
1825 /* If the database may grow as a result of this checkpoint, hint
1826 ** about the eventual size of the db file to the VFS layer.
1828 if( rc
==SQLITE_OK
){
1829 i64 nReq
= ((i64
)mxPage
* szPage
);
1830 rc
= sqlite3OsFileSize(pWal
->pDbFd
, &nSize
);
1831 if( rc
==SQLITE_OK
&& nSize
<nReq
){
1832 sqlite3OsFileControlHint(pWal
->pDbFd
, SQLITE_FCNTL_SIZE_HINT
, &nReq
);
1837 /* Iterate through the contents of the WAL, copying data to the db file */
1838 while( rc
==SQLITE_OK
&& 0==walIteratorNext(pIter
, &iDbpage
, &iFrame
) ){
1840 assert( walFramePgno(pWal
, iFrame
)==iDbpage
);
1841 if( db
->u1
.isInterrupted
){
1842 rc
= db
->mallocFailed
? SQLITE_NOMEM_BKPT
: SQLITE_INTERRUPT
;
1845 if( iFrame
<=nBackfill
|| iFrame
>mxSafeFrame
|| iDbpage
>mxPage
){
1848 iOffset
= walFrameOffset(iFrame
, szPage
) + WAL_FRAME_HDRSIZE
;
1849 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL file */
1850 rc
= sqlite3OsRead(pWal
->pWalFd
, zBuf
, szPage
, iOffset
);
1851 if( rc
!=SQLITE_OK
) break;
1852 iOffset
= (iDbpage
-1)*(i64
)szPage
;
1853 testcase( IS_BIG_INT(iOffset
) );
1854 rc
= sqlite3OsWrite(pWal
->pDbFd
, zBuf
, szPage
, iOffset
);
1855 if( rc
!=SQLITE_OK
) break;
1858 /* If work was actually accomplished... */
1859 if( rc
==SQLITE_OK
){
1860 if( mxSafeFrame
==walIndexHdr(pWal
)->mxFrame
){
1861 i64 szDb
= pWal
->hdr
.nPage
*(i64
)szPage
;
1862 testcase( IS_BIG_INT(szDb
) );
1863 rc
= sqlite3OsTruncate(pWal
->pDbFd
, szDb
);
1864 if( rc
==SQLITE_OK
){
1865 rc
= sqlite3OsSync(pWal
->pDbFd
, CKPT_SYNC_FLAGS(sync_flags
));
1868 if( rc
==SQLITE_OK
){
1869 pInfo
->nBackfill
= mxSafeFrame
;
1873 /* Release the reader lock held while backfilling */
1874 walUnlockExclusive(pWal
, WAL_READ_LOCK(0), 1);
1877 if( rc
==SQLITE_BUSY
){
1878 /* Reset the return code so as not to report a checkpoint failure
1879 ** just because there are active readers. */
1884 /* If this is an SQLITE_CHECKPOINT_RESTART or TRUNCATE operation, and the
1885 ** entire wal file has been copied into the database file, then block
1886 ** until all readers have finished using the wal file. This ensures that
1887 ** the next process to write to the database restarts the wal file.
1889 if( rc
==SQLITE_OK
&& eMode
!=SQLITE_CHECKPOINT_PASSIVE
){
1890 assert( pWal
->writeLock
);
1891 if( pInfo
->nBackfill
<pWal
->hdr
.mxFrame
){
1893 }else if( eMode
>=SQLITE_CHECKPOINT_RESTART
){
1895 sqlite3_randomness(4, &salt1
);
1896 assert( pInfo
->nBackfill
==pWal
->hdr
.mxFrame
);
1897 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1898 if( rc
==SQLITE_OK
){
1899 if( eMode
==SQLITE_CHECKPOINT_TRUNCATE
){
1900 /* IMPLEMENTATION-OF: R-44699-57140 This mode works the same way as
1901 ** SQLITE_CHECKPOINT_RESTART with the addition that it also
1902 ** truncates the log file to zero bytes just prior to a
1903 ** successful return.
1905 ** In theory, it might be safe to do this without updating the
1906 ** wal-index header in shared memory, as all subsequent reader or
1907 ** writer clients should see that the entire log file has been
1908 ** checkpointed and behave accordingly. This seems unsafe though,
1909 ** as it would leave the system in a state where the contents of
1910 ** the wal-index header do not match the contents of the
1911 ** file-system. To avoid this, update the wal-index header to
1912 ** indicate that the log file contains zero valid frames. */
1913 walRestartHdr(pWal
, salt1
);
1914 rc
= sqlite3OsTruncate(pWal
->pWalFd
, 0);
1916 walUnlockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
1922 walIteratorFree(pIter
);
1927 ** If the WAL file is currently larger than nMax bytes in size, truncate
1928 ** it to exactly nMax bytes. If an error occurs while doing so, ignore it.
1930 static void walLimitSize(Wal
*pWal
, i64 nMax
){
1933 sqlite3BeginBenignMalloc();
1934 rx
= sqlite3OsFileSize(pWal
->pWalFd
, &sz
);
1935 if( rx
==SQLITE_OK
&& (sz
> nMax
) ){
1936 rx
= sqlite3OsTruncate(pWal
->pWalFd
, nMax
);
1938 sqlite3EndBenignMalloc();
1940 sqlite3_log(rx
, "cannot limit WAL size: %s", pWal
->zWalName
);
1945 ** Close a connection to a log file.
1947 int sqlite3WalClose(
1948 Wal
*pWal
, /* Wal to close */
1949 sqlite3
*db
, /* For interrupt flag */
1950 int sync_flags
, /* Flags to pass to OsSync() (or 0) */
1952 u8
*zBuf
/* Buffer of at least nBuf bytes */
1956 int isDelete
= 0; /* True to unlink wal and wal-index files */
1958 /* If an EXCLUSIVE lock can be obtained on the database file (using the
1959 ** ordinary, rollback-mode locking methods, this guarantees that the
1960 ** connection associated with this log file is the only connection to
1961 ** the database. In this case checkpoint the database and unlink both
1962 ** the wal and wal-index files.
1964 ** The EXCLUSIVE lock is not released before returning.
1967 && SQLITE_OK
==(rc
= sqlite3OsLock(pWal
->pDbFd
, SQLITE_LOCK_EXCLUSIVE
))
1969 if( pWal
->exclusiveMode
==WAL_NORMAL_MODE
){
1970 pWal
->exclusiveMode
= WAL_EXCLUSIVE_MODE
;
1972 rc
= sqlite3WalCheckpoint(pWal
, db
,
1973 SQLITE_CHECKPOINT_PASSIVE
, 0, 0, sync_flags
, nBuf
, zBuf
, 0, 0
1975 if( rc
==SQLITE_OK
){
1977 sqlite3OsFileControlHint(
1978 pWal
->pDbFd
, SQLITE_FCNTL_PERSIST_WAL
, &bPersist
1981 /* Try to delete the WAL file if the checkpoint completed and
1982 ** fsyned (rc==SQLITE_OK) and if we are not in persistent-wal
1983 ** mode (!bPersist) */
1985 }else if( pWal
->mxWalSize
>=0 ){
1986 /* Try to truncate the WAL file to zero bytes if the checkpoint
1987 ** completed and fsynced (rc==SQLITE_OK) and we are in persistent
1988 ** WAL mode (bPersist) and if the PRAGMA journal_size_limit is a
1989 ** non-negative value (pWal->mxWalSize>=0). Note that we truncate
1990 ** to zero bytes as truncating to the journal_size_limit might
1991 ** leave a corrupt WAL file on disk. */
1992 walLimitSize(pWal
, 0);
1997 walIndexClose(pWal
, isDelete
);
1998 sqlite3OsClose(pWal
->pWalFd
);
2000 sqlite3BeginBenignMalloc();
2001 sqlite3OsDelete(pWal
->pVfs
, pWal
->zWalName
, 0);
2002 sqlite3EndBenignMalloc();
2004 WALTRACE(("WAL%p: closed\n", pWal
));
2005 sqlite3_free((void *)pWal
->apWiData
);
2012 ** Try to read the wal-index header. Return 0 on success and 1 if
2013 ** there is a problem.
2015 ** The wal-index is in shared memory. Another thread or process might
2016 ** be writing the header at the same time this procedure is trying to
2017 ** read it, which might result in inconsistency. A dirty read is detected
2018 ** by verifying that both copies of the header are the same and also by
2019 ** a checksum on the header.
2021 ** If and only if the read is consistent and the header is different from
2022 ** pWal->hdr, then pWal->hdr is updated to the content of the new header
2023 ** and *pChanged is set to 1.
2025 ** If the checksum cannot be verified return non-zero. If the header
2026 ** is read successfully and the checksum verified, return zero.
2028 static int walIndexTryHdr(Wal
*pWal
, int *pChanged
){
2029 u32 aCksum
[2]; /* Checksum on the header content */
2030 WalIndexHdr h1
, h2
; /* Two copies of the header content */
2031 WalIndexHdr
volatile *aHdr
; /* Header in shared memory */
2033 /* The first page of the wal-index must be mapped at this point. */
2034 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
2036 /* Read the header. This might happen concurrently with a write to the
2037 ** same area of shared memory on a different CPU in a SMP,
2038 ** meaning it is possible that an inconsistent snapshot is read
2039 ** from the file. If this happens, return non-zero.
2041 ** There are two copies of the header at the beginning of the wal-index.
2042 ** When reading, read [0] first then [1]. Writes are in the reverse order.
2043 ** Memory barriers are used to prevent the compiler or the hardware from
2044 ** reordering the reads and writes.
2046 aHdr
= walIndexHdr(pWal
);
2047 memcpy(&h1
, (void *)&aHdr
[0], sizeof(h1
));
2048 walShmBarrier(pWal
);
2049 memcpy(&h2
, (void *)&aHdr
[1], sizeof(h2
));
2051 if( memcmp(&h1
, &h2
, sizeof(h1
))!=0 ){
2052 return 1; /* Dirty read */
2055 return 1; /* Malformed header - probably all zeros */
2057 walChecksumBytes(1, (u8
*)&h1
, sizeof(h1
)-sizeof(h1
.aCksum
), 0, aCksum
);
2058 if( aCksum
[0]!=h1
.aCksum
[0] || aCksum
[1]!=h1
.aCksum
[1] ){
2059 return 1; /* Checksum does not match */
2062 if( memcmp(&pWal
->hdr
, &h1
, sizeof(WalIndexHdr
)) ){
2064 memcpy(&pWal
->hdr
, &h1
, sizeof(WalIndexHdr
));
2065 pWal
->szPage
= (pWal
->hdr
.szPage
&0xfe00) + ((pWal
->hdr
.szPage
&0x0001)<<16);
2066 testcase( pWal
->szPage
<=32768 );
2067 testcase( pWal
->szPage
>=65536 );
2070 /* The header was successfully read. Return zero. */
2075 ** This is the value that walTryBeginRead returns when it needs to
2078 #define WAL_RETRY (-1)
2081 ** Read the wal-index header from the wal-index and into pWal->hdr.
2082 ** If the wal-header appears to be corrupt, try to reconstruct the
2083 ** wal-index from the WAL before returning.
2085 ** Set *pChanged to 1 if the wal-index header value in pWal->hdr is
2086 ** changed by this operation. If pWal->hdr is unchanged, set *pChanged
2089 ** If the wal-index header is successfully read, return SQLITE_OK.
2090 ** Otherwise an SQLite error code.
2092 static int walIndexReadHdr(Wal
*pWal
, int *pChanged
){
2093 int rc
; /* Return code */
2094 int badHdr
; /* True if a header read failed */
2095 volatile u32
*page0
; /* Chunk of wal-index containing header */
2097 /* Ensure that page 0 of the wal-index (the page that contains the
2098 ** wal-index header) is mapped. Return early if an error occurs here.
2101 rc
= walIndexPage(pWal
, 0, &page0
);
2102 if( rc
==SQLITE_READONLY_CANTINIT
){
2103 assert( page0
==0 && pWal
->writeLock
==0 );
2104 pWal
->bUnlocked
= 1;
2105 pWal
->exclusiveMode
= WAL_HEAPMEMORY_MODE
;
2108 if( rc
!=SQLITE_OK
){
2111 assert( page0
|| pWal
->writeLock
==0 );
2113 /* If the first page of the wal-index has been mapped, try to read the
2114 ** wal-index header immediately, without holding any lock. This usually
2115 ** works, but may fail if the wal-index header is corrupt or currently
2116 ** being modified by another thread or process.
2118 badHdr
= (page0
? walIndexTryHdr(pWal
, pChanged
) : 1);
2120 /* If the first attempt failed, it might have been due to a race
2121 ** with a writer. So get a WRITE lock and try again.
2123 assert( badHdr
==0 || pWal
->writeLock
==0 );
2125 if( pWal
->bUnlocked
==0 && (pWal
->readOnly
& WAL_SHM_RDONLY
) ){
2126 if( SQLITE_OK
==(rc
= walLockShared(pWal
, WAL_WRITE_LOCK
)) ){
2127 walUnlockShared(pWal
, WAL_WRITE_LOCK
);
2128 rc
= SQLITE_READONLY_RECOVERY
;
2130 }else if( SQLITE_OK
==(rc
= walLockExclusive(pWal
, WAL_WRITE_LOCK
, 1)) ){
2131 pWal
->writeLock
= 1;
2132 if( SQLITE_OK
==(rc
= walIndexPage(pWal
, 0, &page0
)) ){
2133 badHdr
= walIndexTryHdr(pWal
, pChanged
);
2135 /* If the wal-index header is still malformed even while holding
2136 ** a WRITE lock, it can only mean that the header is corrupted and
2137 ** needs to be reconstructed. So run recovery to do exactly that.
2139 rc
= walIndexRecover(pWal
);
2143 pWal
->writeLock
= 0;
2144 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2148 /* If the header is read successfully, check the version number to make
2149 ** sure the wal-index was not constructed with some future format that
2150 ** this version of SQLite cannot understand.
2152 if( badHdr
==0 && pWal
->hdr
.iVersion
!=WALINDEX_MAX_VERSION
){
2153 rc
= SQLITE_CANTOPEN_BKPT
;
2155 if( pWal
->bUnlocked
){
2156 if( rc
!=SQLITE_OK
){
2157 walIndexClose(pWal
, 0);
2158 pWal
->bUnlocked
= 0;
2159 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0]==0 );
2160 if( rc
==SQLITE_IOERR_SHORT_READ
) rc
= WAL_RETRY
;
2162 pWal
->exclusiveMode
= WAL_NORMAL_MODE
;
2169 ** Open an "unlocked" transaction. An unlocked transaction is a read
2170 ** transaction used by a read-only client in cases where the *-shm
2171 ** file cannot be mapped and its contents cannot be trusted. It is
2172 ** assumed that the *-wal file has been read and that a wal-index
2173 ** constructed in heap memory is currently available in Wal.apWiData[].
2175 ** If this function returns SQLITE_OK, then the read transaction has
2176 ** been successfully opened. In this case output variable (*pChanged)
2177 ** is set to true before returning if the caller should discard the
2178 ** contents of the page cache before proceeding. Or, if it returns
2179 ** WAL_RETRY, then the heap memory wal-index has been discarded and
2180 ** the caller should retry opening the read transaction from the
2181 ** beginning (including attempting to map the *-shm file).
2183 ** If an error occurs, an SQLite error code is returned.
2185 static int walBeginUnlocked(Wal
*pWal
, int *pChanged
){
2186 i64 szWal
; /* Size of wal file on disk in bytes */
2187 i64 iOffset
; /* Current offset when reading wal file */
2188 u8 aBuf
[WAL_HDRSIZE
]; /* Buffer to load WAL header into */
2189 u8
*aFrame
= 0; /* Malloc'd buffer to load entire frame */
2190 int szFrame
; /* Number of bytes in buffer aFrame[] */
2191 u8
*aData
; /* Pointer to data part of aFrame buffer */
2192 volatile void *pDummy
; /* Dummy argument for xShmMap */
2193 int rc
; /* Return code */
2194 u32 aSaveCksum
[2]; /* Saved copy of pWal->hdr.aFrameCksum */
2196 assert( pWal
->bUnlocked
);
2197 assert( pWal
->readOnly
& WAL_SHM_RDONLY
);
2198 assert( pWal
->nWiData
>0 && pWal
->apWiData
[0] );
2200 /* Take WAL_READ_LOCK(0). This has the effect of preventing any
2201 ** live clients from running a checkpoint, but does not stop them
2202 ** from running recovery. */
2203 rc
= walLockShared(pWal
, WAL_READ_LOCK(0));
2204 if( rc
!=SQLITE_OK
){
2205 if( rc
==SQLITE_BUSY
) rc
= WAL_RETRY
;
2206 goto begin_unlocked_out
;
2210 /* Try to map the *-shm file again. If it succeeds this time, then
2211 ** a non-readonly_shm connection has already connected to the database.
2212 ** In this case, start over with opening the transaction.
2214 ** The *-shm file was opened read-only, so sqlite3OsShmMap() can never
2215 ** return SQLITE_OK here, as that would imply that it had established
2216 ** a read/write mapping. A return of SQLITE_READONLY means success - that
2217 ** a mapping has been established to a shared-memory segment that is actively
2218 ** maintained by a writer. SQLITE_READONLY_CANTINIT means that all
2219 ** all connections to the -shm file are read-only and hence the content
2220 ** of the -shm file might be out-of-date.
2222 ** The WAL_READ_LOCK(0) lock held by this client prevents a checkpoint
2223 ** from taking place. But it does not prevent the wal from being wrapped
2224 ** if a checkpoint has already taken place. This means that if another
2225 ** client is connected at this point, it may have already checkpointed
2226 ** the entire wal. In that case it would not be safe to continue with
2227 ** the unlocked transaction, as the other client may overwrite wal
2228 ** frames that this client is still using. */
2229 rc
= sqlite3OsShmMap(pWal
->pDbFd
, 0, WALINDEX_PGSZ
, 0, &pDummy
);
2230 assert( rc
!=SQLITE_OK
); /* SQLITE_OK not possible for read-only connection */
2231 if( rc
!=SQLITE_READONLY_CANTINIT
){
2232 rc
= (rc
==SQLITE_READONLY
? WAL_RETRY
: rc
);
2233 goto begin_unlocked_out
;
2236 memcpy(&pWal
->hdr
, (void*)walIndexHdr(pWal
), sizeof(WalIndexHdr
));
2237 rc
= sqlite3OsFileSize(pWal
->pWalFd
, &szWal
);
2238 if( rc
!=SQLITE_OK
){
2239 goto begin_unlocked_out
;
2241 if( szWal
<WAL_HDRSIZE
){
2242 /* If the wal file is too small to contain a wal-header and the
2243 ** wal-index header has mxFrame==0, then it must be safe to proceed
2244 ** reading the database file only. However, the page cache cannot
2245 ** be trusted, as a read/write connection may have connected, written
2246 ** the db, run a checkpoint, truncated the wal file and disconnected
2247 ** since this client's last read transaction. */
2249 rc
= (pWal
->hdr
.mxFrame
==0 ? SQLITE_OK
: WAL_RETRY
);
2250 goto begin_unlocked_out
;
2253 /* Check the salt keys at the start of the wal file still match. */
2254 rc
= sqlite3OsRead(pWal
->pWalFd
, aBuf
, WAL_HDRSIZE
, 0);
2255 if( rc
!=SQLITE_OK
){
2256 goto begin_unlocked_out
;
2258 if( memcmp(&pWal
->hdr
.aSalt
, &aBuf
[16], 8) ){
2260 goto begin_unlocked_out
;
2263 /* Allocate a buffer to read frames into */
2264 szFrame
= pWal
->hdr
.szPage
+ WAL_FRAME_HDRSIZE
;
2265 aFrame
= (u8
*)sqlite3_malloc64(szFrame
);
2267 rc
= SQLITE_NOMEM_BKPT
;
2268 goto begin_unlocked_out
;
2270 aData
= &aFrame
[WAL_FRAME_HDRSIZE
];
2272 /* Check to see if a complete transaction has been appended to the
2273 ** wal file since the heap-memory wal-index was created. If so, the
2274 ** heap-memory wal-index is discarded and WAL_RETRY returned to
2276 aSaveCksum
[0] = pWal
->hdr
.aFrameCksum
[0];
2277 aSaveCksum
[1] = pWal
->hdr
.aFrameCksum
[1];
2278 for(iOffset
=walFrameOffset(pWal
->hdr
.mxFrame
+1, pWal
->hdr
.szPage
);
2279 iOffset
+szFrame
<=szWal
;
2282 u32 pgno
; /* Database page number for frame */
2283 u32 nTruncate
; /* dbsize field from frame header */
2285 /* Read and decode the next log frame. */
2286 rc
= sqlite3OsRead(pWal
->pWalFd
, aFrame
, szFrame
, iOffset
);
2287 if( rc
!=SQLITE_OK
) break;
2288 if( !walDecodeFrame(pWal
, &pgno
, &nTruncate
, aData
, aFrame
) ) break;
2290 /* If nTruncate is non-zero, then a complete transaction has been
2291 ** appended to this wal file. Set rc to WAL_RETRY and break out of
2298 pWal
->hdr
.aFrameCksum
[0] = aSaveCksum
[0];
2299 pWal
->hdr
.aFrameCksum
[1] = aSaveCksum
[1];
2302 sqlite3_free(aFrame
);
2303 if( rc
!=SQLITE_OK
){
2305 for(i
=0; i
<pWal
->nWiData
; i
++){
2306 sqlite3_free((void*)pWal
->apWiData
[i
]);
2307 pWal
->apWiData
[i
] = 0;
2309 pWal
->bUnlocked
= 0;
2310 sqlite3WalEndReadTransaction(pWal
);
2317 ** Attempt to start a read transaction. This might fail due to a race or
2318 ** other transient condition. When that happens, it returns WAL_RETRY to
2319 ** indicate to the caller that it is safe to retry immediately.
2321 ** On success return SQLITE_OK. On a permanent failure (such an
2322 ** I/O error or an SQLITE_BUSY because another process is running
2323 ** recovery) return a positive error code.
2325 ** The useWal parameter is true to force the use of the WAL and disable
2326 ** the case where the WAL is bypassed because it has been completely
2327 ** checkpointed. If useWal==0 then this routine calls walIndexReadHdr()
2328 ** to make a copy of the wal-index header into pWal->hdr. If the
2329 ** wal-index header has changed, *pChanged is set to 1 (as an indication
2330 ** to the caller that the local page cache is obsolete and needs to be
2331 ** flushed.) When useWal==1, the wal-index header is assumed to already
2332 ** be loaded and the pChanged parameter is unused.
2334 ** The caller must set the cnt parameter to the number of prior calls to
2335 ** this routine during the current read attempt that returned WAL_RETRY.
2336 ** This routine will start taking more aggressive measures to clear the
2337 ** race conditions after multiple WAL_RETRY returns, and after an excessive
2338 ** number of errors will ultimately return SQLITE_PROTOCOL. The
2339 ** SQLITE_PROTOCOL return indicates that some other process has gone rogue
2340 ** and is not honoring the locking protocol. There is a vanishingly small
2341 ** chance that SQLITE_PROTOCOL could be returned because of a run of really
2342 ** bad luck when there is lots of contention for the wal-index, but that
2343 ** possibility is so small that it can be safely neglected, we believe.
2345 ** On success, this routine obtains a read lock on
2346 ** WAL_READ_LOCK(pWal->readLock). The pWal->readLock integer is
2347 ** in the range 0 <= pWal->readLock < WAL_NREADER. If pWal->readLock==(-1)
2348 ** that means the Wal does not hold any read lock. The reader must not
2349 ** access any database page that is modified by a WAL frame up to and
2350 ** including frame number aReadMark[pWal->readLock]. The reader will
2351 ** use WAL frames up to and including pWal->hdr.mxFrame if pWal->readLock>0
2352 ** Or if pWal->readLock==0, then the reader will ignore the WAL
2353 ** completely and get all content directly from the database file.
2354 ** If the useWal parameter is 1 then the WAL will never be ignored and
2355 ** this routine will always set pWal->readLock>0 on success.
2356 ** When the read transaction is completed, the caller must release the
2357 ** lock on WAL_READ_LOCK(pWal->readLock) and set pWal->readLock to -1.
2359 ** This routine uses the nBackfill and aReadMark[] fields of the header
2360 ** to select a particular WAL_READ_LOCK() that strives to let the
2361 ** checkpoint process do as much work as possible. This routine might
2362 ** update values of the aReadMark[] array in the header, but if it does
2363 ** so it takes care to hold an exclusive lock on the corresponding
2364 ** WAL_READ_LOCK() while changing values.
2366 static int walTryBeginRead(Wal
*pWal
, int *pChanged
, int useWal
, int cnt
){
2367 volatile WalCkptInfo
*pInfo
; /* Checkpoint information in wal-index */
2368 u32 mxReadMark
; /* Largest aReadMark[] value */
2369 int mxI
; /* Index of largest aReadMark[] value */
2370 int i
; /* Loop counter */
2371 int rc
= SQLITE_OK
; /* Return code */
2372 u32 mxFrame
; /* Wal frame to lock to */
2374 assert( pWal
->readLock
<0 ); /* Not currently locked */
2376 /* Take steps to avoid spinning forever if there is a protocol error.
2378 ** Circumstances that cause a RETRY should only last for the briefest
2379 ** instances of time. No I/O or other system calls are done while the
2380 ** locks are held, so the locks should not be held for very long. But
2381 ** if we are unlucky, another process that is holding a lock might get
2382 ** paged out or take a page-fault that is time-consuming to resolve,
2383 ** during the few nanoseconds that it is holding the lock. In that case,
2384 ** it might take longer than normal for the lock to free.
2386 ** After 5 RETRYs, we begin calling sqlite3OsSleep(). The first few
2387 ** calls to sqlite3OsSleep() have a delay of 1 microsecond. Really this
2388 ** is more of a scheduler yield than an actual delay. But on the 10th
2389 ** an subsequent retries, the delays start becoming longer and longer,
2390 ** so that on the 100th (and last) RETRY we delay for 323 milliseconds.
2391 ** The total delay time before giving up is less than 10 seconds.
2394 int nDelay
= 1; /* Pause time in microseconds */
2396 VVA_ONLY( pWal
->lockError
= 1; )
2397 return SQLITE_PROTOCOL
;
2399 if( cnt
>=10 ) nDelay
= (cnt
-9)*(cnt
-9)*39;
2400 sqlite3OsSleep(pWal
->pVfs
, nDelay
);
2404 assert( rc
==SQLITE_OK
);
2405 if( pWal
->bUnlocked
==0 ){
2406 rc
= walIndexReadHdr(pWal
, pChanged
);
2408 if( rc
==SQLITE_BUSY
){
2409 /* If there is not a recovery running in another thread or process
2410 ** then convert BUSY errors to WAL_RETRY. If recovery is known to
2411 ** be running, convert BUSY to BUSY_RECOVERY. There is a race here
2412 ** which might cause WAL_RETRY to be returned even if BUSY_RECOVERY
2413 ** would be technically correct. But the race is benign since with
2414 ** WAL_RETRY this routine will be called again and will probably be
2415 ** right on the second iteration.
2417 if( pWal
->apWiData
[0]==0 ){
2418 /* This branch is taken when the xShmMap() method returns SQLITE_BUSY.
2419 ** We assume this is a transient condition, so return WAL_RETRY. The
2420 ** xShmMap() implementation used by the default unix and win32 VFS
2421 ** modules may return SQLITE_BUSY due to a race condition in the
2422 ** code that determines whether or not the shared-memory region
2423 ** must be zeroed before the requested page is returned.
2426 }else if( SQLITE_OK
==(rc
= walLockShared(pWal
, WAL_RECOVER_LOCK
)) ){
2427 walUnlockShared(pWal
, WAL_RECOVER_LOCK
);
2429 }else if( rc
==SQLITE_BUSY
){
2430 rc
= SQLITE_BUSY_RECOVERY
;
2433 if( rc
!=SQLITE_OK
){
2436 else if( pWal
->bUnlocked
){
2437 return walBeginUnlocked(pWal
, pChanged
);
2441 assert( pWal
->nWiData
>0 );
2442 assert( pWal
->apWiData
[0] || (pWal
->readOnly
& WAL_SHM_RDONLY
) );
2443 pInfo
= pWal
->apWiData
[0] ? walCkptInfo(pWal
) : 0;
2444 if( !useWal
&& (pInfo
==0 || pInfo
->nBackfill
==pWal
->hdr
.mxFrame
)
2445 #ifdef SQLITE_ENABLE_SNAPSHOT
2446 && (pWal
->pSnapshot
==0 || pWal
->hdr
.mxFrame
==0
2447 || 0==memcmp(&pWal
->hdr
, pWal
->pSnapshot
, sizeof(WalIndexHdr
)))
2450 /* The WAL has been completely backfilled (or it is empty).
2451 ** and can be safely ignored.
2453 rc
= walLockShared(pWal
, WAL_READ_LOCK(0));
2454 walShmBarrier(pWal
);
2455 if( rc
==SQLITE_OK
){
2457 && memcmp((void *)walIndexHdr(pWal
), &pWal
->hdr
, sizeof(WalIndexHdr
))
2459 /* It is not safe to allow the reader to continue here if frames
2460 ** may have been appended to the log before READ_LOCK(0) was obtained.
2461 ** When holding READ_LOCK(0), the reader ignores the entire log file,
2462 ** which implies that the database file contains a trustworthy
2463 ** snapshot. Since holding READ_LOCK(0) prevents a checkpoint from
2464 ** happening, this is usually correct.
2466 ** However, if frames have been appended to the log (or if the log
2467 ** is wrapped and written for that matter) before the READ_LOCK(0)
2468 ** is obtained, that is not necessarily true. A checkpointer may
2469 ** have started to backfill the appended frames but crashed before
2470 ** it finished. Leaving a corrupt image in the database file.
2472 walUnlockShared(pWal
, WAL_READ_LOCK(0));
2477 }else if( rc
!=SQLITE_BUSY
){
2482 /* If we get this far, it means that the reader will want to use
2483 ** the WAL to get at content from recent commits. The job now is
2484 ** to select one of the aReadMark[] entries that is closest to
2485 ** but not exceeding pWal->hdr.mxFrame and lock that entry.
2489 mxFrame
= pWal
->hdr
.mxFrame
;
2490 #ifdef SQLITE_ENABLE_SNAPSHOT
2491 if( pWal
->pSnapshot
&& pWal
->pSnapshot
->mxFrame
<mxFrame
){
2492 mxFrame
= pWal
->pSnapshot
->mxFrame
;
2495 for(i
=1; i
<WAL_NREADER
; i
++){
2496 u32 thisMark
= pInfo
->aReadMark
[i
];
2497 if( mxReadMark
<=thisMark
&& thisMark
<=mxFrame
){
2498 assert( thisMark
!=READMARK_NOT_USED
);
2499 mxReadMark
= thisMark
;
2503 if( (pWal
->readOnly
& WAL_SHM_RDONLY
)==0
2504 && (mxReadMark
<mxFrame
|| mxI
==0)
2506 for(i
=1; i
<WAL_NREADER
; i
++){
2507 rc
= walLockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
2508 if( rc
==SQLITE_OK
){
2509 mxReadMark
= pInfo
->aReadMark
[i
] = mxFrame
;
2511 walUnlockExclusive(pWal
, WAL_READ_LOCK(i
), 1);
2513 }else if( rc
!=SQLITE_BUSY
){
2519 assert( rc
==SQLITE_BUSY
|| (pWal
->readOnly
& WAL_SHM_RDONLY
)!=0 );
2520 return rc
==SQLITE_BUSY
? WAL_RETRY
: SQLITE_READONLY_CANTINIT
;
2523 rc
= walLockShared(pWal
, WAL_READ_LOCK(mxI
));
2525 return rc
==SQLITE_BUSY
? WAL_RETRY
: rc
;
2527 /* Now that the read-lock has been obtained, check that neither the
2528 ** value in the aReadMark[] array or the contents of the wal-index
2529 ** header have changed.
2531 ** It is necessary to check that the wal-index header did not change
2532 ** between the time it was read and when the shared-lock was obtained
2533 ** on WAL_READ_LOCK(mxI) was obtained to account for the possibility
2534 ** that the log file may have been wrapped by a writer, or that frames
2535 ** that occur later in the log than pWal->hdr.mxFrame may have been
2536 ** copied into the database by a checkpointer. If either of these things
2537 ** happened, then reading the database with the current value of
2538 ** pWal->hdr.mxFrame risks reading a corrupted snapshot. So, retry
2541 ** Before checking that the live wal-index header has not changed
2542 ** since it was read, set Wal.minFrame to the first frame in the wal
2543 ** file that has not yet been checkpointed. This client will not need
2544 ** to read any frames earlier than minFrame from the wal file - they
2545 ** can be safely read directly from the database file.
2547 ** Because a ShmBarrier() call is made between taking the copy of
2548 ** nBackfill and checking that the wal-header in shared-memory still
2549 ** matches the one cached in pWal->hdr, it is guaranteed that the
2550 ** checkpointer that set nBackfill was not working with a wal-index
2551 ** header newer than that cached in pWal->hdr. If it were, that could
2552 ** cause a problem. The checkpointer could omit to checkpoint
2553 ** a version of page X that lies before pWal->minFrame (call that version
2554 ** A) on the basis that there is a newer version (version B) of the same
2555 ** page later in the wal file. But if version B happens to like past
2556 ** frame pWal->hdr.mxFrame - then the client would incorrectly assume
2557 ** that it can read version A from the database file. However, since
2558 ** we can guarantee that the checkpointer that set nBackfill could not
2559 ** see any pages past pWal->hdr.mxFrame, this problem does not come up.
2561 pWal
->minFrame
= pInfo
->nBackfill
+1;
2562 walShmBarrier(pWal
);
2563 if( pInfo
->aReadMark
[mxI
]!=mxReadMark
2564 || memcmp((void *)walIndexHdr(pWal
), &pWal
->hdr
, sizeof(WalIndexHdr
))
2566 walUnlockShared(pWal
, WAL_READ_LOCK(mxI
));
2569 assert( mxReadMark
<=pWal
->hdr
.mxFrame
);
2570 pWal
->readLock
= (i16
)mxI
;
2575 #ifdef SQLITE_ENABLE_SNAPSHOT
2577 ** Attempt to reduce the value of the WalCkptInfo.nBackfillAttempted
2578 ** variable so that older snapshots can be accessed. To do this, loop
2579 ** through all wal frames from nBackfillAttempted to (nBackfill+1),
2580 ** comparing their content to the corresponding page with the database
2581 ** file, if any. Set nBackfillAttempted to the frame number of the
2582 ** first frame for which the wal file content matches the db file.
2584 ** This is only really safe if the file-system is such that any page
2585 ** writes made by earlier checkpointers were atomic operations, which
2586 ** is not always true. It is also possible that nBackfillAttempted
2587 ** may be left set to a value larger than expected, if a wal frame
2588 ** contains content that duplicate of an earlier version of the same
2591 ** SQLITE_OK is returned if successful, or an SQLite error code if an
2592 ** error occurs. It is not an error if nBackfillAttempted cannot be
2593 ** decreased at all.
2595 int sqlite3WalSnapshotRecover(Wal
*pWal
){
2598 assert( pWal
->readLock
>=0 );
2599 rc
= walLockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
2600 if( rc
==SQLITE_OK
){
2601 volatile WalCkptInfo
*pInfo
= walCkptInfo(pWal
);
2602 int szPage
= (int)pWal
->szPage
;
2603 i64 szDb
; /* Size of db file in bytes */
2605 rc
= sqlite3OsFileSize(pWal
->pDbFd
, &szDb
);
2606 if( rc
==SQLITE_OK
){
2607 void *pBuf1
= sqlite3_malloc(szPage
);
2608 void *pBuf2
= sqlite3_malloc(szPage
);
2609 if( pBuf1
==0 || pBuf2
==0 ){
2612 u32 i
= pInfo
->nBackfillAttempted
;
2613 for(i
=pInfo
->nBackfillAttempted
; i
>pInfo
->nBackfill
; i
--){
2614 volatile ht_slot
*dummy
;
2615 volatile u32
*aPgno
; /* Array of page numbers */
2616 u32 iZero
; /* Frame corresponding to aPgno[0] */
2617 u32 pgno
; /* Page number in db file */
2618 i64 iDbOff
; /* Offset of db file entry */
2619 i64 iWalOff
; /* Offset of wal file entry */
2621 rc
= walHashGet(pWal
, walFramePage(i
), &dummy
, &aPgno
, &iZero
);
2622 if( rc
!=SQLITE_OK
) break;
2623 pgno
= aPgno
[i
-iZero
];
2624 iDbOff
= (i64
)(pgno
-1) * szPage
;
2626 if( iDbOff
+szPage
<=szDb
){
2627 iWalOff
= walFrameOffset(i
, szPage
) + WAL_FRAME_HDRSIZE
;
2628 rc
= sqlite3OsRead(pWal
->pWalFd
, pBuf1
, szPage
, iWalOff
);
2630 if( rc
==SQLITE_OK
){
2631 rc
= sqlite3OsRead(pWal
->pDbFd
, pBuf2
, szPage
, iDbOff
);
2634 if( rc
!=SQLITE_OK
|| 0==memcmp(pBuf1
, pBuf2
, szPage
) ){
2639 pInfo
->nBackfillAttempted
= i
-1;
2643 sqlite3_free(pBuf1
);
2644 sqlite3_free(pBuf2
);
2646 walUnlockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
2651 #endif /* SQLITE_ENABLE_SNAPSHOT */
2654 ** Begin a read transaction on the database.
2656 ** This routine used to be called sqlite3OpenSnapshot() and with good reason:
2657 ** it takes a snapshot of the state of the WAL and wal-index for the current
2658 ** instant in time. The current thread will continue to use this snapshot.
2659 ** Other threads might append new content to the WAL and wal-index but
2660 ** that extra content is ignored by the current thread.
2662 ** If the database contents have changes since the previous read
2663 ** transaction, then *pChanged is set to 1 before returning. The
2664 ** Pager layer will use this to know that is cache is stale and
2665 ** needs to be flushed.
2667 int sqlite3WalBeginReadTransaction(Wal
*pWal
, int *pChanged
){
2668 int rc
; /* Return code */
2669 int cnt
= 0; /* Number of TryBeginRead attempts */
2671 #ifdef SQLITE_ENABLE_SNAPSHOT
2673 WalIndexHdr
*pSnapshot
= pWal
->pSnapshot
;
2674 if( pSnapshot
&& memcmp(pSnapshot
, &pWal
->hdr
, sizeof(WalIndexHdr
))!=0 ){
2680 rc
= walTryBeginRead(pWal
, pChanged
, 0, ++cnt
);
2681 }while( rc
==WAL_RETRY
);
2682 testcase( (rc
&0xff)==SQLITE_BUSY
);
2683 testcase( (rc
&0xff)==SQLITE_IOERR
);
2684 testcase( rc
==SQLITE_PROTOCOL
);
2685 testcase( rc
==SQLITE_OK
);
2687 #ifdef SQLITE_ENABLE_SNAPSHOT
2688 if( rc
==SQLITE_OK
){
2689 if( pSnapshot
&& memcmp(pSnapshot
, &pWal
->hdr
, sizeof(WalIndexHdr
))!=0 ){
2690 /* At this point the client has a lock on an aReadMark[] slot holding
2691 ** a value equal to or smaller than pSnapshot->mxFrame, but pWal->hdr
2692 ** is populated with the wal-index header corresponding to the head
2693 ** of the wal file. Verify that pSnapshot is still valid before
2694 ** continuing. Reasons why pSnapshot might no longer be valid:
2696 ** (1) The WAL file has been reset since the snapshot was taken.
2697 ** In this case, the salt will have changed.
2699 ** (2) A checkpoint as been attempted that wrote frames past
2700 ** pSnapshot->mxFrame into the database file. Note that the
2701 ** checkpoint need not have completed for this to cause problems.
2703 volatile WalCkptInfo
*pInfo
= walCkptInfo(pWal
);
2705 assert( pWal
->readLock
>0 || pWal
->hdr
.mxFrame
==0 );
2706 assert( pInfo
->aReadMark
[pWal
->readLock
]<=pSnapshot
->mxFrame
);
2708 /* It is possible that there is a checkpointer thread running
2709 ** concurrent with this code. If this is the case, it may be that the
2710 ** checkpointer has already determined that it will checkpoint
2711 ** snapshot X, where X is later in the wal file than pSnapshot, but
2712 ** has not yet set the pInfo->nBackfillAttempted variable to indicate
2713 ** its intent. To avoid the race condition this leads to, ensure that
2714 ** there is no checkpointer process by taking a shared CKPT lock
2715 ** before checking pInfo->nBackfillAttempted.
2717 ** TODO: Does the aReadMark[] lock prevent a checkpointer from doing
2720 rc
= walLockShared(pWal
, WAL_CKPT_LOCK
);
2722 if( rc
==SQLITE_OK
){
2723 /* Check that the wal file has not been wrapped. Assuming that it has
2724 ** not, also check that no checkpointer has attempted to checkpoint any
2725 ** frames beyond pSnapshot->mxFrame. If either of these conditions are
2726 ** true, return SQLITE_BUSY_SNAPSHOT. Otherwise, overwrite pWal->hdr
2727 ** with *pSnapshot and set *pChanged as appropriate for opening the
2729 if( !memcmp(pSnapshot
->aSalt
, pWal
->hdr
.aSalt
, sizeof(pWal
->hdr
.aSalt
))
2730 && pSnapshot
->mxFrame
>=pInfo
->nBackfillAttempted
2732 assert( pWal
->readLock
>0 );
2733 memcpy(&pWal
->hdr
, pSnapshot
, sizeof(WalIndexHdr
));
2734 *pChanged
= bChanged
;
2736 rc
= SQLITE_BUSY_SNAPSHOT
;
2739 /* Release the shared CKPT lock obtained above. */
2740 walUnlockShared(pWal
, WAL_CKPT_LOCK
);
2744 if( rc
!=SQLITE_OK
){
2745 sqlite3WalEndReadTransaction(pWal
);
2754 ** Finish with a read transaction. All this does is release the
2757 void sqlite3WalEndReadTransaction(Wal
*pWal
){
2758 sqlite3WalEndWriteTransaction(pWal
);
2759 if( pWal
->readLock
>=0 ){
2760 walUnlockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
));
2761 pWal
->readLock
= -1;
2766 ** Search the wal file for page pgno. If found, set *piRead to the frame that
2767 ** contains the page. Otherwise, if pgno is not in the wal file, set *piRead
2770 ** Return SQLITE_OK if successful, or an error code if an error occurs. If an
2771 ** error does occur, the final value of *piRead is undefined.
2773 int sqlite3WalFindFrame(
2774 Wal
*pWal
, /* WAL handle */
2775 Pgno pgno
, /* Database page number to read data for */
2776 u32
*piRead
/* OUT: Frame number (or zero) */
2778 u32 iRead
= 0; /* If !=0, WAL frame to return data from */
2779 u32 iLast
= pWal
->hdr
.mxFrame
; /* Last page in WAL for this reader */
2780 int iHash
; /* Used to loop through N hash tables */
2783 /* This routine is only be called from within a read transaction. */
2784 assert( pWal
->readLock
>=0 || pWal
->lockError
);
2786 /* If the "last page" field of the wal-index header snapshot is 0, then
2787 ** no data will be read from the wal under any circumstances. Return early
2788 ** in this case as an optimization. Likewise, if pWal->readLock==0,
2789 ** then the WAL is ignored by the reader so return early, as if the
2792 if( iLast
==0 || (pWal
->readLock
==0 && pWal
->bUnlocked
==0) ){
2797 /* Search the hash table or tables for an entry matching page number
2798 ** pgno. Each iteration of the following for() loop searches one
2799 ** hash table (each hash table indexes up to HASHTABLE_NPAGE frames).
2801 ** This code might run concurrently to the code in walIndexAppend()
2802 ** that adds entries to the wal-index (and possibly to this hash
2803 ** table). This means the value just read from the hash
2804 ** slot (aHash[iKey]) may have been added before or after the
2805 ** current read transaction was opened. Values added after the
2806 ** read transaction was opened may have been written incorrectly -
2807 ** i.e. these slots may contain garbage data. However, we assume
2808 ** that any slots written before the current read transaction was
2809 ** opened remain unmodified.
2811 ** For the reasons above, the if(...) condition featured in the inner
2812 ** loop of the following block is more stringent that would be required
2813 ** if we had exclusive access to the hash-table:
2815 ** (aPgno[iFrame]==pgno):
2816 ** This condition filters out normal hash-table collisions.
2819 ** This condition filters out entries that were added to the hash
2820 ** table after the current read-transaction had started.
2822 iMinHash
= walFramePage(pWal
->minFrame
);
2823 for(iHash
=walFramePage(iLast
); iHash
>=iMinHash
&& iRead
==0; iHash
--){
2824 volatile ht_slot
*aHash
; /* Pointer to hash table */
2825 volatile u32
*aPgno
; /* Pointer to array of page numbers */
2826 u32 iZero
; /* Frame number corresponding to aPgno[0] */
2827 int iKey
; /* Hash slot index */
2828 int nCollide
; /* Number of hash collisions remaining */
2829 int rc
; /* Error code */
2831 rc
= walHashGet(pWal
, iHash
, &aHash
, &aPgno
, &iZero
);
2832 if( rc
!=SQLITE_OK
){
2835 nCollide
= HASHTABLE_NSLOT
;
2836 for(iKey
=walHash(pgno
); aHash
[iKey
]; iKey
=walNextHash(iKey
)){
2837 u32 iFrame
= aHash
[iKey
] + iZero
;
2838 if( iFrame
<=iLast
&& iFrame
>=pWal
->minFrame
&& aPgno
[aHash
[iKey
]]==pgno
){
2839 assert( iFrame
>iRead
|| CORRUPT_DB
);
2842 if( (nCollide
--)==0 ){
2843 return SQLITE_CORRUPT_BKPT
;
2848 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
2849 /* If expensive assert() statements are available, do a linear search
2850 ** of the wal-index file content. Make sure the results agree with the
2851 ** result obtained using the hash indexes above. */
2855 assert( pWal
->bUnlocked
|| pWal
->minFrame
>0 );
2856 for(iTest
=iLast
; iTest
>=pWal
->minFrame
&& iTest
>0; iTest
--){
2857 if( walFramePgno(pWal
, iTest
)==pgno
){
2862 assert( iRead
==iRead2
);
2871 ** Read the contents of frame iRead from the wal file into buffer pOut
2872 ** (which is nOut bytes in size). Return SQLITE_OK if successful, or an
2873 ** error code otherwise.
2875 int sqlite3WalReadFrame(
2876 Wal
*pWal
, /* WAL handle */
2877 u32 iRead
, /* Frame to read */
2878 int nOut
, /* Size of buffer pOut in bytes */
2879 u8
*pOut
/* Buffer to write page data to */
2883 sz
= pWal
->hdr
.szPage
;
2884 sz
= (sz
&0xfe00) + ((sz
&0x0001)<<16);
2885 testcase( sz
<=32768 );
2886 testcase( sz
>=65536 );
2887 iOffset
= walFrameOffset(iRead
, sz
) + WAL_FRAME_HDRSIZE
;
2888 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL */
2889 return sqlite3OsRead(pWal
->pWalFd
, pOut
, (nOut
>sz
? sz
: nOut
), iOffset
);
2893 ** Return the size of the database in pages (or zero, if unknown).
2895 Pgno
sqlite3WalDbsize(Wal
*pWal
){
2896 if( pWal
&& ALWAYS(pWal
->readLock
>=0) ){
2897 return pWal
->hdr
.nPage
;
2904 ** This function starts a write transaction on the WAL.
2906 ** A read transaction must have already been started by a prior call
2907 ** to sqlite3WalBeginReadTransaction().
2909 ** If another thread or process has written into the database since
2910 ** the read transaction was started, then it is not possible for this
2911 ** thread to write as doing so would cause a fork. So this routine
2912 ** returns SQLITE_BUSY in that case and no write transaction is started.
2914 ** There can only be a single writer active at a time.
2916 int sqlite3WalBeginWriteTransaction(Wal
*pWal
){
2919 /* Cannot start a write transaction without first holding a read
2921 assert( pWal
->readLock
>=0 );
2922 assert( pWal
->writeLock
==0 && pWal
->iReCksum
==0 );
2924 if( pWal
->readOnly
){
2925 return SQLITE_READONLY
;
2928 /* Only one writer allowed at a time. Get the write lock. Return
2929 ** SQLITE_BUSY if unable.
2931 rc
= walLockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2935 pWal
->writeLock
= 1;
2937 /* If another connection has written to the database file since the
2938 ** time the read transaction on this connection was started, then
2939 ** the write is disallowed.
2941 if( memcmp(&pWal
->hdr
, (void *)walIndexHdr(pWal
), sizeof(WalIndexHdr
))!=0 ){
2942 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2943 pWal
->writeLock
= 0;
2944 rc
= SQLITE_BUSY_SNAPSHOT
;
2951 ** End a write transaction. The commit has already been done. This
2952 ** routine merely releases the lock.
2954 int sqlite3WalEndWriteTransaction(Wal
*pWal
){
2955 if( pWal
->writeLock
){
2956 walUnlockExclusive(pWal
, WAL_WRITE_LOCK
, 1);
2957 pWal
->writeLock
= 0;
2959 pWal
->truncateOnCommit
= 0;
2965 ** If any data has been written (but not committed) to the log file, this
2966 ** function moves the write-pointer back to the start of the transaction.
2968 ** Additionally, the callback function is invoked for each frame written
2969 ** to the WAL since the start of the transaction. If the callback returns
2970 ** other than SQLITE_OK, it is not invoked again and the error code is
2971 ** returned to the caller.
2973 ** Otherwise, if the callback function does not return an error, this
2974 ** function returns SQLITE_OK.
2976 int sqlite3WalUndo(Wal
*pWal
, int (*xUndo
)(void *, Pgno
), void *pUndoCtx
){
2978 if( ALWAYS(pWal
->writeLock
) ){
2979 Pgno iMax
= pWal
->hdr
.mxFrame
;
2982 /* Restore the clients cache of the wal-index header to the state it
2983 ** was in before the client began writing to the database.
2985 memcpy(&pWal
->hdr
, (void *)walIndexHdr(pWal
), sizeof(WalIndexHdr
));
2987 for(iFrame
=pWal
->hdr
.mxFrame
+1;
2988 ALWAYS(rc
==SQLITE_OK
) && iFrame
<=iMax
;
2991 /* This call cannot fail. Unless the page for which the page number
2992 ** is passed as the second argument is (a) in the cache and
2993 ** (b) has an outstanding reference, then xUndo is either a no-op
2994 ** (if (a) is false) or simply expels the page from the cache (if (b)
2997 ** If the upper layer is doing a rollback, it is guaranteed that there
2998 ** are no outstanding references to any page other than page 1. And
2999 ** page 1 is never written to the log until the transaction is
3000 ** committed. As a result, the call to xUndo may not fail.
3002 assert( walFramePgno(pWal
, iFrame
)!=1 );
3003 rc
= xUndo(pUndoCtx
, walFramePgno(pWal
, iFrame
));
3005 if( iMax
!=pWal
->hdr
.mxFrame
) walCleanupHash(pWal
);
3011 ** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32
3012 ** values. This function populates the array with values required to
3013 ** "rollback" the write position of the WAL handle back to the current
3014 ** point in the event of a savepoint rollback (via WalSavepointUndo()).
3016 void sqlite3WalSavepoint(Wal
*pWal
, u32
*aWalData
){
3017 assert( pWal
->writeLock
);
3018 aWalData
[0] = pWal
->hdr
.mxFrame
;
3019 aWalData
[1] = pWal
->hdr
.aFrameCksum
[0];
3020 aWalData
[2] = pWal
->hdr
.aFrameCksum
[1];
3021 aWalData
[3] = pWal
->nCkpt
;
3025 ** Move the write position of the WAL back to the point identified by
3026 ** the values in the aWalData[] array. aWalData must point to an array
3027 ** of WAL_SAVEPOINT_NDATA u32 values that has been previously populated
3028 ** by a call to WalSavepoint().
3030 int sqlite3WalSavepointUndo(Wal
*pWal
, u32
*aWalData
){
3033 assert( pWal
->writeLock
);
3034 assert( aWalData
[3]!=pWal
->nCkpt
|| aWalData
[0]<=pWal
->hdr
.mxFrame
);
3036 if( aWalData
[3]!=pWal
->nCkpt
){
3037 /* This savepoint was opened immediately after the write-transaction
3038 ** was started. Right after that, the writer decided to wrap around
3039 ** to the start of the log. Update the savepoint values to match.
3042 aWalData
[3] = pWal
->nCkpt
;
3045 if( aWalData
[0]<pWal
->hdr
.mxFrame
){
3046 pWal
->hdr
.mxFrame
= aWalData
[0];
3047 pWal
->hdr
.aFrameCksum
[0] = aWalData
[1];
3048 pWal
->hdr
.aFrameCksum
[1] = aWalData
[2];
3049 walCleanupHash(pWal
);
3056 ** This function is called just before writing a set of frames to the log
3057 ** file (see sqlite3WalFrames()). It checks to see if, instead of appending
3058 ** to the current log file, it is possible to overwrite the start of the
3059 ** existing log file with the new frames (i.e. "reset" the log). If so,
3060 ** it sets pWal->hdr.mxFrame to 0. Otherwise, pWal->hdr.mxFrame is left
3063 ** SQLITE_OK is returned if no error is encountered (regardless of whether
3064 ** or not pWal->hdr.mxFrame is modified). An SQLite error code is returned
3065 ** if an error occurs.
3067 static int walRestartLog(Wal
*pWal
){
3071 if( pWal
->readLock
==0 ){
3072 volatile WalCkptInfo
*pInfo
= walCkptInfo(pWal
);
3073 assert( pInfo
->nBackfill
==pWal
->hdr
.mxFrame
);
3074 if( pInfo
->nBackfill
>0 ){
3076 sqlite3_randomness(4, &salt1
);
3077 rc
= walLockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
3078 if( rc
==SQLITE_OK
){
3079 /* If all readers are using WAL_READ_LOCK(0) (in other words if no
3080 ** readers are currently using the WAL), then the transactions
3081 ** frames will overwrite the start of the existing log. Update the
3082 ** wal-index header to reflect this.
3084 ** In theory it would be Ok to update the cache of the header only
3085 ** at this point. But updating the actual wal-index header is also
3086 ** safe and means there is no special case for sqlite3WalUndo()
3087 ** to handle if this transaction is rolled back. */
3088 walRestartHdr(pWal
, salt1
);
3089 walUnlockExclusive(pWal
, WAL_READ_LOCK(1), WAL_NREADER
-1);
3090 }else if( rc
!=SQLITE_BUSY
){
3094 walUnlockShared(pWal
, WAL_READ_LOCK(0));
3095 pWal
->readLock
= -1;
3099 rc
= walTryBeginRead(pWal
, ¬Used
, 1, ++cnt
);
3100 }while( rc
==WAL_RETRY
);
3101 assert( (rc
&0xff)!=SQLITE_BUSY
); /* BUSY not possible when useWal==1 */
3102 testcase( (rc
&0xff)==SQLITE_IOERR
);
3103 testcase( rc
==SQLITE_PROTOCOL
);
3104 testcase( rc
==SQLITE_OK
);
3110 ** Information about the current state of the WAL file and where
3111 ** the next fsync should occur - passed from sqlite3WalFrames() into
3114 typedef struct WalWriter
{
3115 Wal
*pWal
; /* The complete WAL information */
3116 sqlite3_file
*pFd
; /* The WAL file to which we write */
3117 sqlite3_int64 iSyncPoint
; /* Fsync at this offset */
3118 int syncFlags
; /* Flags for the fsync */
3119 int szPage
; /* Size of one page */
3123 ** Write iAmt bytes of content into the WAL file beginning at iOffset.
3124 ** Do a sync when crossing the p->iSyncPoint boundary.
3126 ** In other words, if iSyncPoint is in between iOffset and iOffset+iAmt,
3127 ** first write the part before iSyncPoint, then sync, then write the
3130 static int walWriteToLog(
3131 WalWriter
*p
, /* WAL to write to */
3132 void *pContent
, /* Content to be written */
3133 int iAmt
, /* Number of bytes to write */
3134 sqlite3_int64 iOffset
/* Start writing at this offset */
3137 if( iOffset
<p
->iSyncPoint
&& iOffset
+iAmt
>=p
->iSyncPoint
){
3138 int iFirstAmt
= (int)(p
->iSyncPoint
- iOffset
);
3139 rc
= sqlite3OsWrite(p
->pFd
, pContent
, iFirstAmt
, iOffset
);
3141 iOffset
+= iFirstAmt
;
3143 pContent
= (void*)(iFirstAmt
+ (char*)pContent
);
3144 assert( WAL_SYNC_FLAGS(p
->syncFlags
)!=0 );
3145 rc
= sqlite3OsSync(p
->pFd
, WAL_SYNC_FLAGS(p
->syncFlags
));
3146 if( iAmt
==0 || rc
) return rc
;
3148 rc
= sqlite3OsWrite(p
->pFd
, pContent
, iAmt
, iOffset
);
3153 ** Write out a single frame of the WAL
3155 static int walWriteOneFrame(
3156 WalWriter
*p
, /* Where to write the frame */
3157 PgHdr
*pPage
, /* The page of the frame to be written */
3158 int nTruncate
, /* The commit flag. Usually 0. >0 for commit */
3159 sqlite3_int64 iOffset
/* Byte offset at which to write */
3161 int rc
; /* Result code from subfunctions */
3162 void *pData
; /* Data actually written */
3163 u8 aFrame
[WAL_FRAME_HDRSIZE
]; /* Buffer to assemble frame-header in */
3164 #if defined(SQLITE_HAS_CODEC)
3165 if( (pData
= sqlite3PagerCodec(pPage
))==0 ) return SQLITE_NOMEM_BKPT
;
3167 pData
= pPage
->pData
;
3169 walEncodeFrame(p
->pWal
, pPage
->pgno
, nTruncate
, pData
, aFrame
);
3170 rc
= walWriteToLog(p
, aFrame
, sizeof(aFrame
), iOffset
);
3172 /* Write the page data */
3173 rc
= walWriteToLog(p
, pData
, p
->szPage
, iOffset
+sizeof(aFrame
));
3178 ** This function is called as part of committing a transaction within which
3179 ** one or more frames have been overwritten. It updates the checksums for
3180 ** all frames written to the wal file by the current transaction starting
3181 ** with the earliest to have been overwritten.
3183 ** SQLITE_OK is returned if successful, or an SQLite error code otherwise.
3185 static int walRewriteChecksums(Wal
*pWal
, u32 iLast
){
3186 const int szPage
= pWal
->szPage
;/* Database page size */
3187 int rc
= SQLITE_OK
; /* Return code */
3188 u8
*aBuf
; /* Buffer to load data from wal file into */
3189 u8 aFrame
[WAL_FRAME_HDRSIZE
]; /* Buffer to assemble frame-headers in */
3190 u32 iRead
; /* Next frame to read from wal file */
3193 aBuf
= sqlite3_malloc(szPage
+ WAL_FRAME_HDRSIZE
);
3194 if( aBuf
==0 ) return SQLITE_NOMEM_BKPT
;
3196 /* Find the checksum values to use as input for the recalculating the
3197 ** first checksum. If the first frame is frame 1 (implying that the current
3198 ** transaction restarted the wal file), these values must be read from the
3199 ** wal-file header. Otherwise, read them from the frame header of the
3200 ** previous frame. */
3201 assert( pWal
->iReCksum
>0 );
3202 if( pWal
->iReCksum
==1 ){
3205 iCksumOff
= walFrameOffset(pWal
->iReCksum
-1, szPage
) + 16;
3207 rc
= sqlite3OsRead(pWal
->pWalFd
, aBuf
, sizeof(u32
)*2, iCksumOff
);
3208 pWal
->hdr
.aFrameCksum
[0] = sqlite3Get4byte(aBuf
);
3209 pWal
->hdr
.aFrameCksum
[1] = sqlite3Get4byte(&aBuf
[sizeof(u32
)]);
3211 iRead
= pWal
->iReCksum
;
3213 for(; rc
==SQLITE_OK
&& iRead
<=iLast
; iRead
++){
3214 i64 iOff
= walFrameOffset(iRead
, szPage
);
3215 rc
= sqlite3OsRead(pWal
->pWalFd
, aBuf
, szPage
+WAL_FRAME_HDRSIZE
, iOff
);
3216 if( rc
==SQLITE_OK
){
3218 iPgno
= sqlite3Get4byte(aBuf
);
3219 nDbSize
= sqlite3Get4byte(&aBuf
[4]);
3221 walEncodeFrame(pWal
, iPgno
, nDbSize
, &aBuf
[WAL_FRAME_HDRSIZE
], aFrame
);
3222 rc
= sqlite3OsWrite(pWal
->pWalFd
, aFrame
, sizeof(aFrame
), iOff
);
3231 ** Write a set of frames to the log. The caller must hold the write-lock
3232 ** on the log file (obtained using sqlite3WalBeginWriteTransaction()).
3234 int sqlite3WalFrames(
3235 Wal
*pWal
, /* Wal handle to write to */
3236 int szPage
, /* Database page-size in bytes */
3237 PgHdr
*pList
, /* List of dirty pages to write */
3238 Pgno nTruncate
, /* Database size after this commit */
3239 int isCommit
, /* True if this is a commit */
3240 int sync_flags
/* Flags to pass to OsSync() (or 0) */
3242 int rc
; /* Used to catch return codes */
3243 u32 iFrame
; /* Next frame address */
3244 PgHdr
*p
; /* Iterator to run through pList with. */
3245 PgHdr
*pLast
= 0; /* Last frame in list */
3246 int nExtra
= 0; /* Number of extra copies of last page */
3247 int szFrame
; /* The size of a single frame */
3248 i64 iOffset
; /* Next byte to write in WAL file */
3249 WalWriter w
; /* The writer */
3250 u32 iFirst
= 0; /* First frame that may be overwritten */
3251 WalIndexHdr
*pLive
; /* Pointer to shared header */
3254 assert( pWal
->writeLock
);
3256 /* If this frame set completes a transaction, then nTruncate>0. If
3257 ** nTruncate==0 then this frame set does not complete the transaction. */
3258 assert( (isCommit
!=0)==(nTruncate
!=0) );
3260 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
3261 { int cnt
; for(cnt
=0, p
=pList
; p
; p
=p
->pDirty
, cnt
++){}
3262 WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n",
3263 pWal
, cnt
, pWal
->hdr
.mxFrame
, isCommit
? "Commit" : "Spill"));
3267 pLive
= (WalIndexHdr
*)walIndexHdr(pWal
);
3268 if( memcmp(&pWal
->hdr
, (void *)pLive
, sizeof(WalIndexHdr
))!=0 ){
3269 iFirst
= pLive
->mxFrame
+1;
3272 /* See if it is possible to write these frames into the start of the
3273 ** log file, instead of appending to it at pWal->hdr.mxFrame.
3275 if( SQLITE_OK
!=(rc
= walRestartLog(pWal
)) ){
3279 /* If this is the first frame written into the log, write the WAL
3280 ** header to the start of the WAL file. See comments at the top of
3281 ** this source file for a description of the WAL header format.
3283 iFrame
= pWal
->hdr
.mxFrame
;
3285 u8 aWalHdr
[WAL_HDRSIZE
]; /* Buffer to assemble wal-header in */
3286 u32 aCksum
[2]; /* Checksum for wal-header */
3288 sqlite3Put4byte(&aWalHdr
[0], (WAL_MAGIC
| SQLITE_BIGENDIAN
));
3289 sqlite3Put4byte(&aWalHdr
[4], WAL_MAX_VERSION
);
3290 sqlite3Put4byte(&aWalHdr
[8], szPage
);
3291 sqlite3Put4byte(&aWalHdr
[12], pWal
->nCkpt
);
3292 if( pWal
->nCkpt
==0 ) sqlite3_randomness(8, pWal
->hdr
.aSalt
);
3293 memcpy(&aWalHdr
[16], pWal
->hdr
.aSalt
, 8);
3294 walChecksumBytes(1, aWalHdr
, WAL_HDRSIZE
-2*4, 0, aCksum
);
3295 sqlite3Put4byte(&aWalHdr
[24], aCksum
[0]);
3296 sqlite3Put4byte(&aWalHdr
[28], aCksum
[1]);
3298 pWal
->szPage
= szPage
;
3299 pWal
->hdr
.bigEndCksum
= SQLITE_BIGENDIAN
;
3300 pWal
->hdr
.aFrameCksum
[0] = aCksum
[0];
3301 pWal
->hdr
.aFrameCksum
[1] = aCksum
[1];
3302 pWal
->truncateOnCommit
= 1;
3304 rc
= sqlite3OsWrite(pWal
->pWalFd
, aWalHdr
, sizeof(aWalHdr
), 0);
3305 WALTRACE(("WAL%p: wal-header write %s\n", pWal
, rc
? "failed" : "ok"));
3306 if( rc
!=SQLITE_OK
){
3310 /* Sync the header (unless SQLITE_IOCAP_SEQUENTIAL is true or unless
3311 ** all syncing is turned off by PRAGMA synchronous=OFF). Otherwise
3312 ** an out-of-order write following a WAL restart could result in
3313 ** database corruption. See the ticket:
3315 ** https://sqlite.org/src/info/ff5be73dee
3317 if( pWal
->syncHeader
){
3318 rc
= sqlite3OsSync(pWal
->pWalFd
, CKPT_SYNC_FLAGS(sync_flags
));
3322 assert( (int)pWal
->szPage
==szPage
);
3324 /* Setup information needed to write frames into the WAL */
3326 w
.pFd
= pWal
->pWalFd
;
3328 w
.syncFlags
= sync_flags
;
3330 iOffset
= walFrameOffset(iFrame
+1, szPage
);
3331 szFrame
= szPage
+ WAL_FRAME_HDRSIZE
;
3333 /* Write all frames into the log file exactly once */
3334 for(p
=pList
; p
; p
=p
->pDirty
){
3335 int nDbSize
; /* 0 normally. Positive == commit flag */
3337 /* Check if this page has already been written into the wal file by
3338 ** the current transaction. If so, overwrite the existing frame and
3339 ** set Wal.writeLock to WAL_WRITELOCK_RECKSUM - indicating that
3340 ** checksums must be recomputed when the transaction is committed. */
3341 if( iFirst
&& (p
->pDirty
|| isCommit
==0) ){
3343 VVA_ONLY(rc
=) sqlite3WalFindFrame(pWal
, p
->pgno
, &iWrite
);
3344 assert( rc
==SQLITE_OK
|| iWrite
==0 );
3345 if( iWrite
>=iFirst
){
3346 i64 iOff
= walFrameOffset(iWrite
, szPage
) + WAL_FRAME_HDRSIZE
;
3348 if( pWal
->iReCksum
==0 || iWrite
<pWal
->iReCksum
){
3349 pWal
->iReCksum
= iWrite
;
3351 #if defined(SQLITE_HAS_CODEC)
3352 if( (pData
= sqlite3PagerCodec(p
))==0 ) return SQLITE_NOMEM
;
3356 rc
= sqlite3OsWrite(pWal
->pWalFd
, pData
, szPage
, iOff
);
3358 p
->flags
&= ~PGHDR_WAL_APPEND
;
3364 assert( iOffset
==walFrameOffset(iFrame
, szPage
) );
3365 nDbSize
= (isCommit
&& p
->pDirty
==0) ? nTruncate
: 0;
3366 rc
= walWriteOneFrame(&w
, p
, nDbSize
, iOffset
);
3370 p
->flags
|= PGHDR_WAL_APPEND
;
3373 /* Recalculate checksums within the wal file if required. */
3374 if( isCommit
&& pWal
->iReCksum
){
3375 rc
= walRewriteChecksums(pWal
, iFrame
);
3379 /* If this is the end of a transaction, then we might need to pad
3380 ** the transaction and/or sync the WAL file.
3382 ** Padding and syncing only occur if this set of frames complete a
3383 ** transaction and if PRAGMA synchronous=FULL. If synchronous==NORMAL
3384 ** or synchronous==OFF, then no padding or syncing are needed.
3386 ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not
3387 ** needed and only the sync is done. If padding is needed, then the
3388 ** final frame is repeated (with its commit mark) until the next sector
3389 ** boundary is crossed. Only the part of the WAL prior to the last
3390 ** sector boundary is synced; the part of the last frame that extends
3391 ** past the sector boundary is written after the sync.
3393 if( isCommit
&& WAL_SYNC_FLAGS(sync_flags
)!=0 ){
3395 if( pWal
->padToSectorBoundary
){
3396 int sectorSize
= sqlite3SectorSize(pWal
->pWalFd
);
3397 w
.iSyncPoint
= ((iOffset
+sectorSize
-1)/sectorSize
)*sectorSize
;
3398 bSync
= (w
.iSyncPoint
==iOffset
);
3400 while( iOffset
<w
.iSyncPoint
){
3401 rc
= walWriteOneFrame(&w
, pLast
, nTruncate
, iOffset
);
3408 assert( rc
==SQLITE_OK
);
3409 rc
= sqlite3OsSync(w
.pFd
, WAL_SYNC_FLAGS(sync_flags
));
3413 /* If this frame set completes the first transaction in the WAL and
3414 ** if PRAGMA journal_size_limit is set, then truncate the WAL to the
3415 ** journal size limit, if possible.
3417 if( isCommit
&& pWal
->truncateOnCommit
&& pWal
->mxWalSize
>=0 ){
3418 i64 sz
= pWal
->mxWalSize
;
3419 if( walFrameOffset(iFrame
+nExtra
+1, szPage
)>pWal
->mxWalSize
){
3420 sz
= walFrameOffset(iFrame
+nExtra
+1, szPage
);
3422 walLimitSize(pWal
, sz
);
3423 pWal
->truncateOnCommit
= 0;
3426 /* Append data to the wal-index. It is not necessary to lock the
3427 ** wal-index to do this as the SQLITE_SHM_WRITE lock held on the wal-index
3428 ** guarantees that there are no other writers, and no data that may
3429 ** be in use by existing readers is being overwritten.
3431 iFrame
= pWal
->hdr
.mxFrame
;
3432 for(p
=pList
; p
&& rc
==SQLITE_OK
; p
=p
->pDirty
){
3433 if( (p
->flags
& PGHDR_WAL_APPEND
)==0 ) continue;
3435 rc
= walIndexAppend(pWal
, iFrame
, p
->pgno
);
3437 while( rc
==SQLITE_OK
&& nExtra
>0 ){
3440 rc
= walIndexAppend(pWal
, iFrame
, pLast
->pgno
);
3443 if( rc
==SQLITE_OK
){
3444 /* Update the private copy of the header. */
3445 pWal
->hdr
.szPage
= (u16
)((szPage
&0xff00) | (szPage
>>16));
3446 testcase( szPage
<=32768 );
3447 testcase( szPage
>=65536 );
3448 pWal
->hdr
.mxFrame
= iFrame
;
3450 pWal
->hdr
.iChange
++;
3451 pWal
->hdr
.nPage
= nTruncate
;
3453 /* If this is a commit, update the wal-index header too. */
3455 walIndexWriteHdr(pWal
);
3456 pWal
->iCallback
= iFrame
;
3460 WALTRACE(("WAL%p: frame write %s\n", pWal
, rc
? "failed" : "ok"));
3465 ** This routine is called to implement sqlite3_wal_checkpoint() and
3466 ** related interfaces.
3468 ** Obtain a CHECKPOINT lock and then backfill as much information as
3469 ** we can from WAL into the database.
3471 ** If parameter xBusy is not NULL, it is a pointer to a busy-handler
3472 ** callback. In this case this function runs a blocking checkpoint.
3474 int sqlite3WalCheckpoint(
3475 Wal
*pWal
, /* Wal connection */
3476 sqlite3
*db
, /* Check this handle's interrupt flag */
3477 int eMode
, /* PASSIVE, FULL, RESTART, or TRUNCATE */
3478 int (*xBusy
)(void*), /* Function to call when busy */
3479 void *pBusyArg
, /* Context argument for xBusyHandler */
3480 int sync_flags
, /* Flags to sync db file with (or 0) */
3481 int nBuf
, /* Size of temporary buffer */
3482 u8
*zBuf
, /* Temporary buffer to use */
3483 int *pnLog
, /* OUT: Number of frames in WAL */
3484 int *pnCkpt
/* OUT: Number of backfilled frames in WAL */
3486 int rc
; /* Return code */
3487 int isChanged
= 0; /* True if a new wal-index header is loaded */
3488 int eMode2
= eMode
; /* Mode to pass to walCheckpoint() */
3489 int (*xBusy2
)(void*) = xBusy
; /* Busy handler for eMode2 */
3491 assert( pWal
->ckptLock
==0 );
3492 assert( pWal
->writeLock
==0 );
3494 /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked
3495 ** in the SQLITE_CHECKPOINT_PASSIVE mode. */
3496 assert( eMode
!=SQLITE_CHECKPOINT_PASSIVE
|| xBusy
==0 );
3498 if( pWal
->readOnly
) return SQLITE_READONLY
;
3499 WALTRACE(("WAL%p: checkpoint begins\n", pWal
));
3501 /* IMPLEMENTATION-OF: R-62028-47212 All calls obtain an exclusive
3502 ** "checkpoint" lock on the database file. */
3503 rc
= walLockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
3505 /* EVIDENCE-OF: R-10421-19736 If any other process is running a
3506 ** checkpoint operation at the same time, the lock cannot be obtained and
3507 ** SQLITE_BUSY is returned.
3508 ** EVIDENCE-OF: R-53820-33897 Even if there is a busy-handler configured,
3509 ** it will not be invoked in this case.
3511 testcase( rc
==SQLITE_BUSY
);
3512 testcase( xBusy
!=0 );
3517 /* IMPLEMENTATION-OF: R-59782-36818 The SQLITE_CHECKPOINT_FULL, RESTART and
3518 ** TRUNCATE modes also obtain the exclusive "writer" lock on the database
3521 ** EVIDENCE-OF: R-60642-04082 If the writer lock cannot be obtained
3522 ** immediately, and a busy-handler is configured, it is invoked and the
3523 ** writer lock retried until either the busy-handler returns 0 or the
3524 ** lock is successfully obtained.
3526 if( eMode
!=SQLITE_CHECKPOINT_PASSIVE
){
3527 rc
= walBusyLock(pWal
, xBusy
, pBusyArg
, WAL_WRITE_LOCK
, 1);
3528 if( rc
==SQLITE_OK
){
3529 pWal
->writeLock
= 1;
3530 }else if( rc
==SQLITE_BUSY
){
3531 eMode2
= SQLITE_CHECKPOINT_PASSIVE
;
3537 /* Read the wal-index header. */
3538 if( rc
==SQLITE_OK
){
3539 rc
= walIndexReadHdr(pWal
, &isChanged
);
3540 if( isChanged
&& pWal
->pDbFd
->pMethods
->iVersion
>=3 ){
3541 sqlite3OsUnfetch(pWal
->pDbFd
, 0, 0);
3545 /* Copy data from the log to the database file. */
3546 if( rc
==SQLITE_OK
){
3548 if( pWal
->hdr
.mxFrame
&& walPagesize(pWal
)!=nBuf
){
3549 rc
= SQLITE_CORRUPT_BKPT
;
3551 rc
= walCheckpoint(pWal
, db
, eMode2
, xBusy2
, pBusyArg
, sync_flags
, zBuf
);
3554 /* If no error occurred, set the output variables. */
3555 if( rc
==SQLITE_OK
|| rc
==SQLITE_BUSY
){
3556 if( pnLog
) *pnLog
= (int)pWal
->hdr
.mxFrame
;
3557 if( pnCkpt
) *pnCkpt
= (int)(walCkptInfo(pWal
)->nBackfill
);
3562 /* If a new wal-index header was loaded before the checkpoint was
3563 ** performed, then the pager-cache associated with pWal is now
3564 ** out of date. So zero the cached wal-index header to ensure that
3565 ** next time the pager opens a snapshot on this database it knows that
3566 ** the cache needs to be reset.
3568 memset(&pWal
->hdr
, 0, sizeof(WalIndexHdr
));
3571 /* Release the locks. */
3572 sqlite3WalEndWriteTransaction(pWal
);
3573 walUnlockExclusive(pWal
, WAL_CKPT_LOCK
, 1);
3575 WALTRACE(("WAL%p: checkpoint %s\n", pWal
, rc
? "failed" : "ok"));
3576 return (rc
==SQLITE_OK
&& eMode
!=eMode2
? SQLITE_BUSY
: rc
);
3579 /* Return the value to pass to a sqlite3_wal_hook callback, the
3580 ** number of frames in the WAL at the point of the last commit since
3581 ** sqlite3WalCallback() was called. If no commits have occurred since
3582 ** the last call, then return 0.
3584 int sqlite3WalCallback(Wal
*pWal
){
3587 ret
= pWal
->iCallback
;
3588 pWal
->iCallback
= 0;
3594 ** This function is called to change the WAL subsystem into or out
3595 ** of locking_mode=EXCLUSIVE.
3597 ** If op is zero, then attempt to change from locking_mode=EXCLUSIVE
3598 ** into locking_mode=NORMAL. This means that we must acquire a lock
3599 ** on the pWal->readLock byte. If the WAL is already in locking_mode=NORMAL
3600 ** or if the acquisition of the lock fails, then return 0. If the
3601 ** transition out of exclusive-mode is successful, return 1. This
3602 ** operation must occur while the pager is still holding the exclusive
3603 ** lock on the main database file.
3605 ** If op is one, then change from locking_mode=NORMAL into
3606 ** locking_mode=EXCLUSIVE. This means that the pWal->readLock must
3607 ** be released. Return 1 if the transition is made and 0 if the
3608 ** WAL is already in exclusive-locking mode - meaning that this
3609 ** routine is a no-op. The pager must already hold the exclusive lock
3610 ** on the main database file before invoking this operation.
3612 ** If op is negative, then do a dry-run of the op==1 case but do
3613 ** not actually change anything. The pager uses this to see if it
3614 ** should acquire the database exclusive lock prior to invoking
3617 int sqlite3WalExclusiveMode(Wal
*pWal
, int op
){
3619 assert( pWal
->writeLock
==0 );
3620 assert( pWal
->exclusiveMode
!=WAL_HEAPMEMORY_MODE
|| op
==-1 );
3622 /* pWal->readLock is usually set, but might be -1 if there was a
3623 ** prior error while attempting to acquire are read-lock. This cannot
3624 ** happen if the connection is actually in exclusive mode (as no xShmLock
3625 ** locks are taken in this case). Nor should the pager attempt to
3626 ** upgrade to exclusive-mode following such an error.
3628 assert( pWal
->readLock
>=0 || pWal
->lockError
);
3629 assert( pWal
->readLock
>=0 || (op
<=0 && pWal
->exclusiveMode
==0) );
3632 if( pWal
->exclusiveMode
){
3633 pWal
->exclusiveMode
= 0;
3634 if( walLockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
))!=SQLITE_OK
){
3635 pWal
->exclusiveMode
= 1;
3637 rc
= pWal
->exclusiveMode
==0;
3639 /* Already in locking_mode=NORMAL */
3643 assert( pWal
->exclusiveMode
==0 );
3644 assert( pWal
->readLock
>=0 );
3645 walUnlockShared(pWal
, WAL_READ_LOCK(pWal
->readLock
));
3646 pWal
->exclusiveMode
= 1;
3649 rc
= pWal
->exclusiveMode
==0;
3655 ** Return true if the argument is non-NULL and the WAL module is using
3656 ** heap-memory for the wal-index. Otherwise, if the argument is NULL or the
3657 ** WAL module is using shared-memory, return false.
3659 int sqlite3WalHeapMemory(Wal
*pWal
){
3660 return (pWal
&& pWal
->exclusiveMode
==WAL_HEAPMEMORY_MODE
);
3663 #ifdef SQLITE_ENABLE_SNAPSHOT
3664 /* Create a snapshot object. The content of a snapshot is opaque to
3665 ** every other subsystem, so the WAL module can put whatever it needs
3668 int sqlite3WalSnapshotGet(Wal
*pWal
, sqlite3_snapshot
**ppSnapshot
){
3671 static const u32 aZero
[4] = { 0, 0, 0, 0 };
3673 assert( pWal
->readLock
>=0 && pWal
->writeLock
==0 );
3675 if( memcmp(&pWal
->hdr
.aFrameCksum
[0],aZero
,16)==0 ){
3677 return SQLITE_ERROR
;
3679 pRet
= (WalIndexHdr
*)sqlite3_malloc(sizeof(WalIndexHdr
));
3681 rc
= SQLITE_NOMEM_BKPT
;
3683 memcpy(pRet
, &pWal
->hdr
, sizeof(WalIndexHdr
));
3684 *ppSnapshot
= (sqlite3_snapshot
*)pRet
;
3690 /* Try to open on pSnapshot when the next read-transaction starts
3692 void sqlite3WalSnapshotOpen(Wal
*pWal
, sqlite3_snapshot
*pSnapshot
){
3693 pWal
->pSnapshot
= (WalIndexHdr
*)pSnapshot
;
3697 ** Return a +ve value if snapshot p1 is newer than p2. A -ve value if
3698 ** p1 is older than p2 and zero if p1 and p2 are the same snapshot.
3700 int sqlite3_snapshot_cmp(sqlite3_snapshot
*p1
, sqlite3_snapshot
*p2
){
3701 WalIndexHdr
*pHdr1
= (WalIndexHdr
*)p1
;
3702 WalIndexHdr
*pHdr2
= (WalIndexHdr
*)p2
;
3704 /* aSalt[0] is a copy of the value stored in the wal file header. It
3705 ** is incremented each time the wal file is restarted. */
3706 if( pHdr1
->aSalt
[0]<pHdr2
->aSalt
[0] ) return -1;
3707 if( pHdr1
->aSalt
[0]>pHdr2
->aSalt
[0] ) return +1;
3708 if( pHdr1
->mxFrame
<pHdr2
->mxFrame
) return -1;
3709 if( pHdr1
->mxFrame
>pHdr2
->mxFrame
) return +1;
3712 #endif /* SQLITE_ENABLE_SNAPSHOT */
3714 #ifdef SQLITE_ENABLE_ZIPVFS
3716 ** If the argument is not NULL, it points to a Wal object that holds a
3717 ** read-lock. This function returns the database page-size if it is known,
3718 ** or zero if it is not (or if pWal is NULL).
3720 int sqlite3WalFramesize(Wal
*pWal
){
3721 assert( pWal
==0 || pWal
->readLock
>=0 );
3722 return (pWal
? pWal
->szPage
: 0);
3726 /* Return the sqlite3_file object for the WAL file
3728 sqlite3_file
*sqlite3WalFile(Wal
*pWal
){
3729 return pWal
->pWalFd
;
3732 #endif /* #ifndef SQLITE_OMIT_WAL */