000001 /*
000002 ** 2010 February 1
000003 **
000004 ** The author disclaims copyright to this source code. In place of
000005 ** a legal notice, here is a blessing:
000006 **
000007 ** May you do good and not evil.
000008 ** May you find forgiveness for yourself and forgive others.
000009 ** May you share freely, never taking more than you give.
000010 **
000011 *************************************************************************
000012 **
000013 ** This file contains the implementation of a write-ahead log (WAL) used in
000014 ** "journal_mode=WAL" mode.
000015 **
000016 ** WRITE-AHEAD LOG (WAL) FILE FORMAT
000017 **
000018 ** A WAL file consists of a header followed by zero or more "frames".
000019 ** Each frame records the revised content of a single page from the
000020 ** database file. All changes to the database are recorded by writing
000021 ** frames into the WAL. Transactions commit when a frame is written that
000022 ** contains a commit marker. A single WAL can and usually does record
000023 ** multiple transactions. Periodically, the content of the WAL is
000024 ** transferred back into the database file in an operation called a
000025 ** "checkpoint".
000026 **
000027 ** A single WAL file can be used multiple times. In other words, the
000028 ** WAL can fill up with frames and then be checkpointed and then new
000029 ** frames can overwrite the old ones. A WAL always grows from beginning
000030 ** toward the end. Checksums and counters attached to each frame are
000031 ** used to determine which frames within the WAL are valid and which
000032 ** are leftovers from prior checkpoints.
000033 **
000034 ** The WAL header is 32 bytes in size and consists of the following eight
000035 ** big-endian 32-bit unsigned integer values:
000036 **
000037 ** 0: Magic number. 0x377f0682 or 0x377f0683
000038 ** 4: File format version. Currently 3007000
000039 ** 8: Database page size. Example: 1024
000040 ** 12: Checkpoint sequence number
000041 ** 16: Salt-1, random integer incremented with each checkpoint
000042 ** 20: Salt-2, a different random integer changing with each ckpt
000043 ** 24: Checksum-1 (first part of checksum for first 24 bytes of header).
000044 ** 28: Checksum-2 (second part of checksum for first 24 bytes of header).
000045 **
000046 ** Immediately following the wal-header are zero or more frames. Each
000047 ** frame consists of a 24-byte frame-header followed by a <page-size> bytes
000048 ** of page data. The frame-header is six big-endian 32-bit unsigned
000049 ** integer values, as follows:
000050 **
000051 ** 0: Page number.
000052 ** 4: For commit records, the size of the database image in pages
000053 ** after the commit. For all other records, zero.
000054 ** 8: Salt-1 (copied from the header)
000055 ** 12: Salt-2 (copied from the header)
000056 ** 16: Checksum-1.
000057 ** 20: Checksum-2.
000058 **
000059 ** A frame is considered valid if and only if the following conditions are
000060 ** true:
000061 **
000062 ** (1) The salt-1 and salt-2 values in the frame-header match
000063 ** salt values in the wal-header
000064 **
000065 ** (2) The checksum values in the final 8 bytes of the frame-header
000066 ** exactly match the checksum computed consecutively on the
000067 ** WAL header and the first 8 bytes and the content of all frames
000068 ** up to and including the current frame.
000069 **
000070 ** The checksum is computed using 32-bit big-endian integers if the
000071 ** magic number in the first 4 bytes of the WAL is 0x377f0683 and it
000072 ** is computed using little-endian if the magic number is 0x377f0682.
000073 ** The checksum values are always stored in the frame header in a
000074 ** big-endian format regardless of which byte order is used to compute
000075 ** the checksum. The checksum is computed by interpreting the input as
000076 ** an even number of unsigned 32-bit integers: x[0] through x[N]. The
000077 ** algorithm used for the checksum is as follows:
000078 **
000079 ** for i from 0 to n-1 step 2:
000080 ** s0 += x[i] + s1;
000081 ** s1 += x[i+1] + s0;
000082 ** endfor
000083 **
000084 ** Note that s0 and s1 are both weighted checksums using fibonacci weights
000085 ** in reverse order (the largest fibonacci weight occurs on the first element
000086 ** of the sequence being summed.) The s1 value spans all 32-bit
000087 ** terms of the sequence whereas s0 omits the final term.
000088 **
000089 ** On a checkpoint, the WAL is first VFS.xSync-ed, then valid content of the
000090 ** WAL is transferred into the database, then the database is VFS.xSync-ed.
000091 ** The VFS.xSync operations serve as write barriers - all writes launched
000092 ** before the xSync must complete before any write that launches after the
000093 ** xSync begins.
000094 **
000095 ** After each checkpoint, the salt-1 value is incremented and the salt-2
000096 ** value is randomized. This prevents old and new frames in the WAL from
000097 ** being considered valid at the same time and being checkpointing together
000098 ** following a crash.
000099 **
000100 ** READER ALGORITHM
000101 **
000102 ** To read a page from the database (call it page number P), a reader
000103 ** first checks the WAL to see if it contains page P. If so, then the
000104 ** last valid instance of page P that is a followed by a commit frame
000105 ** or is a commit frame itself becomes the value read. If the WAL
000106 ** contains no copies of page P that are valid and which are a commit
000107 ** frame or are followed by a commit frame, then page P is read from
000108 ** the database file.
000109 **
000110 ** To start a read transaction, the reader records the index of the last
000111 ** valid frame in the WAL. The reader uses this recorded "mxFrame" value
000112 ** for all subsequent read operations. New transactions can be appended
000113 ** to the WAL, but as long as the reader uses its original mxFrame value
000114 ** and ignores the newly appended content, it will see a consistent snapshot
000115 ** of the database from a single point in time. This technique allows
000116 ** multiple concurrent readers to view different versions of the database
000117 ** content simultaneously.
000118 **
000119 ** The reader algorithm in the previous paragraphs works correctly, but
000120 ** because frames for page P can appear anywhere within the WAL, the
000121 ** reader has to scan the entire WAL looking for page P frames. If the
000122 ** WAL is large (multiple megabytes is typical) that scan can be slow,
000123 ** and read performance suffers. To overcome this problem, a separate
000124 ** data structure called the wal-index is maintained to expedite the
000125 ** search for frames of a particular page.
000126 **
000127 ** WAL-INDEX FORMAT
000128 **
000129 ** Conceptually, the wal-index is shared memory, though VFS implementations
000130 ** might choose to implement the wal-index using a mmapped file. Because
000131 ** the wal-index is shared memory, SQLite does not support journal_mode=WAL
000132 ** on a network filesystem. All users of the database must be able to
000133 ** share memory.
000134 **
000135 ** In the default unix and windows implementation, the wal-index is a mmapped
000136 ** file whose name is the database name with a "-shm" suffix added. For that
000137 ** reason, the wal-index is sometimes called the "shm" file.
000138 **
000139 ** The wal-index is transient. After a crash, the wal-index can (and should
000140 ** be) reconstructed from the original WAL file. In fact, the VFS is required
000141 ** to either truncate or zero the header of the wal-index when the last
000142 ** connection to it closes. Because the wal-index is transient, it can
000143 ** use an architecture-specific format; it does not have to be cross-platform.
000144 ** Hence, unlike the database and WAL file formats which store all values
000145 ** as big endian, the wal-index can store multi-byte values in the native
000146 ** byte order of the host computer.
000147 **
000148 ** The purpose of the wal-index is to answer this question quickly: Given
000149 ** a page number P and a maximum frame index M, return the index of the
000150 ** last frame in the wal before frame M for page P in the WAL, or return
000151 ** NULL if there are no frames for page P in the WAL prior to M.
000152 **
000153 ** The wal-index consists of a header region, followed by an one or
000154 ** more index blocks.
000155 **
000156 ** The wal-index header contains the total number of frames within the WAL
000157 ** in the mxFrame field.
000158 **
000159 ** Each index block except for the first contains information on
000160 ** HASHTABLE_NPAGE frames. The first index block contains information on
000161 ** HASHTABLE_NPAGE_ONE frames. The values of HASHTABLE_NPAGE_ONE and
000162 ** HASHTABLE_NPAGE are selected so that together the wal-index header and
000163 ** first index block are the same size as all other index blocks in the
000164 ** wal-index. The values are:
000165 **
000166 ** HASHTABLE_NPAGE 4096
000167 ** HASHTABLE_NPAGE_ONE 4062
000168 **
000169 ** Each index block contains two sections, a page-mapping that contains the
000170 ** database page number associated with each wal frame, and a hash-table
000171 ** that allows readers to query an index block for a specific page number.
000172 ** The page-mapping is an array of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE
000173 ** for the first index block) 32-bit page numbers. The first entry in the
000174 ** first index-block contains the database page number corresponding to the
000175 ** first frame in the WAL file. The first entry in the second index block
000176 ** in the WAL file corresponds to the (HASHTABLE_NPAGE_ONE+1)th frame in
000177 ** the log, and so on.
000178 **
000179 ** The last index block in a wal-index usually contains less than the full
000180 ** complement of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE) page-numbers,
000181 ** depending on the contents of the WAL file. This does not change the
000182 ** allocated size of the page-mapping array - the page-mapping array merely
000183 ** contains unused entries.
000184 **
000185 ** Even without using the hash table, the last frame for page P
000186 ** can be found by scanning the page-mapping sections of each index block
000187 ** starting with the last index block and moving toward the first, and
000188 ** within each index block, starting at the end and moving toward the
000189 ** beginning. The first entry that equals P corresponds to the frame
000190 ** holding the content for that page.
000191 **
000192 ** The hash table consists of HASHTABLE_NSLOT 16-bit unsigned integers.
000193 ** HASHTABLE_NSLOT = 2*HASHTABLE_NPAGE, and there is one entry in the
000194 ** hash table for each page number in the mapping section, so the hash
000195 ** table is never more than half full. The expected number of collisions
000196 ** prior to finding a match is 1. Each entry of the hash table is an
000197 ** 1-based index of an entry in the mapping section of the same
000198 ** index block. Let K be the 1-based index of the largest entry in
000199 ** the mapping section. (For index blocks other than the last, K will
000200 ** always be exactly HASHTABLE_NPAGE (4096) and for the last index block
000201 ** K will be (mxFrame%HASHTABLE_NPAGE).) Unused slots of the hash table
000202 ** contain a value of 0.
000203 **
000204 ** To look for page P in the hash table, first compute a hash iKey on
000205 ** P as follows:
000206 **
000207 ** iKey = (P * 383) % HASHTABLE_NSLOT
000208 **
000209 ** Then start scanning entries of the hash table, starting with iKey
000210 ** (wrapping around to the beginning when the end of the hash table is
000211 ** reached) until an unused hash slot is found. Let the first unused slot
000212 ** be at index iUnused. (iUnused might be less than iKey if there was
000213 ** wrap-around.) Because the hash table is never more than half full,
000214 ** the search is guaranteed to eventually hit an unused entry. Let
000215 ** iMax be the value between iKey and iUnused, closest to iUnused,
000216 ** where aHash[iMax]==P. If there is no iMax entry (if there exists
000217 ** no hash slot such that aHash[i]==p) then page P is not in the
000218 ** current index block. Otherwise the iMax-th mapping entry of the
000219 ** current index block corresponds to the last entry that references
000220 ** page P.
000221 **
000222 ** A hash search begins with the last index block and moves toward the
000223 ** first index block, looking for entries corresponding to page P. On
000224 ** average, only two or three slots in each index block need to be
000225 ** examined in order to either find the last entry for page P, or to
000226 ** establish that no such entry exists in the block. Each index block
000227 ** holds over 4000 entries. So two or three index blocks are sufficient
000228 ** to cover a typical 10 megabyte WAL file, assuming 1K pages. 8 or 10
000229 ** comparisons (on average) suffice to either locate a frame in the
000230 ** WAL or to establish that the frame does not exist in the WAL. This
000231 ** is much faster than scanning the entire 10MB WAL.
000232 **
000233 ** Note that entries are added in order of increasing K. Hence, one
000234 ** reader might be using some value K0 and a second reader that started
000235 ** at a later time (after additional transactions were added to the WAL
000236 ** and to the wal-index) might be using a different value K1, where K1>K0.
000237 ** Both readers can use the same hash table and mapping section to get
000238 ** the correct result. There may be entries in the hash table with
000239 ** K>K0 but to the first reader, those entries will appear to be unused
000240 ** slots in the hash table and so the first reader will get an answer as
000241 ** if no values greater than K0 had ever been inserted into the hash table
000242 ** in the first place - which is what reader one wants. Meanwhile, the
000243 ** second reader using K1 will see additional values that were inserted
000244 ** later, which is exactly what reader two wants.
000245 **
000246 ** When a rollback occurs, the value of K is decreased. Hash table entries
000247 ** that correspond to frames greater than the new K value are removed
000248 ** from the hash table at this point.
000249 */
000250 #ifndef SQLITE_OMIT_WAL
000251
000252 #include "wal.h"
000253
000254 /*
000255 ** Trace output macros
000256 */
000257 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
000258 int sqlite3WalTrace = 0;
000259 # define WALTRACE(X) if(sqlite3WalTrace) sqlite3DebugPrintf X
000260 #else
000261 # define WALTRACE(X)
000262 #endif
000263
000264 /*
000265 ** The maximum (and only) versions of the wal and wal-index formats
000266 ** that may be interpreted by this version of SQLite.
000267 **
000268 ** If a client begins recovering a WAL file and finds that (a) the checksum
000269 ** values in the wal-header are correct and (b) the version field is not
000270 ** WAL_MAX_VERSION, recovery fails and SQLite returns SQLITE_CANTOPEN.
000271 **
000272 ** Similarly, if a client successfully reads a wal-index header (i.e. the
000273 ** checksum test is successful) and finds that the version field is not
000274 ** WALINDEX_MAX_VERSION, then no read-transaction is opened and SQLite
000275 ** returns SQLITE_CANTOPEN.
000276 */
000277 #define WAL_MAX_VERSION 3007000
000278 #define WALINDEX_MAX_VERSION 3007000
000279
000280 /*
000281 ** Index numbers for various locking bytes. WAL_NREADER is the number
000282 ** of available reader locks and should be at least 3. The default
000283 ** is SQLITE_SHM_NLOCK==8 and WAL_NREADER==5.
000284 **
000285 ** Technically, the various VFSes are free to implement these locks however
000286 ** they see fit. However, compatibility is encouraged so that VFSes can
000287 ** interoperate. The standard implementation used on both unix and windows
000288 ** is for the index number to indicate a byte offset into the
000289 ** WalCkptInfo.aLock[] array in the wal-index header. In other words, all
000290 ** locks are on the shm file. The WALINDEX_LOCK_OFFSET constant (which
000291 ** should be 120) is the location in the shm file for the first locking
000292 ** byte.
000293 */
000294 #define WAL_WRITE_LOCK 0
000295 #define WAL_ALL_BUT_WRITE 1
000296 #define WAL_CKPT_LOCK 1
000297 #define WAL_RECOVER_LOCK 2
000298 #define WAL_READ_LOCK(I) (3+(I))
000299 #define WAL_NREADER (SQLITE_SHM_NLOCK-3)
000300
000301
000302 /* Object declarations */
000303 typedef struct WalIndexHdr WalIndexHdr;
000304 typedef struct WalIterator WalIterator;
000305 typedef struct WalCkptInfo WalCkptInfo;
000306
000307
000308 /*
000309 ** The following object holds a copy of the wal-index header content.
000310 **
000311 ** The actual header in the wal-index consists of two copies of this
000312 ** object followed by one instance of the WalCkptInfo object.
000313 ** For all versions of SQLite through 3.10.0 and probably beyond,
000314 ** the locking bytes (WalCkptInfo.aLock) start at offset 120 and
000315 ** the total header size is 136 bytes.
000316 **
000317 ** The szPage value can be any power of 2 between 512 and 32768, inclusive.
000318 ** Or it can be 1 to represent a 65536-byte page. The latter case was
000319 ** added in 3.7.1 when support for 64K pages was added.
000320 */
000321 struct WalIndexHdr {
000322 u32 iVersion; /* Wal-index version */
000323 u32 unused; /* Unused (padding) field */
000324 u32 iChange; /* Counter incremented each transaction */
000325 u8 isInit; /* 1 when initialized */
000326 u8 bigEndCksum; /* True if checksums in WAL are big-endian */
000327 u16 szPage; /* Database page size in bytes. 1==64K */
000328 u32 mxFrame; /* Index of last valid frame in the WAL */
000329 u32 nPage; /* Size of database in pages */
000330 u32 aFrameCksum[2]; /* Checksum of last frame in log */
000331 u32 aSalt[2]; /* Two salt values copied from WAL header */
000332 u32 aCksum[2]; /* Checksum over all prior fields */
000333 };
000334
000335 /*
000336 ** A copy of the following object occurs in the wal-index immediately
000337 ** following the second copy of the WalIndexHdr. This object stores
000338 ** information used by checkpoint.
000339 **
000340 ** nBackfill is the number of frames in the WAL that have been written
000341 ** back into the database. (We call the act of moving content from WAL to
000342 ** database "backfilling".) The nBackfill number is never greater than
000343 ** WalIndexHdr.mxFrame. nBackfill can only be increased by threads
000344 ** holding the WAL_CKPT_LOCK lock (which includes a recovery thread).
000345 ** However, a WAL_WRITE_LOCK thread can move the value of nBackfill from
000346 ** mxFrame back to zero when the WAL is reset.
000347 **
000348 ** nBackfillAttempted is the largest value of nBackfill that a checkpoint
000349 ** has attempted to achieve. Normally nBackfill==nBackfillAtempted, however
000350 ** the nBackfillAttempted is set before any backfilling is done and the
000351 ** nBackfill is only set after all backfilling completes. So if a checkpoint
000352 ** crashes, nBackfillAttempted might be larger than nBackfill. The
000353 ** WalIndexHdr.mxFrame must never be less than nBackfillAttempted.
000354 **
000355 ** The aLock[] field is a set of bytes used for locking. These bytes should
000356 ** never be read or written.
000357 **
000358 ** There is one entry in aReadMark[] for each reader lock. If a reader
000359 ** holds read-lock K, then the value in aReadMark[K] is no greater than
000360 ** the mxFrame for that reader. The value READMARK_NOT_USED (0xffffffff)
000361 ** for any aReadMark[] means that entry is unused. aReadMark[0] is
000362 ** a special case; its value is never used and it exists as a place-holder
000363 ** to avoid having to offset aReadMark[] indexes by one. Readers holding
000364 ** WAL_READ_LOCK(0) always ignore the entire WAL and read all content
000365 ** directly from the database.
000366 **
000367 ** The value of aReadMark[K] may only be changed by a thread that
000368 ** is holding an exclusive lock on WAL_READ_LOCK(K). Thus, the value of
000369 ** aReadMark[K] cannot changed while there is a reader is using that mark
000370 ** since the reader will be holding a shared lock on WAL_READ_LOCK(K).
000371 **
000372 ** The checkpointer may only transfer frames from WAL to database where
000373 ** the frame numbers are less than or equal to every aReadMark[] that is
000374 ** in use (that is, every aReadMark[j] for which there is a corresponding
000375 ** WAL_READ_LOCK(j)). New readers (usually) pick the aReadMark[] with the
000376 ** largest value and will increase an unused aReadMark[] to mxFrame if there
000377 ** is not already an aReadMark[] equal to mxFrame. The exception to the
000378 ** previous sentence is when nBackfill equals mxFrame (meaning that everything
000379 ** in the WAL has been backfilled into the database) then new readers
000380 ** will choose aReadMark[0] which has value 0 and hence such reader will
000381 ** get all their all content directly from the database file and ignore
000382 ** the WAL.
000383 **
000384 ** Writers normally append new frames to the end of the WAL. However,
000385 ** if nBackfill equals mxFrame (meaning that all WAL content has been
000386 ** written back into the database) and if no readers are using the WAL
000387 ** (in other words, if there are no WAL_READ_LOCK(i) where i>0) then
000388 ** the writer will first "reset" the WAL back to the beginning and start
000389 ** writing new content beginning at frame 1.
000390 **
000391 ** We assume that 32-bit loads are atomic and so no locks are needed in
000392 ** order to read from any aReadMark[] entries.
000393 */
000394 struct WalCkptInfo {
000395 u32 nBackfill; /* Number of WAL frames backfilled into DB */
000396 u32 aReadMark[WAL_NREADER]; /* Reader marks */
000397 u8 aLock[SQLITE_SHM_NLOCK]; /* Reserved space for locks */
000398 u32 nBackfillAttempted; /* WAL frames perhaps written, or maybe not */
000399 u32 notUsed0; /* Available for future enhancements */
000400 };
000401 #define READMARK_NOT_USED 0xffffffff
000402
000403 /*
000404 ** This is a schematic view of the complete 136-byte header of the
000405 ** wal-index file (also known as the -shm file):
000406 **
000407 ** +-----------------------------+
000408 ** 0: | iVersion | \
000409 ** +-----------------------------+ |
000410 ** 4: | (unused padding) | |
000411 ** +-----------------------------+ |
000412 ** 8: | iChange | |
000413 ** +-------+-------+-------------+ |
000414 ** 12: | bInit | bBig | szPage | |
000415 ** +-------+-------+-------------+ |
000416 ** 16: | mxFrame | | First copy of the
000417 ** +-----------------------------+ | WalIndexHdr object
000418 ** 20: | nPage | |
000419 ** +-----------------------------+ |
000420 ** 24: | aFrameCksum | |
000421 ** | | |
000422 ** +-----------------------------+ |
000423 ** 32: | aSalt | |
000424 ** | | |
000425 ** +-----------------------------+ |
000426 ** 40: | aCksum | |
000427 ** | | /
000428 ** +-----------------------------+
000429 ** 48: | iVersion | \
000430 ** +-----------------------------+ |
000431 ** 52: | (unused padding) | |
000432 ** +-----------------------------+ |
000433 ** 56: | iChange | |
000434 ** +-------+-------+-------------+ |
000435 ** 60: | bInit | bBig | szPage | |
000436 ** +-------+-------+-------------+ | Second copy of the
000437 ** 64: | mxFrame | | WalIndexHdr
000438 ** +-----------------------------+ |
000439 ** 68: | nPage | |
000440 ** +-----------------------------+ |
000441 ** 72: | aFrameCksum | |
000442 ** | | |
000443 ** +-----------------------------+ |
000444 ** 80: | aSalt | |
000445 ** | | |
000446 ** +-----------------------------+ |
000447 ** 88: | aCksum | |
000448 ** | | /
000449 ** +-----------------------------+
000450 ** 96: | nBackfill |
000451 ** +-----------------------------+
000452 ** 100: | 5 read marks |
000453 ** | |
000454 ** | |
000455 ** | |
000456 ** | |
000457 ** +-------+-------+------+------+
000458 ** 120: | Write | Ckpt | Rcvr | Rd0 | \
000459 ** +-------+-------+------+------+ ) 8 lock bytes
000460 ** | Read1 | Read2 | Rd3 | Rd4 | /
000461 ** +-------+-------+------+------+
000462 ** 128: | nBackfillAttempted |
000463 ** +-----------------------------+
000464 ** 132: | (unused padding) |
000465 ** +-----------------------------+
000466 */
000467
000468 /* A block of WALINDEX_LOCK_RESERVED bytes beginning at
000469 ** WALINDEX_LOCK_OFFSET is reserved for locks. Since some systems
000470 ** only support mandatory file-locks, we do not read or write data
000471 ** from the region of the file on which locks are applied.
000472 */
000473 #define WALINDEX_LOCK_OFFSET (sizeof(WalIndexHdr)*2+offsetof(WalCkptInfo,aLock))
000474 #define WALINDEX_HDR_SIZE (sizeof(WalIndexHdr)*2+sizeof(WalCkptInfo))
000475
000476 /* Size of header before each frame in wal */
000477 #define WAL_FRAME_HDRSIZE 24
000478
000479 /* Size of write ahead log header, including checksum. */
000480 #define WAL_HDRSIZE 32
000481
000482 /* WAL magic value. Either this value, or the same value with the least
000483 ** significant bit also set (WAL_MAGIC | 0x00000001) is stored in 32-bit
000484 ** big-endian format in the first 4 bytes of a WAL file.
000485 **
000486 ** If the LSB is set, then the checksums for each frame within the WAL
000487 ** file are calculated by treating all data as an array of 32-bit
000488 ** big-endian words. Otherwise, they are calculated by interpreting
000489 ** all data as 32-bit little-endian words.
000490 */
000491 #define WAL_MAGIC 0x377f0682
000492
000493 /*
000494 ** Return the offset of frame iFrame in the write-ahead log file,
000495 ** assuming a database page size of szPage bytes. The offset returned
000496 ** is to the start of the write-ahead log frame-header.
000497 */
000498 #define walFrameOffset(iFrame, szPage) ( \
000499 WAL_HDRSIZE + ((iFrame)-1)*(i64)((szPage)+WAL_FRAME_HDRSIZE) \
000500 )
000501
000502 /*
000503 ** An open write-ahead log file is represented by an instance of the
000504 ** following object.
000505 */
000506 struct Wal {
000507 sqlite3_vfs *pVfs; /* The VFS used to create pDbFd */
000508 sqlite3_file *pDbFd; /* File handle for the database file */
000509 sqlite3_file *pWalFd; /* File handle for WAL file */
000510 u32 iCallback; /* Value to pass to log callback (or 0) */
000511 i64 mxWalSize; /* Truncate WAL to this size upon reset */
000512 int nWiData; /* Size of array apWiData */
000513 int szFirstBlock; /* Size of first block written to WAL file */
000514 volatile u32 **apWiData; /* Pointer to wal-index content in memory */
000515 u32 szPage; /* Database page size */
000516 i16 readLock; /* Which read lock is being held. -1 for none */
000517 u8 syncFlags; /* Flags to use to sync header writes */
000518 u8 exclusiveMode; /* Non-zero if connection is in exclusive mode */
000519 u8 writeLock; /* True if in a write transaction */
000520 u8 ckptLock; /* True if holding a checkpoint lock */
000521 u8 readOnly; /* WAL_RDWR, WAL_RDONLY, or WAL_SHM_RDONLY */
000522 u8 truncateOnCommit; /* True to truncate WAL file on commit */
000523 u8 syncHeader; /* Fsync the WAL header if true */
000524 u8 padToSectorBoundary; /* Pad transactions out to the next sector */
000525 u8 bShmUnreliable; /* SHM content is read-only and unreliable */
000526 WalIndexHdr hdr; /* Wal-index header for current transaction */
000527 u32 minFrame; /* Ignore wal frames before this one */
000528 u32 iReCksum; /* On commit, recalculate checksums from here */
000529 const char *zWalName; /* Name of WAL file */
000530 u32 nCkpt; /* Checkpoint sequence counter in the wal-header */
000531 #ifdef SQLITE_USE_SEH
000532 u32 lockMask; /* Mask of locks held */
000533 void *pFree; /* Pointer to sqlite3_free() if exception thrown */
000534 u32 *pWiValue; /* Value to write into apWiData[iWiPg] */
000535 int iWiPg; /* Write pWiValue into apWiData[iWiPg] */
000536 int iSysErrno; /* System error code following exception */
000537 #endif
000538 #ifdef SQLITE_DEBUG
000539 int nSehTry; /* Number of nested SEH_TRY{} blocks */
000540 u8 lockError; /* True if a locking error has occurred */
000541 #endif
000542 #ifdef SQLITE_ENABLE_SNAPSHOT
000543 WalIndexHdr *pSnapshot; /* Start transaction here if not NULL */
000544 #endif
000545 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
000546 sqlite3 *db;
000547 #endif
000548 };
000549
000550 /*
000551 ** Candidate values for Wal.exclusiveMode.
000552 */
000553 #define WAL_NORMAL_MODE 0
000554 #define WAL_EXCLUSIVE_MODE 1
000555 #define WAL_HEAPMEMORY_MODE 2
000556
000557 /*
000558 ** Possible values for WAL.readOnly
000559 */
000560 #define WAL_RDWR 0 /* Normal read/write connection */
000561 #define WAL_RDONLY 1 /* The WAL file is readonly */
000562 #define WAL_SHM_RDONLY 2 /* The SHM file is readonly */
000563
000564 /*
000565 ** Each page of the wal-index mapping contains a hash-table made up of
000566 ** an array of HASHTABLE_NSLOT elements of the following type.
000567 */
000568 typedef u16 ht_slot;
000569
000570 /*
000571 ** This structure is used to implement an iterator that loops through
000572 ** all frames in the WAL in database page order. Where two or more frames
000573 ** correspond to the same database page, the iterator visits only the
000574 ** frame most recently written to the WAL (in other words, the frame with
000575 ** the largest index).
000576 **
000577 ** The internals of this structure are only accessed by:
000578 **
000579 ** walIteratorInit() - Create a new iterator,
000580 ** walIteratorNext() - Step an iterator,
000581 ** walIteratorFree() - Free an iterator.
000582 **
000583 ** This functionality is used by the checkpoint code (see walCheckpoint()).
000584 */
000585 struct WalIterator {
000586 u32 iPrior; /* Last result returned from the iterator */
000587 int nSegment; /* Number of entries in aSegment[] */
000588 struct WalSegment {
000589 int iNext; /* Next slot in aIndex[] not yet returned */
000590 ht_slot *aIndex; /* i0, i1, i2... such that aPgno[iN] ascend */
000591 u32 *aPgno; /* Array of page numbers. */
000592 int nEntry; /* Nr. of entries in aPgno[] and aIndex[] */
000593 int iZero; /* Frame number associated with aPgno[0] */
000594 } aSegment[1]; /* One for every 32KB page in the wal-index */
000595 };
000596
000597 /*
000598 ** Define the parameters of the hash tables in the wal-index file. There
000599 ** is a hash-table following every HASHTABLE_NPAGE page numbers in the
000600 ** wal-index.
000601 **
000602 ** Changing any of these constants will alter the wal-index format and
000603 ** create incompatibilities.
000604 */
000605 #define HASHTABLE_NPAGE 4096 /* Must be power of 2 */
000606 #define HASHTABLE_HASH_1 383 /* Should be prime */
000607 #define HASHTABLE_NSLOT (HASHTABLE_NPAGE*2) /* Must be a power of 2 */
000608
000609 /*
000610 ** The block of page numbers associated with the first hash-table in a
000611 ** wal-index is smaller than usual. This is so that there is a complete
000612 ** hash-table on each aligned 32KB page of the wal-index.
000613 */
000614 #define HASHTABLE_NPAGE_ONE (HASHTABLE_NPAGE - (WALINDEX_HDR_SIZE/sizeof(u32)))
000615
000616 /* The wal-index is divided into pages of WALINDEX_PGSZ bytes each. */
000617 #define WALINDEX_PGSZ ( \
000618 sizeof(ht_slot)*HASHTABLE_NSLOT + HASHTABLE_NPAGE*sizeof(u32) \
000619 )
000620
000621 /*
000622 ** Structured Exception Handling (SEH) is a Windows-specific technique
000623 ** for catching exceptions raised while accessing memory-mapped files.
000624 **
000625 ** The -DSQLITE_USE_SEH compile-time option means to use SEH to catch and
000626 ** deal with system-level errors that arise during WAL -shm file processing.
000627 ** Without this compile-time option, any system-level faults that appear
000628 ** while accessing the memory-mapped -shm file will cause a process-wide
000629 ** signal to be deliver, which will more than likely cause the entire
000630 ** process to exit.
000631 */
000632 #ifdef SQLITE_USE_SEH
000633 #include <Windows.h>
000634
000635 /* Beginning of a block of code in which an exception might occur */
000636 # define SEH_TRY __try { \
000637 assert( walAssertLockmask(pWal) && pWal->nSehTry==0 ); \
000638 VVA_ONLY(pWal->nSehTry++);
000639
000640 /* The end of a block of code in which an exception might occur */
000641 # define SEH_EXCEPT(X) \
000642 VVA_ONLY(pWal->nSehTry--); \
000643 assert( pWal->nSehTry==0 ); \
000644 } __except( sehExceptionFilter(pWal, GetExceptionCode(), GetExceptionInformation() ) ){ X }
000645
000646 /* Simulate a memory-mapping fault in the -shm file for testing purposes */
000647 # define SEH_INJECT_FAULT sehInjectFault(pWal)
000648
000649 /*
000650 ** The second argument is the return value of GetExceptionCode() for the
000651 ** current exception. Return EXCEPTION_EXECUTE_HANDLER if the exception code
000652 ** indicates that the exception may have been caused by accessing the *-shm
000653 ** file mapping. Or EXCEPTION_CONTINUE_SEARCH otherwise.
000654 */
000655 static int sehExceptionFilter(Wal *pWal, int eCode, EXCEPTION_POINTERS *p){
000656 VVA_ONLY(pWal->nSehTry--);
000657 if( eCode==EXCEPTION_IN_PAGE_ERROR ){
000658 if( p && p->ExceptionRecord && p->ExceptionRecord->NumberParameters>=3 ){
000659 /* From MSDN: For this type of exception, the first element of the
000660 ** ExceptionInformation[] array is a read-write flag - 0 if the exception
000661 ** was thrown while reading, 1 if while writing. The second element is
000662 ** the virtual address being accessed. The "third array element specifies
000663 ** the underlying NTSTATUS code that resulted in the exception". */
000664 pWal->iSysErrno = (int)p->ExceptionRecord->ExceptionInformation[2];
000665 }
000666 return EXCEPTION_EXECUTE_HANDLER;
000667 }
000668 return EXCEPTION_CONTINUE_SEARCH;
000669 }
000670
000671 /*
000672 ** If one is configured, invoke the xTestCallback callback with 650 as
000673 ** the argument. If it returns true, throw the same exception that is
000674 ** thrown by the system if the *-shm file mapping is accessed after it
000675 ** has been invalidated.
000676 */
000677 static void sehInjectFault(Wal *pWal){
000678 int res;
000679 assert( pWal->nSehTry>0 );
000680
000681 res = sqlite3FaultSim(650);
000682 if( res!=0 ){
000683 ULONG_PTR aArg[3];
000684 aArg[0] = 0;
000685 aArg[1] = 0;
000686 aArg[2] = (ULONG_PTR)res;
000687 RaiseException(EXCEPTION_IN_PAGE_ERROR, 0, 3, (const ULONG_PTR*)aArg);
000688 }
000689 }
000690
000691 /*
000692 ** There are two ways to use this macro. To set a pointer to be freed
000693 ** if an exception is thrown:
000694 **
000695 ** SEH_FREE_ON_ERROR(0, pPtr);
000696 **
000697 ** and to cancel the same:
000698 **
000699 ** SEH_FREE_ON_ERROR(pPtr, 0);
000700 **
000701 ** In the first case, there must not already be a pointer registered to
000702 ** be freed. In the second case, pPtr must be the registered pointer.
000703 */
000704 #define SEH_FREE_ON_ERROR(X,Y) \
000705 assert( (X==0 || Y==0) && pWal->pFree==X ); pWal->pFree = Y
000706
000707 /*
000708 ** There are two ways to use this macro. To arrange for pWal->apWiData[iPg]
000709 ** to be set to pValue if an exception is thrown:
000710 **
000711 ** SEH_SET_ON_ERROR(iPg, pValue);
000712 **
000713 ** and to cancel the same:
000714 **
000715 ** SEH_SET_ON_ERROR(0, 0);
000716 */
000717 #define SEH_SET_ON_ERROR(X,Y) pWal->iWiPg = X; pWal->pWiValue = Y
000718
000719 #else
000720 # define SEH_TRY VVA_ONLY(pWal->nSehTry++);
000721 # define SEH_EXCEPT(X) VVA_ONLY(pWal->nSehTry--); assert( pWal->nSehTry==0 );
000722 # define SEH_INJECT_FAULT assert( pWal->nSehTry>0 );
000723 # define SEH_FREE_ON_ERROR(X,Y)
000724 # define SEH_SET_ON_ERROR(X,Y)
000725 #endif /* ifdef SQLITE_USE_SEH */
000726
000727
000728 /*
000729 ** Obtain a pointer to the iPage'th page of the wal-index. The wal-index
000730 ** is broken into pages of WALINDEX_PGSZ bytes. Wal-index pages are
000731 ** numbered from zero.
000732 **
000733 ** If the wal-index is currently smaller the iPage pages then the size
000734 ** of the wal-index might be increased, but only if it is safe to do
000735 ** so. It is safe to enlarge the wal-index if pWal->writeLock is true
000736 ** or pWal->exclusiveMode==WAL_HEAPMEMORY_MODE.
000737 **
000738 ** Three possible result scenarios:
000739 **
000740 ** (1) rc==SQLITE_OK and *ppPage==Requested-Wal-Index-Page
000741 ** (2) rc>=SQLITE_ERROR and *ppPage==NULL
000742 ** (3) rc==SQLITE_OK and *ppPage==NULL // only if iPage==0
000743 **
000744 ** Scenario (3) can only occur when pWal->writeLock is false and iPage==0
000745 */
000746 static SQLITE_NOINLINE int walIndexPageRealloc(
000747 Wal *pWal, /* The WAL context */
000748 int iPage, /* The page we seek */
000749 volatile u32 **ppPage /* Write the page pointer here */
000750 ){
000751 int rc = SQLITE_OK;
000752
000753 /* Enlarge the pWal->apWiData[] array if required */
000754 if( pWal->nWiData<=iPage ){
000755 sqlite3_int64 nByte = sizeof(u32*)*(iPage+1);
000756 volatile u32 **apNew;
000757 apNew = (volatile u32 **)sqlite3Realloc((void *)pWal->apWiData, nByte);
000758 if( !apNew ){
000759 *ppPage = 0;
000760 return SQLITE_NOMEM_BKPT;
000761 }
000762 memset((void*)&apNew[pWal->nWiData], 0,
000763 sizeof(u32*)*(iPage+1-pWal->nWiData));
000764 pWal->apWiData = apNew;
000765 pWal->nWiData = iPage+1;
000766 }
000767
000768 /* Request a pointer to the required page from the VFS */
000769 assert( pWal->apWiData[iPage]==0 );
000770 if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE ){
000771 pWal->apWiData[iPage] = (u32 volatile *)sqlite3MallocZero(WALINDEX_PGSZ);
000772 if( !pWal->apWiData[iPage] ) rc = SQLITE_NOMEM_BKPT;
000773 }else{
000774 rc = sqlite3OsShmMap(pWal->pDbFd, iPage, WALINDEX_PGSZ,
000775 pWal->writeLock, (void volatile **)&pWal->apWiData[iPage]
000776 );
000777 assert( pWal->apWiData[iPage]!=0
000778 || rc!=SQLITE_OK
000779 || (pWal->writeLock==0 && iPage==0) );
000780 testcase( pWal->apWiData[iPage]==0 && rc==SQLITE_OK );
000781 if( rc==SQLITE_OK ){
000782 if( iPage>0 && sqlite3FaultSim(600) ) rc = SQLITE_NOMEM;
000783 }else if( (rc&0xff)==SQLITE_READONLY ){
000784 pWal->readOnly |= WAL_SHM_RDONLY;
000785 if( rc==SQLITE_READONLY ){
000786 rc = SQLITE_OK;
000787 }
000788 }
000789 }
000790
000791 *ppPage = pWal->apWiData[iPage];
000792 assert( iPage==0 || *ppPage || rc!=SQLITE_OK );
000793 return rc;
000794 }
000795 static int walIndexPage(
000796 Wal *pWal, /* The WAL context */
000797 int iPage, /* The page we seek */
000798 volatile u32 **ppPage /* Write the page pointer here */
000799 ){
000800 SEH_INJECT_FAULT;
000801 if( pWal->nWiData<=iPage || (*ppPage = pWal->apWiData[iPage])==0 ){
000802 return walIndexPageRealloc(pWal, iPage, ppPage);
000803 }
000804 return SQLITE_OK;
000805 }
000806
000807 /*
000808 ** Return a pointer to the WalCkptInfo structure in the wal-index.
000809 */
000810 static volatile WalCkptInfo *walCkptInfo(Wal *pWal){
000811 assert( pWal->nWiData>0 && pWal->apWiData[0] );
000812 SEH_INJECT_FAULT;
000813 return (volatile WalCkptInfo*)&(pWal->apWiData[0][sizeof(WalIndexHdr)/2]);
000814 }
000815
000816 /*
000817 ** Return a pointer to the WalIndexHdr structure in the wal-index.
000818 */
000819 static volatile WalIndexHdr *walIndexHdr(Wal *pWal){
000820 assert( pWal->nWiData>0 && pWal->apWiData[0] );
000821 SEH_INJECT_FAULT;
000822 return (volatile WalIndexHdr*)pWal->apWiData[0];
000823 }
000824
000825 /*
000826 ** The argument to this macro must be of type u32. On a little-endian
000827 ** architecture, it returns the u32 value that results from interpreting
000828 ** the 4 bytes as a big-endian value. On a big-endian architecture, it
000829 ** returns the value that would be produced by interpreting the 4 bytes
000830 ** of the input value as a little-endian integer.
000831 */
000832 #define BYTESWAP32(x) ( \
000833 (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8) \
000834 + (((x)&0x00FF0000)>>8) + (((x)&0xFF000000)>>24) \
000835 )
000836
000837 /*
000838 ** Generate or extend an 8 byte checksum based on the data in
000839 ** array aByte[] and the initial values of aIn[0] and aIn[1] (or
000840 ** initial values of 0 and 0 if aIn==NULL).
000841 **
000842 ** The checksum is written back into aOut[] before returning.
000843 **
000844 ** nByte must be a positive multiple of 8.
000845 */
000846 static void walChecksumBytes(
000847 int nativeCksum, /* True for native byte-order, false for non-native */
000848 u8 *a, /* Content to be checksummed */
000849 int nByte, /* Bytes of content in a[]. Must be a multiple of 8. */
000850 const u32 *aIn, /* Initial checksum value input */
000851 u32 *aOut /* OUT: Final checksum value output */
000852 ){
000853 u32 s1, s2;
000854 u32 *aData = (u32 *)a;
000855 u32 *aEnd = (u32 *)&a[nByte];
000856
000857 if( aIn ){
000858 s1 = aIn[0];
000859 s2 = aIn[1];
000860 }else{
000861 s1 = s2 = 0;
000862 }
000863
000864 assert( nByte>=8 );
000865 assert( (nByte&0x00000007)==0 );
000866 assert( nByte<=65536 );
000867 assert( nByte%4==0 );
000868
000869 if( !nativeCksum ){
000870 do {
000871 s1 += BYTESWAP32(aData[0]) + s2;
000872 s2 += BYTESWAP32(aData[1]) + s1;
000873 aData += 2;
000874 }while( aData<aEnd );
000875 }else if( nByte%64==0 ){
000876 do {
000877 s1 += *aData++ + s2;
000878 s2 += *aData++ + s1;
000879 s1 += *aData++ + s2;
000880 s2 += *aData++ + s1;
000881 s1 += *aData++ + s2;
000882 s2 += *aData++ + s1;
000883 s1 += *aData++ + s2;
000884 s2 += *aData++ + s1;
000885 s1 += *aData++ + s2;
000886 s2 += *aData++ + s1;
000887 s1 += *aData++ + s2;
000888 s2 += *aData++ + s1;
000889 s1 += *aData++ + s2;
000890 s2 += *aData++ + s1;
000891 s1 += *aData++ + s2;
000892 s2 += *aData++ + s1;
000893 }while( aData<aEnd );
000894 }else{
000895 do {
000896 s1 += *aData++ + s2;
000897 s2 += *aData++ + s1;
000898 }while( aData<aEnd );
000899 }
000900 assert( aData==aEnd );
000901
000902 aOut[0] = s1;
000903 aOut[1] = s2;
000904 }
000905
000906 /*
000907 ** If there is the possibility of concurrent access to the SHM file
000908 ** from multiple threads and/or processes, then do a memory barrier.
000909 */
000910 static void walShmBarrier(Wal *pWal){
000911 if( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE ){
000912 sqlite3OsShmBarrier(pWal->pDbFd);
000913 }
000914 }
000915
000916 /*
000917 ** Add the SQLITE_NO_TSAN as part of the return-type of a function
000918 ** definition as a hint that the function contains constructs that
000919 ** might give false-positive TSAN warnings.
000920 **
000921 ** See tag-20200519-1.
000922 */
000923 #if defined(__clang__) && !defined(SQLITE_NO_TSAN)
000924 # define SQLITE_NO_TSAN __attribute__((no_sanitize_thread))
000925 #else
000926 # define SQLITE_NO_TSAN
000927 #endif
000928
000929 /*
000930 ** Write the header information in pWal->hdr into the wal-index.
000931 **
000932 ** The checksum on pWal->hdr is updated before it is written.
000933 */
000934 static SQLITE_NO_TSAN void walIndexWriteHdr(Wal *pWal){
000935 volatile WalIndexHdr *aHdr = walIndexHdr(pWal);
000936 const int nCksum = offsetof(WalIndexHdr, aCksum);
000937
000938 assert( pWal->writeLock );
000939 pWal->hdr.isInit = 1;
000940 pWal->hdr.iVersion = WALINDEX_MAX_VERSION;
000941 walChecksumBytes(1, (u8*)&pWal->hdr, nCksum, 0, pWal->hdr.aCksum);
000942 /* Possible TSAN false-positive. See tag-20200519-1 */
000943 memcpy((void*)&aHdr[1], (const void*)&pWal->hdr, sizeof(WalIndexHdr));
000944 walShmBarrier(pWal);
000945 memcpy((void*)&aHdr[0], (const void*)&pWal->hdr, sizeof(WalIndexHdr));
000946 }
000947
000948 /*
000949 ** This function encodes a single frame header and writes it to a buffer
000950 ** supplied by the caller. A frame-header is made up of a series of
000951 ** 4-byte big-endian integers, as follows:
000952 **
000953 ** 0: Page number.
000954 ** 4: For commit records, the size of the database image in pages
000955 ** after the commit. For all other records, zero.
000956 ** 8: Salt-1 (copied from the wal-header)
000957 ** 12: Salt-2 (copied from the wal-header)
000958 ** 16: Checksum-1.
000959 ** 20: Checksum-2.
000960 */
000961 static void walEncodeFrame(
000962 Wal *pWal, /* The write-ahead log */
000963 u32 iPage, /* Database page number for frame */
000964 u32 nTruncate, /* New db size (or 0 for non-commit frames) */
000965 u8 *aData, /* Pointer to page data */
000966 u8 *aFrame /* OUT: Write encoded frame here */
000967 ){
000968 int nativeCksum; /* True for native byte-order checksums */
000969 u32 *aCksum = pWal->hdr.aFrameCksum;
000970 assert( WAL_FRAME_HDRSIZE==24 );
000971 sqlite3Put4byte(&aFrame[0], iPage);
000972 sqlite3Put4byte(&aFrame[4], nTruncate);
000973 if( pWal->iReCksum==0 ){
000974 memcpy(&aFrame[8], pWal->hdr.aSalt, 8);
000975
000976 nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN);
000977 walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum);
000978 walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum);
000979
000980 sqlite3Put4byte(&aFrame[16], aCksum[0]);
000981 sqlite3Put4byte(&aFrame[20], aCksum[1]);
000982 }else{
000983 memset(&aFrame[8], 0, 16);
000984 }
000985 }
000986
000987 /*
000988 ** Check to see if the frame with header in aFrame[] and content
000989 ** in aData[] is valid. If it is a valid frame, fill *piPage and
000990 ** *pnTruncate and return true. Return if the frame is not valid.
000991 */
000992 static int walDecodeFrame(
000993 Wal *pWal, /* The write-ahead log */
000994 u32 *piPage, /* OUT: Database page number for frame */
000995 u32 *pnTruncate, /* OUT: New db size (or 0 if not commit) */
000996 u8 *aData, /* Pointer to page data (for checksum) */
000997 u8 *aFrame /* Frame data */
000998 ){
000999 int nativeCksum; /* True for native byte-order checksums */
001000 u32 *aCksum = pWal->hdr.aFrameCksum;
001001 u32 pgno; /* Page number of the frame */
001002 assert( WAL_FRAME_HDRSIZE==24 );
001003
001004 /* A frame is only valid if the salt values in the frame-header
001005 ** match the salt values in the wal-header.
001006 */
001007 if( memcmp(&pWal->hdr.aSalt, &aFrame[8], 8)!=0 ){
001008 return 0;
001009 }
001010
001011 /* A frame is only valid if the page number is greater than zero.
001012 */
001013 pgno = sqlite3Get4byte(&aFrame[0]);
001014 if( pgno==0 ){
001015 return 0;
001016 }
001017
001018 /* A frame is only valid if a checksum of the WAL header,
001019 ** all prior frames, the first 16 bytes of this frame-header,
001020 ** and the frame-data matches the checksum in the last 8
001021 ** bytes of this frame-header.
001022 */
001023 nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN);
001024 walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum);
001025 walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum);
001026 if( aCksum[0]!=sqlite3Get4byte(&aFrame[16])
001027 || aCksum[1]!=sqlite3Get4byte(&aFrame[20])
001028 ){
001029 /* Checksum failed. */
001030 return 0;
001031 }
001032
001033 /* If we reach this point, the frame is valid. Return the page number
001034 ** and the new database size.
001035 */
001036 *piPage = pgno;
001037 *pnTruncate = sqlite3Get4byte(&aFrame[4]);
001038 return 1;
001039 }
001040
001041
001042 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
001043 /*
001044 ** Names of locks. This routine is used to provide debugging output and is not
001045 ** a part of an ordinary build.
001046 */
001047 static const char *walLockName(int lockIdx){
001048 if( lockIdx==WAL_WRITE_LOCK ){
001049 return "WRITE-LOCK";
001050 }else if( lockIdx==WAL_CKPT_LOCK ){
001051 return "CKPT-LOCK";
001052 }else if( lockIdx==WAL_RECOVER_LOCK ){
001053 return "RECOVER-LOCK";
001054 }else{
001055 static char zName[15];
001056 sqlite3_snprintf(sizeof(zName), zName, "READ-LOCK[%d]",
001057 lockIdx-WAL_READ_LOCK(0));
001058 return zName;
001059 }
001060 }
001061 #endif /*defined(SQLITE_TEST) || defined(SQLITE_DEBUG) */
001062
001063
001064 /*
001065 ** Set or release locks on the WAL. Locks are either shared or exclusive.
001066 ** A lock cannot be moved directly between shared and exclusive - it must go
001067 ** through the unlocked state first.
001068 **
001069 ** In locking_mode=EXCLUSIVE, all of these routines become no-ops.
001070 */
001071 static int walLockShared(Wal *pWal, int lockIdx){
001072 int rc;
001073 if( pWal->exclusiveMode ) return SQLITE_OK;
001074 rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1,
001075 SQLITE_SHM_LOCK | SQLITE_SHM_SHARED);
001076 WALTRACE(("WAL%p: acquire SHARED-%s %s\n", pWal,
001077 walLockName(lockIdx), rc ? "failed" : "ok"));
001078 VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); )
001079 #ifdef SQLITE_USE_SEH
001080 if( rc==SQLITE_OK ) pWal->lockMask |= (1 << lockIdx);
001081 #endif
001082 return rc;
001083 }
001084 static void walUnlockShared(Wal *pWal, int lockIdx){
001085 if( pWal->exclusiveMode ) return;
001086 (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1,
001087 SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED);
001088 #ifdef SQLITE_USE_SEH
001089 pWal->lockMask &= ~(1 << lockIdx);
001090 #endif
001091 WALTRACE(("WAL%p: release SHARED-%s\n", pWal, walLockName(lockIdx)));
001092 }
001093 static int walLockExclusive(Wal *pWal, int lockIdx, int n){
001094 int rc;
001095 if( pWal->exclusiveMode ) return SQLITE_OK;
001096 rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, n,
001097 SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE);
001098 WALTRACE(("WAL%p: acquire EXCLUSIVE-%s cnt=%d %s\n", pWal,
001099 walLockName(lockIdx), n, rc ? "failed" : "ok"));
001100 VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); )
001101 #ifdef SQLITE_USE_SEH
001102 if( rc==SQLITE_OK ){
001103 pWal->lockMask |= (((1<<n)-1) << (SQLITE_SHM_NLOCK+lockIdx));
001104 }
001105 #endif
001106 return rc;
001107 }
001108 static void walUnlockExclusive(Wal *pWal, int lockIdx, int n){
001109 if( pWal->exclusiveMode ) return;
001110 (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, n,
001111 SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE);
001112 #ifdef SQLITE_USE_SEH
001113 pWal->lockMask &= ~(((1<<n)-1) << (SQLITE_SHM_NLOCK+lockIdx));
001114 #endif
001115 WALTRACE(("WAL%p: release EXCLUSIVE-%s cnt=%d\n", pWal,
001116 walLockName(lockIdx), n));
001117 }
001118
001119 /*
001120 ** Compute a hash on a page number. The resulting hash value must land
001121 ** between 0 and (HASHTABLE_NSLOT-1). The walHashNext() function advances
001122 ** the hash to the next value in the event of a collision.
001123 */
001124 static int walHash(u32 iPage){
001125 assert( iPage>0 );
001126 assert( (HASHTABLE_NSLOT & (HASHTABLE_NSLOT-1))==0 );
001127 return (iPage*HASHTABLE_HASH_1) & (HASHTABLE_NSLOT-1);
001128 }
001129 static int walNextHash(int iPriorHash){
001130 return (iPriorHash+1)&(HASHTABLE_NSLOT-1);
001131 }
001132
001133 /*
001134 ** An instance of the WalHashLoc object is used to describe the location
001135 ** of a page hash table in the wal-index. This becomes the return value
001136 ** from walHashGet().
001137 */
001138 typedef struct WalHashLoc WalHashLoc;
001139 struct WalHashLoc {
001140 volatile ht_slot *aHash; /* Start of the wal-index hash table */
001141 volatile u32 *aPgno; /* aPgno[1] is the page of first frame indexed */
001142 u32 iZero; /* One less than the frame number of first indexed*/
001143 };
001144
001145 /*
001146 ** Return pointers to the hash table and page number array stored on
001147 ** page iHash of the wal-index. The wal-index is broken into 32KB pages
001148 ** numbered starting from 0.
001149 **
001150 ** Set output variable pLoc->aHash to point to the start of the hash table
001151 ** in the wal-index file. Set pLoc->iZero to one less than the frame
001152 ** number of the first frame indexed by this hash table. If a
001153 ** slot in the hash table is set to N, it refers to frame number
001154 ** (pLoc->iZero+N) in the log.
001155 **
001156 ** Finally, set pLoc->aPgno so that pLoc->aPgno[0] is the page number of the
001157 ** first frame indexed by the hash table, frame (pLoc->iZero).
001158 */
001159 static int walHashGet(
001160 Wal *pWal, /* WAL handle */
001161 int iHash, /* Find the iHash'th table */
001162 WalHashLoc *pLoc /* OUT: Hash table location */
001163 ){
001164 int rc; /* Return code */
001165
001166 rc = walIndexPage(pWal, iHash, &pLoc->aPgno);
001167 assert( rc==SQLITE_OK || iHash>0 );
001168
001169 if( pLoc->aPgno ){
001170 pLoc->aHash = (volatile ht_slot *)&pLoc->aPgno[HASHTABLE_NPAGE];
001171 if( iHash==0 ){
001172 pLoc->aPgno = &pLoc->aPgno[WALINDEX_HDR_SIZE/sizeof(u32)];
001173 pLoc->iZero = 0;
001174 }else{
001175 pLoc->iZero = HASHTABLE_NPAGE_ONE + (iHash-1)*HASHTABLE_NPAGE;
001176 }
001177 }else if( NEVER(rc==SQLITE_OK) ){
001178 rc = SQLITE_ERROR;
001179 }
001180 return rc;
001181 }
001182
001183 /*
001184 ** Return the number of the wal-index page that contains the hash-table
001185 ** and page-number array that contain entries corresponding to WAL frame
001186 ** iFrame. The wal-index is broken up into 32KB pages. Wal-index pages
001187 ** are numbered starting from 0.
001188 */
001189 static int walFramePage(u32 iFrame){
001190 int iHash = (iFrame+HASHTABLE_NPAGE-HASHTABLE_NPAGE_ONE-1) / HASHTABLE_NPAGE;
001191 assert( (iHash==0 || iFrame>HASHTABLE_NPAGE_ONE)
001192 && (iHash>=1 || iFrame<=HASHTABLE_NPAGE_ONE)
001193 && (iHash<=1 || iFrame>(HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE))
001194 && (iHash>=2 || iFrame<=HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE)
001195 && (iHash<=2 || iFrame>(HASHTABLE_NPAGE_ONE+2*HASHTABLE_NPAGE))
001196 );
001197 assert( iHash>=0 );
001198 return iHash;
001199 }
001200
001201 /*
001202 ** Return the page number associated with frame iFrame in this WAL.
001203 */
001204 static u32 walFramePgno(Wal *pWal, u32 iFrame){
001205 int iHash = walFramePage(iFrame);
001206 SEH_INJECT_FAULT;
001207 if( iHash==0 ){
001208 return pWal->apWiData[0][WALINDEX_HDR_SIZE/sizeof(u32) + iFrame - 1];
001209 }
001210 return pWal->apWiData[iHash][(iFrame-1-HASHTABLE_NPAGE_ONE)%HASHTABLE_NPAGE];
001211 }
001212
001213 /*
001214 ** Remove entries from the hash table that point to WAL slots greater
001215 ** than pWal->hdr.mxFrame.
001216 **
001217 ** This function is called whenever pWal->hdr.mxFrame is decreased due
001218 ** to a rollback or savepoint.
001219 **
001220 ** At most only the hash table containing pWal->hdr.mxFrame needs to be
001221 ** updated. Any later hash tables will be automatically cleared when
001222 ** pWal->hdr.mxFrame advances to the point where those hash tables are
001223 ** actually needed.
001224 */
001225 static void walCleanupHash(Wal *pWal){
001226 WalHashLoc sLoc; /* Hash table location */
001227 int iLimit = 0; /* Zero values greater than this */
001228 int nByte; /* Number of bytes to zero in aPgno[] */
001229 int i; /* Used to iterate through aHash[] */
001230
001231 assert( pWal->writeLock );
001232 testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE-1 );
001233 testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE );
001234 testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE+1 );
001235
001236 if( pWal->hdr.mxFrame==0 ) return;
001237
001238 /* Obtain pointers to the hash-table and page-number array containing
001239 ** the entry that corresponds to frame pWal->hdr.mxFrame. It is guaranteed
001240 ** that the page said hash-table and array reside on is already mapped.(1)
001241 */
001242 assert( pWal->nWiData>walFramePage(pWal->hdr.mxFrame) );
001243 assert( pWal->apWiData[walFramePage(pWal->hdr.mxFrame)] );
001244 i = walHashGet(pWal, walFramePage(pWal->hdr.mxFrame), &sLoc);
001245 if( NEVER(i) ) return; /* Defense-in-depth, in case (1) above is wrong */
001246
001247 /* Zero all hash-table entries that correspond to frame numbers greater
001248 ** than pWal->hdr.mxFrame.
001249 */
001250 iLimit = pWal->hdr.mxFrame - sLoc.iZero;
001251 assert( iLimit>0 );
001252 for(i=0; i<HASHTABLE_NSLOT; i++){
001253 if( sLoc.aHash[i]>iLimit ){
001254 sLoc.aHash[i] = 0;
001255 }
001256 }
001257
001258 /* Zero the entries in the aPgno array that correspond to frames with
001259 ** frame numbers greater than pWal->hdr.mxFrame.
001260 */
001261 nByte = (int)((char *)sLoc.aHash - (char *)&sLoc.aPgno[iLimit]);
001262 assert( nByte>=0 );
001263 memset((void *)&sLoc.aPgno[iLimit], 0, nByte);
001264
001265 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
001266 /* Verify that the every entry in the mapping region is still reachable
001267 ** via the hash table even after the cleanup.
001268 */
001269 if( iLimit ){
001270 int j; /* Loop counter */
001271 int iKey; /* Hash key */
001272 for(j=0; j<iLimit; j++){
001273 for(iKey=walHash(sLoc.aPgno[j]);sLoc.aHash[iKey];iKey=walNextHash(iKey)){
001274 if( sLoc.aHash[iKey]==j+1 ) break;
001275 }
001276 assert( sLoc.aHash[iKey]==j+1 );
001277 }
001278 }
001279 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
001280 }
001281
001282
001283 /*
001284 ** Set an entry in the wal-index that will map database page number
001285 ** pPage into WAL frame iFrame.
001286 */
001287 static int walIndexAppend(Wal *pWal, u32 iFrame, u32 iPage){
001288 int rc; /* Return code */
001289 WalHashLoc sLoc; /* Wal-index hash table location */
001290
001291 rc = walHashGet(pWal, walFramePage(iFrame), &sLoc);
001292
001293 /* Assuming the wal-index file was successfully mapped, populate the
001294 ** page number array and hash table entry.
001295 */
001296 if( rc==SQLITE_OK ){
001297 int iKey; /* Hash table key */
001298 int idx; /* Value to write to hash-table slot */
001299 int nCollide; /* Number of hash collisions */
001300
001301 idx = iFrame - sLoc.iZero;
001302 assert( idx <= HASHTABLE_NSLOT/2 + 1 );
001303
001304 /* If this is the first entry to be added to this hash-table, zero the
001305 ** entire hash table and aPgno[] array before proceeding.
001306 */
001307 if( idx==1 ){
001308 int nByte = (int)((u8*)&sLoc.aHash[HASHTABLE_NSLOT] - (u8*)sLoc.aPgno);
001309 assert( nByte>=0 );
001310 memset((void*)sLoc.aPgno, 0, nByte);
001311 }
001312
001313 /* If the entry in aPgno[] is already set, then the previous writer
001314 ** must have exited unexpectedly in the middle of a transaction (after
001315 ** writing one or more dirty pages to the WAL to free up memory).
001316 ** Remove the remnants of that writers uncommitted transaction from
001317 ** the hash-table before writing any new entries.
001318 */
001319 if( sLoc.aPgno[idx-1] ){
001320 walCleanupHash(pWal);
001321 assert( !sLoc.aPgno[idx-1] );
001322 }
001323
001324 /* Write the aPgno[] array entry and the hash-table slot. */
001325 nCollide = idx;
001326 for(iKey=walHash(iPage); sLoc.aHash[iKey]; iKey=walNextHash(iKey)){
001327 if( (nCollide--)==0 ) return SQLITE_CORRUPT_BKPT;
001328 }
001329 sLoc.aPgno[idx-1] = iPage;
001330 AtomicStore(&sLoc.aHash[iKey], (ht_slot)idx);
001331
001332 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
001333 /* Verify that the number of entries in the hash table exactly equals
001334 ** the number of entries in the mapping region.
001335 */
001336 {
001337 int i; /* Loop counter */
001338 int nEntry = 0; /* Number of entries in the hash table */
001339 for(i=0; i<HASHTABLE_NSLOT; i++){ if( sLoc.aHash[i] ) nEntry++; }
001340 assert( nEntry==idx );
001341 }
001342
001343 /* Verify that the every entry in the mapping region is reachable
001344 ** via the hash table. This turns out to be a really, really expensive
001345 ** thing to check, so only do this occasionally - not on every
001346 ** iteration.
001347 */
001348 if( (idx&0x3ff)==0 ){
001349 int i; /* Loop counter */
001350 for(i=0; i<idx; i++){
001351 for(iKey=walHash(sLoc.aPgno[i]);
001352 sLoc.aHash[iKey];
001353 iKey=walNextHash(iKey)){
001354 if( sLoc.aHash[iKey]==i+1 ) break;
001355 }
001356 assert( sLoc.aHash[iKey]==i+1 );
001357 }
001358 }
001359 #endif /* SQLITE_ENABLE_EXPENSIVE_ASSERT */
001360 }
001361
001362 return rc;
001363 }
001364
001365
001366 /*
001367 ** Recover the wal-index by reading the write-ahead log file.
001368 **
001369 ** This routine first tries to establish an exclusive lock on the
001370 ** wal-index to prevent other threads/processes from doing anything
001371 ** with the WAL or wal-index while recovery is running. The
001372 ** WAL_RECOVER_LOCK is also held so that other threads will know
001373 ** that this thread is running recovery. If unable to establish
001374 ** the necessary locks, this routine returns SQLITE_BUSY.
001375 */
001376 static int walIndexRecover(Wal *pWal){
001377 int rc; /* Return Code */
001378 i64 nSize; /* Size of log file */
001379 u32 aFrameCksum[2] = {0, 0};
001380 int iLock; /* Lock offset to lock for checkpoint */
001381
001382 /* Obtain an exclusive lock on all byte in the locking range not already
001383 ** locked by the caller. The caller is guaranteed to have locked the
001384 ** WAL_WRITE_LOCK byte, and may have also locked the WAL_CKPT_LOCK byte.
001385 ** If successful, the same bytes that are locked here are unlocked before
001386 ** this function returns.
001387 */
001388 assert( pWal->ckptLock==1 || pWal->ckptLock==0 );
001389 assert( WAL_ALL_BUT_WRITE==WAL_WRITE_LOCK+1 );
001390 assert( WAL_CKPT_LOCK==WAL_ALL_BUT_WRITE );
001391 assert( pWal->writeLock );
001392 iLock = WAL_ALL_BUT_WRITE + pWal->ckptLock;
001393 rc = walLockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock);
001394 if( rc ){
001395 return rc;
001396 }
001397
001398 WALTRACE(("WAL%p: recovery begin...\n", pWal));
001399
001400 memset(&pWal->hdr, 0, sizeof(WalIndexHdr));
001401
001402 rc = sqlite3OsFileSize(pWal->pWalFd, &nSize);
001403 if( rc!=SQLITE_OK ){
001404 goto recovery_error;
001405 }
001406
001407 if( nSize>WAL_HDRSIZE ){
001408 u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */
001409 u32 *aPrivate = 0; /* Heap copy of *-shm hash being populated */
001410 u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */
001411 int szFrame; /* Number of bytes in buffer aFrame[] */
001412 u8 *aData; /* Pointer to data part of aFrame buffer */
001413 int szPage; /* Page size according to the log */
001414 u32 magic; /* Magic value read from WAL header */
001415 u32 version; /* Magic value read from WAL header */
001416 int isValid; /* True if this frame is valid */
001417 u32 iPg; /* Current 32KB wal-index page */
001418 u32 iLastFrame; /* Last frame in wal, based on nSize alone */
001419
001420 /* Read in the WAL header. */
001421 rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0);
001422 if( rc!=SQLITE_OK ){
001423 goto recovery_error;
001424 }
001425
001426 /* If the database page size is not a power of two, or is greater than
001427 ** SQLITE_MAX_PAGE_SIZE, conclude that the WAL file contains no valid
001428 ** data. Similarly, if the 'magic' value is invalid, ignore the whole
001429 ** WAL file.
001430 */
001431 magic = sqlite3Get4byte(&aBuf[0]);
001432 szPage = sqlite3Get4byte(&aBuf[8]);
001433 if( (magic&0xFFFFFFFE)!=WAL_MAGIC
001434 || szPage&(szPage-1)
001435 || szPage>SQLITE_MAX_PAGE_SIZE
001436 || szPage<512
001437 ){
001438 goto finished;
001439 }
001440 pWal->hdr.bigEndCksum = (u8)(magic&0x00000001);
001441 pWal->szPage = szPage;
001442 pWal->nCkpt = sqlite3Get4byte(&aBuf[12]);
001443 memcpy(&pWal->hdr.aSalt, &aBuf[16], 8);
001444
001445 /* Verify that the WAL header checksum is correct */
001446 walChecksumBytes(pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN,
001447 aBuf, WAL_HDRSIZE-2*4, 0, pWal->hdr.aFrameCksum
001448 );
001449 if( pWal->hdr.aFrameCksum[0]!=sqlite3Get4byte(&aBuf[24])
001450 || pWal->hdr.aFrameCksum[1]!=sqlite3Get4byte(&aBuf[28])
001451 ){
001452 goto finished;
001453 }
001454
001455 /* Verify that the version number on the WAL format is one that
001456 ** are able to understand */
001457 version = sqlite3Get4byte(&aBuf[4]);
001458 if( version!=WAL_MAX_VERSION ){
001459 rc = SQLITE_CANTOPEN_BKPT;
001460 goto finished;
001461 }
001462
001463 /* Malloc a buffer to read frames into. */
001464 szFrame = szPage + WAL_FRAME_HDRSIZE;
001465 aFrame = (u8 *)sqlite3_malloc64(szFrame + WALINDEX_PGSZ);
001466 SEH_FREE_ON_ERROR(0, aFrame);
001467 if( !aFrame ){
001468 rc = SQLITE_NOMEM_BKPT;
001469 goto recovery_error;
001470 }
001471 aData = &aFrame[WAL_FRAME_HDRSIZE];
001472 aPrivate = (u32*)&aData[szPage];
001473
001474 /* Read all frames from the log file. */
001475 iLastFrame = (nSize - WAL_HDRSIZE) / szFrame;
001476 for(iPg=0; iPg<=(u32)walFramePage(iLastFrame); iPg++){
001477 u32 *aShare;
001478 u32 iFrame; /* Index of last frame read */
001479 u32 iLast = MIN(iLastFrame, HASHTABLE_NPAGE_ONE+iPg*HASHTABLE_NPAGE);
001480 u32 iFirst = 1 + (iPg==0?0:HASHTABLE_NPAGE_ONE+(iPg-1)*HASHTABLE_NPAGE);
001481 u32 nHdr, nHdr32;
001482 rc = walIndexPage(pWal, iPg, (volatile u32**)&aShare);
001483 assert( aShare!=0 || rc!=SQLITE_OK );
001484 if( aShare==0 ) break;
001485 SEH_SET_ON_ERROR(iPg, aShare);
001486 pWal->apWiData[iPg] = aPrivate;
001487
001488 for(iFrame=iFirst; iFrame<=iLast; iFrame++){
001489 i64 iOffset = walFrameOffset(iFrame, szPage);
001490 u32 pgno; /* Database page number for frame */
001491 u32 nTruncate; /* dbsize field from frame header */
001492
001493 /* Read and decode the next log frame. */
001494 rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset);
001495 if( rc!=SQLITE_OK ) break;
001496 isValid = walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame);
001497 if( !isValid ) break;
001498 rc = walIndexAppend(pWal, iFrame, pgno);
001499 if( NEVER(rc!=SQLITE_OK) ) break;
001500
001501 /* If nTruncate is non-zero, this is a commit record. */
001502 if( nTruncate ){
001503 pWal->hdr.mxFrame = iFrame;
001504 pWal->hdr.nPage = nTruncate;
001505 pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16));
001506 testcase( szPage<=32768 );
001507 testcase( szPage>=65536 );
001508 aFrameCksum[0] = pWal->hdr.aFrameCksum[0];
001509 aFrameCksum[1] = pWal->hdr.aFrameCksum[1];
001510 }
001511 }
001512 pWal->apWiData[iPg] = aShare;
001513 SEH_SET_ON_ERROR(0,0);
001514 nHdr = (iPg==0 ? WALINDEX_HDR_SIZE : 0);
001515 nHdr32 = nHdr / sizeof(u32);
001516 #ifndef SQLITE_SAFER_WALINDEX_RECOVERY
001517 /* Memcpy() should work fine here, on all reasonable implementations.
001518 ** Technically, memcpy() might change the destination to some
001519 ** intermediate value before setting to the final value, and that might
001520 ** cause a concurrent reader to malfunction. Memcpy() is allowed to
001521 ** do that, according to the spec, but no memcpy() implementation that
001522 ** we know of actually does that, which is why we say that memcpy()
001523 ** is safe for this. Memcpy() is certainly a lot faster.
001524 */
001525 memcpy(&aShare[nHdr32], &aPrivate[nHdr32], WALINDEX_PGSZ-nHdr);
001526 #else
001527 /* In the event that some platform is found for which memcpy()
001528 ** changes the destination to some intermediate value before
001529 ** setting the final value, this alternative copy routine is
001530 ** provided.
001531 */
001532 {
001533 int i;
001534 for(i=nHdr32; i<WALINDEX_PGSZ/sizeof(u32); i++){
001535 if( aShare[i]!=aPrivate[i] ){
001536 /* Atomic memory operations are not required here because if
001537 ** the value needs to be changed, that means it is not being
001538 ** accessed concurrently. */
001539 aShare[i] = aPrivate[i];
001540 }
001541 }
001542 }
001543 #endif
001544 SEH_INJECT_FAULT;
001545 if( iFrame<=iLast ) break;
001546 }
001547
001548 SEH_FREE_ON_ERROR(aFrame, 0);
001549 sqlite3_free(aFrame);
001550 }
001551
001552 finished:
001553 if( rc==SQLITE_OK ){
001554 volatile WalCkptInfo *pInfo;
001555 int i;
001556 pWal->hdr.aFrameCksum[0] = aFrameCksum[0];
001557 pWal->hdr.aFrameCksum[1] = aFrameCksum[1];
001558 walIndexWriteHdr(pWal);
001559
001560 /* Reset the checkpoint-header. This is safe because this thread is
001561 ** currently holding locks that exclude all other writers and
001562 ** checkpointers. Then set the values of read-mark slots 1 through N.
001563 */
001564 pInfo = walCkptInfo(pWal);
001565 pInfo->nBackfill = 0;
001566 pInfo->nBackfillAttempted = pWal->hdr.mxFrame;
001567 pInfo->aReadMark[0] = 0;
001568 for(i=1; i<WAL_NREADER; i++){
001569 rc = walLockExclusive(pWal, WAL_READ_LOCK(i), 1);
001570 if( rc==SQLITE_OK ){
001571 if( i==1 && pWal->hdr.mxFrame ){
001572 pInfo->aReadMark[i] = pWal->hdr.mxFrame;
001573 }else{
001574 pInfo->aReadMark[i] = READMARK_NOT_USED;
001575 }
001576 SEH_INJECT_FAULT;
001577 walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1);
001578 }else if( rc!=SQLITE_BUSY ){
001579 goto recovery_error;
001580 }
001581 }
001582
001583 /* If more than one frame was recovered from the log file, report an
001584 ** event via sqlite3_log(). This is to help with identifying performance
001585 ** problems caused by applications routinely shutting down without
001586 ** checkpointing the log file.
001587 */
001588 if( pWal->hdr.nPage ){
001589 sqlite3_log(SQLITE_NOTICE_RECOVER_WAL,
001590 "recovered %d frames from WAL file %s",
001591 pWal->hdr.mxFrame, pWal->zWalName
001592 );
001593 }
001594 }
001595
001596 recovery_error:
001597 WALTRACE(("WAL%p: recovery %s\n", pWal, rc ? "failed" : "ok"));
001598 walUnlockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock);
001599 return rc;
001600 }
001601
001602 /*
001603 ** Close an open wal-index.
001604 */
001605 static void walIndexClose(Wal *pWal, int isDelete){
001606 if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE || pWal->bShmUnreliable ){
001607 int i;
001608 for(i=0; i<pWal->nWiData; i++){
001609 sqlite3_free((void *)pWal->apWiData[i]);
001610 pWal->apWiData[i] = 0;
001611 }
001612 }
001613 if( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE ){
001614 sqlite3OsShmUnmap(pWal->pDbFd, isDelete);
001615 }
001616 }
001617
001618 /*
001619 ** Open a connection to the WAL file zWalName. The database file must
001620 ** already be opened on connection pDbFd. The buffer that zWalName points
001621 ** to must remain valid for the lifetime of the returned Wal* handle.
001622 **
001623 ** A SHARED lock should be held on the database file when this function
001624 ** is called. The purpose of this SHARED lock is to prevent any other
001625 ** client from unlinking the WAL or wal-index file. If another process
001626 ** were to do this just after this client opened one of these files, the
001627 ** system would be badly broken.
001628 **
001629 ** If the log file is successfully opened, SQLITE_OK is returned and
001630 ** *ppWal is set to point to a new WAL handle. If an error occurs,
001631 ** an SQLite error code is returned and *ppWal is left unmodified.
001632 */
001633 int sqlite3WalOpen(
001634 sqlite3_vfs *pVfs, /* vfs module to open wal and wal-index */
001635 sqlite3_file *pDbFd, /* The open database file */
001636 const char *zWalName, /* Name of the WAL file */
001637 int bNoShm, /* True to run in heap-memory mode */
001638 i64 mxWalSize, /* Truncate WAL to this size on reset */
001639 Wal **ppWal /* OUT: Allocated Wal handle */
001640 ){
001641 int rc; /* Return Code */
001642 Wal *pRet; /* Object to allocate and return */
001643 int flags; /* Flags passed to OsOpen() */
001644
001645 assert( zWalName && zWalName[0] );
001646 assert( pDbFd );
001647
001648 /* Verify the values of various constants. Any changes to the values
001649 ** of these constants would result in an incompatible on-disk format
001650 ** for the -shm file. Any change that causes one of these asserts to
001651 ** fail is a backward compatibility problem, even if the change otherwise
001652 ** works.
001653 **
001654 ** This table also serves as a helpful cross-reference when trying to
001655 ** interpret hex dumps of the -shm file.
001656 */
001657 assert( 48 == sizeof(WalIndexHdr) );
001658 assert( 40 == sizeof(WalCkptInfo) );
001659 assert( 120 == WALINDEX_LOCK_OFFSET );
001660 assert( 136 == WALINDEX_HDR_SIZE );
001661 assert( 4096 == HASHTABLE_NPAGE );
001662 assert( 4062 == HASHTABLE_NPAGE_ONE );
001663 assert( 8192 == HASHTABLE_NSLOT );
001664 assert( 383 == HASHTABLE_HASH_1 );
001665 assert( 32768 == WALINDEX_PGSZ );
001666 assert( 8 == SQLITE_SHM_NLOCK );
001667 assert( 5 == WAL_NREADER );
001668 assert( 24 == WAL_FRAME_HDRSIZE );
001669 assert( 32 == WAL_HDRSIZE );
001670 assert( 120 == WALINDEX_LOCK_OFFSET + WAL_WRITE_LOCK );
001671 assert( 121 == WALINDEX_LOCK_OFFSET + WAL_CKPT_LOCK );
001672 assert( 122 == WALINDEX_LOCK_OFFSET + WAL_RECOVER_LOCK );
001673 assert( 123 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(0) );
001674 assert( 124 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(1) );
001675 assert( 125 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(2) );
001676 assert( 126 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(3) );
001677 assert( 127 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(4) );
001678
001679 /* In the amalgamation, the os_unix.c and os_win.c source files come before
001680 ** this source file. Verify that the #defines of the locking byte offsets
001681 ** in os_unix.c and os_win.c agree with the WALINDEX_LOCK_OFFSET value.
001682 ** For that matter, if the lock offset ever changes from its initial design
001683 ** value of 120, we need to know that so there is an assert() to check it.
001684 */
001685 #ifdef WIN_SHM_BASE
001686 assert( WIN_SHM_BASE==WALINDEX_LOCK_OFFSET );
001687 #endif
001688 #ifdef UNIX_SHM_BASE
001689 assert( UNIX_SHM_BASE==WALINDEX_LOCK_OFFSET );
001690 #endif
001691
001692
001693 /* Allocate an instance of struct Wal to return. */
001694 *ppWal = 0;
001695 pRet = (Wal*)sqlite3MallocZero(sizeof(Wal) + pVfs->szOsFile);
001696 if( !pRet ){
001697 return SQLITE_NOMEM_BKPT;
001698 }
001699
001700 pRet->pVfs = pVfs;
001701 pRet->pWalFd = (sqlite3_file *)&pRet[1];
001702 pRet->pDbFd = pDbFd;
001703 pRet->readLock = -1;
001704 pRet->mxWalSize = mxWalSize;
001705 pRet->zWalName = zWalName;
001706 pRet->syncHeader = 1;
001707 pRet->padToSectorBoundary = 1;
001708 pRet->exclusiveMode = (bNoShm ? WAL_HEAPMEMORY_MODE: WAL_NORMAL_MODE);
001709
001710 /* Open file handle on the write-ahead log file. */
001711 flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE|SQLITE_OPEN_WAL);
001712 rc = sqlite3OsOpen(pVfs, zWalName, pRet->pWalFd, flags, &flags);
001713 if( rc==SQLITE_OK && flags&SQLITE_OPEN_READONLY ){
001714 pRet->readOnly = WAL_RDONLY;
001715 }
001716
001717 if( rc!=SQLITE_OK ){
001718 walIndexClose(pRet, 0);
001719 sqlite3OsClose(pRet->pWalFd);
001720 sqlite3_free(pRet);
001721 }else{
001722 int iDC = sqlite3OsDeviceCharacteristics(pDbFd);
001723 if( iDC & SQLITE_IOCAP_SEQUENTIAL ){ pRet->syncHeader = 0; }
001724 if( iDC & SQLITE_IOCAP_POWERSAFE_OVERWRITE ){
001725 pRet->padToSectorBoundary = 0;
001726 }
001727 *ppWal = pRet;
001728 WALTRACE(("WAL%d: opened\n", pRet));
001729 }
001730 return rc;
001731 }
001732
001733 /*
001734 ** Change the size to which the WAL file is truncated on each reset.
001735 */
001736 void sqlite3WalLimit(Wal *pWal, i64 iLimit){
001737 if( pWal ) pWal->mxWalSize = iLimit;
001738 }
001739
001740 /*
001741 ** Find the smallest page number out of all pages held in the WAL that
001742 ** has not been returned by any prior invocation of this method on the
001743 ** same WalIterator object. Write into *piFrame the frame index where
001744 ** that page was last written into the WAL. Write into *piPage the page
001745 ** number.
001746 **
001747 ** Return 0 on success. If there are no pages in the WAL with a page
001748 ** number larger than *piPage, then return 1.
001749 */
001750 static int walIteratorNext(
001751 WalIterator *p, /* Iterator */
001752 u32 *piPage, /* OUT: The page number of the next page */
001753 u32 *piFrame /* OUT: Wal frame index of next page */
001754 ){
001755 u32 iMin; /* Result pgno must be greater than iMin */
001756 u32 iRet = 0xFFFFFFFF; /* 0xffffffff is never a valid page number */
001757 int i; /* For looping through segments */
001758
001759 iMin = p->iPrior;
001760 assert( iMin<0xffffffff );
001761 for(i=p->nSegment-1; i>=0; i--){
001762 struct WalSegment *pSegment = &p->aSegment[i];
001763 while( pSegment->iNext<pSegment->nEntry ){
001764 u32 iPg = pSegment->aPgno[pSegment->aIndex[pSegment->iNext]];
001765 if( iPg>iMin ){
001766 if( iPg<iRet ){
001767 iRet = iPg;
001768 *piFrame = pSegment->iZero + pSegment->aIndex[pSegment->iNext];
001769 }
001770 break;
001771 }
001772 pSegment->iNext++;
001773 }
001774 }
001775
001776 *piPage = p->iPrior = iRet;
001777 return (iRet==0xFFFFFFFF);
001778 }
001779
001780 /*
001781 ** This function merges two sorted lists into a single sorted list.
001782 **
001783 ** aLeft[] and aRight[] are arrays of indices. The sort key is
001784 ** aContent[aLeft[]] and aContent[aRight[]]. Upon entry, the following
001785 ** is guaranteed for all J<K:
001786 **
001787 ** aContent[aLeft[J]] < aContent[aLeft[K]]
001788 ** aContent[aRight[J]] < aContent[aRight[K]]
001789 **
001790 ** This routine overwrites aRight[] with a new (probably longer) sequence
001791 ** of indices such that the aRight[] contains every index that appears in
001792 ** either aLeft[] or the old aRight[] and such that the second condition
001793 ** above is still met.
001794 **
001795 ** The aContent[aLeft[X]] values will be unique for all X. And the
001796 ** aContent[aRight[X]] values will be unique too. But there might be
001797 ** one or more combinations of X and Y such that
001798 **
001799 ** aLeft[X]!=aRight[Y] && aContent[aLeft[X]] == aContent[aRight[Y]]
001800 **
001801 ** When that happens, omit the aLeft[X] and use the aRight[Y] index.
001802 */
001803 static void walMerge(
001804 const u32 *aContent, /* Pages in wal - keys for the sort */
001805 ht_slot *aLeft, /* IN: Left hand input list */
001806 int nLeft, /* IN: Elements in array *paLeft */
001807 ht_slot **paRight, /* IN/OUT: Right hand input list */
001808 int *pnRight, /* IN/OUT: Elements in *paRight */
001809 ht_slot *aTmp /* Temporary buffer */
001810 ){
001811 int iLeft = 0; /* Current index in aLeft */
001812 int iRight = 0; /* Current index in aRight */
001813 int iOut = 0; /* Current index in output buffer */
001814 int nRight = *pnRight;
001815 ht_slot *aRight = *paRight;
001816
001817 assert( nLeft>0 && nRight>0 );
001818 while( iRight<nRight || iLeft<nLeft ){
001819 ht_slot logpage;
001820 Pgno dbpage;
001821
001822 if( (iLeft<nLeft)
001823 && (iRight>=nRight || aContent[aLeft[iLeft]]<aContent[aRight[iRight]])
001824 ){
001825 logpage = aLeft[iLeft++];
001826 }else{
001827 logpage = aRight[iRight++];
001828 }
001829 dbpage = aContent[logpage];
001830
001831 aTmp[iOut++] = logpage;
001832 if( iLeft<nLeft && aContent[aLeft[iLeft]]==dbpage ) iLeft++;
001833
001834 assert( iLeft>=nLeft || aContent[aLeft[iLeft]]>dbpage );
001835 assert( iRight>=nRight || aContent[aRight[iRight]]>dbpage );
001836 }
001837
001838 *paRight = aLeft;
001839 *pnRight = iOut;
001840 memcpy(aLeft, aTmp, sizeof(aTmp[0])*iOut);
001841 }
001842
001843 /*
001844 ** Sort the elements in list aList using aContent[] as the sort key.
001845 ** Remove elements with duplicate keys, preferring to keep the
001846 ** larger aList[] values.
001847 **
001848 ** The aList[] entries are indices into aContent[]. The values in
001849 ** aList[] are to be sorted so that for all J<K:
001850 **
001851 ** aContent[aList[J]] < aContent[aList[K]]
001852 **
001853 ** For any X and Y such that
001854 **
001855 ** aContent[aList[X]] == aContent[aList[Y]]
001856 **
001857 ** Keep the larger of the two values aList[X] and aList[Y] and discard
001858 ** the smaller.
001859 */
001860 static void walMergesort(
001861 const u32 *aContent, /* Pages in wal */
001862 ht_slot *aBuffer, /* Buffer of at least *pnList items to use */
001863 ht_slot *aList, /* IN/OUT: List to sort */
001864 int *pnList /* IN/OUT: Number of elements in aList[] */
001865 ){
001866 struct Sublist {
001867 int nList; /* Number of elements in aList */
001868 ht_slot *aList; /* Pointer to sub-list content */
001869 };
001870
001871 const int nList = *pnList; /* Size of input list */
001872 int nMerge = 0; /* Number of elements in list aMerge */
001873 ht_slot *aMerge = 0; /* List to be merged */
001874 int iList; /* Index into input list */
001875 u32 iSub = 0; /* Index into aSub array */
001876 struct Sublist aSub[13]; /* Array of sub-lists */
001877
001878 memset(aSub, 0, sizeof(aSub));
001879 assert( nList<=HASHTABLE_NPAGE && nList>0 );
001880 assert( HASHTABLE_NPAGE==(1<<(ArraySize(aSub)-1)) );
001881
001882 for(iList=0; iList<nList; iList++){
001883 nMerge = 1;
001884 aMerge = &aList[iList];
001885 for(iSub=0; iList & (1<<iSub); iSub++){
001886 struct Sublist *p;
001887 assert( iSub<ArraySize(aSub) );
001888 p = &aSub[iSub];
001889 assert( p->aList && p->nList<=(1<<iSub) );
001890 assert( p->aList==&aList[iList&~((2<<iSub)-1)] );
001891 walMerge(aContent, p->aList, p->nList, &aMerge, &nMerge, aBuffer);
001892 }
001893 aSub[iSub].aList = aMerge;
001894 aSub[iSub].nList = nMerge;
001895 }
001896
001897 for(iSub++; iSub<ArraySize(aSub); iSub++){
001898 if( nList & (1<<iSub) ){
001899 struct Sublist *p;
001900 assert( iSub<ArraySize(aSub) );
001901 p = &aSub[iSub];
001902 assert( p->nList<=(1<<iSub) );
001903 assert( p->aList==&aList[nList&~((2<<iSub)-1)] );
001904 walMerge(aContent, p->aList, p->nList, &aMerge, &nMerge, aBuffer);
001905 }
001906 }
001907 assert( aMerge==aList );
001908 *pnList = nMerge;
001909
001910 #ifdef SQLITE_DEBUG
001911 {
001912 int i;
001913 for(i=1; i<*pnList; i++){
001914 assert( aContent[aList[i]] > aContent[aList[i-1]] );
001915 }
001916 }
001917 #endif
001918 }
001919
001920 /*
001921 ** Free an iterator allocated by walIteratorInit().
001922 */
001923 static void walIteratorFree(WalIterator *p){
001924 sqlite3_free(p);
001925 }
001926
001927 /*
001928 ** Construct a WalInterator object that can be used to loop over all
001929 ** pages in the WAL following frame nBackfill in ascending order. Frames
001930 ** nBackfill or earlier may be included - excluding them is an optimization
001931 ** only. The caller must hold the checkpoint lock.
001932 **
001933 ** On success, make *pp point to the newly allocated WalInterator object
001934 ** return SQLITE_OK. Otherwise, return an error code. If this routine
001935 ** returns an error, the value of *pp is undefined.
001936 **
001937 ** The calling routine should invoke walIteratorFree() to destroy the
001938 ** WalIterator object when it has finished with it.
001939 */
001940 static int walIteratorInit(Wal *pWal, u32 nBackfill, WalIterator **pp){
001941 WalIterator *p; /* Return value */
001942 int nSegment; /* Number of segments to merge */
001943 u32 iLast; /* Last frame in log */
001944 sqlite3_int64 nByte; /* Number of bytes to allocate */
001945 int i; /* Iterator variable */
001946 ht_slot *aTmp; /* Temp space used by merge-sort */
001947 int rc = SQLITE_OK; /* Return Code */
001948
001949 /* This routine only runs while holding the checkpoint lock. And
001950 ** it only runs if there is actually content in the log (mxFrame>0).
001951 */
001952 assert( pWal->ckptLock && pWal->hdr.mxFrame>0 );
001953 iLast = pWal->hdr.mxFrame;
001954
001955 /* Allocate space for the WalIterator object. */
001956 nSegment = walFramePage(iLast) + 1;
001957 nByte = sizeof(WalIterator)
001958 + (nSegment-1)*sizeof(struct WalSegment)
001959 + iLast*sizeof(ht_slot);
001960 p = (WalIterator *)sqlite3_malloc64(nByte
001961 + sizeof(ht_slot) * (iLast>HASHTABLE_NPAGE?HASHTABLE_NPAGE:iLast)
001962 );
001963 if( !p ){
001964 return SQLITE_NOMEM_BKPT;
001965 }
001966 memset(p, 0, nByte);
001967 p->nSegment = nSegment;
001968 aTmp = (ht_slot*)&(((u8*)p)[nByte]);
001969 SEH_FREE_ON_ERROR(0, p);
001970 for(i=walFramePage(nBackfill+1); rc==SQLITE_OK && i<nSegment; i++){
001971 WalHashLoc sLoc;
001972
001973 rc = walHashGet(pWal, i, &sLoc);
001974 if( rc==SQLITE_OK ){
001975 int j; /* Counter variable */
001976 int nEntry; /* Number of entries in this segment */
001977 ht_slot *aIndex; /* Sorted index for this segment */
001978
001979 if( (i+1)==nSegment ){
001980 nEntry = (int)(iLast - sLoc.iZero);
001981 }else{
001982 nEntry = (int)((u32*)sLoc.aHash - (u32*)sLoc.aPgno);
001983 }
001984 aIndex = &((ht_slot *)&p->aSegment[p->nSegment])[sLoc.iZero];
001985 sLoc.iZero++;
001986
001987 for(j=0; j<nEntry; j++){
001988 aIndex[j] = (ht_slot)j;
001989 }
001990 walMergesort((u32 *)sLoc.aPgno, aTmp, aIndex, &nEntry);
001991 p->aSegment[i].iZero = sLoc.iZero;
001992 p->aSegment[i].nEntry = nEntry;
001993 p->aSegment[i].aIndex = aIndex;
001994 p->aSegment[i].aPgno = (u32 *)sLoc.aPgno;
001995 }
001996 }
001997 if( rc!=SQLITE_OK ){
001998 SEH_FREE_ON_ERROR(p, 0);
001999 walIteratorFree(p);
002000 p = 0;
002001 }
002002 *pp = p;
002003 return rc;
002004 }
002005
002006 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
002007
002008
002009 /*
002010 ** Attempt to enable blocking locks that block for nMs ms. Return 1 if
002011 ** blocking locks are successfully enabled, or 0 otherwise.
002012 */
002013 static int walEnableBlockingMs(Wal *pWal, int nMs){
002014 int rc = sqlite3OsFileControl(
002015 pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&nMs
002016 );
002017 return (rc==SQLITE_OK);
002018 }
002019
002020 /*
002021 ** Attempt to enable blocking locks. Blocking locks are enabled only if (a)
002022 ** they are supported by the VFS, and (b) the database handle is configured
002023 ** with a busy-timeout. Return 1 if blocking locks are successfully enabled,
002024 ** or 0 otherwise.
002025 */
002026 static int walEnableBlocking(Wal *pWal){
002027 int res = 0;
002028 if( pWal->db ){
002029 int tmout = pWal->db->busyTimeout;
002030 if( tmout ){
002031 res = walEnableBlockingMs(pWal, tmout);
002032 }
002033 }
002034 return res;
002035 }
002036
002037 /*
002038 ** Disable blocking locks.
002039 */
002040 static void walDisableBlocking(Wal *pWal){
002041 int tmout = 0;
002042 sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout);
002043 }
002044
002045 /*
002046 ** If parameter bLock is true, attempt to enable blocking locks, take
002047 ** the WRITER lock, and then disable blocking locks. If blocking locks
002048 ** cannot be enabled, no attempt to obtain the WRITER lock is made. Return
002049 ** an SQLite error code if an error occurs, or SQLITE_OK otherwise. It is not
002050 ** an error if blocking locks can not be enabled.
002051 **
002052 ** If the bLock parameter is false and the WRITER lock is held, release it.
002053 */
002054 int sqlite3WalWriteLock(Wal *pWal, int bLock){
002055 int rc = SQLITE_OK;
002056 assert( pWal->readLock<0 || bLock==0 );
002057 if( bLock ){
002058 assert( pWal->db );
002059 if( walEnableBlocking(pWal) ){
002060 rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1);
002061 if( rc==SQLITE_OK ){
002062 pWal->writeLock = 1;
002063 }
002064 walDisableBlocking(pWal);
002065 }
002066 }else if( pWal->writeLock ){
002067 walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1);
002068 pWal->writeLock = 0;
002069 }
002070 return rc;
002071 }
002072
002073 /*
002074 ** Set the database handle used to determine if blocking locks are required.
002075 */
002076 void sqlite3WalDb(Wal *pWal, sqlite3 *db){
002077 pWal->db = db;
002078 }
002079
002080 #else
002081 # define walEnableBlocking(x) 0
002082 # define walDisableBlocking(x)
002083 # define walEnableBlockingMs(pWal, ms) 0
002084 # define sqlite3WalDb(pWal, db)
002085 #endif /* ifdef SQLITE_ENABLE_SETLK_TIMEOUT */
002086
002087
002088 /*
002089 ** Attempt to obtain the exclusive WAL lock defined by parameters lockIdx and
002090 ** n. If the attempt fails and parameter xBusy is not NULL, then it is a
002091 ** busy-handler function. Invoke it and retry the lock until either the
002092 ** lock is successfully obtained or the busy-handler returns 0.
002093 */
002094 static int walBusyLock(
002095 Wal *pWal, /* WAL connection */
002096 int (*xBusy)(void*), /* Function to call when busy */
002097 void *pBusyArg, /* Context argument for xBusyHandler */
002098 int lockIdx, /* Offset of first byte to lock */
002099 int n /* Number of bytes to lock */
002100 ){
002101 int rc;
002102 do {
002103 rc = walLockExclusive(pWal, lockIdx, n);
002104 }while( xBusy && rc==SQLITE_BUSY && xBusy(pBusyArg) );
002105 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
002106 if( rc==SQLITE_BUSY_TIMEOUT ){
002107 walDisableBlocking(pWal);
002108 rc = SQLITE_BUSY;
002109 }
002110 #endif
002111 return rc;
002112 }
002113
002114 /*
002115 ** The cache of the wal-index header must be valid to call this function.
002116 ** Return the page-size in bytes used by the database.
002117 */
002118 static int walPagesize(Wal *pWal){
002119 return (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16);
002120 }
002121
002122 /*
002123 ** The following is guaranteed when this function is called:
002124 **
002125 ** a) the WRITER lock is held,
002126 ** b) the entire log file has been checkpointed, and
002127 ** c) any existing readers are reading exclusively from the database
002128 ** file - there are no readers that may attempt to read a frame from
002129 ** the log file.
002130 **
002131 ** This function updates the shared-memory structures so that the next
002132 ** client to write to the database (which may be this one) does so by
002133 ** writing frames into the start of the log file.
002134 **
002135 ** The value of parameter salt1 is used as the aSalt[1] value in the
002136 ** new wal-index header. It should be passed a pseudo-random value (i.e.
002137 ** one obtained from sqlite3_randomness()).
002138 */
002139 static void walRestartHdr(Wal *pWal, u32 salt1){
002140 volatile WalCkptInfo *pInfo = walCkptInfo(pWal);
002141 int i; /* Loop counter */
002142 u32 *aSalt = pWal->hdr.aSalt; /* Big-endian salt values */
002143 pWal->nCkpt++;
002144 pWal->hdr.mxFrame = 0;
002145 sqlite3Put4byte((u8*)&aSalt[0], 1 + sqlite3Get4byte((u8*)&aSalt[0]));
002146 memcpy(&pWal->hdr.aSalt[1], &salt1, 4);
002147 walIndexWriteHdr(pWal);
002148 AtomicStore(&pInfo->nBackfill, 0);
002149 pInfo->nBackfillAttempted = 0;
002150 pInfo->aReadMark[1] = 0;
002151 for(i=2; i<WAL_NREADER; i++) pInfo->aReadMark[i] = READMARK_NOT_USED;
002152 assert( pInfo->aReadMark[0]==0 );
002153 }
002154
002155 /*
002156 ** Copy as much content as we can from the WAL back into the database file
002157 ** in response to an sqlite3_wal_checkpoint() request or the equivalent.
002158 **
002159 ** The amount of information copies from WAL to database might be limited
002160 ** by active readers. This routine will never overwrite a database page
002161 ** that a concurrent reader might be using.
002162 **
002163 ** All I/O barrier operations (a.k.a fsyncs) occur in this routine when
002164 ** SQLite is in WAL-mode in synchronous=NORMAL. That means that if
002165 ** checkpoints are always run by a background thread or background
002166 ** process, foreground threads will never block on a lengthy fsync call.
002167 **
002168 ** Fsync is called on the WAL before writing content out of the WAL and
002169 ** into the database. This ensures that if the new content is persistent
002170 ** in the WAL and can be recovered following a power-loss or hard reset.
002171 **
002172 ** Fsync is also called on the database file if (and only if) the entire
002173 ** WAL content is copied into the database file. This second fsync makes
002174 ** it safe to delete the WAL since the new content will persist in the
002175 ** database file.
002176 **
002177 ** This routine uses and updates the nBackfill field of the wal-index header.
002178 ** This is the only routine that will increase the value of nBackfill.
002179 ** (A WAL reset or recovery will revert nBackfill to zero, but not increase
002180 ** its value.)
002181 **
002182 ** The caller must be holding sufficient locks to ensure that no other
002183 ** checkpoint is running (in any other thread or process) at the same
002184 ** time.
002185 */
002186 static int walCheckpoint(
002187 Wal *pWal, /* Wal connection */
002188 sqlite3 *db, /* Check for interrupts on this handle */
002189 int eMode, /* One of PASSIVE, FULL or RESTART */
002190 int (*xBusy)(void*), /* Function to call when busy */
002191 void *pBusyArg, /* Context argument for xBusyHandler */
002192 int sync_flags, /* Flags for OsSync() (or 0) */
002193 u8 *zBuf /* Temporary buffer to use */
002194 ){
002195 int rc = SQLITE_OK; /* Return code */
002196 int szPage; /* Database page-size */
002197 WalIterator *pIter = 0; /* Wal iterator context */
002198 u32 iDbpage = 0; /* Next database page to write */
002199 u32 iFrame = 0; /* Wal frame containing data for iDbpage */
002200 u32 mxSafeFrame; /* Max frame that can be backfilled */
002201 u32 mxPage; /* Max database page to write */
002202 int i; /* Loop counter */
002203 volatile WalCkptInfo *pInfo; /* The checkpoint status information */
002204
002205 szPage = walPagesize(pWal);
002206 testcase( szPage<=32768 );
002207 testcase( szPage>=65536 );
002208 pInfo = walCkptInfo(pWal);
002209 if( pInfo->nBackfill<pWal->hdr.mxFrame ){
002210
002211 /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked
002212 ** in the SQLITE_CHECKPOINT_PASSIVE mode. */
002213 assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 );
002214
002215 /* Compute in mxSafeFrame the index of the last frame of the WAL that is
002216 ** safe to write into the database. Frames beyond mxSafeFrame might
002217 ** overwrite database pages that are in use by active readers and thus
002218 ** cannot be backfilled from the WAL.
002219 */
002220 mxSafeFrame = pWal->hdr.mxFrame;
002221 mxPage = pWal->hdr.nPage;
002222 for(i=1; i<WAL_NREADER; i++){
002223 u32 y = AtomicLoad(pInfo->aReadMark+i); SEH_INJECT_FAULT;
002224 if( mxSafeFrame>y ){
002225 assert( y<=pWal->hdr.mxFrame );
002226 rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(i), 1);
002227 if( rc==SQLITE_OK ){
002228 u32 iMark = (i==1 ? mxSafeFrame : READMARK_NOT_USED);
002229 AtomicStore(pInfo->aReadMark+i, iMark); SEH_INJECT_FAULT;
002230 walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1);
002231 }else if( rc==SQLITE_BUSY ){
002232 mxSafeFrame = y;
002233 xBusy = 0;
002234 }else{
002235 goto walcheckpoint_out;
002236 }
002237 }
002238 }
002239
002240 /* Allocate the iterator */
002241 if( pInfo->nBackfill<mxSafeFrame ){
002242 rc = walIteratorInit(pWal, pInfo->nBackfill, &pIter);
002243 assert( rc==SQLITE_OK || pIter==0 );
002244 }
002245
002246 if( pIter
002247 && (rc = walBusyLock(pWal,xBusy,pBusyArg,WAL_READ_LOCK(0),1))==SQLITE_OK
002248 ){
002249 u32 nBackfill = pInfo->nBackfill;
002250 pInfo->nBackfillAttempted = mxSafeFrame; SEH_INJECT_FAULT;
002251
002252 /* Sync the WAL to disk */
002253 rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags));
002254
002255 /* If the database may grow as a result of this checkpoint, hint
002256 ** about the eventual size of the db file to the VFS layer.
002257 */
002258 if( rc==SQLITE_OK ){
002259 i64 nReq = ((i64)mxPage * szPage);
002260 i64 nSize; /* Current size of database file */
002261 sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_CKPT_START, 0);
002262 rc = sqlite3OsFileSize(pWal->pDbFd, &nSize);
002263 if( rc==SQLITE_OK && nSize<nReq ){
002264 if( (nSize+65536+(i64)pWal->hdr.mxFrame*szPage)<nReq ){
002265 /* If the size of the final database is larger than the current
002266 ** database plus the amount of data in the wal file, plus the
002267 ** maximum size of the pending-byte page (65536 bytes), then
002268 ** must be corruption somewhere. */
002269 rc = SQLITE_CORRUPT_BKPT;
002270 }else{
002271 sqlite3OsFileControlHint(pWal->pDbFd, SQLITE_FCNTL_SIZE_HINT,&nReq);
002272 }
002273 }
002274
002275 }
002276
002277 /* Iterate through the contents of the WAL, copying data to the db file */
002278 while( rc==SQLITE_OK && 0==walIteratorNext(pIter, &iDbpage, &iFrame) ){
002279 i64 iOffset;
002280 assert( walFramePgno(pWal, iFrame)==iDbpage );
002281 SEH_INJECT_FAULT;
002282 if( AtomicLoad(&db->u1.isInterrupted) ){
002283 rc = db->mallocFailed ? SQLITE_NOMEM_BKPT : SQLITE_INTERRUPT;
002284 break;
002285 }
002286 if( iFrame<=nBackfill || iFrame>mxSafeFrame || iDbpage>mxPage ){
002287 continue;
002288 }
002289 iOffset = walFrameOffset(iFrame, szPage) + WAL_FRAME_HDRSIZE;
002290 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL file */
002291 rc = sqlite3OsRead(pWal->pWalFd, zBuf, szPage, iOffset);
002292 if( rc!=SQLITE_OK ) break;
002293 iOffset = (iDbpage-1)*(i64)szPage;
002294 testcase( IS_BIG_INT(iOffset) );
002295 rc = sqlite3OsWrite(pWal->pDbFd, zBuf, szPage, iOffset);
002296 if( rc!=SQLITE_OK ) break;
002297 }
002298 sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_CKPT_DONE, 0);
002299
002300 /* If work was actually accomplished... */
002301 if( rc==SQLITE_OK ){
002302 if( mxSafeFrame==walIndexHdr(pWal)->mxFrame ){
002303 i64 szDb = pWal->hdr.nPage*(i64)szPage;
002304 testcase( IS_BIG_INT(szDb) );
002305 rc = sqlite3OsTruncate(pWal->pDbFd, szDb);
002306 if( rc==SQLITE_OK ){
002307 rc = sqlite3OsSync(pWal->pDbFd, CKPT_SYNC_FLAGS(sync_flags));
002308 }
002309 }
002310 if( rc==SQLITE_OK ){
002311 AtomicStore(&pInfo->nBackfill, mxSafeFrame); SEH_INJECT_FAULT;
002312 }
002313 }
002314
002315 /* Release the reader lock held while backfilling */
002316 walUnlockExclusive(pWal, WAL_READ_LOCK(0), 1);
002317 }
002318
002319 if( rc==SQLITE_BUSY ){
002320 /* Reset the return code so as not to report a checkpoint failure
002321 ** just because there are active readers. */
002322 rc = SQLITE_OK;
002323 }
002324 }
002325
002326 /* If this is an SQLITE_CHECKPOINT_RESTART or TRUNCATE operation, and the
002327 ** entire wal file has been copied into the database file, then block
002328 ** until all readers have finished using the wal file. This ensures that
002329 ** the next process to write to the database restarts the wal file.
002330 */
002331 if( rc==SQLITE_OK && eMode!=SQLITE_CHECKPOINT_PASSIVE ){
002332 assert( pWal->writeLock );
002333 SEH_INJECT_FAULT;
002334 if( pInfo->nBackfill<pWal->hdr.mxFrame ){
002335 rc = SQLITE_BUSY;
002336 }else if( eMode>=SQLITE_CHECKPOINT_RESTART ){
002337 u32 salt1;
002338 sqlite3_randomness(4, &salt1);
002339 assert( pInfo->nBackfill==pWal->hdr.mxFrame );
002340 rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(1), WAL_NREADER-1);
002341 if( rc==SQLITE_OK ){
002342 if( eMode==SQLITE_CHECKPOINT_TRUNCATE ){
002343 /* IMPLEMENTATION-OF: R-44699-57140 This mode works the same way as
002344 ** SQLITE_CHECKPOINT_RESTART with the addition that it also
002345 ** truncates the log file to zero bytes just prior to a
002346 ** successful return.
002347 **
002348 ** In theory, it might be safe to do this without updating the
002349 ** wal-index header in shared memory, as all subsequent reader or
002350 ** writer clients should see that the entire log file has been
002351 ** checkpointed and behave accordingly. This seems unsafe though,
002352 ** as it would leave the system in a state where the contents of
002353 ** the wal-index header do not match the contents of the
002354 ** file-system. To avoid this, update the wal-index header to
002355 ** indicate that the log file contains zero valid frames. */
002356 walRestartHdr(pWal, salt1);
002357 rc = sqlite3OsTruncate(pWal->pWalFd, 0);
002358 }
002359 walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1);
002360 }
002361 }
002362 }
002363
002364 walcheckpoint_out:
002365 SEH_FREE_ON_ERROR(pIter, 0);
002366 walIteratorFree(pIter);
002367 return rc;
002368 }
002369
002370 /*
002371 ** If the WAL file is currently larger than nMax bytes in size, truncate
002372 ** it to exactly nMax bytes. If an error occurs while doing so, ignore it.
002373 */
002374 static void walLimitSize(Wal *pWal, i64 nMax){
002375 i64 sz;
002376 int rx;
002377 sqlite3BeginBenignMalloc();
002378 rx = sqlite3OsFileSize(pWal->pWalFd, &sz);
002379 if( rx==SQLITE_OK && (sz > nMax ) ){
002380 rx = sqlite3OsTruncate(pWal->pWalFd, nMax);
002381 }
002382 sqlite3EndBenignMalloc();
002383 if( rx ){
002384 sqlite3_log(rx, "cannot limit WAL size: %s", pWal->zWalName);
002385 }
002386 }
002387
002388 #ifdef SQLITE_USE_SEH
002389 /*
002390 ** This is the "standard" exception handler used in a few places to handle
002391 ** an exception thrown by reading from the *-shm mapping after it has become
002392 ** invalid in SQLITE_USE_SEH builds. It is used as follows:
002393 **
002394 ** SEH_TRY { ... }
002395 ** SEH_EXCEPT( rc = walHandleException(pWal); )
002396 **
002397 ** This function does three things:
002398 **
002399 ** 1) Determines the locks that should be held, based on the contents of
002400 ** the Wal.readLock, Wal.writeLock and Wal.ckptLock variables. All other
002401 ** held locks are assumed to be transient locks that would have been
002402 ** released had the exception not been thrown and are dropped.
002403 **
002404 ** 2) Frees the pointer at Wal.pFree, if any, using sqlite3_free().
002405 **
002406 ** 3) Set pWal->apWiData[pWal->iWiPg] to pWal->pWiValue if not NULL
002407 **
002408 ** 4) Returns SQLITE_IOERR.
002409 */
002410 static int walHandleException(Wal *pWal){
002411 if( pWal->exclusiveMode==0 ){
002412 static const int S = 1;
002413 static const int E = (1<<SQLITE_SHM_NLOCK);
002414 int ii;
002415 u32 mUnlock = pWal->lockMask & ~(
002416 (pWal->readLock<0 ? 0 : (S << WAL_READ_LOCK(pWal->readLock)))
002417 | (pWal->writeLock ? (E << WAL_WRITE_LOCK) : 0)
002418 | (pWal->ckptLock ? (E << WAL_CKPT_LOCK) : 0)
002419 );
002420 for(ii=0; ii<SQLITE_SHM_NLOCK; ii++){
002421 if( (S<<ii) & mUnlock ) walUnlockShared(pWal, ii);
002422 if( (E<<ii) & mUnlock ) walUnlockExclusive(pWal, ii, 1);
002423 }
002424 }
002425 sqlite3_free(pWal->pFree);
002426 pWal->pFree = 0;
002427 if( pWal->pWiValue ){
002428 pWal->apWiData[pWal->iWiPg] = pWal->pWiValue;
002429 pWal->pWiValue = 0;
002430 }
002431 return SQLITE_IOERR_IN_PAGE;
002432 }
002433
002434 /*
002435 ** Assert that the Wal.lockMask mask, which indicates the locks held
002436 ** by the connenction, is consistent with the Wal.readLock, Wal.writeLock
002437 ** and Wal.ckptLock variables. To be used as:
002438 **
002439 ** assert( walAssertLockmask(pWal) );
002440 */
002441 static int walAssertLockmask(Wal *pWal){
002442 if( pWal->exclusiveMode==0 ){
002443 static const int S = 1;
002444 static const int E = (1<<SQLITE_SHM_NLOCK);
002445 u32 mExpect = (
002446 (pWal->readLock<0 ? 0 : (S << WAL_READ_LOCK(pWal->readLock)))
002447 | (pWal->writeLock ? (E << WAL_WRITE_LOCK) : 0)
002448 | (pWal->ckptLock ? (E << WAL_CKPT_LOCK) : 0)
002449 #ifdef SQLITE_ENABLE_SNAPSHOT
002450 | (pWal->pSnapshot ? (pWal->lockMask & (1 << WAL_CKPT_LOCK)) : 0)
002451 #endif
002452 );
002453 assert( mExpect==pWal->lockMask );
002454 }
002455 return 1;
002456 }
002457
002458 /*
002459 ** Return and zero the "system error" field set when an
002460 ** EXCEPTION_IN_PAGE_ERROR exception is caught.
002461 */
002462 int sqlite3WalSystemErrno(Wal *pWal){
002463 int iRet = 0;
002464 if( pWal ){
002465 iRet = pWal->iSysErrno;
002466 pWal->iSysErrno = 0;
002467 }
002468 return iRet;
002469 }
002470
002471 #else
002472 # define walAssertLockmask(x) 1
002473 #endif /* ifdef SQLITE_USE_SEH */
002474
002475 /*
002476 ** Close a connection to a log file.
002477 */
002478 int sqlite3WalClose(
002479 Wal *pWal, /* Wal to close */
002480 sqlite3 *db, /* For interrupt flag */
002481 int sync_flags, /* Flags to pass to OsSync() (or 0) */
002482 int nBuf,
002483 u8 *zBuf /* Buffer of at least nBuf bytes */
002484 ){
002485 int rc = SQLITE_OK;
002486 if( pWal ){
002487 int isDelete = 0; /* True to unlink wal and wal-index files */
002488
002489 assert( walAssertLockmask(pWal) );
002490
002491 /* If an EXCLUSIVE lock can be obtained on the database file (using the
002492 ** ordinary, rollback-mode locking methods, this guarantees that the
002493 ** connection associated with this log file is the only connection to
002494 ** the database. In this case checkpoint the database and unlink both
002495 ** the wal and wal-index files.
002496 **
002497 ** The EXCLUSIVE lock is not released before returning.
002498 */
002499 if( zBuf!=0
002500 && SQLITE_OK==(rc = sqlite3OsLock(pWal->pDbFd, SQLITE_LOCK_EXCLUSIVE))
002501 ){
002502 if( pWal->exclusiveMode==WAL_NORMAL_MODE ){
002503 pWal->exclusiveMode = WAL_EXCLUSIVE_MODE;
002504 }
002505 rc = sqlite3WalCheckpoint(pWal, db,
002506 SQLITE_CHECKPOINT_PASSIVE, 0, 0, sync_flags, nBuf, zBuf, 0, 0
002507 );
002508 if( rc==SQLITE_OK ){
002509 int bPersist = -1;
002510 sqlite3OsFileControlHint(
002511 pWal->pDbFd, SQLITE_FCNTL_PERSIST_WAL, &bPersist
002512 );
002513 if( bPersist!=1 ){
002514 /* Try to delete the WAL file if the checkpoint completed and
002515 ** fsynced (rc==SQLITE_OK) and if we are not in persistent-wal
002516 ** mode (!bPersist) */
002517 isDelete = 1;
002518 }else if( pWal->mxWalSize>=0 ){
002519 /* Try to truncate the WAL file to zero bytes if the checkpoint
002520 ** completed and fsynced (rc==SQLITE_OK) and we are in persistent
002521 ** WAL mode (bPersist) and if the PRAGMA journal_size_limit is a
002522 ** non-negative value (pWal->mxWalSize>=0). Note that we truncate
002523 ** to zero bytes as truncating to the journal_size_limit might
002524 ** leave a corrupt WAL file on disk. */
002525 walLimitSize(pWal, 0);
002526 }
002527 }
002528 }
002529
002530 walIndexClose(pWal, isDelete);
002531 sqlite3OsClose(pWal->pWalFd);
002532 if( isDelete ){
002533 sqlite3BeginBenignMalloc();
002534 sqlite3OsDelete(pWal->pVfs, pWal->zWalName, 0);
002535 sqlite3EndBenignMalloc();
002536 }
002537 WALTRACE(("WAL%p: closed\n", pWal));
002538 sqlite3_free((void *)pWal->apWiData);
002539 sqlite3_free(pWal);
002540 }
002541 return rc;
002542 }
002543
002544 /*
002545 ** Try to read the wal-index header. Return 0 on success and 1 if
002546 ** there is a problem.
002547 **
002548 ** The wal-index is in shared memory. Another thread or process might
002549 ** be writing the header at the same time this procedure is trying to
002550 ** read it, which might result in inconsistency. A dirty read is detected
002551 ** by verifying that both copies of the header are the same and also by
002552 ** a checksum on the header.
002553 **
002554 ** If and only if the read is consistent and the header is different from
002555 ** pWal->hdr, then pWal->hdr is updated to the content of the new header
002556 ** and *pChanged is set to 1.
002557 **
002558 ** If the checksum cannot be verified return non-zero. If the header
002559 ** is read successfully and the checksum verified, return zero.
002560 */
002561 static SQLITE_NO_TSAN int walIndexTryHdr(Wal *pWal, int *pChanged){
002562 u32 aCksum[2]; /* Checksum on the header content */
002563 WalIndexHdr h1, h2; /* Two copies of the header content */
002564 WalIndexHdr volatile *aHdr; /* Header in shared memory */
002565
002566 /* The first page of the wal-index must be mapped at this point. */
002567 assert( pWal->nWiData>0 && pWal->apWiData[0] );
002568
002569 /* Read the header. This might happen concurrently with a write to the
002570 ** same area of shared memory on a different CPU in a SMP,
002571 ** meaning it is possible that an inconsistent snapshot is read
002572 ** from the file. If this happens, return non-zero.
002573 **
002574 ** tag-20200519-1:
002575 ** There are two copies of the header at the beginning of the wal-index.
002576 ** When reading, read [0] first then [1]. Writes are in the reverse order.
002577 ** Memory barriers are used to prevent the compiler or the hardware from
002578 ** reordering the reads and writes. TSAN and similar tools can sometimes
002579 ** give false-positive warnings about these accesses because the tools do not
002580 ** account for the double-read and the memory barrier. The use of mutexes
002581 ** here would be problematic as the memory being accessed is potentially
002582 ** shared among multiple processes and not all mutex implementations work
002583 ** reliably in that environment.
002584 */
002585 aHdr = walIndexHdr(pWal);
002586 memcpy(&h1, (void *)&aHdr[0], sizeof(h1)); /* Possible TSAN false-positive */
002587 walShmBarrier(pWal);
002588 memcpy(&h2, (void *)&aHdr[1], sizeof(h2));
002589
002590 if( memcmp(&h1, &h2, sizeof(h1))!=0 ){
002591 return 1; /* Dirty read */
002592 }
002593 if( h1.isInit==0 ){
002594 return 1; /* Malformed header - probably all zeros */
002595 }
002596 walChecksumBytes(1, (u8*)&h1, sizeof(h1)-sizeof(h1.aCksum), 0, aCksum);
002597 if( aCksum[0]!=h1.aCksum[0] || aCksum[1]!=h1.aCksum[1] ){
002598 return 1; /* Checksum does not match */
002599 }
002600
002601 if( memcmp(&pWal->hdr, &h1, sizeof(WalIndexHdr)) ){
002602 *pChanged = 1;
002603 memcpy(&pWal->hdr, &h1, sizeof(WalIndexHdr));
002604 pWal->szPage = (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16);
002605 testcase( pWal->szPage<=32768 );
002606 testcase( pWal->szPage>=65536 );
002607 }
002608
002609 /* The header was successfully read. Return zero. */
002610 return 0;
002611 }
002612
002613 /*
002614 ** This is the value that walTryBeginRead returns when it needs to
002615 ** be retried.
002616 */
002617 #define WAL_RETRY (-1)
002618
002619 /*
002620 ** Read the wal-index header from the wal-index and into pWal->hdr.
002621 ** If the wal-header appears to be corrupt, try to reconstruct the
002622 ** wal-index from the WAL before returning.
002623 **
002624 ** Set *pChanged to 1 if the wal-index header value in pWal->hdr is
002625 ** changed by this operation. If pWal->hdr is unchanged, set *pChanged
002626 ** to 0.
002627 **
002628 ** If the wal-index header is successfully read, return SQLITE_OK.
002629 ** Otherwise an SQLite error code.
002630 */
002631 static int walIndexReadHdr(Wal *pWal, int *pChanged){
002632 int rc; /* Return code */
002633 int badHdr; /* True if a header read failed */
002634 volatile u32 *page0; /* Chunk of wal-index containing header */
002635
002636 /* Ensure that page 0 of the wal-index (the page that contains the
002637 ** wal-index header) is mapped. Return early if an error occurs here.
002638 */
002639 assert( pChanged );
002640 rc = walIndexPage(pWal, 0, &page0);
002641 if( rc!=SQLITE_OK ){
002642 assert( rc!=SQLITE_READONLY ); /* READONLY changed to OK in walIndexPage */
002643 if( rc==SQLITE_READONLY_CANTINIT ){
002644 /* The SQLITE_READONLY_CANTINIT return means that the shared-memory
002645 ** was openable but is not writable, and this thread is unable to
002646 ** confirm that another write-capable connection has the shared-memory
002647 ** open, and hence the content of the shared-memory is unreliable,
002648 ** since the shared-memory might be inconsistent with the WAL file
002649 ** and there is no writer on hand to fix it. */
002650 assert( page0==0 );
002651 assert( pWal->writeLock==0 );
002652 assert( pWal->readOnly & WAL_SHM_RDONLY );
002653 pWal->bShmUnreliable = 1;
002654 pWal->exclusiveMode = WAL_HEAPMEMORY_MODE;
002655 *pChanged = 1;
002656 }else{
002657 return rc; /* Any other non-OK return is just an error */
002658 }
002659 }else{
002660 /* page0 can be NULL if the SHM is zero bytes in size and pWal->writeLock
002661 ** is zero, which prevents the SHM from growing */
002662 testcase( page0!=0 );
002663 }
002664 assert( page0!=0 || pWal->writeLock==0 );
002665
002666 /* If the first page of the wal-index has been mapped, try to read the
002667 ** wal-index header immediately, without holding any lock. This usually
002668 ** works, but may fail if the wal-index header is corrupt or currently
002669 ** being modified by another thread or process.
002670 */
002671 badHdr = (page0 ? walIndexTryHdr(pWal, pChanged) : 1);
002672
002673 /* If the first attempt failed, it might have been due to a race
002674 ** with a writer. So get a WRITE lock and try again.
002675 */
002676 if( badHdr ){
002677 if( pWal->bShmUnreliable==0 && (pWal->readOnly & WAL_SHM_RDONLY) ){
002678 if( SQLITE_OK==(rc = walLockShared(pWal, WAL_WRITE_LOCK)) ){
002679 walUnlockShared(pWal, WAL_WRITE_LOCK);
002680 rc = SQLITE_READONLY_RECOVERY;
002681 }
002682 }else{
002683 int bWriteLock = pWal->writeLock;
002684 if( bWriteLock
002685 || SQLITE_OK==(rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1))
002686 ){
002687 pWal->writeLock = 1;
002688 if( SQLITE_OK==(rc = walIndexPage(pWal, 0, &page0)) ){
002689 badHdr = walIndexTryHdr(pWal, pChanged);
002690 if( badHdr ){
002691 /* If the wal-index header is still malformed even while holding
002692 ** a WRITE lock, it can only mean that the header is corrupted and
002693 ** needs to be reconstructed. So run recovery to do exactly that.
002694 ** Disable blocking locks first. */
002695 walDisableBlocking(pWal);
002696 rc = walIndexRecover(pWal);
002697 *pChanged = 1;
002698 }
002699 }
002700 if( bWriteLock==0 ){
002701 pWal->writeLock = 0;
002702 walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1);
002703 }
002704 }
002705 }
002706 }
002707
002708 /* If the header is read successfully, check the version number to make
002709 ** sure the wal-index was not constructed with some future format that
002710 ** this version of SQLite cannot understand.
002711 */
002712 if( badHdr==0 && pWal->hdr.iVersion!=WALINDEX_MAX_VERSION ){
002713 rc = SQLITE_CANTOPEN_BKPT;
002714 }
002715 if( pWal->bShmUnreliable ){
002716 if( rc!=SQLITE_OK ){
002717 walIndexClose(pWal, 0);
002718 pWal->bShmUnreliable = 0;
002719 assert( pWal->nWiData>0 && pWal->apWiData[0]==0 );
002720 /* walIndexRecover() might have returned SHORT_READ if a concurrent
002721 ** writer truncated the WAL out from under it. If that happens, it
002722 ** indicates that a writer has fixed the SHM file for us, so retry */
002723 if( rc==SQLITE_IOERR_SHORT_READ ) rc = WAL_RETRY;
002724 }
002725 pWal->exclusiveMode = WAL_NORMAL_MODE;
002726 }
002727
002728 return rc;
002729 }
002730
002731 /*
002732 ** Open a transaction in a connection where the shared-memory is read-only
002733 ** and where we cannot verify that there is a separate write-capable connection
002734 ** on hand to keep the shared-memory up-to-date with the WAL file.
002735 **
002736 ** This can happen, for example, when the shared-memory is implemented by
002737 ** memory-mapping a *-shm file, where a prior writer has shut down and
002738 ** left the *-shm file on disk, and now the present connection is trying
002739 ** to use that database but lacks write permission on the *-shm file.
002740 ** Other scenarios are also possible, depending on the VFS implementation.
002741 **
002742 ** Precondition:
002743 **
002744 ** The *-wal file has been read and an appropriate wal-index has been
002745 ** constructed in pWal->apWiData[] using heap memory instead of shared
002746 ** memory.
002747 **
002748 ** If this function returns SQLITE_OK, then the read transaction has
002749 ** been successfully opened. In this case output variable (*pChanged)
002750 ** is set to true before returning if the caller should discard the
002751 ** contents of the page cache before proceeding. Or, if it returns
002752 ** WAL_RETRY, then the heap memory wal-index has been discarded and
002753 ** the caller should retry opening the read transaction from the
002754 ** beginning (including attempting to map the *-shm file).
002755 **
002756 ** If an error occurs, an SQLite error code is returned.
002757 */
002758 static int walBeginShmUnreliable(Wal *pWal, int *pChanged){
002759 i64 szWal; /* Size of wal file on disk in bytes */
002760 i64 iOffset; /* Current offset when reading wal file */
002761 u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */
002762 u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */
002763 int szFrame; /* Number of bytes in buffer aFrame[] */
002764 u8 *aData; /* Pointer to data part of aFrame buffer */
002765 volatile void *pDummy; /* Dummy argument for xShmMap */
002766 int rc; /* Return code */
002767 u32 aSaveCksum[2]; /* Saved copy of pWal->hdr.aFrameCksum */
002768
002769 assert( pWal->bShmUnreliable );
002770 assert( pWal->readOnly & WAL_SHM_RDONLY );
002771 assert( pWal->nWiData>0 && pWal->apWiData[0] );
002772
002773 /* Take WAL_READ_LOCK(0). This has the effect of preventing any
002774 ** writers from running a checkpoint, but does not stop them
002775 ** from running recovery. */
002776 rc = walLockShared(pWal, WAL_READ_LOCK(0));
002777 if( rc!=SQLITE_OK ){
002778 if( rc==SQLITE_BUSY ) rc = WAL_RETRY;
002779 goto begin_unreliable_shm_out;
002780 }
002781 pWal->readLock = 0;
002782
002783 /* Check to see if a separate writer has attached to the shared-memory area,
002784 ** thus making the shared-memory "reliable" again. Do this by invoking
002785 ** the xShmMap() routine of the VFS and looking to see if the return
002786 ** is SQLITE_READONLY instead of SQLITE_READONLY_CANTINIT.
002787 **
002788 ** If the shared-memory is now "reliable" return WAL_RETRY, which will
002789 ** cause the heap-memory WAL-index to be discarded and the actual
002790 ** shared memory to be used in its place.
002791 **
002792 ** This step is important because, even though this connection is holding
002793 ** the WAL_READ_LOCK(0) which prevents a checkpoint, a writer might
002794 ** have already checkpointed the WAL file and, while the current
002795 ** is active, wrap the WAL and start overwriting frames that this
002796 ** process wants to use.
002797 **
002798 ** Once sqlite3OsShmMap() has been called for an sqlite3_file and has
002799 ** returned any SQLITE_READONLY value, it must return only SQLITE_READONLY
002800 ** or SQLITE_READONLY_CANTINIT or some error for all subsequent invocations,
002801 ** even if some external agent does a "chmod" to make the shared-memory
002802 ** writable by us, until sqlite3OsShmUnmap() has been called.
002803 ** This is a requirement on the VFS implementation.
002804 */
002805 rc = sqlite3OsShmMap(pWal->pDbFd, 0, WALINDEX_PGSZ, 0, &pDummy);
002806 assert( rc!=SQLITE_OK ); /* SQLITE_OK not possible for read-only connection */
002807 if( rc!=SQLITE_READONLY_CANTINIT ){
002808 rc = (rc==SQLITE_READONLY ? WAL_RETRY : rc);
002809 goto begin_unreliable_shm_out;
002810 }
002811
002812 /* We reach this point only if the real shared-memory is still unreliable.
002813 ** Assume the in-memory WAL-index substitute is correct and load it
002814 ** into pWal->hdr.
002815 */
002816 memcpy(&pWal->hdr, (void*)walIndexHdr(pWal), sizeof(WalIndexHdr));
002817
002818 /* Make sure some writer hasn't come in and changed the WAL file out
002819 ** from under us, then disconnected, while we were not looking.
002820 */
002821 rc = sqlite3OsFileSize(pWal->pWalFd, &szWal);
002822 if( rc!=SQLITE_OK ){
002823 goto begin_unreliable_shm_out;
002824 }
002825 if( szWal<WAL_HDRSIZE ){
002826 /* If the wal file is too small to contain a wal-header and the
002827 ** wal-index header has mxFrame==0, then it must be safe to proceed
002828 ** reading the database file only. However, the page cache cannot
002829 ** be trusted, as a read/write connection may have connected, written
002830 ** the db, run a checkpoint, truncated the wal file and disconnected
002831 ** since this client's last read transaction. */
002832 *pChanged = 1;
002833 rc = (pWal->hdr.mxFrame==0 ? SQLITE_OK : WAL_RETRY);
002834 goto begin_unreliable_shm_out;
002835 }
002836
002837 /* Check the salt keys at the start of the wal file still match. */
002838 rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0);
002839 if( rc!=SQLITE_OK ){
002840 goto begin_unreliable_shm_out;
002841 }
002842 if( memcmp(&pWal->hdr.aSalt, &aBuf[16], 8) ){
002843 /* Some writer has wrapped the WAL file while we were not looking.
002844 ** Return WAL_RETRY which will cause the in-memory WAL-index to be
002845 ** rebuilt. */
002846 rc = WAL_RETRY;
002847 goto begin_unreliable_shm_out;
002848 }
002849
002850 /* Allocate a buffer to read frames into */
002851 assert( (pWal->szPage & (pWal->szPage-1))==0 );
002852 assert( pWal->szPage>=512 && pWal->szPage<=65536 );
002853 szFrame = pWal->szPage + WAL_FRAME_HDRSIZE;
002854 aFrame = (u8 *)sqlite3_malloc64(szFrame);
002855 if( aFrame==0 ){
002856 rc = SQLITE_NOMEM_BKPT;
002857 goto begin_unreliable_shm_out;
002858 }
002859 aData = &aFrame[WAL_FRAME_HDRSIZE];
002860
002861 /* Check to see if a complete transaction has been appended to the
002862 ** wal file since the heap-memory wal-index was created. If so, the
002863 ** heap-memory wal-index is discarded and WAL_RETRY returned to
002864 ** the caller. */
002865 aSaveCksum[0] = pWal->hdr.aFrameCksum[0];
002866 aSaveCksum[1] = pWal->hdr.aFrameCksum[1];
002867 for(iOffset=walFrameOffset(pWal->hdr.mxFrame+1, pWal->szPage);
002868 iOffset+szFrame<=szWal;
002869 iOffset+=szFrame
002870 ){
002871 u32 pgno; /* Database page number for frame */
002872 u32 nTruncate; /* dbsize field from frame header */
002873
002874 /* Read and decode the next log frame. */
002875 rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset);
002876 if( rc!=SQLITE_OK ) break;
002877 if( !walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame) ) break;
002878
002879 /* If nTruncate is non-zero, then a complete transaction has been
002880 ** appended to this wal file. Set rc to WAL_RETRY and break out of
002881 ** the loop. */
002882 if( nTruncate ){
002883 rc = WAL_RETRY;
002884 break;
002885 }
002886 }
002887 pWal->hdr.aFrameCksum[0] = aSaveCksum[0];
002888 pWal->hdr.aFrameCksum[1] = aSaveCksum[1];
002889
002890 begin_unreliable_shm_out:
002891 sqlite3_free(aFrame);
002892 if( rc!=SQLITE_OK ){
002893 int i;
002894 for(i=0; i<pWal->nWiData; i++){
002895 sqlite3_free((void*)pWal->apWiData[i]);
002896 pWal->apWiData[i] = 0;
002897 }
002898 pWal->bShmUnreliable = 0;
002899 sqlite3WalEndReadTransaction(pWal);
002900 *pChanged = 1;
002901 }
002902 return rc;
002903 }
002904
002905 /*
002906 ** The final argument passed to walTryBeginRead() is of type (int*). The
002907 ** caller should invoke walTryBeginRead as follows:
002908 **
002909 ** int cnt = 0;
002910 ** do {
002911 ** rc = walTryBeginRead(..., &cnt);
002912 ** }while( rc==WAL_RETRY );
002913 **
002914 ** The final value of "cnt" is of no use to the caller. It is used by
002915 ** the implementation of walTryBeginRead() as follows:
002916 **
002917 ** + Each time walTryBeginRead() is called, it is incremented. Once
002918 ** it reaches WAL_RETRY_PROTOCOL_LIMIT - indicating that walTryBeginRead()
002919 ** has many times been invoked and failed with WAL_RETRY - walTryBeginRead()
002920 ** returns SQLITE_PROTOCOL.
002921 **
002922 ** + If SQLITE_ENABLE_SETLK_TIMEOUT is defined and walTryBeginRead() failed
002923 ** because a blocking lock timed out (SQLITE_BUSY_TIMEOUT from the OS
002924 ** layer), the WAL_RETRY_BLOCKED_MASK bit is set in "cnt". In this case
002925 ** the next invocation of walTryBeginRead() may omit an expected call to
002926 ** sqlite3OsSleep(). There has already been a delay when the previous call
002927 ** waited on a lock.
002928 */
002929 #define WAL_RETRY_PROTOCOL_LIMIT 100
002930 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
002931 # define WAL_RETRY_BLOCKED_MASK 0x10000000
002932 #else
002933 # define WAL_RETRY_BLOCKED_MASK 0
002934 #endif
002935
002936 /*
002937 ** Attempt to start a read transaction. This might fail due to a race or
002938 ** other transient condition. When that happens, it returns WAL_RETRY to
002939 ** indicate to the caller that it is safe to retry immediately.
002940 **
002941 ** On success return SQLITE_OK. On a permanent failure (such an
002942 ** I/O error or an SQLITE_BUSY because another process is running
002943 ** recovery) return a positive error code.
002944 **
002945 ** The useWal parameter is true to force the use of the WAL and disable
002946 ** the case where the WAL is bypassed because it has been completely
002947 ** checkpointed. If useWal==0 then this routine calls walIndexReadHdr()
002948 ** to make a copy of the wal-index header into pWal->hdr. If the
002949 ** wal-index header has changed, *pChanged is set to 1 (as an indication
002950 ** to the caller that the local page cache is obsolete and needs to be
002951 ** flushed.) When useWal==1, the wal-index header is assumed to already
002952 ** be loaded and the pChanged parameter is unused.
002953 **
002954 ** The caller must set the cnt parameter to the number of prior calls to
002955 ** this routine during the current read attempt that returned WAL_RETRY.
002956 ** This routine will start taking more aggressive measures to clear the
002957 ** race conditions after multiple WAL_RETRY returns, and after an excessive
002958 ** number of errors will ultimately return SQLITE_PROTOCOL. The
002959 ** SQLITE_PROTOCOL return indicates that some other process has gone rogue
002960 ** and is not honoring the locking protocol. There is a vanishingly small
002961 ** chance that SQLITE_PROTOCOL could be returned because of a run of really
002962 ** bad luck when there is lots of contention for the wal-index, but that
002963 ** possibility is so small that it can be safely neglected, we believe.
002964 **
002965 ** On success, this routine obtains a read lock on
002966 ** WAL_READ_LOCK(pWal->readLock). The pWal->readLock integer is
002967 ** in the range 0 <= pWal->readLock < WAL_NREADER. If pWal->readLock==(-1)
002968 ** that means the Wal does not hold any read lock. The reader must not
002969 ** access any database page that is modified by a WAL frame up to and
002970 ** including frame number aReadMark[pWal->readLock]. The reader will
002971 ** use WAL frames up to and including pWal->hdr.mxFrame if pWal->readLock>0
002972 ** Or if pWal->readLock==0, then the reader will ignore the WAL
002973 ** completely and get all content directly from the database file.
002974 ** If the useWal parameter is 1 then the WAL will never be ignored and
002975 ** this routine will always set pWal->readLock>0 on success.
002976 ** When the read transaction is completed, the caller must release the
002977 ** lock on WAL_READ_LOCK(pWal->readLock) and set pWal->readLock to -1.
002978 **
002979 ** This routine uses the nBackfill and aReadMark[] fields of the header
002980 ** to select a particular WAL_READ_LOCK() that strives to let the
002981 ** checkpoint process do as much work as possible. This routine might
002982 ** update values of the aReadMark[] array in the header, but if it does
002983 ** so it takes care to hold an exclusive lock on the corresponding
002984 ** WAL_READ_LOCK() while changing values.
002985 */
002986 static int walTryBeginRead(Wal *pWal, int *pChanged, int useWal, int *pCnt){
002987 volatile WalCkptInfo *pInfo; /* Checkpoint information in wal-index */
002988 u32 mxReadMark; /* Largest aReadMark[] value */
002989 int mxI; /* Index of largest aReadMark[] value */
002990 int i; /* Loop counter */
002991 int rc = SQLITE_OK; /* Return code */
002992 u32 mxFrame; /* Wal frame to lock to */
002993 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
002994 int nBlockTmout = 0;
002995 #endif
002996
002997 assert( pWal->readLock<0 ); /* Not currently locked */
002998
002999 /* useWal may only be set for read/write connections */
003000 assert( (pWal->readOnly & WAL_SHM_RDONLY)==0 || useWal==0 );
003001
003002 /* Take steps to avoid spinning forever if there is a protocol error.
003003 **
003004 ** Circumstances that cause a RETRY should only last for the briefest
003005 ** instances of time. No I/O or other system calls are done while the
003006 ** locks are held, so the locks should not be held for very long. But
003007 ** if we are unlucky, another process that is holding a lock might get
003008 ** paged out or take a page-fault that is time-consuming to resolve,
003009 ** during the few nanoseconds that it is holding the lock. In that case,
003010 ** it might take longer than normal for the lock to free.
003011 **
003012 ** After 5 RETRYs, we begin calling sqlite3OsSleep(). The first few
003013 ** calls to sqlite3OsSleep() have a delay of 1 microsecond. Really this
003014 ** is more of a scheduler yield than an actual delay. But on the 10th
003015 ** an subsequent retries, the delays start becoming longer and longer,
003016 ** so that on the 100th (and last) RETRY we delay for 323 milliseconds.
003017 ** The total delay time before giving up is less than 10 seconds.
003018 */
003019 (*pCnt)++;
003020 if( *pCnt>5 ){
003021 int nDelay = 1; /* Pause time in microseconds */
003022 int cnt = (*pCnt & ~WAL_RETRY_BLOCKED_MASK);
003023 if( cnt>WAL_RETRY_PROTOCOL_LIMIT ){
003024 VVA_ONLY( pWal->lockError = 1; )
003025 return SQLITE_PROTOCOL;
003026 }
003027 if( *pCnt>=10 ) nDelay = (cnt-9)*(cnt-9)*39;
003028 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
003029 /* In SQLITE_ENABLE_SETLK_TIMEOUT builds, configure the file-descriptor
003030 ** to block for locks for approximately nDelay us. This affects three
003031 ** locks: (a) the shared lock taken on the DMS slot in os_unix.c (if
003032 ** using os_unix.c), (b) the WRITER lock taken in walIndexReadHdr() if the
003033 ** first attempted read fails, and (c) the shared lock taken on the
003034 ** read-mark.
003035 **
003036 ** If the previous call failed due to an SQLITE_BUSY_TIMEOUT error,
003037 ** then sleep for the minimum of 1us. The previous call already provided
003038 ** an extra delay while it was blocking on the lock.
003039 */
003040 nBlockTmout = (nDelay+998) / 1000;
003041 if( !useWal && walEnableBlockingMs(pWal, nBlockTmout) ){
003042 if( *pCnt & WAL_RETRY_BLOCKED_MASK ) nDelay = 1;
003043 }
003044 #endif
003045 sqlite3OsSleep(pWal->pVfs, nDelay);
003046 *pCnt &= ~WAL_RETRY_BLOCKED_MASK;
003047 }
003048
003049 if( !useWal ){
003050 assert( rc==SQLITE_OK );
003051 if( pWal->bShmUnreliable==0 ){
003052 rc = walIndexReadHdr(pWal, pChanged);
003053 }
003054 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
003055 walDisableBlocking(pWal);
003056 if( rc==SQLITE_BUSY_TIMEOUT ){
003057 rc = SQLITE_BUSY;
003058 *pCnt |= WAL_RETRY_BLOCKED_MASK;
003059 }
003060 #endif
003061 if( rc==SQLITE_BUSY ){
003062 /* If there is not a recovery running in another thread or process
003063 ** then convert BUSY errors to WAL_RETRY. If recovery is known to
003064 ** be running, convert BUSY to BUSY_RECOVERY. There is a race here
003065 ** which might cause WAL_RETRY to be returned even if BUSY_RECOVERY
003066 ** would be technically correct. But the race is benign since with
003067 ** WAL_RETRY this routine will be called again and will probably be
003068 ** right on the second iteration.
003069 */
003070 if( pWal->apWiData[0]==0 ){
003071 /* This branch is taken when the xShmMap() method returns SQLITE_BUSY.
003072 ** We assume this is a transient condition, so return WAL_RETRY. The
003073 ** xShmMap() implementation used by the default unix and win32 VFS
003074 ** modules may return SQLITE_BUSY due to a race condition in the
003075 ** code that determines whether or not the shared-memory region
003076 ** must be zeroed before the requested page is returned.
003077 */
003078 rc = WAL_RETRY;
003079 }else if( SQLITE_OK==(rc = walLockShared(pWal, WAL_RECOVER_LOCK)) ){
003080 walUnlockShared(pWal, WAL_RECOVER_LOCK);
003081 rc = WAL_RETRY;
003082 }else if( rc==SQLITE_BUSY ){
003083 rc = SQLITE_BUSY_RECOVERY;
003084 }
003085 }
003086 if( rc!=SQLITE_OK ){
003087 return rc;
003088 }
003089 else if( pWal->bShmUnreliable ){
003090 return walBeginShmUnreliable(pWal, pChanged);
003091 }
003092 }
003093
003094 assert( pWal->nWiData>0 );
003095 assert( pWal->apWiData[0]!=0 );
003096 pInfo = walCkptInfo(pWal);
003097 SEH_INJECT_FAULT;
003098 if( !useWal && AtomicLoad(&pInfo->nBackfill)==pWal->hdr.mxFrame
003099 #ifdef SQLITE_ENABLE_SNAPSHOT
003100 && (pWal->pSnapshot==0 || pWal->hdr.mxFrame==0)
003101 #endif
003102 ){
003103 /* The WAL has been completely backfilled (or it is empty).
003104 ** and can be safely ignored.
003105 */
003106 rc = walLockShared(pWal, WAL_READ_LOCK(0));
003107 walShmBarrier(pWal);
003108 if( rc==SQLITE_OK ){
003109 if( memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr)) ){
003110 /* It is not safe to allow the reader to continue here if frames
003111 ** may have been appended to the log before READ_LOCK(0) was obtained.
003112 ** When holding READ_LOCK(0), the reader ignores the entire log file,
003113 ** which implies that the database file contains a trustworthy
003114 ** snapshot. Since holding READ_LOCK(0) prevents a checkpoint from
003115 ** happening, this is usually correct.
003116 **
003117 ** However, if frames have been appended to the log (or if the log
003118 ** is wrapped and written for that matter) before the READ_LOCK(0)
003119 ** is obtained, that is not necessarily true. A checkpointer may
003120 ** have started to backfill the appended frames but crashed before
003121 ** it finished. Leaving a corrupt image in the database file.
003122 */
003123 walUnlockShared(pWal, WAL_READ_LOCK(0));
003124 return WAL_RETRY;
003125 }
003126 pWal->readLock = 0;
003127 return SQLITE_OK;
003128 }else if( rc!=SQLITE_BUSY ){
003129 return rc;
003130 }
003131 }
003132
003133 /* If we get this far, it means that the reader will want to use
003134 ** the WAL to get at content from recent commits. The job now is
003135 ** to select one of the aReadMark[] entries that is closest to
003136 ** but not exceeding pWal->hdr.mxFrame and lock that entry.
003137 */
003138 mxReadMark = 0;
003139 mxI = 0;
003140 mxFrame = pWal->hdr.mxFrame;
003141 #ifdef SQLITE_ENABLE_SNAPSHOT
003142 if( pWal->pSnapshot && pWal->pSnapshot->mxFrame<mxFrame ){
003143 mxFrame = pWal->pSnapshot->mxFrame;
003144 }
003145 #endif
003146 for(i=1; i<WAL_NREADER; i++){
003147 u32 thisMark = AtomicLoad(pInfo->aReadMark+i); SEH_INJECT_FAULT;
003148 if( mxReadMark<=thisMark && thisMark<=mxFrame ){
003149 assert( thisMark!=READMARK_NOT_USED );
003150 mxReadMark = thisMark;
003151 mxI = i;
003152 }
003153 }
003154 if( (pWal->readOnly & WAL_SHM_RDONLY)==0
003155 && (mxReadMark<mxFrame || mxI==0)
003156 ){
003157 for(i=1; i<WAL_NREADER; i++){
003158 rc = walLockExclusive(pWal, WAL_READ_LOCK(i), 1);
003159 if( rc==SQLITE_OK ){
003160 AtomicStore(pInfo->aReadMark+i,mxFrame);
003161 mxReadMark = mxFrame;
003162 mxI = i;
003163 walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1);
003164 break;
003165 }else if( rc!=SQLITE_BUSY ){
003166 return rc;
003167 }
003168 }
003169 }
003170 if( mxI==0 ){
003171 assert( rc==SQLITE_BUSY || (pWal->readOnly & WAL_SHM_RDONLY)!=0 );
003172 return rc==SQLITE_BUSY ? WAL_RETRY : SQLITE_READONLY_CANTINIT;
003173 }
003174
003175 (void)walEnableBlockingMs(pWal, nBlockTmout);
003176 rc = walLockShared(pWal, WAL_READ_LOCK(mxI));
003177 walDisableBlocking(pWal);
003178 if( rc ){
003179 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
003180 if( rc==SQLITE_BUSY_TIMEOUT ){
003181 *pCnt |= WAL_RETRY_BLOCKED_MASK;
003182 }
003183 #else
003184 assert( rc!=SQLITE_BUSY_TIMEOUT );
003185 #endif
003186 assert( (rc&0xFF)!=SQLITE_BUSY||rc==SQLITE_BUSY||rc==SQLITE_BUSY_TIMEOUT );
003187 return (rc&0xFF)==SQLITE_BUSY ? WAL_RETRY : rc;
003188 }
003189 /* Now that the read-lock has been obtained, check that neither the
003190 ** value in the aReadMark[] array or the contents of the wal-index
003191 ** header have changed.
003192 **
003193 ** It is necessary to check that the wal-index header did not change
003194 ** between the time it was read and when the shared-lock was obtained
003195 ** on WAL_READ_LOCK(mxI) was obtained to account for the possibility
003196 ** that the log file may have been wrapped by a writer, or that frames
003197 ** that occur later in the log than pWal->hdr.mxFrame may have been
003198 ** copied into the database by a checkpointer. If either of these things
003199 ** happened, then reading the database with the current value of
003200 ** pWal->hdr.mxFrame risks reading a corrupted snapshot. So, retry
003201 ** instead.
003202 **
003203 ** Before checking that the live wal-index header has not changed
003204 ** since it was read, set Wal.minFrame to the first frame in the wal
003205 ** file that has not yet been checkpointed. This client will not need
003206 ** to read any frames earlier than minFrame from the wal file - they
003207 ** can be safely read directly from the database file.
003208 **
003209 ** Because a ShmBarrier() call is made between taking the copy of
003210 ** nBackfill and checking that the wal-header in shared-memory still
003211 ** matches the one cached in pWal->hdr, it is guaranteed that the
003212 ** checkpointer that set nBackfill was not working with a wal-index
003213 ** header newer than that cached in pWal->hdr. If it were, that could
003214 ** cause a problem. The checkpointer could omit to checkpoint
003215 ** a version of page X that lies before pWal->minFrame (call that version
003216 ** A) on the basis that there is a newer version (version B) of the same
003217 ** page later in the wal file. But if version B happens to like past
003218 ** frame pWal->hdr.mxFrame - then the client would incorrectly assume
003219 ** that it can read version A from the database file. However, since
003220 ** we can guarantee that the checkpointer that set nBackfill could not
003221 ** see any pages past pWal->hdr.mxFrame, this problem does not come up.
003222 */
003223 pWal->minFrame = AtomicLoad(&pInfo->nBackfill)+1; SEH_INJECT_FAULT;
003224 walShmBarrier(pWal);
003225 if( AtomicLoad(pInfo->aReadMark+mxI)!=mxReadMark
003226 || memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr))
003227 ){
003228 walUnlockShared(pWal, WAL_READ_LOCK(mxI));
003229 return WAL_RETRY;
003230 }else{
003231 assert( mxReadMark<=pWal->hdr.mxFrame );
003232 pWal->readLock = (i16)mxI;
003233 }
003234 return rc;
003235 }
003236
003237 #ifdef SQLITE_ENABLE_SNAPSHOT
003238 /*
003239 ** This function does the work of sqlite3WalSnapshotRecover().
003240 */
003241 static int walSnapshotRecover(
003242 Wal *pWal, /* WAL handle */
003243 void *pBuf1, /* Temp buffer pWal->szPage bytes in size */
003244 void *pBuf2 /* Temp buffer pWal->szPage bytes in size */
003245 ){
003246 int szPage = (int)pWal->szPage;
003247 int rc;
003248 i64 szDb; /* Size of db file in bytes */
003249
003250 rc = sqlite3OsFileSize(pWal->pDbFd, &szDb);
003251 if( rc==SQLITE_OK ){
003252 volatile WalCkptInfo *pInfo = walCkptInfo(pWal);
003253 u32 i = pInfo->nBackfillAttempted;
003254 for(i=pInfo->nBackfillAttempted; i>AtomicLoad(&pInfo->nBackfill); i--){
003255 WalHashLoc sLoc; /* Hash table location */
003256 u32 pgno; /* Page number in db file */
003257 i64 iDbOff; /* Offset of db file entry */
003258 i64 iWalOff; /* Offset of wal file entry */
003259
003260 rc = walHashGet(pWal, walFramePage(i), &sLoc);
003261 if( rc!=SQLITE_OK ) break;
003262 assert( i - sLoc.iZero - 1 >=0 );
003263 pgno = sLoc.aPgno[i-sLoc.iZero-1];
003264 iDbOff = (i64)(pgno-1) * szPage;
003265
003266 if( iDbOff+szPage<=szDb ){
003267 iWalOff = walFrameOffset(i, szPage) + WAL_FRAME_HDRSIZE;
003268 rc = sqlite3OsRead(pWal->pWalFd, pBuf1, szPage, iWalOff);
003269
003270 if( rc==SQLITE_OK ){
003271 rc = sqlite3OsRead(pWal->pDbFd, pBuf2, szPage, iDbOff);
003272 }
003273
003274 if( rc!=SQLITE_OK || 0==memcmp(pBuf1, pBuf2, szPage) ){
003275 break;
003276 }
003277 }
003278
003279 pInfo->nBackfillAttempted = i-1;
003280 }
003281 }
003282
003283 return rc;
003284 }
003285
003286 /*
003287 ** Attempt to reduce the value of the WalCkptInfo.nBackfillAttempted
003288 ** variable so that older snapshots can be accessed. To do this, loop
003289 ** through all wal frames from nBackfillAttempted to (nBackfill+1),
003290 ** comparing their content to the corresponding page with the database
003291 ** file, if any. Set nBackfillAttempted to the frame number of the
003292 ** first frame for which the wal file content matches the db file.
003293 **
003294 ** This is only really safe if the file-system is such that any page
003295 ** writes made by earlier checkpointers were atomic operations, which
003296 ** is not always true. It is also possible that nBackfillAttempted
003297 ** may be left set to a value larger than expected, if a wal frame
003298 ** contains content that duplicate of an earlier version of the same
003299 ** page.
003300 **
003301 ** SQLITE_OK is returned if successful, or an SQLite error code if an
003302 ** error occurs. It is not an error if nBackfillAttempted cannot be
003303 ** decreased at all.
003304 */
003305 int sqlite3WalSnapshotRecover(Wal *pWal){
003306 int rc;
003307
003308 assert( pWal->readLock>=0 );
003309 rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1);
003310 if( rc==SQLITE_OK ){
003311 void *pBuf1 = sqlite3_malloc(pWal->szPage);
003312 void *pBuf2 = sqlite3_malloc(pWal->szPage);
003313 if( pBuf1==0 || pBuf2==0 ){
003314 rc = SQLITE_NOMEM;
003315 }else{
003316 pWal->ckptLock = 1;
003317 SEH_TRY {
003318 rc = walSnapshotRecover(pWal, pBuf1, pBuf2);
003319 }
003320 SEH_EXCEPT( rc = SQLITE_IOERR_IN_PAGE; )
003321 pWal->ckptLock = 0;
003322 }
003323
003324 sqlite3_free(pBuf1);
003325 sqlite3_free(pBuf2);
003326 walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1);
003327 }
003328
003329 return rc;
003330 }
003331 #endif /* SQLITE_ENABLE_SNAPSHOT */
003332
003333 /*
003334 ** This function does the work of sqlite3WalBeginReadTransaction() (see
003335 ** below). That function simply calls this one inside an SEH_TRY{...} block.
003336 */
003337 static int walBeginReadTransaction(Wal *pWal, int *pChanged){
003338 int rc; /* Return code */
003339 int cnt = 0; /* Number of TryBeginRead attempts */
003340 #ifdef SQLITE_ENABLE_SNAPSHOT
003341 int ckptLock = 0;
003342 int bChanged = 0;
003343 WalIndexHdr *pSnapshot = pWal->pSnapshot;
003344 #endif
003345
003346 assert( pWal->ckptLock==0 );
003347 assert( pWal->nSehTry>0 );
003348
003349 #ifdef SQLITE_ENABLE_SNAPSHOT
003350 if( pSnapshot ){
003351 if( memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){
003352 bChanged = 1;
003353 }
003354
003355 /* It is possible that there is a checkpointer thread running
003356 ** concurrent with this code. If this is the case, it may be that the
003357 ** checkpointer has already determined that it will checkpoint
003358 ** snapshot X, where X is later in the wal file than pSnapshot, but
003359 ** has not yet set the pInfo->nBackfillAttempted variable to indicate
003360 ** its intent. To avoid the race condition this leads to, ensure that
003361 ** there is no checkpointer process by taking a shared CKPT lock
003362 ** before checking pInfo->nBackfillAttempted. */
003363 (void)walEnableBlocking(pWal);
003364 rc = walLockShared(pWal, WAL_CKPT_LOCK);
003365 walDisableBlocking(pWal);
003366
003367 if( rc!=SQLITE_OK ){
003368 return rc;
003369 }
003370 ckptLock = 1;
003371 }
003372 #endif
003373
003374 do{
003375 rc = walTryBeginRead(pWal, pChanged, 0, &cnt);
003376 }while( rc==WAL_RETRY );
003377 testcase( (rc&0xff)==SQLITE_BUSY );
003378 testcase( (rc&0xff)==SQLITE_IOERR );
003379 testcase( rc==SQLITE_PROTOCOL );
003380 testcase( rc==SQLITE_OK );
003381
003382 #ifdef SQLITE_ENABLE_SNAPSHOT
003383 if( rc==SQLITE_OK ){
003384 if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){
003385 /* At this point the client has a lock on an aReadMark[] slot holding
003386 ** a value equal to or smaller than pSnapshot->mxFrame, but pWal->hdr
003387 ** is populated with the wal-index header corresponding to the head
003388 ** of the wal file. Verify that pSnapshot is still valid before
003389 ** continuing. Reasons why pSnapshot might no longer be valid:
003390 **
003391 ** (1) The WAL file has been reset since the snapshot was taken.
003392 ** In this case, the salt will have changed.
003393 **
003394 ** (2) A checkpoint as been attempted that wrote frames past
003395 ** pSnapshot->mxFrame into the database file. Note that the
003396 ** checkpoint need not have completed for this to cause problems.
003397 */
003398 volatile WalCkptInfo *pInfo = walCkptInfo(pWal);
003399
003400 assert( pWal->readLock>0 || pWal->hdr.mxFrame==0 );
003401 assert( pInfo->aReadMark[pWal->readLock]<=pSnapshot->mxFrame );
003402
003403 /* Check that the wal file has not been wrapped. Assuming that it has
003404 ** not, also check that no checkpointer has attempted to checkpoint any
003405 ** frames beyond pSnapshot->mxFrame. If either of these conditions are
003406 ** true, return SQLITE_ERROR_SNAPSHOT. Otherwise, overwrite pWal->hdr
003407 ** with *pSnapshot and set *pChanged as appropriate for opening the
003408 ** snapshot. */
003409 if( !memcmp(pSnapshot->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt))
003410 && pSnapshot->mxFrame>=pInfo->nBackfillAttempted
003411 ){
003412 assert( pWal->readLock>0 );
003413 memcpy(&pWal->hdr, pSnapshot, sizeof(WalIndexHdr));
003414 *pChanged = bChanged;
003415 }else{
003416 rc = SQLITE_ERROR_SNAPSHOT;
003417 }
003418
003419 /* A client using a non-current snapshot may not ignore any frames
003420 ** from the start of the wal file. This is because, for a system
003421 ** where (minFrame < iSnapshot < maxFrame), a checkpointer may
003422 ** have omitted to checkpoint a frame earlier than minFrame in
003423 ** the file because there exists a frame after iSnapshot that
003424 ** is the same database page. */
003425 pWal->minFrame = 1;
003426
003427 if( rc!=SQLITE_OK ){
003428 sqlite3WalEndReadTransaction(pWal);
003429 }
003430 }
003431 }
003432
003433 /* Release the shared CKPT lock obtained above. */
003434 if( ckptLock ){
003435 assert( pSnapshot );
003436 walUnlockShared(pWal, WAL_CKPT_LOCK);
003437 }
003438 #endif
003439 return rc;
003440 }
003441
003442 /*
003443 ** Begin a read transaction on the database.
003444 **
003445 ** This routine used to be called sqlite3OpenSnapshot() and with good reason:
003446 ** it takes a snapshot of the state of the WAL and wal-index for the current
003447 ** instant in time. The current thread will continue to use this snapshot.
003448 ** Other threads might append new content to the WAL and wal-index but
003449 ** that extra content is ignored by the current thread.
003450 **
003451 ** If the database contents have changes since the previous read
003452 ** transaction, then *pChanged is set to 1 before returning. The
003453 ** Pager layer will use this to know that its cache is stale and
003454 ** needs to be flushed.
003455 */
003456 int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){
003457 int rc;
003458 SEH_TRY {
003459 rc = walBeginReadTransaction(pWal, pChanged);
003460 }
003461 SEH_EXCEPT( rc = walHandleException(pWal); )
003462 return rc;
003463 }
003464
003465 /*
003466 ** Finish with a read transaction. All this does is release the
003467 ** read-lock.
003468 */
003469 void sqlite3WalEndReadTransaction(Wal *pWal){
003470 sqlite3WalEndWriteTransaction(pWal);
003471 if( pWal->readLock>=0 ){
003472 walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock));
003473 pWal->readLock = -1;
003474 }
003475 }
003476
003477 /*
003478 ** Search the wal file for page pgno. If found, set *piRead to the frame that
003479 ** contains the page. Otherwise, if pgno is not in the wal file, set *piRead
003480 ** to zero.
003481 **
003482 ** Return SQLITE_OK if successful, or an error code if an error occurs. If an
003483 ** error does occur, the final value of *piRead is undefined.
003484 */
003485 static int walFindFrame(
003486 Wal *pWal, /* WAL handle */
003487 Pgno pgno, /* Database page number to read data for */
003488 u32 *piRead /* OUT: Frame number (or zero) */
003489 ){
003490 u32 iRead = 0; /* If !=0, WAL frame to return data from */
003491 u32 iLast = pWal->hdr.mxFrame; /* Last page in WAL for this reader */
003492 int iHash; /* Used to loop through N hash tables */
003493 int iMinHash;
003494
003495 /* This routine is only be called from within a read transaction. */
003496 assert( pWal->readLock>=0 || pWal->lockError );
003497
003498 /* If the "last page" field of the wal-index header snapshot is 0, then
003499 ** no data will be read from the wal under any circumstances. Return early
003500 ** in this case as an optimization. Likewise, if pWal->readLock==0,
003501 ** then the WAL is ignored by the reader so return early, as if the
003502 ** WAL were empty.
003503 */
003504 if( iLast==0 || (pWal->readLock==0 && pWal->bShmUnreliable==0) ){
003505 *piRead = 0;
003506 return SQLITE_OK;
003507 }
003508
003509 /* Search the hash table or tables for an entry matching page number
003510 ** pgno. Each iteration of the following for() loop searches one
003511 ** hash table (each hash table indexes up to HASHTABLE_NPAGE frames).
003512 **
003513 ** This code might run concurrently to the code in walIndexAppend()
003514 ** that adds entries to the wal-index (and possibly to this hash
003515 ** table). This means the value just read from the hash
003516 ** slot (aHash[iKey]) may have been added before or after the
003517 ** current read transaction was opened. Values added after the
003518 ** read transaction was opened may have been written incorrectly -
003519 ** i.e. these slots may contain garbage data. However, we assume
003520 ** that any slots written before the current read transaction was
003521 ** opened remain unmodified.
003522 **
003523 ** For the reasons above, the if(...) condition featured in the inner
003524 ** loop of the following block is more stringent that would be required
003525 ** if we had exclusive access to the hash-table:
003526 **
003527 ** (aPgno[iFrame]==pgno):
003528 ** This condition filters out normal hash-table collisions.
003529 **
003530 ** (iFrame<=iLast):
003531 ** This condition filters out entries that were added to the hash
003532 ** table after the current read-transaction had started.
003533 */
003534 iMinHash = walFramePage(pWal->minFrame);
003535 for(iHash=walFramePage(iLast); iHash>=iMinHash; iHash--){
003536 WalHashLoc sLoc; /* Hash table location */
003537 int iKey; /* Hash slot index */
003538 int nCollide; /* Number of hash collisions remaining */
003539 int rc; /* Error code */
003540 u32 iH;
003541
003542 rc = walHashGet(pWal, iHash, &sLoc);
003543 if( rc!=SQLITE_OK ){
003544 return rc;
003545 }
003546 nCollide = HASHTABLE_NSLOT;
003547 iKey = walHash(pgno);
003548 SEH_INJECT_FAULT;
003549 while( (iH = AtomicLoad(&sLoc.aHash[iKey]))!=0 ){
003550 u32 iFrame = iH + sLoc.iZero;
003551 if( iFrame<=iLast && iFrame>=pWal->minFrame && sLoc.aPgno[iH-1]==pgno ){
003552 assert( iFrame>iRead || CORRUPT_DB );
003553 iRead = iFrame;
003554 }
003555 if( (nCollide--)==0 ){
003556 *piRead = 0;
003557 return SQLITE_CORRUPT_BKPT;
003558 }
003559 iKey = walNextHash(iKey);
003560 }
003561 if( iRead ) break;
003562 }
003563
003564 #ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT
003565 /* If expensive assert() statements are available, do a linear search
003566 ** of the wal-index file content. Make sure the results agree with the
003567 ** result obtained using the hash indexes above. */
003568 {
003569 u32 iRead2 = 0;
003570 u32 iTest;
003571 assert( pWal->bShmUnreliable || pWal->minFrame>0 );
003572 for(iTest=iLast; iTest>=pWal->minFrame && iTest>0; iTest--){
003573 if( walFramePgno(pWal, iTest)==pgno ){
003574 iRead2 = iTest;
003575 break;
003576 }
003577 }
003578 assert( iRead==iRead2 );
003579 }
003580 #endif
003581
003582 *piRead = iRead;
003583 return SQLITE_OK;
003584 }
003585
003586 /*
003587 ** Search the wal file for page pgno. If found, set *piRead to the frame that
003588 ** contains the page. Otherwise, if pgno is not in the wal file, set *piRead
003589 ** to zero.
003590 **
003591 ** Return SQLITE_OK if successful, or an error code if an error occurs. If an
003592 ** error does occur, the final value of *piRead is undefined.
003593 **
003594 ** The difference between this function and walFindFrame() is that this
003595 ** function wraps walFindFrame() in an SEH_TRY{...} block.
003596 */
003597 int sqlite3WalFindFrame(
003598 Wal *pWal, /* WAL handle */
003599 Pgno pgno, /* Database page number to read data for */
003600 u32 *piRead /* OUT: Frame number (or zero) */
003601 ){
003602 int rc;
003603 SEH_TRY {
003604 rc = walFindFrame(pWal, pgno, piRead);
003605 }
003606 SEH_EXCEPT( rc = SQLITE_IOERR_IN_PAGE; )
003607 return rc;
003608 }
003609
003610 /*
003611 ** Read the contents of frame iRead from the wal file into buffer pOut
003612 ** (which is nOut bytes in size). Return SQLITE_OK if successful, or an
003613 ** error code otherwise.
003614 */
003615 int sqlite3WalReadFrame(
003616 Wal *pWal, /* WAL handle */
003617 u32 iRead, /* Frame to read */
003618 int nOut, /* Size of buffer pOut in bytes */
003619 u8 *pOut /* Buffer to write page data to */
003620 ){
003621 int sz;
003622 i64 iOffset;
003623 sz = pWal->hdr.szPage;
003624 sz = (sz&0xfe00) + ((sz&0x0001)<<16);
003625 testcase( sz<=32768 );
003626 testcase( sz>=65536 );
003627 iOffset = walFrameOffset(iRead, sz) + WAL_FRAME_HDRSIZE;
003628 /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL */
003629 return sqlite3OsRead(pWal->pWalFd, pOut, (nOut>sz ? sz : nOut), iOffset);
003630 }
003631
003632 /*
003633 ** Return the size of the database in pages (or zero, if unknown).
003634 */
003635 Pgno sqlite3WalDbsize(Wal *pWal){
003636 if( pWal && ALWAYS(pWal->readLock>=0) ){
003637 return pWal->hdr.nPage;
003638 }
003639 return 0;
003640 }
003641
003642
003643 /*
003644 ** This function starts a write transaction on the WAL.
003645 **
003646 ** A read transaction must have already been started by a prior call
003647 ** to sqlite3WalBeginReadTransaction().
003648 **
003649 ** If another thread or process has written into the database since
003650 ** the read transaction was started, then it is not possible for this
003651 ** thread to write as doing so would cause a fork. So this routine
003652 ** returns SQLITE_BUSY in that case and no write transaction is started.
003653 **
003654 ** There can only be a single writer active at a time.
003655 */
003656 int sqlite3WalBeginWriteTransaction(Wal *pWal){
003657 int rc;
003658
003659 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
003660 /* If the write-lock is already held, then it was obtained before the
003661 ** read-transaction was even opened, making this call a no-op.
003662 ** Return early. */
003663 if( pWal->writeLock ){
003664 assert( !memcmp(&pWal->hdr,(void *)walIndexHdr(pWal),sizeof(WalIndexHdr)) );
003665 return SQLITE_OK;
003666 }
003667 #endif
003668
003669 /* Cannot start a write transaction without first holding a read
003670 ** transaction. */
003671 assert( pWal->readLock>=0 );
003672 assert( pWal->writeLock==0 && pWal->iReCksum==0 );
003673
003674 if( pWal->readOnly ){
003675 return SQLITE_READONLY;
003676 }
003677
003678 /* Only one writer allowed at a time. Get the write lock. Return
003679 ** SQLITE_BUSY if unable.
003680 */
003681 rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1);
003682 if( rc ){
003683 return rc;
003684 }
003685 pWal->writeLock = 1;
003686
003687 /* If another connection has written to the database file since the
003688 ** time the read transaction on this connection was started, then
003689 ** the write is disallowed.
003690 */
003691 SEH_TRY {
003692 if( memcmp(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr))!=0 ){
003693 rc = SQLITE_BUSY_SNAPSHOT;
003694 }
003695 }
003696 SEH_EXCEPT( rc = SQLITE_IOERR_IN_PAGE; )
003697
003698 if( rc!=SQLITE_OK ){
003699 walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1);
003700 pWal->writeLock = 0;
003701 }
003702 return rc;
003703 }
003704
003705 /*
003706 ** End a write transaction. The commit has already been done. This
003707 ** routine merely releases the lock.
003708 */
003709 int sqlite3WalEndWriteTransaction(Wal *pWal){
003710 if( pWal->writeLock ){
003711 walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1);
003712 pWal->writeLock = 0;
003713 pWal->iReCksum = 0;
003714 pWal->truncateOnCommit = 0;
003715 }
003716 return SQLITE_OK;
003717 }
003718
003719 /*
003720 ** If any data has been written (but not committed) to the log file, this
003721 ** function moves the write-pointer back to the start of the transaction.
003722 **
003723 ** Additionally, the callback function is invoked for each frame written
003724 ** to the WAL since the start of the transaction. If the callback returns
003725 ** other than SQLITE_OK, it is not invoked again and the error code is
003726 ** returned to the caller.
003727 **
003728 ** Otherwise, if the callback function does not return an error, this
003729 ** function returns SQLITE_OK.
003730 */
003731 int sqlite3WalUndo(Wal *pWal, int (*xUndo)(void *, Pgno), void *pUndoCtx){
003732 int rc = SQLITE_OK;
003733 if( ALWAYS(pWal->writeLock) ){
003734 Pgno iMax = pWal->hdr.mxFrame;
003735 Pgno iFrame;
003736
003737 SEH_TRY {
003738 /* Restore the clients cache of the wal-index header to the state it
003739 ** was in before the client began writing to the database.
003740 */
003741 memcpy(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr));
003742
003743 for(iFrame=pWal->hdr.mxFrame+1;
003744 ALWAYS(rc==SQLITE_OK) && iFrame<=iMax;
003745 iFrame++
003746 ){
003747 /* This call cannot fail. Unless the page for which the page number
003748 ** is passed as the second argument is (a) in the cache and
003749 ** (b) has an outstanding reference, then xUndo is either a no-op
003750 ** (if (a) is false) or simply expels the page from the cache (if (b)
003751 ** is false).
003752 **
003753 ** If the upper layer is doing a rollback, it is guaranteed that there
003754 ** are no outstanding references to any page other than page 1. And
003755 ** page 1 is never written to the log until the transaction is
003756 ** committed. As a result, the call to xUndo may not fail.
003757 */
003758 assert( walFramePgno(pWal, iFrame)!=1 );
003759 rc = xUndo(pUndoCtx, walFramePgno(pWal, iFrame));
003760 }
003761 if( iMax!=pWal->hdr.mxFrame ) walCleanupHash(pWal);
003762 }
003763 SEH_EXCEPT( rc = SQLITE_IOERR_IN_PAGE; )
003764 }
003765 return rc;
003766 }
003767
003768 /*
003769 ** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32
003770 ** values. This function populates the array with values required to
003771 ** "rollback" the write position of the WAL handle back to the current
003772 ** point in the event of a savepoint rollback (via WalSavepointUndo()).
003773 */
003774 void sqlite3WalSavepoint(Wal *pWal, u32 *aWalData){
003775 assert( pWal->writeLock );
003776 aWalData[0] = pWal->hdr.mxFrame;
003777 aWalData[1] = pWal->hdr.aFrameCksum[0];
003778 aWalData[2] = pWal->hdr.aFrameCksum[1];
003779 aWalData[3] = pWal->nCkpt;
003780 }
003781
003782 /*
003783 ** Move the write position of the WAL back to the point identified by
003784 ** the values in the aWalData[] array. aWalData must point to an array
003785 ** of WAL_SAVEPOINT_NDATA u32 values that has been previously populated
003786 ** by a call to WalSavepoint().
003787 */
003788 int sqlite3WalSavepointUndo(Wal *pWal, u32 *aWalData){
003789 int rc = SQLITE_OK;
003790
003791 assert( pWal->writeLock );
003792 assert( aWalData[3]!=pWal->nCkpt || aWalData[0]<=pWal->hdr.mxFrame );
003793
003794 if( aWalData[3]!=pWal->nCkpt ){
003795 /* This savepoint was opened immediately after the write-transaction
003796 ** was started. Right after that, the writer decided to wrap around
003797 ** to the start of the log. Update the savepoint values to match.
003798 */
003799 aWalData[0] = 0;
003800 aWalData[3] = pWal->nCkpt;
003801 }
003802
003803 if( aWalData[0]<pWal->hdr.mxFrame ){
003804 pWal->hdr.mxFrame = aWalData[0];
003805 pWal->hdr.aFrameCksum[0] = aWalData[1];
003806 pWal->hdr.aFrameCksum[1] = aWalData[2];
003807 SEH_TRY {
003808 walCleanupHash(pWal);
003809 }
003810 SEH_EXCEPT( rc = SQLITE_IOERR_IN_PAGE; )
003811 }
003812
003813 return rc;
003814 }
003815
003816 /*
003817 ** This function is called just before writing a set of frames to the log
003818 ** file (see sqlite3WalFrames()). It checks to see if, instead of appending
003819 ** to the current log file, it is possible to overwrite the start of the
003820 ** existing log file with the new frames (i.e. "reset" the log). If so,
003821 ** it sets pWal->hdr.mxFrame to 0. Otherwise, pWal->hdr.mxFrame is left
003822 ** unchanged.
003823 **
003824 ** SQLITE_OK is returned if no error is encountered (regardless of whether
003825 ** or not pWal->hdr.mxFrame is modified). An SQLite error code is returned
003826 ** if an error occurs.
003827 */
003828 static int walRestartLog(Wal *pWal){
003829 int rc = SQLITE_OK;
003830 int cnt;
003831
003832 if( pWal->readLock==0 ){
003833 volatile WalCkptInfo *pInfo = walCkptInfo(pWal);
003834 assert( pInfo->nBackfill==pWal->hdr.mxFrame );
003835 if( pInfo->nBackfill>0 ){
003836 u32 salt1;
003837 sqlite3_randomness(4, &salt1);
003838 rc = walLockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1);
003839 if( rc==SQLITE_OK ){
003840 /* If all readers are using WAL_READ_LOCK(0) (in other words if no
003841 ** readers are currently using the WAL), then the transactions
003842 ** frames will overwrite the start of the existing log. Update the
003843 ** wal-index header to reflect this.
003844 **
003845 ** In theory it would be Ok to update the cache of the header only
003846 ** at this point. But updating the actual wal-index header is also
003847 ** safe and means there is no special case for sqlite3WalUndo()
003848 ** to handle if this transaction is rolled back. */
003849 walRestartHdr(pWal, salt1);
003850 walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1);
003851 }else if( rc!=SQLITE_BUSY ){
003852 return rc;
003853 }
003854 }
003855 walUnlockShared(pWal, WAL_READ_LOCK(0));
003856 pWal->readLock = -1;
003857 cnt = 0;
003858 do{
003859 int notUsed;
003860 rc = walTryBeginRead(pWal, ¬Used, 1, &cnt);
003861 }while( rc==WAL_RETRY );
003862 assert( (rc&0xff)!=SQLITE_BUSY ); /* BUSY not possible when useWal==1 */
003863 testcase( (rc&0xff)==SQLITE_IOERR );
003864 testcase( rc==SQLITE_PROTOCOL );
003865 testcase( rc==SQLITE_OK );
003866 }
003867 return rc;
003868 }
003869
003870 /*
003871 ** Information about the current state of the WAL file and where
003872 ** the next fsync should occur - passed from sqlite3WalFrames() into
003873 ** walWriteToLog().
003874 */
003875 typedef struct WalWriter {
003876 Wal *pWal; /* The complete WAL information */
003877 sqlite3_file *pFd; /* The WAL file to which we write */
003878 sqlite3_int64 iSyncPoint; /* Fsync at this offset */
003879 int syncFlags; /* Flags for the fsync */
003880 int szPage; /* Size of one page */
003881 } WalWriter;
003882
003883 /*
003884 ** Write iAmt bytes of content into the WAL file beginning at iOffset.
003885 ** Do a sync when crossing the p->iSyncPoint boundary.
003886 **
003887 ** In other words, if iSyncPoint is in between iOffset and iOffset+iAmt,
003888 ** first write the part before iSyncPoint, then sync, then write the
003889 ** rest.
003890 */
003891 static int walWriteToLog(
003892 WalWriter *p, /* WAL to write to */
003893 void *pContent, /* Content to be written */
003894 int iAmt, /* Number of bytes to write */
003895 sqlite3_int64 iOffset /* Start writing at this offset */
003896 ){
003897 int rc;
003898 if( iOffset<p->iSyncPoint && iOffset+iAmt>=p->iSyncPoint ){
003899 int iFirstAmt = (int)(p->iSyncPoint - iOffset);
003900 rc = sqlite3OsWrite(p->pFd, pContent, iFirstAmt, iOffset);
003901 if( rc ) return rc;
003902 iOffset += iFirstAmt;
003903 iAmt -= iFirstAmt;
003904 pContent = (void*)(iFirstAmt + (char*)pContent);
003905 assert( WAL_SYNC_FLAGS(p->syncFlags)!=0 );
003906 rc = sqlite3OsSync(p->pFd, WAL_SYNC_FLAGS(p->syncFlags));
003907 if( iAmt==0 || rc ) return rc;
003908 }
003909 rc = sqlite3OsWrite(p->pFd, pContent, iAmt, iOffset);
003910 return rc;
003911 }
003912
003913 /*
003914 ** Write out a single frame of the WAL
003915 */
003916 static int walWriteOneFrame(
003917 WalWriter *p, /* Where to write the frame */
003918 PgHdr *pPage, /* The page of the frame to be written */
003919 int nTruncate, /* The commit flag. Usually 0. >0 for commit */
003920 sqlite3_int64 iOffset /* Byte offset at which to write */
003921 ){
003922 int rc; /* Result code from subfunctions */
003923 void *pData; /* Data actually written */
003924 u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-header in */
003925 pData = pPage->pData;
003926 walEncodeFrame(p->pWal, pPage->pgno, nTruncate, pData, aFrame);
003927 rc = walWriteToLog(p, aFrame, sizeof(aFrame), iOffset);
003928 if( rc ) return rc;
003929 /* Write the page data */
003930 rc = walWriteToLog(p, pData, p->szPage, iOffset+sizeof(aFrame));
003931 return rc;
003932 }
003933
003934 /*
003935 ** This function is called as part of committing a transaction within which
003936 ** one or more frames have been overwritten. It updates the checksums for
003937 ** all frames written to the wal file by the current transaction starting
003938 ** with the earliest to have been overwritten.
003939 **
003940 ** SQLITE_OK is returned if successful, or an SQLite error code otherwise.
003941 */
003942 static int walRewriteChecksums(Wal *pWal, u32 iLast){
003943 const int szPage = pWal->szPage;/* Database page size */
003944 int rc = SQLITE_OK; /* Return code */
003945 u8 *aBuf; /* Buffer to load data from wal file into */
003946 u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-headers in */
003947 u32 iRead; /* Next frame to read from wal file */
003948 i64 iCksumOff;
003949
003950 aBuf = sqlite3_malloc(szPage + WAL_FRAME_HDRSIZE);
003951 if( aBuf==0 ) return SQLITE_NOMEM_BKPT;
003952
003953 /* Find the checksum values to use as input for the recalculating the
003954 ** first checksum. If the first frame is frame 1 (implying that the current
003955 ** transaction restarted the wal file), these values must be read from the
003956 ** wal-file header. Otherwise, read them from the frame header of the
003957 ** previous frame. */
003958 assert( pWal->iReCksum>0 );
003959 if( pWal->iReCksum==1 ){
003960 iCksumOff = 24;
003961 }else{
003962 iCksumOff = walFrameOffset(pWal->iReCksum-1, szPage) + 16;
003963 }
003964 rc = sqlite3OsRead(pWal->pWalFd, aBuf, sizeof(u32)*2, iCksumOff);
003965 pWal->hdr.aFrameCksum[0] = sqlite3Get4byte(aBuf);
003966 pWal->hdr.aFrameCksum[1] = sqlite3Get4byte(&aBuf[sizeof(u32)]);
003967
003968 iRead = pWal->iReCksum;
003969 pWal->iReCksum = 0;
003970 for(; rc==SQLITE_OK && iRead<=iLast; iRead++){
003971 i64 iOff = walFrameOffset(iRead, szPage);
003972 rc = sqlite3OsRead(pWal->pWalFd, aBuf, szPage+WAL_FRAME_HDRSIZE, iOff);
003973 if( rc==SQLITE_OK ){
003974 u32 iPgno, nDbSize;
003975 iPgno = sqlite3Get4byte(aBuf);
003976 nDbSize = sqlite3Get4byte(&aBuf[4]);
003977
003978 walEncodeFrame(pWal, iPgno, nDbSize, &aBuf[WAL_FRAME_HDRSIZE], aFrame);
003979 rc = sqlite3OsWrite(pWal->pWalFd, aFrame, sizeof(aFrame), iOff);
003980 }
003981 }
003982
003983 sqlite3_free(aBuf);
003984 return rc;
003985 }
003986
003987 /*
003988 ** Write a set of frames to the log. The caller must hold the write-lock
003989 ** on the log file (obtained using sqlite3WalBeginWriteTransaction()).
003990 */
003991 static int walFrames(
003992 Wal *pWal, /* Wal handle to write to */
003993 int szPage, /* Database page-size in bytes */
003994 PgHdr *pList, /* List of dirty pages to write */
003995 Pgno nTruncate, /* Database size after this commit */
003996 int isCommit, /* True if this is a commit */
003997 int sync_flags /* Flags to pass to OsSync() (or 0) */
003998 ){
003999 int rc; /* Used to catch return codes */
004000 u32 iFrame; /* Next frame address */
004001 PgHdr *p; /* Iterator to run through pList with. */
004002 PgHdr *pLast = 0; /* Last frame in list */
004003 int nExtra = 0; /* Number of extra copies of last page */
004004 int szFrame; /* The size of a single frame */
004005 i64 iOffset; /* Next byte to write in WAL file */
004006 WalWriter w; /* The writer */
004007 u32 iFirst = 0; /* First frame that may be overwritten */
004008 WalIndexHdr *pLive; /* Pointer to shared header */
004009
004010 assert( pList );
004011 assert( pWal->writeLock );
004012
004013 /* If this frame set completes a transaction, then nTruncate>0. If
004014 ** nTruncate==0 then this frame set does not complete the transaction. */
004015 assert( (isCommit!=0)==(nTruncate!=0) );
004016
004017 #if defined(SQLITE_TEST) && defined(SQLITE_DEBUG)
004018 { int cnt; for(cnt=0, p=pList; p; p=p->pDirty, cnt++){}
004019 WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n",
004020 pWal, cnt, pWal->hdr.mxFrame, isCommit ? "Commit" : "Spill"));
004021 }
004022 #endif
004023
004024 pLive = (WalIndexHdr*)walIndexHdr(pWal);
004025 if( memcmp(&pWal->hdr, (void *)pLive, sizeof(WalIndexHdr))!=0 ){
004026 iFirst = pLive->mxFrame+1;
004027 }
004028
004029 /* See if it is possible to write these frames into the start of the
004030 ** log file, instead of appending to it at pWal->hdr.mxFrame.
004031 */
004032 if( SQLITE_OK!=(rc = walRestartLog(pWal)) ){
004033 return rc;
004034 }
004035
004036 /* If this is the first frame written into the log, write the WAL
004037 ** header to the start of the WAL file. See comments at the top of
004038 ** this source file for a description of the WAL header format.
004039 */
004040 iFrame = pWal->hdr.mxFrame;
004041 if( iFrame==0 ){
004042 u8 aWalHdr[WAL_HDRSIZE]; /* Buffer to assemble wal-header in */
004043 u32 aCksum[2]; /* Checksum for wal-header */
004044
004045 sqlite3Put4byte(&aWalHdr[0], (WAL_MAGIC | SQLITE_BIGENDIAN));
004046 sqlite3Put4byte(&aWalHdr[4], WAL_MAX_VERSION);
004047 sqlite3Put4byte(&aWalHdr[8], szPage);
004048 sqlite3Put4byte(&aWalHdr[12], pWal->nCkpt);
004049 if( pWal->nCkpt==0 ) sqlite3_randomness(8, pWal->hdr.aSalt);
004050 memcpy(&aWalHdr[16], pWal->hdr.aSalt, 8);
004051 walChecksumBytes(1, aWalHdr, WAL_HDRSIZE-2*4, 0, aCksum);
004052 sqlite3Put4byte(&aWalHdr[24], aCksum[0]);
004053 sqlite3Put4byte(&aWalHdr[28], aCksum[1]);
004054
004055 pWal->szPage = szPage;
004056 pWal->hdr.bigEndCksum = SQLITE_BIGENDIAN;
004057 pWal->hdr.aFrameCksum[0] = aCksum[0];
004058 pWal->hdr.aFrameCksum[1] = aCksum[1];
004059 pWal->truncateOnCommit = 1;
004060
004061 rc = sqlite3OsWrite(pWal->pWalFd, aWalHdr, sizeof(aWalHdr), 0);
004062 WALTRACE(("WAL%p: wal-header write %s\n", pWal, rc ? "failed" : "ok"));
004063 if( rc!=SQLITE_OK ){
004064 return rc;
004065 }
004066
004067 /* Sync the header (unless SQLITE_IOCAP_SEQUENTIAL is true or unless
004068 ** all syncing is turned off by PRAGMA synchronous=OFF). Otherwise
004069 ** an out-of-order write following a WAL restart could result in
004070 ** database corruption. See the ticket:
004071 **
004072 ** https://sqlite.org/src/info/ff5be73dee
004073 */
004074 if( pWal->syncHeader ){
004075 rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags));
004076 if( rc ) return rc;
004077 }
004078 }
004079 if( (int)pWal->szPage!=szPage ){
004080 return SQLITE_CORRUPT_BKPT; /* TH3 test case: cov1/corrupt155.test */
004081 }
004082
004083 /* Setup information needed to write frames into the WAL */
004084 w.pWal = pWal;
004085 w.pFd = pWal->pWalFd;
004086 w.iSyncPoint = 0;
004087 w.syncFlags = sync_flags;
004088 w.szPage = szPage;
004089 iOffset = walFrameOffset(iFrame+1, szPage);
004090 szFrame = szPage + WAL_FRAME_HDRSIZE;
004091
004092 /* Write all frames into the log file exactly once */
004093 for(p=pList; p; p=p->pDirty){
004094 int nDbSize; /* 0 normally. Positive == commit flag */
004095
004096 /* Check if this page has already been written into the wal file by
004097 ** the current transaction. If so, overwrite the existing frame and
004098 ** set Wal.writeLock to WAL_WRITELOCK_RECKSUM - indicating that
004099 ** checksums must be recomputed when the transaction is committed. */
004100 if( iFirst && (p->pDirty || isCommit==0) ){
004101 u32 iWrite = 0;
004102 VVA_ONLY(rc =) walFindFrame(pWal, p->pgno, &iWrite);
004103 assert( rc==SQLITE_OK || iWrite==0 );
004104 if( iWrite>=iFirst ){
004105 i64 iOff = walFrameOffset(iWrite, szPage) + WAL_FRAME_HDRSIZE;
004106 void *pData;
004107 if( pWal->iReCksum==0 || iWrite<pWal->iReCksum ){
004108 pWal->iReCksum = iWrite;
004109 }
004110 pData = p->pData;
004111 rc = sqlite3OsWrite(pWal->pWalFd, pData, szPage, iOff);
004112 if( rc ) return rc;
004113 p->flags &= ~PGHDR_WAL_APPEND;
004114 continue;
004115 }
004116 }
004117
004118 iFrame++;
004119 assert( iOffset==walFrameOffset(iFrame, szPage) );
004120 nDbSize = (isCommit && p->pDirty==0) ? nTruncate : 0;
004121 rc = walWriteOneFrame(&w, p, nDbSize, iOffset);
004122 if( rc ) return rc;
004123 pLast = p;
004124 iOffset += szFrame;
004125 p->flags |= PGHDR_WAL_APPEND;
004126 }
004127
004128 /* Recalculate checksums within the wal file if required. */
004129 if( isCommit && pWal->iReCksum ){
004130 rc = walRewriteChecksums(pWal, iFrame);
004131 if( rc ) return rc;
004132 }
004133
004134 /* If this is the end of a transaction, then we might need to pad
004135 ** the transaction and/or sync the WAL file.
004136 **
004137 ** Padding and syncing only occur if this set of frames complete a
004138 ** transaction and if PRAGMA synchronous=FULL. If synchronous==NORMAL
004139 ** or synchronous==OFF, then no padding or syncing are needed.
004140 **
004141 ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not
004142 ** needed and only the sync is done. If padding is needed, then the
004143 ** final frame is repeated (with its commit mark) until the next sector
004144 ** boundary is crossed. Only the part of the WAL prior to the last
004145 ** sector boundary is synced; the part of the last frame that extends
004146 ** past the sector boundary is written after the sync.
004147 */
004148 if( isCommit && WAL_SYNC_FLAGS(sync_flags)!=0 ){
004149 int bSync = 1;
004150 if( pWal->padToSectorBoundary ){
004151 int sectorSize = sqlite3SectorSize(pWal->pWalFd);
004152 w.iSyncPoint = ((iOffset+sectorSize-1)/sectorSize)*sectorSize;
004153 bSync = (w.iSyncPoint==iOffset);
004154 testcase( bSync );
004155 while( iOffset<w.iSyncPoint ){
004156 rc = walWriteOneFrame(&w, pLast, nTruncate, iOffset);
004157 if( rc ) return rc;
004158 iOffset += szFrame;
004159 nExtra++;
004160 assert( pLast!=0 );
004161 }
004162 }
004163 if( bSync ){
004164 assert( rc==SQLITE_OK );
004165 rc = sqlite3OsSync(w.pFd, WAL_SYNC_FLAGS(sync_flags));
004166 }
004167 }
004168
004169 /* If this frame set completes the first transaction in the WAL and
004170 ** if PRAGMA journal_size_limit is set, then truncate the WAL to the
004171 ** journal size limit, if possible.
004172 */
004173 if( isCommit && pWal->truncateOnCommit && pWal->mxWalSize>=0 ){
004174 i64 sz = pWal->mxWalSize;
004175 if( walFrameOffset(iFrame+nExtra+1, szPage)>pWal->mxWalSize ){
004176 sz = walFrameOffset(iFrame+nExtra+1, szPage);
004177 }
004178 walLimitSize(pWal, sz);
004179 pWal->truncateOnCommit = 0;
004180 }
004181
004182 /* Append data to the wal-index. It is not necessary to lock the
004183 ** wal-index to do this as the SQLITE_SHM_WRITE lock held on the wal-index
004184 ** guarantees that there are no other writers, and no data that may
004185 ** be in use by existing readers is being overwritten.
004186 */
004187 iFrame = pWal->hdr.mxFrame;
004188 for(p=pList; p && rc==SQLITE_OK; p=p->pDirty){
004189 if( (p->flags & PGHDR_WAL_APPEND)==0 ) continue;
004190 iFrame++;
004191 rc = walIndexAppend(pWal, iFrame, p->pgno);
004192 }
004193 assert( pLast!=0 || nExtra==0 );
004194 while( rc==SQLITE_OK && nExtra>0 ){
004195 iFrame++;
004196 nExtra--;
004197 rc = walIndexAppend(pWal, iFrame, pLast->pgno);
004198 }
004199
004200 if( rc==SQLITE_OK ){
004201 /* Update the private copy of the header. */
004202 pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16));
004203 testcase( szPage<=32768 );
004204 testcase( szPage>=65536 );
004205 pWal->hdr.mxFrame = iFrame;
004206 if( isCommit ){
004207 pWal->hdr.iChange++;
004208 pWal->hdr.nPage = nTruncate;
004209 }
004210 /* If this is a commit, update the wal-index header too. */
004211 if( isCommit ){
004212 walIndexWriteHdr(pWal);
004213 pWal->iCallback = iFrame;
004214 }
004215 }
004216
004217 WALTRACE(("WAL%p: frame write %s\n", pWal, rc ? "failed" : "ok"));
004218 return rc;
004219 }
004220
004221 /*
004222 ** Write a set of frames to the log. The caller must hold the write-lock
004223 ** on the log file (obtained using sqlite3WalBeginWriteTransaction()).
004224 **
004225 ** The difference between this function and walFrames() is that this
004226 ** function wraps walFrames() in an SEH_TRY{...} block.
004227 */
004228 int sqlite3WalFrames(
004229 Wal *pWal, /* Wal handle to write to */
004230 int szPage, /* Database page-size in bytes */
004231 PgHdr *pList, /* List of dirty pages to write */
004232 Pgno nTruncate, /* Database size after this commit */
004233 int isCommit, /* True if this is a commit */
004234 int sync_flags /* Flags to pass to OsSync() (or 0) */
004235 ){
004236 int rc;
004237 SEH_TRY {
004238 rc = walFrames(pWal, szPage, pList, nTruncate, isCommit, sync_flags);
004239 }
004240 SEH_EXCEPT( rc = walHandleException(pWal); )
004241 return rc;
004242 }
004243
004244 /*
004245 ** This routine is called to implement sqlite3_wal_checkpoint() and
004246 ** related interfaces.
004247 **
004248 ** Obtain a CHECKPOINT lock and then backfill as much information as
004249 ** we can from WAL into the database.
004250 **
004251 ** If parameter xBusy is not NULL, it is a pointer to a busy-handler
004252 ** callback. In this case this function runs a blocking checkpoint.
004253 */
004254 int sqlite3WalCheckpoint(
004255 Wal *pWal, /* Wal connection */
004256 sqlite3 *db, /* Check this handle's interrupt flag */
004257 int eMode, /* PASSIVE, FULL, RESTART, or TRUNCATE */
004258 int (*xBusy)(void*), /* Function to call when busy */
004259 void *pBusyArg, /* Context argument for xBusyHandler */
004260 int sync_flags, /* Flags to sync db file with (or 0) */
004261 int nBuf, /* Size of temporary buffer */
004262 u8 *zBuf, /* Temporary buffer to use */
004263 int *pnLog, /* OUT: Number of frames in WAL */
004264 int *pnCkpt /* OUT: Number of backfilled frames in WAL */
004265 ){
004266 int rc; /* Return code */
004267 int isChanged = 0; /* True if a new wal-index header is loaded */
004268 int eMode2 = eMode; /* Mode to pass to walCheckpoint() */
004269 int (*xBusy2)(void*) = xBusy; /* Busy handler for eMode2 */
004270
004271 assert( pWal->ckptLock==0 );
004272 assert( pWal->writeLock==0 );
004273
004274 /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked
004275 ** in the SQLITE_CHECKPOINT_PASSIVE mode. */
004276 assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 );
004277
004278 if( pWal->readOnly ) return SQLITE_READONLY;
004279 WALTRACE(("WAL%p: checkpoint begins\n", pWal));
004280
004281 /* Enable blocking locks, if possible. */
004282 sqlite3WalDb(pWal, db);
004283 if( xBusy2 ) (void)walEnableBlocking(pWal);
004284
004285 /* IMPLEMENTATION-OF: R-62028-47212 All calls obtain an exclusive
004286 ** "checkpoint" lock on the database file.
004287 ** EVIDENCE-OF: R-10421-19736 If any other process is running a
004288 ** checkpoint operation at the same time, the lock cannot be obtained and
004289 ** SQLITE_BUSY is returned.
004290 ** EVIDENCE-OF: R-53820-33897 Even if there is a busy-handler configured,
004291 ** it will not be invoked in this case.
004292 */
004293 rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1);
004294 testcase( rc==SQLITE_BUSY );
004295 testcase( rc!=SQLITE_OK && xBusy2!=0 );
004296 if( rc==SQLITE_OK ){
004297 pWal->ckptLock = 1;
004298
004299 /* IMPLEMENTATION-OF: R-59782-36818 The SQLITE_CHECKPOINT_FULL, RESTART and
004300 ** TRUNCATE modes also obtain the exclusive "writer" lock on the database
004301 ** file.
004302 **
004303 ** EVIDENCE-OF: R-60642-04082 If the writer lock cannot be obtained
004304 ** immediately, and a busy-handler is configured, it is invoked and the
004305 ** writer lock retried until either the busy-handler returns 0 or the
004306 ** lock is successfully obtained.
004307 */
004308 if( eMode!=SQLITE_CHECKPOINT_PASSIVE ){
004309 rc = walBusyLock(pWal, xBusy2, pBusyArg, WAL_WRITE_LOCK, 1);
004310 if( rc==SQLITE_OK ){
004311 pWal->writeLock = 1;
004312 }else if( rc==SQLITE_BUSY ){
004313 eMode2 = SQLITE_CHECKPOINT_PASSIVE;
004314 xBusy2 = 0;
004315 rc = SQLITE_OK;
004316 }
004317 }
004318 }
004319
004320
004321 /* Read the wal-index header. */
004322 SEH_TRY {
004323 if( rc==SQLITE_OK ){
004324 /* For a passive checkpoint, do not re-enable blocking locks after
004325 ** reading the wal-index header. A passive checkpoint should not block
004326 ** or invoke the busy handler. The only lock such a checkpoint may
004327 ** attempt to obtain is a lock on a read-slot, and it should give up
004328 ** immediately and do a partial checkpoint if it cannot obtain it. */
004329 walDisableBlocking(pWal);
004330 rc = walIndexReadHdr(pWal, &isChanged);
004331 if( eMode2!=SQLITE_CHECKPOINT_PASSIVE ) (void)walEnableBlocking(pWal);
004332 if( isChanged && pWal->pDbFd->pMethods->iVersion>=3 ){
004333 sqlite3OsUnfetch(pWal->pDbFd, 0, 0);
004334 }
004335 }
004336
004337 /* Copy data from the log to the database file. */
004338 if( rc==SQLITE_OK ){
004339 if( pWal->hdr.mxFrame && walPagesize(pWal)!=nBuf ){
004340 rc = SQLITE_CORRUPT_BKPT;
004341 }else{
004342 rc = walCheckpoint(pWal, db, eMode2, xBusy2, pBusyArg, sync_flags,zBuf);
004343 }
004344
004345 /* If no error occurred, set the output variables. */
004346 if( rc==SQLITE_OK || rc==SQLITE_BUSY ){
004347 if( pnLog ) *pnLog = (int)pWal->hdr.mxFrame;
004348 SEH_INJECT_FAULT;
004349 if( pnCkpt ) *pnCkpt = (int)(walCkptInfo(pWal)->nBackfill);
004350 }
004351 }
004352 }
004353 SEH_EXCEPT( rc = walHandleException(pWal); )
004354
004355 if( isChanged ){
004356 /* If a new wal-index header was loaded before the checkpoint was
004357 ** performed, then the pager-cache associated with pWal is now
004358 ** out of date. So zero the cached wal-index header to ensure that
004359 ** next time the pager opens a snapshot on this database it knows that
004360 ** the cache needs to be reset.
004361 */
004362 memset(&pWal->hdr, 0, sizeof(WalIndexHdr));
004363 }
004364
004365 walDisableBlocking(pWal);
004366 sqlite3WalDb(pWal, 0);
004367
004368 /* Release the locks. */
004369 sqlite3WalEndWriteTransaction(pWal);
004370 if( pWal->ckptLock ){
004371 walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1);
004372 pWal->ckptLock = 0;
004373 }
004374 WALTRACE(("WAL%p: checkpoint %s\n", pWal, rc ? "failed" : "ok"));
004375 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
004376 if( rc==SQLITE_BUSY_TIMEOUT ) rc = SQLITE_BUSY;
004377 #endif
004378 return (rc==SQLITE_OK && eMode!=eMode2 ? SQLITE_BUSY : rc);
004379 }
004380
004381 /* Return the value to pass to a sqlite3_wal_hook callback, the
004382 ** number of frames in the WAL at the point of the last commit since
004383 ** sqlite3WalCallback() was called. If no commits have occurred since
004384 ** the last call, then return 0.
004385 */
004386 int sqlite3WalCallback(Wal *pWal){
004387 u32 ret = 0;
004388 if( pWal ){
004389 ret = pWal->iCallback;
004390 pWal->iCallback = 0;
004391 }
004392 return (int)ret;
004393 }
004394
004395 /*
004396 ** This function is called to change the WAL subsystem into or out
004397 ** of locking_mode=EXCLUSIVE.
004398 **
004399 ** If op is zero, then attempt to change from locking_mode=EXCLUSIVE
004400 ** into locking_mode=NORMAL. This means that we must acquire a lock
004401 ** on the pWal->readLock byte. If the WAL is already in locking_mode=NORMAL
004402 ** or if the acquisition of the lock fails, then return 0. If the
004403 ** transition out of exclusive-mode is successful, return 1. This
004404 ** operation must occur while the pager is still holding the exclusive
004405 ** lock on the main database file.
004406 **
004407 ** If op is one, then change from locking_mode=NORMAL into
004408 ** locking_mode=EXCLUSIVE. This means that the pWal->readLock must
004409 ** be released. Return 1 if the transition is made and 0 if the
004410 ** WAL is already in exclusive-locking mode - meaning that this
004411 ** routine is a no-op. The pager must already hold the exclusive lock
004412 ** on the main database file before invoking this operation.
004413 **
004414 ** If op is negative, then do a dry-run of the op==1 case but do
004415 ** not actually change anything. The pager uses this to see if it
004416 ** should acquire the database exclusive lock prior to invoking
004417 ** the op==1 case.
004418 */
004419 int sqlite3WalExclusiveMode(Wal *pWal, int op){
004420 int rc;
004421 assert( pWal->writeLock==0 );
004422 assert( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE || op==-1 );
004423
004424 /* pWal->readLock is usually set, but might be -1 if there was a
004425 ** prior error while attempting to acquire are read-lock. This cannot
004426 ** happen if the connection is actually in exclusive mode (as no xShmLock
004427 ** locks are taken in this case). Nor should the pager attempt to
004428 ** upgrade to exclusive-mode following such an error.
004429 */
004430 #ifndef SQLITE_USE_SEH
004431 assert( pWal->readLock>=0 || pWal->lockError );
004432 #endif
004433 assert( pWal->readLock>=0 || (op<=0 && pWal->exclusiveMode==0) );
004434
004435 if( op==0 ){
004436 if( pWal->exclusiveMode!=WAL_NORMAL_MODE ){
004437 pWal->exclusiveMode = WAL_NORMAL_MODE;
004438 if( walLockShared(pWal, WAL_READ_LOCK(pWal->readLock))!=SQLITE_OK ){
004439 pWal->exclusiveMode = WAL_EXCLUSIVE_MODE;
004440 }
004441 rc = pWal->exclusiveMode==WAL_NORMAL_MODE;
004442 }else{
004443 /* Already in locking_mode=NORMAL */
004444 rc = 0;
004445 }
004446 }else if( op>0 ){
004447 assert( pWal->exclusiveMode==WAL_NORMAL_MODE );
004448 assert( pWal->readLock>=0 );
004449 walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock));
004450 pWal->exclusiveMode = WAL_EXCLUSIVE_MODE;
004451 rc = 1;
004452 }else{
004453 rc = pWal->exclusiveMode==WAL_NORMAL_MODE;
004454 }
004455 return rc;
004456 }
004457
004458 /*
004459 ** Return true if the argument is non-NULL and the WAL module is using
004460 ** heap-memory for the wal-index. Otherwise, if the argument is NULL or the
004461 ** WAL module is using shared-memory, return false.
004462 */
004463 int sqlite3WalHeapMemory(Wal *pWal){
004464 return (pWal && pWal->exclusiveMode==WAL_HEAPMEMORY_MODE );
004465 }
004466
004467 #ifdef SQLITE_ENABLE_SNAPSHOT
004468 /* Create a snapshot object. The content of a snapshot is opaque to
004469 ** every other subsystem, so the WAL module can put whatever it needs
004470 ** in the object.
004471 */
004472 int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot){
004473 int rc = SQLITE_OK;
004474 WalIndexHdr *pRet;
004475 static const u32 aZero[4] = { 0, 0, 0, 0 };
004476
004477 assert( pWal->readLock>=0 && pWal->writeLock==0 );
004478
004479 if( memcmp(&pWal->hdr.aFrameCksum[0],aZero,16)==0 ){
004480 *ppSnapshot = 0;
004481 return SQLITE_ERROR;
004482 }
004483 pRet = (WalIndexHdr*)sqlite3_malloc(sizeof(WalIndexHdr));
004484 if( pRet==0 ){
004485 rc = SQLITE_NOMEM_BKPT;
004486 }else{
004487 memcpy(pRet, &pWal->hdr, sizeof(WalIndexHdr));
004488 *ppSnapshot = (sqlite3_snapshot*)pRet;
004489 }
004490
004491 return rc;
004492 }
004493
004494 /* Try to open on pSnapshot when the next read-transaction starts
004495 */
004496 void sqlite3WalSnapshotOpen(
004497 Wal *pWal,
004498 sqlite3_snapshot *pSnapshot
004499 ){
004500 pWal->pSnapshot = (WalIndexHdr*)pSnapshot;
004501 }
004502
004503 /*
004504 ** Return a +ve value if snapshot p1 is newer than p2. A -ve value if
004505 ** p1 is older than p2 and zero if p1 and p2 are the same snapshot.
004506 */
004507 int sqlite3_snapshot_cmp(sqlite3_snapshot *p1, sqlite3_snapshot *p2){
004508 WalIndexHdr *pHdr1 = (WalIndexHdr*)p1;
004509 WalIndexHdr *pHdr2 = (WalIndexHdr*)p2;
004510
004511 /* aSalt[0] is a copy of the value stored in the wal file header. It
004512 ** is incremented each time the wal file is restarted. */
004513 if( pHdr1->aSalt[0]<pHdr2->aSalt[0] ) return -1;
004514 if( pHdr1->aSalt[0]>pHdr2->aSalt[0] ) return +1;
004515 if( pHdr1->mxFrame<pHdr2->mxFrame ) return -1;
004516 if( pHdr1->mxFrame>pHdr2->mxFrame ) return +1;
004517 return 0;
004518 }
004519
004520 /*
004521 ** The caller currently has a read transaction open on the database.
004522 ** This function takes a SHARED lock on the CHECKPOINTER slot and then
004523 ** checks if the snapshot passed as the second argument is still
004524 ** available. If so, SQLITE_OK is returned.
004525 **
004526 ** If the snapshot is not available, SQLITE_ERROR is returned. Or, if
004527 ** the CHECKPOINTER lock cannot be obtained, SQLITE_BUSY. If any error
004528 ** occurs (any value other than SQLITE_OK is returned), the CHECKPOINTER
004529 ** lock is released before returning.
004530 */
004531 int sqlite3WalSnapshotCheck(Wal *pWal, sqlite3_snapshot *pSnapshot){
004532 int rc;
004533 SEH_TRY {
004534 rc = walLockShared(pWal, WAL_CKPT_LOCK);
004535 if( rc==SQLITE_OK ){
004536 WalIndexHdr *pNew = (WalIndexHdr*)pSnapshot;
004537 if( memcmp(pNew->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt))
004538 || pNew->mxFrame<walCkptInfo(pWal)->nBackfillAttempted
004539 ){
004540 rc = SQLITE_ERROR_SNAPSHOT;
004541 walUnlockShared(pWal, WAL_CKPT_LOCK);
004542 }
004543 }
004544 }
004545 SEH_EXCEPT( rc = walHandleException(pWal); )
004546 return rc;
004547 }
004548
004549 /*
004550 ** Release a lock obtained by an earlier successful call to
004551 ** sqlite3WalSnapshotCheck().
004552 */
004553 void sqlite3WalSnapshotUnlock(Wal *pWal){
004554 assert( pWal );
004555 walUnlockShared(pWal, WAL_CKPT_LOCK);
004556 }
004557
004558
004559 #endif /* SQLITE_ENABLE_SNAPSHOT */
004560
004561 #ifdef SQLITE_ENABLE_ZIPVFS
004562 /*
004563 ** If the argument is not NULL, it points to a Wal object that holds a
004564 ** read-lock. This function returns the database page-size if it is known,
004565 ** or zero if it is not (or if pWal is NULL).
004566 */
004567 int sqlite3WalFramesize(Wal *pWal){
004568 assert( pWal==0 || pWal->readLock>=0 );
004569 return (pWal ? pWal->szPage : 0);
004570 }
004571 #endif
004572
004573 /* Return the sqlite3_file object for the WAL file
004574 */
004575 sqlite3_file *sqlite3WalFile(Wal *pWal){
004576 return pWal->pWalFd;
004577 }
004578
004579 #endif /* #ifndef SQLITE_OMIT_WAL */