| 1 |
Backport of upstream commit db7d62c8d5: |
| 2 |
|
| 3 |
Avoid attempting to mmap memory from an offset that is not a multiple of |
| 4 |
the system page size on systems with page sizes larger than 32KB. |
| 5 |
|
| 6 |
https://www.sqlite.org/src/info/db7d62c8d58eb1e8654a762c9b199ae4e2759038 |
| 7 |
|
| 8 |
Index: src/os_unix.c |
| 9 |
================================================================== |
| 10 |
--- src/os_unix.c |
| 11 |
+++ src/os_unix.c |
| 12 |
@@ -321,10 +321,11 @@ |
| 13 |
return geteuid() ? 0 : fchown(fd,uid,gid); |
| 14 |
} |
| 15 |
|
| 16 |
/* Forward reference */ |
| 17 |
static int openDirectory(const char*, int*); |
| 18 |
+static int unixGetpagesize(void); |
| 19 |
|
| 20 |
/* |
| 21 |
** Many system calls are accessed through pointer-to-functions so that |
| 22 |
** they may be overridden at runtime to facilitate fault injection during |
| 23 |
** testing and sandboxing. The following array holds the names and pointers |
| 24 |
@@ -443,10 +444,13 @@ |
| 25 |
{ "mremap", (sqlite3_syscall_ptr)mremap, 0 }, |
| 26 |
#else |
| 27 |
{ "mremap", (sqlite3_syscall_ptr)0, 0 }, |
| 28 |
#endif |
| 29 |
#define osMremap ((void*(*)(void*,size_t,size_t,int,...))aSyscall[23].pCurrent) |
| 30 |
+ |
| 31 |
+ { "getpagesize", (sqlite3_syscall_ptr)unixGetpagesize, 0 }, |
| 32 |
+#define osGetpagesize ((int(*)(void))aSyscall[24].pCurrent) |
| 33 |
|
| 34 |
}; /* End of the overrideable system calls */ |
| 35 |
|
| 36 |
/* |
| 37 |
** This is the xSetSystemCall() method of sqlite3_vfs for all of the |
| 38 |
@@ -4103,10 +4107,40 @@ |
| 39 |
#endif |
| 40 |
|
| 41 |
return rc; |
| 42 |
} |
| 43 |
|
| 44 |
+/* |
| 45 |
+** Return the system page size. |
| 46 |
+** |
| 47 |
+** This function should not be called directly by other code in this file. |
| 48 |
+** Instead, it should be called via macro osGetpagesize(). |
| 49 |
+*/ |
| 50 |
+static int unixGetpagesize(void){ |
| 51 |
+#if defined(_BSD_SOURCE) |
| 52 |
+ return getpagesize(); |
| 53 |
+#else |
| 54 |
+ return (int)sysconf(_SC_PAGESIZE); |
| 55 |
+#endif |
| 56 |
+} |
| 57 |
+ |
| 58 |
+/* |
| 59 |
+** Return the minimum number of 32KB shm regions that should be mapped at |
| 60 |
+** a time, assuming that each mapping must be an integer multiple of the |
| 61 |
+** current system page-size. |
| 62 |
+** |
| 63 |
+** Usually, this is 1. The exception seems to be systems that are configured |
| 64 |
+** to use 64KB pages - in this case each mapping must cover at least two |
| 65 |
+** shm regions. |
| 66 |
+*/ |
| 67 |
+static int unixShmRegionPerMap(void){ |
| 68 |
+ int shmsz = 32*1024; /* SHM region size */ |
| 69 |
+ int pgsz = osGetpagesize(); /* System page size */ |
| 70 |
+ assert( ((pgsz-1)&pgsz)==0 ); /* Page size must be a power of 2 */ |
| 71 |
+ if( pgsz<shmsz ) return 1; |
| 72 |
+ return pgsz/shmsz; |
| 73 |
+} |
| 74 |
|
| 75 |
/* |
| 76 |
** Purge the unixShmNodeList list of all entries with unixShmNode.nRef==0. |
| 77 |
** |
| 78 |
** This is not a VFS shared-memory method; it is a utility function called |
| 79 |
@@ -4114,14 +4148,15 @@ |
| 80 |
*/ |
| 81 |
static void unixShmPurge(unixFile *pFd){ |
| 82 |
unixShmNode *p = pFd->pInode->pShmNode; |
| 83 |
assert( unixMutexHeld() ); |
| 84 |
if( p && p->nRef==0 ){ |
| 85 |
+ int nShmPerMap = unixShmRegionPerMap(); |
| 86 |
int i; |
| 87 |
assert( p->pInode==pFd->pInode ); |
| 88 |
sqlite3_mutex_free(p->mutex); |
| 89 |
- for(i=0; i<p->nRegion; i++){ |
| 90 |
+ for(i=0; i<p->nRegion; i+=nShmPerMap){ |
| 91 |
if( p->h>=0 ){ |
| 92 |
osMunmap(p->apRegion[i], p->szRegion); |
| 93 |
}else{ |
| 94 |
sqlite3_free(p->apRegion[i]); |
| 95 |
} |
| 96 |
@@ -4324,10 +4359,12 @@ |
| 97 |
){ |
| 98 |
unixFile *pDbFd = (unixFile*)fd; |
| 99 |
unixShm *p; |
| 100 |
unixShmNode *pShmNode; |
| 101 |
int rc = SQLITE_OK; |
| 102 |
+ int nShmPerMap = unixShmRegionPerMap(); |
| 103 |
+ int nReqRegion; |
| 104 |
|
| 105 |
/* If the shared-memory file has not yet been opened, open it now. */ |
| 106 |
if( pDbFd->pShm==0 ){ |
| 107 |
rc = unixOpenSharedMemory(pDbFd); |
| 108 |
if( rc!=SQLITE_OK ) return rc; |
| 109 |
@@ -4339,13 +4376,16 @@ |
| 110 |
assert( szRegion==pShmNode->szRegion || pShmNode->nRegion==0 ); |
| 111 |
assert( pShmNode->pInode==pDbFd->pInode ); |
| 112 |
assert( pShmNode->h>=0 || pDbFd->pInode->bProcessLock==1 ); |
| 113 |
assert( pShmNode->h<0 || pDbFd->pInode->bProcessLock==0 ); |
| 114 |
|
| 115 |
- if( pShmNode->nRegion<=iRegion ){ |
| 116 |
+ /* Minimum number of regions required to be mapped. */ |
| 117 |
+ nReqRegion = ((iRegion+nShmPerMap) / nShmPerMap) * nShmPerMap; |
| 118 |
+ |
| 119 |
+ if( pShmNode->nRegion<nReqRegion ){ |
| 120 |
char **apNew; /* New apRegion[] array */ |
| 121 |
- int nByte = (iRegion+1)*szRegion; /* Minimum required file size */ |
| 122 |
+ int nByte = nReqRegion*szRegion; /* Minimum required file size */ |
| 123 |
struct stat sStat; /* Used by fstat() */ |
| 124 |
|
| 125 |
pShmNode->szRegion = szRegion; |
| 126 |
|
| 127 |
if( pShmNode->h>=0 ){ |
| 128 |
@@ -4390,21 +4430,23 @@ |
| 129 |
} |
| 130 |
} |
| 131 |
|
| 132 |
/* Map the requested memory region into this processes address space. */ |
| 133 |
apNew = (char **)sqlite3_realloc( |
| 134 |
- pShmNode->apRegion, (iRegion+1)*sizeof(char *) |
| 135 |
+ pShmNode->apRegion, nReqRegion*sizeof(char *) |
| 136 |
); |
| 137 |
if( !apNew ){ |
| 138 |
rc = SQLITE_IOERR_NOMEM; |
| 139 |
goto shmpage_out; |
| 140 |
} |
| 141 |
pShmNode->apRegion = apNew; |
| 142 |
- while(pShmNode->nRegion<=iRegion){ |
| 143 |
+ while( pShmNode->nRegion<nReqRegion ){ |
| 144 |
+ int nMap = szRegion*nShmPerMap; |
| 145 |
+ int i; |
| 146 |
void *pMem; |
| 147 |
if( pShmNode->h>=0 ){ |
| 148 |
- pMem = osMmap(0, szRegion, |
| 149 |
+ pMem = osMmap(0, nMap, |
| 150 |
pShmNode->isReadonly ? PROT_READ : PROT_READ|PROT_WRITE, |
| 151 |
MAP_SHARED, pShmNode->h, szRegion*(i64)pShmNode->nRegion |
| 152 |
); |
| 153 |
if( pMem==MAP_FAILED ){ |
| 154 |
rc = unixLogError(SQLITE_IOERR_SHMMAP, "mmap", pShmNode->zFilename); |
| 155 |
@@ -4416,12 +4458,15 @@ |
| 156 |
rc = SQLITE_NOMEM; |
| 157 |
goto shmpage_out; |
| 158 |
} |
| 159 |
memset(pMem, 0, szRegion); |
| 160 |
} |
| 161 |
- pShmNode->apRegion[pShmNode->nRegion] = pMem; |
| 162 |
- pShmNode->nRegion++; |
| 163 |
+ |
| 164 |
+ for(i=0; i<nShmPerMap; i++){ |
| 165 |
+ pShmNode->apRegion[pShmNode->nRegion+i] = &((char*)pMem)[szRegion*i]; |
| 166 |
+ } |
| 167 |
+ pShmNode->nRegion += nShmPerMap; |
| 168 |
} |
| 169 |
} |
| 170 |
|
| 171 |
shmpage_out: |
| 172 |
if( pShmNode->nRegion>iRegion ){ |
| 173 |
@@ -4631,25 +4676,10 @@ |
| 174 |
#endif |
| 175 |
} |
| 176 |
|
| 177 |
#if SQLITE_MAX_MMAP_SIZE>0 |
| 178 |
/* |
| 179 |
-** Return the system page size. |
| 180 |
-*/ |
| 181 |
-static int unixGetPagesize(void){ |
| 182 |
-#if HAVE_MREMAP |
| 183 |
- return 512; |
| 184 |
-#elif defined(_BSD_SOURCE) |
| 185 |
- return getpagesize(); |
| 186 |
-#else |
| 187 |
- return (int)sysconf(_SC_PAGESIZE); |
| 188 |
-#endif |
| 189 |
-} |
| 190 |
-#endif /* SQLITE_MAX_MMAP_SIZE>0 */ |
| 191 |
- |
| 192 |
-#if SQLITE_MAX_MMAP_SIZE>0 |
| 193 |
-/* |
| 194 |
** Attempt to set the size of the memory mapping maintained by file |
| 195 |
** descriptor pFd to nNew bytes. Any existing mapping is discarded. |
| 196 |
** |
| 197 |
** If successful, this function sets the following variables: |
| 198 |
** |
| 199 |
@@ -4680,12 +4712,16 @@ |
| 200 |
assert( MAP_FAILED!=0 ); |
| 201 |
|
| 202 |
if( (pFd->ctrlFlags & UNIXFILE_RDONLY)==0 ) flags |= PROT_WRITE; |
| 203 |
|
| 204 |
if( pOrig ){ |
| 205 |
- const int szSyspage = unixGetPagesize(); |
| 206 |
+#if HAVE_MREMAP |
| 207 |
+ i64 nReuse = pFd->mmapSize; |
| 208 |
+#else |
| 209 |
+ const int szSyspage = osGetpagesize(); |
| 210 |
i64 nReuse = (pFd->mmapSize & ~(szSyspage-1)); |
| 211 |
+#endif |
| 212 |
u8 *pReq = &pOrig[nReuse]; |
| 213 |
|
| 214 |
/* Unmap any pages of the existing mapping that cannot be reused. */ |
| 215 |
if( nReuse!=nOrig ){ |
| 216 |
osMunmap(pReq, nOrig-nReuse); |
| 217 |
@@ -7427,11 +7463,11 @@ |
| 218 |
}; |
| 219 |
unsigned int i; /* Loop counter */ |
| 220 |
|
| 221 |
/* Double-check that the aSyscall[] array has been constructed |
| 222 |
** correctly. See ticket [bb3a86e890c8e96ab] */ |
| 223 |
- assert( ArraySize(aSyscall)==24 ); |
| 224 |
+ assert( ArraySize(aSyscall)==25 ); |
| 225 |
|
| 226 |
/* Register all VFSes defined in the aVfs[] array */ |
| 227 |
for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){ |
| 228 |
sqlite3_vfs_register(&aVfs[i], i==0); |
| 229 |
} |
| 230 |
|
| 231 |
Index: src/test_syscall.c |
| 232 |
================================================================== |
| 233 |
--- src/test_syscall.c |
| 234 |
+++ src/test_syscall.c |
| 235 |
@@ -65,10 +65,15 @@ |
| 236 |
** Return true if the named system call exists. Or false otherwise. |
| 237 |
** |
| 238 |
** test_syscall list |
| 239 |
** Return a list of all system calls. The list is constructed using |
| 240 |
** the xNextSystemCall() VFS method. |
| 241 |
+** |
| 242 |
+** test_syscall pagesize PGSZ |
| 243 |
+** If PGSZ is a power of two greater than 256, install a wrapper around |
| 244 |
+** OS function getpagesize() that reports the system page size as PGSZ. |
| 245 |
+** Or, if PGSZ is less than zero, remove any wrapper already installed. |
| 246 |
*/ |
| 247 |
|
| 248 |
#include "sqliteInt.h" |
| 249 |
#include "sqlite3.h" |
| 250 |
#include "tcl.h" |
| 251 |
@@ -87,11 +92,13 @@ |
| 252 |
|
| 253 |
static struct TestSyscallGlobal { |
| 254 |
int bPersist; /* 1 for persistent errors, 0 for transient */ |
| 255 |
int nCount; /* Fail after this many more calls */ |
| 256 |
int nFail; /* Number of failures that have occurred */ |
| 257 |
-} gSyscall = { 0, 0 }; |
| 258 |
+ int pgsz; |
| 259 |
+ sqlite3_syscall_ptr orig_getpagesize; |
| 260 |
+} gSyscall = { 0, 0, 0, 0, 0 }; |
| 261 |
|
| 262 |
static int ts_open(const char *, int, int); |
| 263 |
static int ts_close(int fd); |
| 264 |
static int ts_access(const char *zPath, int mode); |
| 265 |
static char *ts_getcwd(char *zPath, size_t nPath); |
| 266 |
@@ -647,10 +654,49 @@ |
| 267 |
|
| 268 |
pVfs = sqlite3_vfs_find(0); |
| 269 |
Tcl_SetObjResult(interp, Tcl_NewStringObj(pVfs->zName, -1)); |
| 270 |
return TCL_OK; |
| 271 |
} |
| 272 |
+ |
| 273 |
+static int ts_getpagesize(void){ |
| 274 |
+ return gSyscall.pgsz; |
| 275 |
+} |
| 276 |
+ |
| 277 |
+static int test_syscall_pagesize( |
| 278 |
+ void * clientData, |
| 279 |
+ Tcl_Interp *interp, |
| 280 |
+ int objc, |
| 281 |
+ Tcl_Obj *CONST objv[] |
| 282 |
+){ |
| 283 |
+ sqlite3_vfs *pVfs = sqlite3_vfs_find(0); |
| 284 |
+ int pgsz; |
| 285 |
+ if( objc!=3 ){ |
| 286 |
+ Tcl_WrongNumArgs(interp, 2, objv, "PGSZ"); |
| 287 |
+ return TCL_ERROR; |
| 288 |
+ } |
| 289 |
+ if( Tcl_GetIntFromObj(interp, objv[2], &pgsz) ){ |
| 290 |
+ return TCL_ERROR; |
| 291 |
+ } |
| 292 |
+ |
| 293 |
+ if( pgsz<0 ){ |
| 294 |
+ if( gSyscall.orig_getpagesize ){ |
| 295 |
+ pVfs->xSetSystemCall(pVfs, "getpagesize", gSyscall.orig_getpagesize); |
| 296 |
+ } |
| 297 |
+ }else{ |
| 298 |
+ if( pgsz<512 || (pgsz & (pgsz-1)) ){ |
| 299 |
+ Tcl_AppendResult(interp, "pgsz out of range", 0); |
| 300 |
+ return TCL_ERROR; |
| 301 |
+ } |
| 302 |
+ gSyscall.orig_getpagesize = pVfs->xGetSystemCall(pVfs, "getpagesize"); |
| 303 |
+ gSyscall.pgsz = pgsz; |
| 304 |
+ pVfs->xSetSystemCall( |
| 305 |
+ pVfs, "getpagesize", (sqlite3_syscall_ptr)ts_getpagesize |
| 306 |
+ ); |
| 307 |
+ } |
| 308 |
+ |
| 309 |
+ return TCL_OK; |
| 310 |
+} |
| 311 |
|
| 312 |
static int test_syscall( |
| 313 |
void * clientData, |
| 314 |
Tcl_Interp *interp, |
| 315 |
int objc, |
| 316 |
@@ -666,10 +712,11 @@ |
| 317 |
{ "reset", test_syscall_reset }, |
| 318 |
{ "errno", test_syscall_errno }, |
| 319 |
{ "exists", test_syscall_exists }, |
| 320 |
{ "list", test_syscall_list }, |
| 321 |
{ "defaultvfs", test_syscall_defaultvfs }, |
| 322 |
+ { "pagesize", test_syscall_pagesize }, |
| 323 |
{ 0, 0 } |
| 324 |
}; |
| 325 |
int iCmd; |
| 326 |
int rc; |
| 327 |
|
| 328 |
|
| 329 |
Index: test/syscall.test |
| 330 |
================================================================== |
| 331 |
--- test/syscall.test |
| 332 |
+++ test/syscall.test |
| 333 |
@@ -59,10 +59,11 @@ |
| 334 |
foreach s { |
| 335 |
open close access getcwd stat fstat ftruncate |
| 336 |
fcntl read pread write pwrite fchmod fallocate |
| 337 |
pread64 pwrite64 unlink openDirectory mkdir rmdir |
| 338 |
statvfs fchown umask mmap munmap mremap |
| 339 |
+ getpagesize |
| 340 |
} { |
| 341 |
if {[test_syscall exists $s]} {lappend syscall_list $s} |
| 342 |
} |
| 343 |
do_test 3.1 { lsort [test_syscall list] } [lsort $syscall_list] |
| 344 |
|
| 345 |
|
| 346 |
ADDED test/wal64k.test |
| 347 |
Index: test/wal64k.test |
| 348 |
================================================================== |
| 349 |
--- test/wal64k.test |
| 350 |
+++ test/wal64k.test |
| 351 |
@@ -0,0 +1,47 @@ |
| 352 |
+# 2010 April 13 |
| 353 |
+# |
| 354 |
+# The author disclaims copyright to this source code. In place of |
| 355 |
+# a legal notice, here is a blessing: |
| 356 |
+# |
| 357 |
+# May you do good and not evil. |
| 358 |
+# May you find forgiveness for yourself and forgive others. |
| 359 |
+# May you share freely, never taking more than you give. |
| 360 |
+# |
| 361 |
+#*********************************************************************** |
| 362 |
+# This file implements regression tests for SQLite library. The |
| 363 |
+# focus of this file is testing the operation of the library in |
| 364 |
+# "PRAGMA journal_mode=WAL" mode. |
| 365 |
+# |
| 366 |
+ |
| 367 |
+set testdir [file dirname $argv0] |
| 368 |
+source $testdir/tester.tcl |
| 369 |
+set testprefix wal64k |
| 370 |
+ |
| 371 |
+ifcapable !wal {finish_test ; return } |
| 372 |
+ |
| 373 |
+db close |
| 374 |
+test_syscall pagesize 65536 |
| 375 |
+sqlite3 db test.db |
| 376 |
+ |
| 377 |
+do_execsql_test 1.0 { |
| 378 |
+ PRAGMA journal_mode = WAL; |
| 379 |
+ CREATE TABLE t1(x); |
| 380 |
+ CREATE INDEX i1 ON t1(x); |
| 381 |
+} {wal} |
| 382 |
+do_test 1.1 { file size test.db-shm } {65536} |
| 383 |
+ |
| 384 |
+do_test 1.2 { |
| 385 |
+ execsql BEGIN |
| 386 |
+ while {[file size test.db-shm]==65536} { |
| 387 |
+ execsql { INSERT INTO t1 VALUES( randstr(900,1100) ) } |
| 388 |
+ } |
| 389 |
+ execsql COMMIT |
| 390 |
+ file size test.db-shm |
| 391 |
+} {131072} |
| 392 |
+ |
| 393 |
+integrity_check 1.3 |
| 394 |
+ |
| 395 |
+db close |
| 396 |
+test_syscall pagesize -1 |
| 397 |
+finish_test |
| 398 |
+ |
| 399 |
|
Parent Directory
| 
Revision Log
| 
Revision Graph
