Take advantage of atomic-write capabilities in the F2FS filesystem when the

database is stored on such a filesystem.  This is a compile-time option
activated using SQLITE_ENABLE_BATCH_ATOMIC_WRITE.

FossilOrigin-Name: 24190b221f73472dafaead6de101b4debc2c91c1ca28d70b45a38df5bb61fb39

manifest

@@ -1,5 +1,5 @@
-C Allow\sATTACH\sand\sDETACH\sinside\sof\sa\stransaction.
+C Take\sadvantage\sof\satomic-write\scapabilities\sin\sthe\sF2FS\sfilesystem\swhen\sthe\ndatabase\sis\sstored\son\ssuch\sa\sfilesystem.\s\sThis\sis\sa\scompile-time\soption\nactivated\susing\sSQLITE_ENABLE_BATCH_ATOMIC_WRITE.
-D 2017-08-01T00:20:34.182
+D 2017-08-01T14:16:15.157
 F Makefile.in d9873c9925917cca9990ee24be17eb9613a668012c85a343aef7e5536ae266e8
 F Makefile.linux-gcc 7bc79876b875010e8c8f9502eb935ca92aa3c434
 F Makefile.msc 02b469e9dcd5b7ee63fc1fb05babc174260ee4cfa4e0ef2e48c3c6801567a016
@@ -401,7 +401,7 @@ F src/btreeInt.h 97700795edf8a43245720414798b7b29d8e465aef46bf301ffacd431910c0da
 F src/build.c f65f86520aa877853125565e42c59c5c49851a4733392931777fb1aace4aedfd
 F src/callback.c 930648a084a3adc741c6471adfbdc50ba47ba3542421cb80a26f259f467de65e
 F src/complete.c a3634ab1e687055cd002e11b8f43eb75c17da23e
-F src/ctime.c 928954802b1397d9fb1378c7eb702c94b4735bbab1d5793e21b6a77734f56a1b
+F src/ctime.c ff1be3eed7bdd75aaca61ca8dc848f7c9f850ef2fb9cb56f2734e922a098f9c0
 F src/date.c 48f743d88bbe88f848532d333cca84f26e52a4f217e86f86be7fc1b919c33d74
 F src/dbstat.c 7a4ba8518b6369ef3600c49cf9c918ad979acba610b2aebef1b656d649b96720
 F src/delete.c 939bd15e6b54b82b951e1c0ffc2ff2b4ab579196780a1f6d394e47bd6f799b6c
@@ -424,7 +424,7 @@ F src/mem1.c c12a42539b1ba105e3707d0e628ad70e611040d8f5e38cf942cee30c867083de
 F src/mem2.c f1940d9e91948dd6a908fbb9ce3835c36b5d83c3
 F src/mem3.c 8768ac94694f31ffaf8b4d0ea5dc08af7010a35a
 F src/mem5.c 9bf955937b07f8c32541c8a9991f33ce3173d944
-F src/memjournal.c 95752936c11dc6995672d1dd783cd633eea0cc95
+F src/memjournal.c 6f3d36a0a8f72f48f6c3c722f04301ac64f2515435fa42924293e46fc7994661
 F src/msvc.h 4942752b6a253116baaa8de75256c51a459a5e81
 F src/mutex.c 8e45800ee78e0cd1f1f3fe8e398853307f4a085c
 F src/mutex.h 779d588e3b7756ec3ecf7d78cde1d84aba414f85
@@ -436,10 +436,10 @@ F src/os.c add02933b1dce7a39a005b00a2f5364b763e9a24
 F src/os.h 8e976e59eb4ca1c0fca6d35ee803e38951cb0343
 F src/os_common.h b2f4707a603e36811d9b1a13278bffd757857b85
 F src/os_setup.h 0dbaea40a7d36bf311613d31342e0b99e2536586
-F src/os_unix.c 30e2c43e4955db990e5b5a81e901f8aa74cc8820
+F src/os_unix.c 0681c6ef336fcb6a111f45b60a5faea38992ed6c4ae9fbd57a6f8e247724fa68
 F src/os_win.c 2a6c73eef01c51a048cc4ddccd57f981afbec18a
 F src/os_win.h 7b073010f1451abe501be30d12f6bc599824944a
-F src/pager.c 14f6982c470c05b8e85575c69e9c1712010602e20400f8670d8699e21283e0e4
+F src/pager.c 1e63b0299cf123cf38c48413ec03190f56c1e7d0ccc6573c467d8ac240b898e9
 F src/pager.h f2a99646c5533ffe11afa43e9e0bea74054e4efa
 F src/parse.y e384cb73f99e1b074085c974b37f4d830e885359e4b60837e30f7d67c16ba65b
 F src/pcache.c 62835bed959e2914edd26afadfecce29ece0e870
@@ -455,20 +455,20 @@ F src/rowset.c 7b7e7e479212e65b723bf40128c7b36dc5afdfac
 F src/select.c 31b35ddf55f1021f7148a01306984b057c11ebb6e3463d94677225e0a1e301a3
 F src/shell.c bd6a37cbe8bf64ef6a6a74fdc50f067d3148149b4ce2b4d03154663e66ded55f
 F src/shell.c.in b5725acacba95ccefa57b6d068f710e29ba8239c3aa704628a1902a1f729c175
-F src/sqlite.h.in 0e2603c23f0747c5660669f946e231730af000c76d1653b153dcf2c26fce0a6b
+F src/sqlite.h.in 72f1775c7a134f9e358eedafe1ebc703c28b0d705d976464ddbf6a9219448952
 F src/sqlite3.rc 5121c9e10c3964d5755191c80dd1180c122fc3a8
 F src/sqlite3ext.h 0f9f72b86a3792314f5db7a1dfbc2c82376bcd8d0919ceb80637bca126ec3c68
-F src/sqliteInt.h a1b8df420e8fa80fda9414ab7784d6e62271e1f7d65034ffd3e906ee6f014def
+F src/sqliteInt.h fe648fe59c71f7f44b5e89cf7cff0b96d81bd718263517c6895014632357df7b
 F src/sqliteLimit.h 1513bfb7b20378aa0041e7022d04acb73525de35b80b252f1b83fedb4de6a76b
 F src/status.c a9e66593dfb28a9e746cba7153f84d49c1ddc4b1
 F src/table.c b46ad567748f24a326d9de40e5b9659f96ffff34
 F src/tclsqlite.c 487951d81f9704800fd9f0ffdaa2f935a83ccb6be3575c2c4ef83e4789b4c828
-F src/test1.c cfb78b728b37ae3a2b14fe1b3a6c766e0da41370eda112594e698c94011b622e
+F src/test1.c 8513b17ca4a7a9ba28748535d178b6e472ec7394ae0eea53907f2d3bcdbab2df
 F src/test2.c 3efb99ab7f1fc8d154933e02ae1378bac9637da5
 F src/test3.c b8434949dfb8aff8dfa082c8b592109e77844c2135ed3c492113839b6956255b
 F src/test4.c 18ec393bb4d0ad1de729f0b94da7267270f3d8e6
 F src/test5.c 328aae2c010c57a9829d255dc099d6899311672d
-F src/test6.c 004ad42f121f693b8cbe060d1a330678abc61620
+F src/test6.c e8d839fbc552ce044bec8234561a2d5b8819b48e29548ad0ba400471697946a8
 F src/test7.c 5612e9aecf934d6df7bba6ce861fdf5ba5456010
 F src/test8.c 4f4904721167b32f7a4fa8c7b32a07a673d6cc86
 F src/test9.c 12e5ba554d2d1cbe0158f6ab3f7ffcd7a86ee4e5
@@ -481,7 +481,7 @@ F src/test_btree.c 8b2dc8b8848cf3a4db93f11578f075e82252a274
 F src/test_config.c abf6fc1fe9d041b699578c42e3db81f8831c4f5b804f1927958102ee8f2b773e
 F src/test_delete.c e2fe07646dff6300b48d49b2fee2fe192ed389e834dd635e3b3bac0ce0bf9f8f
 F src/test_demovfs.c a0c3bdd45ed044115c2c9f7779e56eafff18741e
-F src/test_devsym.c 4e58dec2602d8e139ca08659f62a62450587cb58
+F src/test_devsym.c 1960abbb234b97e9b920f07e99503fc04b443f62bbc3c6ff2c2cea2133e3b8a2
 F src/test_fs.c 35a2f7dd8a915900873386331386d9ba1ae1b5026d74fd20c2807bc76221f291
 F src/test_func.c a4fdab3363b436c1b12660e9362ce3f3782b7b5e
 F src/test_hexio.c 1d4469ca61ab202a1fcec6543f584d2407205e8d
@@ -572,6 +572,7 @@ F test/async3.test d73a062002376d7edc1fe3edff493edbec1fc2f7
 F test/async4.test 1787e3952128aa10238bf39945126de7ca23685a
 F test/async5.test 383ab533fdb9f7ad228cc99ee66e1acb34cc0dc0
 F test/atof1.test ff0b0156fd705b67c506e1f2bfe9e26102bea9bd
+F test/atomic.test 065a453dde33c77ff586d91ccaa6ed419829d492dbb1a5694b8a09f3f9d7d061
 F test/attach.test f4b8918ba2f3e88e6883b8452340545f10a1388af808343c37fc5c577be8281c
 F test/attach2.test 567047a7607aae8ebb3794642ebb168abe66b4af366fcd0cf7f616a1495cd43f
 F test/attach3.test c59d92791070c59272e00183b7353eeb94915976
@@ -754,7 +755,7 @@ F test/exists.test 79a75323c78f02bbe9c251ea502a092f9ef63dac
 F test/expr.test 66a2c9ac34f74f036faa4092f5402c7d3162fc93
 F test/extension01.test 00d13cec817f331a687a243e0e5a2d87b0e358c9
 F test/extraquick.test cb254400bd42bfb777ff675356aabf3287978f79
-F test/fallocate.test 3e979af17dfa7e5e9dda5eba1a696c04fa9d47f7
+F test/fallocate.test 87b5e43c872b7e69cd80b7b8813eb102b571a75d45dda24e38b65537bcc85733
 F test/filectrl.test 6e871c2d35dead1d9a88e176e8d2ca094fec6bb3
 F test/filefmt.test f393e80c4b8d493b7a7f8f3809a8425bbf4292af1f5140f01cb1427798a2bbd4
 F test/fkey1.test ba64806ff9a04eecab2679caad377ae99a5e94e4
@@ -1029,7 +1030,7 @@ F test/minmax.test 6751e87b409fe11b02e70a306d846fa544e25a41
 F test/minmax2.test b44bae787fc7b227597b01b0ca5575c7cb54d3bc
 F test/minmax3.test cc1e8b010136db0d01a6f2a29ba5a9f321034354
 F test/minmax4.test 936941484ebdceb8adec7c86b6cd9b6e5e897c1f
-F test/misc1.test 6430dabfb4b4fa480633590118964201f94d3ccc
+F test/misc1.test 51ec3f56a2a7965de226964cff856695e743186826561536647f1e8b7d1d0eb3
 F test/misc2.test 00d7de54eda90e237fc9a38b9e5ccc769ebf6d4d
 F test/misc3.test cf3dda47d5dda3e53fc5804a100d3c82be736c9d
 F test/misc4.test 0d8be3466adf123a7791a66ba2bc8e8d229e87f3
@@ -1087,7 +1088,7 @@ F test/parser1.test 391b9bf9a229547a129c61ac345ed1a6f5eb1854
 F test/pcache.test c8acbedd3b6fd0f9a7ca887a83b11d24a007972b
 F test/pcache2.test af7f3deb1a819f77a6d0d81534e97d1cf62cd442
 F test/percentile.test 4243af26b8f3f4555abe166f723715a1f74c77ff
-F test/permutations.test 5e2e5439642898e0947ced066ad09b82bd817ddfb83dc71291b4c957efc84b62
+F test/permutations.test 3b94f8fd431d39fac4952eb5dc38e1bb2b4518e1ac967d66f5abc815c104aeb6
 F test/pragma.test f274259d6393b6681eb433beb8dd39a26ec06a4431052a4880b43b84912a3f58
 F test/pragma2.test e5d5c176360c321344249354c0c16aec46214c9f
 F test/pragma3.test 14c12bc5352b1e100e0b6b44f371053a81ccf8ed
@@ -1113,7 +1114,7 @@ F test/regexp2.test 40e894223b3d6672655481493f1be12012f2b33c
 F test/reindex.test 44edd3966b474468b823d481eafef0c305022254
 F test/releasetest.tcl 7bb585433ce7fb2a2c255ae4b5e24f1bc27fe177ec1120f886cc4852f48f5ee9 x
 F test/resolver01.test f4022acafda7f4d40eca94dbf16bc5fc4ac30ceb
-F test/rollback.test 458fe73eb3ffdfdf9f6ba3e9b7350a6220414dea
+F test/rollback.test f580934279800d480a19176c6b44909df31ce7ad45267ea475a541daa522f3d3
 F test/rollback2.test 8435d6ff0f13f51d2a4181c232e706005fa90fc5
 F test/rollbackfault.test 0e646aeab8840c399cfbfa43daab46fd609cf04a
 F test/rowallock.test 3f88ec6819489d0b2341c7a7528ae17c053ab7cc
@@ -1233,7 +1234,7 @@ F test/superlock.test ec94f0556b6488d97f71c79f9061ae08d9ab8f12
 F test/symlink.test c9ebe7330d228249e447038276bfc8a7b22f4849
 F test/sync.test 2f84bdbc2b2df1fcb0220575b4b9f8cea94b7529
 F test/sync2.test 6be8ed007fa063b147773c1982b5bdba97a32badc536bdc6077eff5cf8710ece
-F test/syscall.test 7a60601770172a8014a4d222d5f3d95a5d2b5c47fbb0374e2698e89c99e37256
+F test/syscall.test a39d9a36f852ae6e4800f861bc2f2e83f68bbc2112d9399931ecfadeabd2d69d
 F test/sysfault.test c9f2b0d8d677558f74de750c75e12a5454719d04
 F test/tabfunc01.test c47171c36b3d411df2bd49719dcaa5d034f8d277477fd41d253940723b969a51
 F test/table.test b708f3e5fa2542fa51dfab21fc07b36ea445cb2f
@@ -1247,7 +1248,7 @@ F test/temptable.test d2c9b87a54147161bcd1822e30c1d1cd891e5b30
 F test/temptable2.test cd396beb41117a5302fff61767c35fa4270a0d5e
 F test/temptable3.test d11a0974e52b347e45ee54ef1923c91ed91e4637
 F test/temptrigger.test 38f0ca479b1822d3117069e014daabcaacefffcc
-F test/tester.tcl 581f0185434daf7026ccede4c07e8d1479186ec5
+F test/tester.tcl eb7ec55fe074a909423c1de701f7c545417b8aa96787b8c3e7a79203f2cebec8
 F test/thread001.test 9f22fd3525a307ff42a326b6bc7b0465be1745a5
 F test/thread002.test e630504f8a06c00bf8bbe68528774dd96aeb2e58
 F test/thread003.test ee4c9efc3b86a6a2767516a37bd64251272560a7
@@ -1543,6 +1544,7 @@ F test/without_rowid4.test 4e08bcbaee0399f35d58b5581881e7a6243d458a
 F test/without_rowid5.test 89b1c587bd92a0590e440da33e7666bf4891572a
 F test/without_rowid6.test 1f99644e6508447fb050f73697350c7ceca3392e
 F test/wordcount.c 06efb84b7c48a4973c2c24ea06c93d00bce24389
+F test/writecrash.test f1da7f7adfe8d7f09ea79b42e5ca6dcc41102f27f8e334ad71539501ddd910cc
 F test/zeroblob.test 3857870fe681b8185654414a9bccfde80b62a0fa
 F test/zerodamage.test e59a56443d6298ecf7435f618f0b27654f0c849e
 F tool/GetFile.cs a15e08acb5dd7539b75ba23501581d7c2b462cb5
@@ -1638,8 +1640,7 @@ F vsixtest/vsixtest.tcl 6a9a6ab600c25a91a7acc6293828957a386a8a93
 F vsixtest/vsixtest.vcxproj.data 2ed517e100c66dc455b492e1a33350c1b20fbcdc
 F vsixtest/vsixtest.vcxproj.filters 37e51ffedcdb064aad6ff33b6148725226cd608e
 F vsixtest/vsixtest_TemporaryKey.pfx e5b1b036facdb453873e7084e1cae9102ccc67a0
-P 0c77935cf9949099d834ec51384c1d4dcdaf7b4422c859c9fce6d3cb3bde2645 ac1fd6beb6c804af5faf1e06a51177a8316007ff9e718c398bd7a24d2ecc4ed3
+P 95e8f31658254dd2df3eeaae337aff0fe2125d170ae966c74f4fc70400e099b1 4c0520d4df7473eb4cc764774df7d99bb96cf067ac224755e09f0df47fb2a810
-R 0baf627cb47530fd453d60ae728f6e16
+R 785ca20553bf2e0c19860d1453d0c98e
-T +closed ac1fd6beb6c804af5faf1e06a51177a8316007ff9e718c398bd7a24d2ecc4ed3
 U drh
-Z 5c8b473184a93b34c81627bf252d49c0
+Z 2380a4cbd6dbbedaa37eda54fb01fd11

manifest.uuid

@@ -1 +1 @@
-95e8f31658254dd2df3eeaae337aff0fe2125d170ae966c74f4fc70400e099b1
+24190b221f73472dafaead6de101b4debc2c91c1ca28d70b45a38df5bb61fb39

src/ctime.c

@@ -184,6 +184,9 @@ static const char * const sqlite3azCompileOpt[] = {
 #if SQLITE_ENABLE_ATOMIC_WRITE
   "ENABLE_ATOMIC_WRITE",
 #endif
+#if SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+  "ENABLE_BATCH_ATOMIC_WRITE",
+#endif
 #if SQLITE_ENABLE_CEROD
   "ENABLE_CEROD",
 #endif

src/memjournal.c

@@ -96,7 +96,8 @@ static int memjrnlRead(
   int iChunkOffset;
   FileChunk *pChunk;
 
-#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \
+ || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
   if( (iAmt+iOfst)>p->endpoint.iOffset ){
     return SQLITE_IOERR_SHORT_READ;
   }
@@ -215,7 +216,8 @@ static int memjrnlWrite(
     ** atomic-write optimization. In this case the first 28 bytes of the
     ** journal file may be written as part of committing the transaction. */ 
     assert( iOfst==p->endpoint.iOffset || iOfst==0 );
-#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \
+ || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
     if( iOfst==0 && p->pFirst ){
       assert( p->nChunkSize>iAmt );
       memcpy((u8*)p->pFirst->zChunk, zBuf, iAmt);
@@ -384,17 +386,31 @@ void sqlite3MemJournalOpen(sqlite3_file *pJfd){
   sqlite3JournalOpen(0, 0, pJfd, 0, -1);
 }
 
-#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \
+ || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
 /*
 ** If the argument p points to a MemJournal structure that is not an 
 ** in-memory-only journal file (i.e. is one that was opened with a +ve
-** nSpill parameter), and the underlying file has not yet been created, 
+** nSpill parameter or as SQLITE_OPEN_MAIN_JOURNAL), and the underlying 
-** create it now.
+** file has not yet been created, create it now.
 */
-int sqlite3JournalCreate(sqlite3_file *p){
+int sqlite3JournalCreate(sqlite3_file *pJfd){
   int rc = SQLITE_OK;
-  if( p->pMethods==&MemJournalMethods && ((MemJournal*)p)->nSpill>0 ){
+  MemJournal *p = (MemJournal*)pJfd;
-    rc = memjrnlCreateFile((MemJournal*)p);
+  if( p->pMethod==&MemJournalMethods && (
+#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+     p->nSpill>0
+#else
+     /* While this appears to not be possible without ATOMIC_WRITE, the
+     ** paths are complex, so it seems prudent to leave the test in as
+     ** a NEVER(), in case our analysis is subtly flawed. */
+     NEVER(p->nSpill>0)
+#endif
+#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+     || (p->flags & SQLITE_OPEN_MAIN_JOURNAL)
+#endif
+  )){
+    rc = memjrnlCreateFile(p);
   }
   return rc;
 }

src/os_unix.c

@@ -90,6 +90,7 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
+#include <sys/ioctl.h>
 #include <unistd.h>
 #include <time.h>
 #include <sys/time.h>
@@ -220,10 +221,8 @@ struct unixFile {
   sqlite3_int64 mmapSizeMax;          /* Configured FCNTL_MMAP_SIZE value */
   void *pMapRegion;                   /* Memory mapped region */
 #endif
-#ifdef __QNXNTO__
   int sectorSize;                     /* Device sector size */
   int deviceCharacteristics;          /* Precomputed device characteristics */
-#endif
 #if SQLITE_ENABLE_LOCKING_STYLE
   int openFlags;                      /* The flags specified at open() */
 #endif
@@ -328,6 +327,20 @@ static pid_t randomnessPid = 0;
 # define lseek lseek64
 #endif
 
+#ifdef __linux__
+/*
+** Linux-specific IOCTL magic numbers used for controlling F2FS
+*/
+#define F2FS_IOCTL_MAGIC        0xf5
+#define F2FS_IOC_START_ATOMIC_WRITE     _IO(F2FS_IOCTL_MAGIC, 1)
+#define F2FS_IOC_COMMIT_ATOMIC_WRITE    _IO(F2FS_IOCTL_MAGIC, 2)
+#define F2FS_IOC_START_VOLATILE_WRITE   _IO(F2FS_IOCTL_MAGIC, 3)
+#define F2FS_IOC_ABORT_VOLATILE_WRITE   _IO(F2FS_IOCTL_MAGIC, 5)
+#define F2FS_IOC_GET_FEATURES           _IOR(F2FS_IOCTL_MAGIC, 12, u32)
+#define F2FS_FEATURE_ATOMIC_WRITE 0x0004
+#endif /* __linux__ */
+
+
 /*
 ** Different Unix systems declare open() in different ways.  Same use
 ** open(const char*,int,mode_t).  Others use open(const char*,int,...).
@@ -500,6 +513,9 @@ static struct unix_syscall {
 #endif
 #define osLstat      ((int(*)(const char*,struct stat*))aSyscall[27].pCurrent)
 
+  { "ioctl",         (sqlite3_syscall_ptr)ioctl,          0 },
+#define osIoctl ((int(*)(int,int,...))aSyscall[28].pCurrent)
+
 }; /* End of the overrideable system calls */
 
 
@@ -3777,6 +3793,21 @@ static int unixGetTempname(int nBuf, char *zBuf);
 static int unixFileControl(sqlite3_file *id, int op, void *pArg){
   unixFile *pFile = (unixFile*)id;
   switch( op ){
+#if defined(__linux__) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
+    case SQLITE_FCNTL_BEGIN_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_START_ATOMIC_WRITE);
+      return rc ? SQLITE_IOERR_BEGIN_ATOMIC : SQLITE_OK;
+    }
+    case SQLITE_FCNTL_COMMIT_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_COMMIT_ATOMIC_WRITE);
+      return rc ? SQLITE_IOERR_COMMIT_ATOMIC : SQLITE_OK;
+    }
+    case SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_ABORT_VOLATILE_WRITE);
+      return rc ? SQLITE_IOERR_ROLLBACK_ATOMIC : SQLITE_OK;
+    }
+#endif /* __linux__ && SQLITE_ENABLE_BATCH_ATOMIC_WRITE */
+
     case SQLITE_FCNTL_LOCKSTATE: {
       *(int*)pArg = pFile->eFileLock;
       return SQLITE_OK;
@@ -3860,30 +3891,41 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
 }
 
 /*
-** Return the sector size in bytes of the underlying block device for
+** If pFd->sectorSize is non-zero when this function is called, it is a
-** the specified file. This is almost always 512 bytes, but may be
+** no-op. Otherwise, the values of pFd->sectorSize and 
-** larger for some devices.
+** pFd->deviceCharacteristics are set according to the file-system 
+** characteristics. 
 **
-** SQLite code assumes this function cannot fail. It also assumes that
+** There are two versions of this function. One for QNX and one for all
-** if two files are created in the same file-system directory (i.e.
+** other systems.
-** a database and its journal file) that the sector size will be the
-** same for both.
 */
-#ifndef __QNXNTO__ 
+#ifndef __QNXNTO__
-static int unixSectorSize(sqlite3_file *NotUsed){
+static void setDeviceCharacteristics(unixFile *pFd){
-  UNUSED_PARAMETER(NotUsed);
+  assert( pFd->deviceCharacteristics==0 || pFd->sectorSize!=0 );
-  return SQLITE_DEFAULT_SECTOR_SIZE;
+  if( pFd->sectorSize==0 ){
-}
+#if defined(__linux__) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
-#endif
+    int res;
+    u32 f = 0;
 
-/*
+    /* Check for support for F2FS atomic batch writes. */
-** The following version of unixSectorSize() is optimized for QNX.
+    res = osIoctl(pFd->h, F2FS_IOC_GET_FEATURES, &f);
-*/
+    if( res==0 && (f & F2FS_FEATURE_ATOMIC_WRITE) ){
-#ifdef __QNXNTO__
+      pFd->deviceCharacteristics = SQLITE_IOCAP_BATCH_ATOMIC;
+    }
+#endif /* __linux__ && SQLITE_ENABLE_BATCH_ATOMIC_WRITE */
+
+    /* Set the POWERSAFE_OVERWRITE flag if requested. */
+    if( pFd->ctrlFlags & UNIXFILE_PSOW ){
+      pFd->deviceCharacteristics |= SQLITE_IOCAP_POWERSAFE_OVERWRITE;
+    }
+
+    pFd->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
+  }
+}
+#else
 #include <sys/dcmd_blk.h>
 #include <sys/statvfs.h>
-static int unixSectorSize(sqlite3_file *id){
+static void setDeviceCharacteristics(unixFile *pFile){
-  unixFile *pFile = (unixFile*)id;
   if( pFile->sectorSize == 0 ){
     struct statvfs fsInfo;
        
@@ -3952,9 +3994,24 @@ static int unixSectorSize(sqlite3_file *id){
     pFile->deviceCharacteristics = 0;
     pFile->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
   }
-  return pFile->sectorSize;
 }
-#endif /* __QNXNTO__ */
+#endif
+
+/*
+** Return the sector size in bytes of the underlying block device for
+** the specified file. This is almost always 512 bytes, but may be
+** larger for some devices.
+**
+** SQLite code assumes this function cannot fail. It also assumes that
+** if two files are created in the same file-system directory (i.e.
+** a database and its journal file) that the sector size will be the
+** same for both.
+*/
+static int unixSectorSize(sqlite3_file *id){
+  unixFile *pFd = (unixFile*)id;
+  setDeviceCharacteristics(pFd);
+  return pFd->sectorSize;
+}
 
 /*
 ** Return the device characteristics for the file.
@@ -3970,16 +4027,9 @@ static int unixSectorSize(sqlite3_file *id){
 ** available to turn it off and URI query parameter available to turn it off.
 */
 static int unixDeviceCharacteristics(sqlite3_file *id){
-  unixFile *p = (unixFile*)id;
+  unixFile *pFd = (unixFile*)id;
-  int rc = 0;
+  setDeviceCharacteristics(pFd);
-#ifdef __QNXNTO__
+  return pFd->deviceCharacteristics;
-  if( p->sectorSize==0 ) unixSectorSize(id);
-  rc = p->deviceCharacteristics;
-#endif
-  if( p->ctrlFlags & UNIXFILE_PSOW ){
-    rc |= SQLITE_IOCAP_POWERSAFE_OVERWRITE;
-  }
-  return rc;
 }
 
 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
@@ -7598,7 +7648,7 @@ int sqlite3_os_init(void){
 
   /* Double-check that the aSyscall[] array has been constructed
   ** correctly.  See ticket [bb3a86e890c8e96ab] */
-  assert( ArraySize(aSyscall)==28 );
+  assert( ArraySize(aSyscall)==29 );
 
   /* Register all VFSes defined in the aVfs[] array */
   for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){

src/pager.c

@@ -947,6 +947,7 @@ static int assert_pager_state(Pager *p){
       assert( isOpen(p->jfd) 
            || p->journalMode==PAGER_JOURNALMODE_OFF 
            || p->journalMode==PAGER_JOURNALMODE_WAL 
+           || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC)
       );
       assert( pPager->dbOrigSize<=pPager->dbHintSize );
       break;
@@ -958,6 +959,7 @@ static int assert_pager_state(Pager *p){
       assert( isOpen(p->jfd) 
            || p->journalMode==PAGER_JOURNALMODE_OFF 
            || p->journalMode==PAGER_JOURNALMODE_WAL 
+           || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC)
       );
       break;
 
@@ -1168,34 +1170,45 @@ static int pagerLockDb(Pager *pPager, int eLock){
 }
 
 /*
-** This function determines whether or not the atomic-write optimization
+** This function determines whether or not the atomic-write or
-** can be used with this pager. The optimization can be used if:
+** atomic-batch-write optimizations can be used with this pager. The
+** atomic-write optimization can be used if:
 **
 **  (a) the value returned by OsDeviceCharacteristics() indicates that
 **      a database page may be written atomically, and
 **  (b) the value returned by OsSectorSize() is less than or equal
 **      to the page size.
 **
-** The optimization is also always enabled for temporary files. It is
+** If it can be used, then the value returned is the size of the journal 
-** an error to call this function if pPager is opened on an in-memory
+** file when it contains rollback data for exactly one page.
-** database.
 **
-** If the optimization cannot be used, 0 is returned. If it can be used,
+** The atomic-batch-write optimization can be used if OsDeviceCharacteristics()
-** then the value returned is the size of the journal file when it
+** returns a value with the SQLITE_IOCAP_BATCH_ATOMIC bit set. -1 is
-** contains rollback data for exactly one page.
+** returned in this case.
+**
+** If neither optimization can be used, 0 is returned.
 */
-#ifdef SQLITE_ENABLE_ATOMIC_WRITE
 static int jrnlBufferSize(Pager *pPager){
   assert( !MEMDB );
-  if( !pPager->tempFile ){
-    int dc;                           /* Device characteristics */
-    int nSector;                      /* Sector size */
-    int szPage;                       /* Page size */
 
-    assert( isOpen(pPager->fd) );
+#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \
-    dc = sqlite3OsDeviceCharacteristics(pPager->fd);
+ || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
-    nSector = pPager->sectorSize;
+  int dc;                           /* Device characteristics */
-    szPage = pPager->pageSize;
+
+  assert( isOpen(pPager->fd) );
+  dc = sqlite3OsDeviceCharacteristics(pPager->fd);
+#endif
+
+#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+  if( dc&SQLITE_IOCAP_BATCH_ATOMIC ){
+    return -1;
+  }
+#endif
+
+#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+  {
+    int nSector = pPager->sectorSize;
+    int szPage = pPager->pageSize;
 
     assert(SQLITE_IOCAP_ATOMIC512==(512>>8));
     assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8));
@@ -1205,11 +1218,11 @@ static int jrnlBufferSize(Pager *pPager){
   }
 
   return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager);
-}
-#else
-# define jrnlBufferSize(x) 0
 #endif
 
+  return 0;
+}
+
 /*
 ** If SQLITE_CHECK_PAGES is defined then we do some sanity checking
 ** on the cache using a hash function.  This is used for testing
@@ -2012,7 +2025,9 @@ static int pager_end_transaction(Pager *pPager, int hasMaster, int bCommit){
   }
 
   releaseAllSavepoints(pPager);
-  assert( isOpen(pPager->jfd) || pPager->pInJournal==0 );
+  assert( isOpen(pPager->jfd) || pPager->pInJournal==0 
+      || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_BATCH_ATOMIC)
+  );
   if( isOpen(pPager->jfd) ){
     assert( !pagerUseWal(pPager) );
 
@@ -4567,6 +4582,13 @@ static int pagerStress(void *p, PgHdr *pPg){
       rc = pagerWalFrames(pPager, pPg, 0, 0);
     }
   }else{
+    
+#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+    if( pPager->tempFile==0 ){
+      rc = sqlite3JournalCreate(pPager->jfd);
+      if( rc!=SQLITE_OK ) return pager_error(pPager, rc);
+    }
+#endif
   
     /* Sync the journal file if required. */
     if( pPg->flags&PGHDR_NEED_SYNC 
@@ -6347,6 +6369,21 @@ int sqlite3PagerCommitPhaseOne(
         sqlite3PcacheCleanAll(pPager->pPCache);
       }
     }else{
+      /* The bBatch boolean is true if the batch-atomic-write commit method
+      ** should be used.  No rollback journal is created if batch-atomic-write
+      ** is enabled.
+      */
+      sqlite3_file *fd = pPager->fd;
+#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+      const int bBatch = zMaster==0    /* An SQLITE_IOCAP_BATCH_ATOMIC commit */
+        && (sqlite3OsDeviceCharacteristics(fd) & SQLITE_IOCAP_BATCH_ATOMIC)
+        && !pPager->noSync
+        && sqlite3JournalIsInMemory(pPager->jfd);
+#else
+# define bBatch 0
+#endif
+
+#ifdef SQLITE_ENABLE_ATOMIC_WRITE
       /* The following block updates the change-counter. Exactly how it
       ** does this depends on whether or not the atomic-update optimization
       ** was enabled at compile time, and if this transaction meets the 
@@ -6370,33 +6407,40 @@ int sqlite3PagerCommitPhaseOne(
       ** in 'direct' mode. In this case the journal file will never be
       ** created for this transaction.
       */
-  #ifdef SQLITE_ENABLE_ATOMIC_WRITE
+      if( bBatch==0 ){
-      PgHdr *pPg;
+        PgHdr *pPg;
-      assert( isOpen(pPager->jfd) 
+        assert( isOpen(pPager->jfd) 
-           || pPager->journalMode==PAGER_JOURNALMODE_OFF 
+            || pPager->journalMode==PAGER_JOURNALMODE_OFF 
-           || pPager->journalMode==PAGER_JOURNALMODE_WAL 
+            || pPager->journalMode==PAGER_JOURNALMODE_WAL 
-      );
+            );
-      if( !zMaster && isOpen(pPager->jfd) 
+        if( !zMaster && isOpen(pPager->jfd) 
-       && pPager->journalOff==jrnlBufferSize(pPager) 
+         && pPager->journalOff==jrnlBufferSize(pPager) 
-       && pPager->dbSize>=pPager->dbOrigSize
+         && pPager->dbSize>=pPager->dbOrigSize
-       && (0==(pPg = sqlite3PcacheDirtyList(pPager->pPCache)) || 0==pPg->pDirty)
+         && (!(pPg = sqlite3PcacheDirtyList(pPager->pPCache)) || 0==pPg->pDirty)
-      ){
+        ){
-        /* Update the db file change counter via the direct-write method. The 
+          /* Update the db file change counter via the direct-write method. The 
-        ** following call will modify the in-memory representation of page 1 
+          ** following call will modify the in-memory representation of page 1 
-        ** to include the updated change counter and then write page 1 
+          ** to include the updated change counter and then write page 1 
-        ** directly to the database file. Because of the atomic-write 
+          ** directly to the database file. Because of the atomic-write 
-        ** property of the host file-system, this is safe.
+          ** property of the host file-system, this is safe.
-        */
+          */
-        rc = pager_incr_changecounter(pPager, 1);
+          rc = pager_incr_changecounter(pPager, 1);
-      }else{
+        }else{
-        rc = sqlite3JournalCreate(pPager->jfd);
+          rc = sqlite3JournalCreate(pPager->jfd);
-        if( rc==SQLITE_OK ){
+          if( rc==SQLITE_OK ){
-          rc = pager_incr_changecounter(pPager, 0);
+            rc = pager_incr_changecounter(pPager, 0);
+          }
         }
       }
-  #else
+#else 
+#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+      if( zMaster ){
+        rc = sqlite3JournalCreate(pPager->jfd);
+        if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
+      }
+#endif
       rc = pager_incr_changecounter(pPager, 0);
-  #endif
+#endif
       if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
   
       /* Write the master journal name into the journal file. If a master 
@@ -6419,8 +6463,24 @@ int sqlite3PagerCommitPhaseOne(
       */
       rc = syncJournal(pPager, 0);
       if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
-  
+
+      if( bBatch ){
+        /* The pager is now in DBMOD state. But regardless of what happens
+        ** next, attempting to play the journal back into the database would
+        ** be unsafe. Close it now to make sure that does not happen.  */
+        sqlite3OsClose(pPager->jfd);
+        rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_BEGIN_ATOMIC_WRITE, 0);
+        if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
+      }
       rc = pager_write_pagelist(pPager,sqlite3PcacheDirtyList(pPager->pPCache));
+      if( bBatch ){
+        if( rc==SQLITE_OK ){
+          rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_COMMIT_ATOMIC_WRITE, 0);
+        }else{
+          sqlite3OsFileControl(fd, SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE, 0);
+        }
+      }
+
       if( rc!=SQLITE_OK ){
         assert( rc!=SQLITE_IOERR_BLOCKED );
         goto commit_phase_one_exit;

src/sqlite.h.in

@@ -494,6 +494,9 @@ int sqlite3_exec(
 #define SQLITE_IOERR_CONVPATH          (SQLITE_IOERR | (26<<8))
 #define SQLITE_IOERR_VNODE             (SQLITE_IOERR | (27<<8))
 #define SQLITE_IOERR_AUTH              (SQLITE_IOERR | (28<<8))
+#define SQLITE_IOERR_BEGIN_ATOMIC      (SQLITE_IOERR | (29<<8))
+#define SQLITE_IOERR_COMMIT_ATOMIC     (SQLITE_IOERR | (30<<8))
+#define SQLITE_IOERR_ROLLBACK_ATOMIC   (SQLITE_IOERR | (31<<8))
 #define SQLITE_LOCKED_SHAREDCACHE      (SQLITE_LOCKED |  (1<<8))
 #define SQLITE_BUSY_RECOVERY           (SQLITE_BUSY   |  (1<<8))
 #define SQLITE_BUSY_SNAPSHOT           (SQLITE_BUSY   |  (2<<8))
@@ -580,6 +583,11 @@ int sqlite3_exec(
 ** SQLITE_IOCAP_IMMUTABLE flag indicates that the file is on
 ** read-only media and cannot be changed even by processes with
 ** elevated privileges.
+**
+** The SQLITE_IOCAP_BATCH_ATOMIC property means that the underlying
+** filesystem supports doing multiple write operations atomically when those
+** write operations are bracketed by [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] and
+** [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE].
 */
 #define SQLITE_IOCAP_ATOMIC                 0x00000001
 #define SQLITE_IOCAP_ATOMIC512              0x00000002
@@ -595,6 +603,7 @@ int sqlite3_exec(
 #define SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN  0x00000800
 #define SQLITE_IOCAP_POWERSAFE_OVERWRITE    0x00001000
 #define SQLITE_IOCAP_IMMUTABLE              0x00002000
+#define SQLITE_IOCAP_BATCH_ATOMIC           0x00004000
 
 /*
 ** CAPI3REF: File Locking Levels
@@ -729,6 +738,7 @@ struct sqlite3_file {
 ** <li> [SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN]
 ** <li> [SQLITE_IOCAP_POWERSAFE_OVERWRITE]
 ** <li> [SQLITE_IOCAP_IMMUTABLE]
+** <li> [SQLITE_IOCAP_BATCH_ATOMIC]
 ** </ul>
 **
 ** The SQLITE_IOCAP_ATOMIC property means that all writes of
@@ -1012,6 +1022,40 @@ struct sqlite3_io_methods {
 ** The [SQLITE_FCNTL_RBU] opcode is implemented by the special VFS used by
 ** the RBU extension only.  All other VFS should return SQLITE_NOTFOUND for
 ** this opcode.  
+**
+** <li>[[SQLITE_FCNTL_BEGIN_ATOMIC_WRITE]]
+** If the [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] opcode returns SQLITE_OK, then
+** the file descriptor is placed in "batch write mode", which
+** means all subsequent write operations will be deferred and done
+** atomically at the next [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE].  Systems
+** that do not support batch atomic writes will return SQLITE_NOTFOUND.
+** ^Following a successful SQLITE_FCNTL_BEGIN_ATOMIC_WRITE and prior to
+** the closing [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE] or
+** [SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE], SQLite will make
+** no VFS interface calls on the same [sqlite3_file] file descriptor
+** except for calls to the xWrite method and the xFileControl method
+** with [SQLITE_FCNTL_SIZE_HINT].
+**
+** <li>[[SQLITE_FCNTL_COMMIT_ATOMIC_WRITE]]
+** The [SQLITE_FCNTL_COMMIT_ATOMIC_WRITE] opcode causes all write
+** operations since the previous successful call to 
+** [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] to be performed atomically.
+** This file control returns [SQLITE_OK] if and only if the writes were
+** all performed successfully and have been committed to persistent storage.
+** ^Regardless of whether or not it is successful, this file control takes
+** the file descriptor out of batch write mode so that all subsequent
+** write operations are independent.
+** ^SQLite will never invoke SQLITE_FCNTL_COMMIT_ATOMIC_WRITE without
+** a prior successful call to [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE].
+**
+** <li>[[SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE]]
+** The [SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE] opcode causes all write
+** operations since the previous successful call to 
+** [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE] to be rolled back.
+** ^This file control takes the file descriptor out of batch write mode
+** so that all subsequent write operations are independent.
+** ^SQLite will never invoke SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE without
+** a prior successful call to [SQLITE_FCNTL_BEGIN_ATOMIC_WRITE].
 ** </ul>
 */
 #define SQLITE_FCNTL_LOCKSTATE               1
@@ -1043,6 +1087,9 @@ struct sqlite3_io_methods {
 #define SQLITE_FCNTL_JOURNAL_POINTER        28
 #define SQLITE_FCNTL_WIN32_GET_HANDLE       29
 #define SQLITE_FCNTL_PDB                    30
+#define SQLITE_FCNTL_BEGIN_ATOMIC_WRITE     31
+#define SQLITE_FCNTL_COMMIT_ATOMIC_WRITE    32
+#define SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE  33
 
 /* deprecated names */
 #define SQLITE_GET_LOCKPROXYFILE      SQLITE_FCNTL_GET_LOCKPROXYFILE

src/sqliteInt.h

@@ -630,6 +630,15 @@
 # define SQLITE_DEFAULT_PCACHE_INITSZ 20
 #endif
 
+/*
+** The compile-time options SQLITE_MMAP_READWRITE and 
+** SQLITE_ENABLE_BATCH_ATOMIC_WRITE are not compatible with one another.
+** You must choose one or the other (or neither) but not both.
+*/
+#if defined(SQLITE_MMAP_READWRITE) && defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
+#error Cannot use both SQLITE_MMAP_READWRITE and SQLITE_ENABLE_BATCH_ATOMIC_WRITE
+#endif
+
 /*
 ** GCC does not define the offsetof() macro so we'll have to do it
 ** ourselves.
@@ -4287,7 +4296,8 @@ int sqlite3FindInIndex(Parse *, Expr *, u32, int*, int*);
 
 int sqlite3JournalOpen(sqlite3_vfs *, const char *, sqlite3_file *, int, int);
 int sqlite3JournalSize(sqlite3_vfs *);
-#ifdef SQLITE_ENABLE_ATOMIC_WRITE
+#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \
+ || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE)
   int sqlite3JournalCreate(sqlite3_file *);
 #endif
 

src/test1.c

@@ -2553,6 +2553,46 @@ static int SQLITE_TCLAPI test_delete_database(
   return TCL_OK;
 }
 
+/*
+** Usage: atomic_batch_write PATH
+*/
+static int SQLITE_TCLAPI test_atomic_batch_write(
+  void * clientData,
+  Tcl_Interp *interp,
+  int objc,
+  Tcl_Obj *CONST objv[]
+){
+  char *zFile = 0;                /* Path to file to test */
+  sqlite3 *db = 0;                /* Database handle */
+  sqlite3_file *pFd = 0;          /* SQLite fd open on zFile */
+  int bRes = 0;                   /* Integer result of this command */
+  int dc = 0;                     /* Device-characteristics mask */
+  int rc;                         /* sqlite3_open() return code */
+
+  if( objc!=2 ){
+    Tcl_WrongNumArgs(interp, 1, objv, "PATH");
+    return TCL_ERROR;
+  }
+  zFile = Tcl_GetString(objv[1]);
+
+  rc = sqlite3_open(zFile, &db);
+  if( rc!=SQLITE_OK ){
+    Tcl_AppendResult(interp, sqlite3_errmsg(db), 0);
+    sqlite3_close(db);
+    return TCL_ERROR;
+  }
+
+  rc = sqlite3_file_control(db, "main", SQLITE_FCNTL_FILE_POINTER, (void*)&pFd);
+  dc = pFd->pMethods->xDeviceCharacteristics(pFd);
+  if( dc & SQLITE_IOCAP_BATCH_ATOMIC ){
+    bRes = 1;
+  }
+
+  Tcl_SetObjResult(interp, Tcl_NewIntObj(bRes));
+  sqlite3_close(db);
+  return TCL_OK;
+}
+
 /*
 ** Usage:  sqlite3_next_stmt  DB  STMT
 **
@@ -7645,6 +7685,7 @@ int Sqlitetest1_Init(Tcl_Interp *interp){
      { "sqlite3_snapshot_cmp_blob", test_snapshot_cmp_blob, 0 },
 #endif
      { "sqlite3_delete_database", test_delete_database, 0 },
+     { "atomic_batch_write",      test_atomic_batch_write,     0   },
   };
   static int bitmask_size = sizeof(Bitmask)*8;
   static int longdouble_size = sizeof(LONGDOUBLE_TYPE);

src/test6.c

@@ -736,6 +736,7 @@ static int processDevSymArgs(
     { "sequential",          SQLITE_IOCAP_SEQUENTIAL            },
     { "safe_append",         SQLITE_IOCAP_SAFE_APPEND           },
     { "powersafe_overwrite", SQLITE_IOCAP_POWERSAFE_OVERWRITE   },
+    { "batch-atomic",        SQLITE_IOCAP_BATCH_ATOMIC          },
     { 0, 0 }
   };
 
@@ -976,7 +977,30 @@ static int SQLITE_TCLAPI devSymObjCmd(
   devsym_register(iDc, iSectorSize);
 
   return TCL_OK;
+}
 
+/*
+** tclcmd: sqlite3_crash_on_write N
+*/
+static int SQLITE_TCLAPI writeCrashObjCmd(
+  void * clientData,
+  Tcl_Interp *interp,
+  int objc,
+  Tcl_Obj *CONST objv[]
+){
+  void devsym_crash_on_write(int);
+  int nWrite = 0;
+
+  if( objc!=2 ){
+    Tcl_WrongNumArgs(interp, 1, objv, "NWRITE");
+    return TCL_ERROR;
+  }
+  if( Tcl_GetIntFromObj(interp, objv[1], &nWrite) ){
+    return TCL_ERROR;
+  }
+
+  devsym_crash_on_write(nWrite);
+  return TCL_OK;
 }
 
 /*
@@ -1068,6 +1092,7 @@ int Sqlitetest6_Init(Tcl_Interp *interp){
   Tcl_CreateObjCommand(interp, "sqlite3_crashparams", crashParamsObjCmd, 0, 0);
   Tcl_CreateObjCommand(interp, "sqlite3_crash_now", crashNowCmd, 0, 0);
   Tcl_CreateObjCommand(interp, "sqlite3_simulate_device", devSymObjCmd, 0, 0);
+  Tcl_CreateObjCommand(interp, "sqlite3_crash_on_write", writeCrashObjCmd,0,0);
   Tcl_CreateObjCommand(interp, "unregister_devsim", dsUnregisterObjCmd, 0, 0);
   Tcl_CreateObjCommand(interp, "register_jt_vfs", jtObjCmd, 0, 0);
   Tcl_CreateObjCommand(interp, "unregister_jt_vfs", jtUnregisterObjCmd, 0, 0);

src/test_devsym.c

@@ -28,6 +28,7 @@
 ** Name used to identify this VFS.
 */
 #define DEVSYM_VFS_NAME "devsym"
+#define WRITECRASH_NAME "writecrash"
 
 typedef struct devsym_file devsym_file;
 struct devsym_file {
@@ -72,61 +73,13 @@ static int devsymRandomness(sqlite3_vfs*, int nByte, char *zOut);
 static int devsymSleep(sqlite3_vfs*, int microseconds);
 static int devsymCurrentTime(sqlite3_vfs*, double*);
 
-static sqlite3_vfs devsym_vfs = {
-  2,                     /* iVersion */
-  sizeof(devsym_file),      /* szOsFile */
-  DEVSYM_MAX_PATHNAME,      /* mxPathname */
-  0,                     /* pNext */
-  DEVSYM_VFS_NAME,          /* zName */
-  0,                     /* pAppData */
-  devsymOpen,               /* xOpen */
-  devsymDelete,             /* xDelete */
-  devsymAccess,             /* xAccess */
-  devsymFullPathname,       /* xFullPathname */
-#ifndef SQLITE_OMIT_LOAD_EXTENSION
-  devsymDlOpen,             /* xDlOpen */
-  devsymDlError,            /* xDlError */
-  devsymDlSym,              /* xDlSym */
-  devsymDlClose,            /* xDlClose */
-#else
-  0,                        /* xDlOpen */
-  0,                        /* xDlError */
-  0,                        /* xDlSym */
-  0,                        /* xDlClose */
-#endif /* SQLITE_OMIT_LOAD_EXTENSION */
-  devsymRandomness,         /* xRandomness */
-  devsymSleep,              /* xSleep */
-  devsymCurrentTime,        /* xCurrentTime */
-  0,                        /* xGetLastError */
-  0                         /* xCurrentTimeInt64 */
-};
-
-static sqlite3_io_methods devsym_io_methods = {
-  2,                                /* iVersion */
-  devsymClose,                      /* xClose */
-  devsymRead,                       /* xRead */
-  devsymWrite,                      /* xWrite */
-  devsymTruncate,                   /* xTruncate */
-  devsymSync,                       /* xSync */
-  devsymFileSize,                   /* xFileSize */
-  devsymLock,                       /* xLock */
-  devsymUnlock,                     /* xUnlock */
-  devsymCheckReservedLock,          /* xCheckReservedLock */
-  devsymFileControl,                /* xFileControl */
-  devsymSectorSize,                 /* xSectorSize */
-  devsymDeviceCharacteristics,      /* xDeviceCharacteristics */
-  devsymShmMap,                     /* xShmMap */
-  devsymShmLock,                    /* xShmLock */
-  devsymShmBarrier,                 /* xShmBarrier */
-  devsymShmUnmap                    /* xShmUnmap */
-};
-
 struct DevsymGlobal {
   sqlite3_vfs *pVfs;
   int iDeviceChar;
   int iSectorSize;
+  int nWriteCrash;
 };
-struct DevsymGlobal g = {0, 0, 512};
+struct DevsymGlobal g = {0, 0, 512, 0};
 
 /*
 ** Close an devsym-file.
@@ -271,6 +224,26 @@ static int devsymOpen(
   int flags,
   int *pOutFlags
 ){
+static sqlite3_io_methods devsym_io_methods = {
+  2,                                /* iVersion */
+  devsymClose,                      /* xClose */
+  devsymRead,                       /* xRead */
+  devsymWrite,                      /* xWrite */
+  devsymTruncate,                   /* xTruncate */
+  devsymSync,                       /* xSync */
+  devsymFileSize,                   /* xFileSize */
+  devsymLock,                       /* xLock */
+  devsymUnlock,                     /* xUnlock */
+  devsymCheckReservedLock,          /* xCheckReservedLock */
+  devsymFileControl,                /* xFileControl */
+  devsymSectorSize,                 /* xSectorSize */
+  devsymDeviceCharacteristics,      /* xDeviceCharacteristics */
+  devsymShmMap,                     /* xShmMap */
+  devsymShmLock,                    /* xShmLock */
+  devsymShmBarrier,                 /* xShmBarrier */
+  devsymShmUnmap                    /* xShmUnmap */
+};
+
   int rc;
   devsym_file *p = (devsym_file *)pFile;
   p->pReal = (sqlite3_file *)&p[1];
@@ -372,6 +345,137 @@ static int devsymCurrentTime(sqlite3_vfs *pVfs, double *pTimeOut){
   return g.pVfs->xCurrentTime(g.pVfs, pTimeOut);
 }
 
+/*
+** Return the sector-size in bytes for an writecrash-file.
+*/
+static int writecrashSectorSize(sqlite3_file *pFile){
+  devsym_file *p = (devsym_file *)pFile;
+  return sqlite3OsSectorSize(p->pReal);
+}
+
+/*
+** Return the device characteristic flags supported by an writecrash-file.
+*/
+static int writecrashDeviceCharacteristics(sqlite3_file *pFile){
+  devsym_file *p = (devsym_file *)pFile;
+  return sqlite3OsDeviceCharacteristics(p->pReal);
+}
+
+/*
+** Write data to an writecrash-file.
+*/
+static int writecrashWrite(
+  sqlite3_file *pFile, 
+  const void *zBuf, 
+  int iAmt, 
+  sqlite_int64 iOfst
+){
+  devsym_file *p = (devsym_file *)pFile;
+  if( g.nWriteCrash>0 ){
+    g.nWriteCrash--;
+    if( g.nWriteCrash==0 ) abort();
+  }
+  return sqlite3OsWrite(p->pReal, zBuf, iAmt, iOfst);
+}
+
+/*
+** Open an writecrash file handle.
+*/
+static int writecrashOpen(
+  sqlite3_vfs *pVfs,
+  const char *zName,
+  sqlite3_file *pFile,
+  int flags,
+  int *pOutFlags
+){
+static sqlite3_io_methods writecrash_io_methods = {
+  2,                                /* iVersion */
+  devsymClose,                      /* xClose */
+  devsymRead,                       /* xRead */
+  writecrashWrite,                  /* xWrite */
+  devsymTruncate,                   /* xTruncate */
+  devsymSync,                       /* xSync */
+  devsymFileSize,                   /* xFileSize */
+  devsymLock,                       /* xLock */
+  devsymUnlock,                     /* xUnlock */
+  devsymCheckReservedLock,          /* xCheckReservedLock */
+  devsymFileControl,                /* xFileControl */
+  writecrashSectorSize,             /* xSectorSize */
+  writecrashDeviceCharacteristics,  /* xDeviceCharacteristics */
+  devsymShmMap,                     /* xShmMap */
+  devsymShmLock,                    /* xShmLock */
+  devsymShmBarrier,                 /* xShmBarrier */
+  devsymShmUnmap                    /* xShmUnmap */
+};
+
+  int rc;
+  devsym_file *p = (devsym_file *)pFile;
+  p->pReal = (sqlite3_file *)&p[1];
+  rc = sqlite3OsOpen(g.pVfs, zName, p->pReal, flags, pOutFlags);
+  if( p->pReal->pMethods ){
+    pFile->pMethods = &writecrash_io_methods;
+  }
+  return rc;
+}
+
+static sqlite3_vfs devsym_vfs = {
+  2,                     /* iVersion */
+  sizeof(devsym_file),      /* szOsFile */
+  DEVSYM_MAX_PATHNAME,      /* mxPathname */
+  0,                     /* pNext */
+  DEVSYM_VFS_NAME,          /* zName */
+  0,                     /* pAppData */
+  devsymOpen,               /* xOpen */
+  devsymDelete,             /* xDelete */
+  devsymAccess,             /* xAccess */
+  devsymFullPathname,       /* xFullPathname */
+#ifndef SQLITE_OMIT_LOAD_EXTENSION
+  devsymDlOpen,             /* xDlOpen */
+  devsymDlError,            /* xDlError */
+  devsymDlSym,              /* xDlSym */
+  devsymDlClose,            /* xDlClose */
+#else
+  0,                        /* xDlOpen */
+  0,                        /* xDlError */
+  0,                        /* xDlSym */
+  0,                        /* xDlClose */
+#endif /* SQLITE_OMIT_LOAD_EXTENSION */
+  devsymRandomness,         /* xRandomness */
+  devsymSleep,              /* xSleep */
+  devsymCurrentTime,        /* xCurrentTime */
+  0,                        /* xGetLastError */
+  0                         /* xCurrentTimeInt64 */
+};
+
+static sqlite3_vfs writecrash_vfs = {
+  2,                     /* iVersion */
+  sizeof(devsym_file),      /* szOsFile */
+  DEVSYM_MAX_PATHNAME,      /* mxPathname */
+  0,                     /* pNext */
+  WRITECRASH_NAME,          /* zName */
+  0,                     /* pAppData */
+  writecrashOpen,           /* xOpen */
+  devsymDelete,             /* xDelete */
+  devsymAccess,             /* xAccess */
+  devsymFullPathname,       /* xFullPathname */
+#ifndef SQLITE_OMIT_LOAD_EXTENSION
+  devsymDlOpen,             /* xDlOpen */
+  devsymDlError,            /* xDlError */
+  devsymDlSym,              /* xDlSym */
+  devsymDlClose,            /* xDlClose */
+#else
+  0,                        /* xDlOpen */
+  0,                        /* xDlError */
+  0,                        /* xDlSym */
+  0,                        /* xDlClose */
+#endif /* SQLITE_OMIT_LOAD_EXTENSION */
+  devsymRandomness,         /* xRandomness */
+  devsymSleep,              /* xSleep */
+  devsymCurrentTime,        /* xCurrentTime */
+  0,                        /* xGetLastError */
+  0                         /* xCurrentTimeInt64 */
+};
+
 
 /*
 ** This procedure registers the devsym vfs with SQLite. If the argument is
@@ -379,10 +483,13 @@ static int devsymCurrentTime(sqlite3_vfs *pVfs, double *pTimeOut){
 ** available function in this file.
 */
 void devsym_register(int iDeviceChar, int iSectorSize){
+
   if( g.pVfs==0 ){
     g.pVfs = sqlite3_vfs_find(0);
     devsym_vfs.szOsFile += g.pVfs->szOsFile;
+    writecrash_vfs.szOsFile += g.pVfs->szOsFile;
     sqlite3_vfs_register(&devsym_vfs, 0);
+    sqlite3_vfs_register(&writecrash_vfs, 0);
   }
   if( iDeviceChar>=0 ){
     g.iDeviceChar = iDeviceChar;
@@ -403,4 +510,15 @@ void devsym_unregister(){
   g.iSectorSize = 0;
 }
 
+void devsym_crash_on_write(int nWrite){
+  if( g.pVfs==0 ){
+    g.pVfs = sqlite3_vfs_find(0);
+    devsym_vfs.szOsFile += g.pVfs->szOsFile;
+    writecrash_vfs.szOsFile += g.pVfs->szOsFile;
+    sqlite3_vfs_register(&devsym_vfs, 0);
+    sqlite3_vfs_register(&writecrash_vfs, 0);
+  }
+  g.nWriteCrash = nWrite;
+}
+
 #endif

test/atomic.test

@@ -0,0 +1,41 @@
+# 2015-11-07
+#
+# The author disclaims copyright to this source code.  In place of
+# a legal notice, here is a blessing:
+#
+#    May you do good and not evil.
+#    May you find forgiveness for yourself and forgive others.
+#    May you share freely, never taking more than you give.
+#
+#***********************************************************************
+# This file implements regression tests for SQLite library.  The
+# focus of this file is testing the WITH clause.
+#
+
+set testdir [file dirname $argv0]
+source $testdir/tester.tcl
+set ::testprefix atomic
+
+db close
+if {[atomic_batch_write test.db]==0} {
+  puts "No f2fs atomic-batch-write support. Skipping tests..."
+  finish_test
+  return
+}
+
+reset_db
+
+do_execsql_test 1.0 {
+  CREATE TABLE t1(x, y);
+  BEGIN;
+    INSERT INTO t1 VALUES(1, 2);
+}
+
+do_test 1.1 { file exists test.db-journal } {0}
+
+do_execsql_test 1.2 {
+  COMMIT;
+}
+
+
+finish_test

test/fallocate.test

@@ -59,7 +59,9 @@ do_test fallocate-1.6 {
 #
 do_test fallocate-1.7 {
   execsql { BEGIN; INSERT INTO t1 VALUES(1, 2); }
-  if {[permutation] != "inmemory_journal"} {
+  if {[permutation] != "inmemory_journal"
+   && [permutation] != "atomic-batch-write"
+  } {
     hexio_get_int [hexio_read test.db-journal 16 4]
   } else {
     set {} 1024

test/misc1.test

@@ -479,26 +479,28 @@ ifcapable curdir {
 # Make sure a database connection still works after changing the
 # working directory.
 #
-do_test misc1-14.1 {
+if {[atomic_batch_write test.db]==0} {
-  file mkdir tempdir
+  do_test misc1-14.1 {
-  cd tempdir
+    file mkdir tempdir
-  execsql {BEGIN}
+    cd tempdir
-  file exists ./test.db-journal
+    execsql {BEGIN}
-} {0}
+    file exists ./test.db-journal
-do_test misc1-14.2a {
+  } {0}
-  execsql {UPDATE t1 SET a=a||'x' WHERE 0}
+  do_test misc1-14.2a {
-  file exists ../test.db-journal
+    execsql {UPDATE t1 SET a=a||'x' WHERE 0}
-} {0}
+    file exists ../test.db-journal
-do_test misc1-14.2b {
+  } {0}
-  execsql {UPDATE t1 SET a=a||'y' WHERE 1}
+  do_test misc1-14.2b {
-  file exists ../test.db-journal
+    execsql {UPDATE t1 SET a=a||'y' WHERE 1}
-} {1}
+    file exists ../test.db-journal
-do_test misc1-14.3 {
+  } {1}
-  cd ..
+  do_test misc1-14.3 {
-  forcedelete tempdir
+    cd ..
-  execsql {COMMIT}
+    forcedelete tempdir
-  file exists ./test.db-journal
+    execsql {COMMIT}
-} {0}
+    file exists ./test.db-journal
+  } {0}
+}
 }
 
 # A failed create table should not leave the table in the internal

test/permutations.test

@@ -389,6 +389,30 @@ test_suite "vfslog" -prefix "" -description {
   wal* mmap*
 ]
 
+test_suite "atomic-batch-write" -prefix "" -description {
+  Like veryquick.test, but must be run on a file-system that supports
+  atomic-batch-writes. Tests that depend on the journal file being present
+  are omitted.
+} -files [
+  test_set $allquicktests -exclude *malloc* *ioerr* *fault* *bigfile* *_err* \
+      *fts5corrupt* *fts5big* *fts5aj*  \
+      crash8.test delete_db.test        \
+      exclusive.test journal3.test      \
+      journal1.test                     \
+      jrnlmode.test jrnlmode2.test      \
+      lock4.test pager1.test            \
+      pager3.test sharedA.test          \
+      symlink.test stmt.test            \
+      sync.test sync2.test              \
+      tempdb.test tkt3457.test          \
+      vacuum5.test wal2.test            \
+      walmode.test zerodamage.test
+] -initialize {
+  if {[atomic_batch_write test.db]==0} {
+    error "File system does NOT support atomic-batch-write"
+  }
+}
+
 lappend ::testsuitelist xxx
 #-------------------------------------------------------------------------
 # Define the coverage related test suites:

test/rollback.test

@@ -82,6 +82,7 @@ do_test rollback-1.9 {
 if {$tcl_platform(platform) == "unix" 
  && [permutation] ne "onefile"
  && [permutation] ne "inmemory_journal"
+ && [permutation] ne "atomic-batch-write"
 } {
   do_test rollback-2.1 {
     execsql {

test/syscall.test

@@ -61,7 +61,7 @@ foreach s {
     fcntl read pread write pwrite fchmod fallocate
     pread64 pwrite64 unlink openDirectory mkdir rmdir 
     statvfs fchown geteuid umask mmap munmap mremap
-    getpagesize readlink lstat
+    getpagesize readlink lstat ioctl
 } {
   if {[test_syscall exists $s]} {lappend syscall_list $s}
 }

test/tester.tcl

@@ -1608,6 +1608,54 @@ proc crashsql {args} {
   lappend r $msg
 }
 
+#   crash_on_write ?-devchar DEVCHAR? CRASHDELAY SQL
+#
+proc crash_on_write {args} {
+
+  set nArg [llength $args]
+  if {$nArg<2 || $nArg%2} {
+    error "bad args: $args"
+  }
+  set zSql [lindex $args end]
+  set nDelay [lindex $args end-1]
+
+  set devchar {}
+  for {set ii 0} {$ii < $nArg-2} {incr ii 2} {
+    set opt [lindex $args $ii]
+    switch -- [lindex $args $ii] {
+      -devchar {
+        set devchar [lindex $args [expr $ii+1]]
+      }
+
+      default { error "unrecognized option: $opt" }
+    }
+  }
+
+  set f [open crash.tcl w]
+  puts $f "sqlite3_crash_on_write $nDelay"
+  puts $f "sqlite3_test_control_pending_byte $::sqlite_pending_byte"
+  puts $f "sqlite3 db test.db -vfs writecrash"
+  puts $f "db eval {$zSql}"
+  puts $f "set {} {}"
+
+  close $f
+  set r [catch {
+    exec [info nameofexec] crash.tcl >@stdout
+  } msg]
+
+  # Windows/ActiveState TCL returns a slightly different
+  # error message.  We map that to the expected message
+  # so that we don't have to change all of the test
+  # cases.
+  if {$::tcl_platform(platform)=="windows"} {
+    if {$msg=="child killed: unknown signal"} {
+      set msg "child process exited abnormally"
+    }
+  }
+
+  lappend r $msg
+}
+
 proc run_ioerr_prep {} {
   set ::sqlite_io_error_pending 0
   catch {db close}

test/writecrash.test

@@ -0,0 +1,68 @@
+# 2009 January 8
+#
+# The author disclaims copyright to this source code.  In place of
+# a legal notice, here is a blessing:
+#
+#    May you do good and not evil.
+#    May you find forgiveness for yourself and forgive others.
+#    May you share freely, never taking more than you give.
+#
+#***********************************************************************
+#
+# Test the outcome of a writer crashing within a call to the VFS
+# xWrite function.
+#
+
+
+set testdir [file dirname $argv0]
+source $testdir/tester.tcl
+set testprefix writecrash
+
+do_not_use_codec
+
+
+if {$tcl_platform(platform)=="windows"} {
+  finish_test
+  return
+}
+
+do_execsql_test 1.0 {
+  CREATE TABLE t1(a INTEGER PRIMARY KEY, b BLOB UNIQUE);
+  WITH s(i) AS (
+    VALUES(1) UNION ALL SELECT i+1 FROM s WHERE i<100
+  )
+  INSERT INTO t1 SELECT NULL, randomblob(900) FROM s;
+} {}
+
+set bGo 1
+for {set tn 1} {$bGo} {incr tn} {
+
+db close
+sqlite3 db test.db
+
+  do_test 1.$tn.1 {
+    set res [crash_on_write $tn {
+      UPDATE t1 SET b = randomblob(899) WHERE (a%3)==0
+    }]
+    set bGo 0
+    if {[string match {1 {child killed:*}} $res]} {
+      set res {0 {}}
+      set bGo 1
+    }
+    set res
+  } {0 {}}
+
+#db close
+#sqlite3 db test.db
+
+  do_execsql_test 1.$tn.2 { PRAGMA integrity_check } {ok}
+
+db close
+sqlite3 db test.db
+
+  do_execsql_test 1.$tn.3 { PRAGMA integrity_check } {ok}
+}
+
+
+
+finish_test