/ Check-in [e3ea43da]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Stop requiring the global VFS mutex to access the unixInodeInfo.pUnused field. The unixInodeInfo mutex is sufficient.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA3-256:e3ea43dabf099dc2954c23d348638e7b2a8b9122d2994154bc649a2c09260c46
User & Date: drh 2018-08-13 22:50:34
Context
2018-08-14
15:12
Fix UPSERT so that it checks the target-constraint first and fires the DO UPDATE if that constraint is violated regardless of whether or not other constraints are in violation. This aligns SQLite behavior with what PostgreSQL does. Fix for ticket [908f001483982c43cdb476dfb590a1a]. check-in: 529fb55e user: drh tags: trunk
2018-08-13
22:50
Stop requiring the global VFS mutex to access the unixInodeInfo.pUnused field. The unixInodeInfo mutex is sufficient. check-in: e3ea43da user: drh tags: trunk
20:46
Take extra precautions to ensure access to unixInodeInfo.pUnused is protected by all necessary mutexes. check-in: 8b1e0010 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/os_unix.c.

1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
....
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
....
1310
1311
1312
1313
1314
1315
1316

1317

1318
1319
1320
1321
1322
1323
1324
....
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
....
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
....
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
....
2110
2111
2112
2113
2114
2115
2116



2117
2118
2119
2120
2121
2122
2123
2124
2125

2126
2127
2128
2129
2130
2131
2132
2133

2134
2135
2136
2137
2138
2139
2140
....
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
....
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
....
3192
3193
3194
3195
3196
3197
3198



3199
3200
3201
3202
3203
3204
3205


3206
3207
3208
3209
3210
3211
3212
** structure contains a pointer to an instance of this object and this
** object keeps a count of the number of unixFile pointing to it.
**
** Mutex rules:
**
**  (1) Only the pLockMutex mutex must be held in order to read or write
**      any of the locking fields:
**          nShared, nLock, eFileLock, or bProcessLock
**
**  (2) When nRef>0, then the following fields are unchanging and can
**      be read (but not written) without holding any mutex:
**          fileId, pLockMutex
**
**  (3) The pUnused field may only be changed while holding bo the
**      pLockMutex and the bigUnixLock mutex.  But it may be read
**      while holding either.
**
**  (4) With the exceptions above, all the fields may only be read
**      or written while holding the global unixBigLock mutex.
**
** Deadlock prevention:  The global unixBigLock mutex may not
** be acquired while holding the pLockMutex mutex.  If both unixBigLock
** and pLockMutex are needed, then unixBigLock must be acquired first.
*/
struct unixInodeInfo {
  struct unixFileId fileId;       /* The lookup key */
  sqlite3_mutex *pLockMutex;      /* Hold this mutex for... */
  int nShared;                      /* Number of SHARED locks held */
  int nLock;                        /* Number of outstanding file locks */
  unsigned char eFileLock;          /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  unsigned char bProcessLock;       /* An exclusive process lock is held */
  UnixUnusedFd *pUnused;          /* Unused file descriptors to close */
  int nRef;                       /* Number of pointers to this structure */
  unixShmNode *pShmNode;          /* Shared memory associated with this inode */
  unixInodeInfo *pNext;           /* List of all unixInodeInfo objects */
  unixInodeInfo *pPrev;           /*    .... doubly linked */
#if SQLITE_ENABLE_LOCKING_STYLE
  unsigned long long sharedByte;  /* for AFP simulated shared lock */
#endif
................................................................................
/*
** Close all file descriptors accumuated in the unixInodeInfo->pUnused list.
*/ 
static void closePendingFds(unixFile *pFile){
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p;
  UnixUnusedFd *pNext;
  assert( unixMutexHeld() );
  assert( unixFileMutexNotheld(pFile) );
  sqlite3_mutex_enter(pInode->pLockMutex);
  for(p=pInode->pUnused; p; p=pNext){
    pNext = p->pNext;
    robust_close(pFile, p->fd, __LINE__);
    sqlite3_free(p);
  }
  pInode->pUnused = 0;
  sqlite3_mutex_leave(pInode->pLockMutex);
}

/*
** Release a unixInodeInfo structure previously allocated by findInodeInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
................................................................................
  unixInodeInfo *pInode = pFile->pInode;
  assert( unixMutexHeld() );
  assert( unixFileMutexNotheld(pFile) );
  if( ALWAYS(pInode) ){
    pInode->nRef--;
    if( pInode->nRef==0 ){
      assert( pInode->pShmNode==0 );

      closePendingFds(pFile);

      if( pInode->pPrev ){
        assert( pInode->pPrev->pNext==pInode );
        pInode->pPrev->pNext = pInode->pNext;
      }else{
        assert( inodeList==pInode );
        inodeList = pInode->pNext;
      }
................................................................................
/*
** Add the file descriptor used by file handle pFile to the corresponding
** pUnused list.
*/
static void setPendingFd(unixFile *pFile){
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p = pFile->pPreallocatedUnused;
  assert( unixMutexHeld() );
  assert( unixFileMutexNotheld(pFile) );
  sqlite3_mutex_enter(pInode->pLockMutex);
  p->pNext = pInode->pUnused;
  pInode->pUnused = p;
  sqlite3_mutex_leave(pInode->pLockMutex);
  pFile->h = -1;
  pFile->pPreallocatedUnused = 0;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................
** remove the write lock on a region when a read lock is set.
*/
static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  struct flock lock;
  int rc = SQLITE_OK;
  int wantToClosePending = 0;  /* True to try to close file old descriptors */

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock,
      pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
      osGetpid(0)));

  assert( eFileLock<=SHARED_LOCK );
................................................................................

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */
    pInode->nLock--;
    assert( pInode->nLock>=0 );
    if( pInode->nLock==0 && pInode->pUnused!=0 ){
      wantToClosePending = 1;
    }
  }

end_unlock:
  sqlite3_mutex_leave(pInode->pLockMutex);
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    if( wantToClosePending ){
      unixEnterMutex();
      if( pInode->nLock==0 ) closePendingFds(pFile);
      unixLeaveMutex();
    }
  }
  return rc;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................

/*
** Close a file.
*/
static int unixClose(sqlite3_file *id){
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile *)id;



  verifyDbFile(pFile);
  unixUnlock(id, NO_LOCK);
  assert( unixFileMutexNotheld(pFile) );
  unixEnterMutex();

  /* unixFile.pInode is always valid here. Otherwise, a different close
  ** routine (e.g. nolockClose()) would be called instead.
  */
  assert( pFile->pInode->nLock>0 || pFile->pInode->bProcessLock==0 );

  if( ALWAYS(pFile->pInode) && pFile->pInode->nLock ){
    /* If there are outstanding locks, do not actually close the file just
    ** yet because that would clear those locks.  Instead, add the file
    ** descriptor to pInode->pUnused list.  It will be automatically closed 
    ** when the last lock is cleared.
    */
    setPendingFd(pFile);
  }

  releaseInodeInfo(pFile);
  rc = closeUnixFile(id);
  unixLeaveMutex();
  return rc;
}

/************** End of the posix advisory lock implementation *****************
................................................................................
*/
static int afpUnlock(sqlite3_file *id, int eFileLock) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int skipShared = 0;
  int wantToClosePending = 0;
#ifdef SQLITE_TEST
  int h = pFile->h;
#endif

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (afp)\n", pFile->h, eFileLock,
           pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
................................................................................
        pInode->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }
    if( rc==SQLITE_OK ){
      pInode->nLock--;
      assert( pInode->nLock>=0 );
      if( pInode->nLock==0 && pInode->pUnused!=0 ) wantToClosePending = 1;
    }
  }
  
  sqlite3_mutex_leave(pInode->pLockMutex);
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    if( wantToClosePending ){
      unixEnterMutex();
      if( pInode->nLock==0 ) closePendingFds(pFile);
      unixLeaveMutex();
    }
  }
  return rc;
}

/*
** Close a file & cleanup AFP specific locking context 
*/
................................................................................
static int afpClose(sqlite3_file *id) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  assert( id!=0 );
  afpUnlock(id, NO_LOCK);
  assert( unixFileMutexNotheld(pFile) );
  unixEnterMutex();



  if( pFile->pInode && pFile->pInode->nLock ){
    /* If there are outstanding locks, do not actually close the file just
    ** yet because that would clear those locks.  Instead, add the file
    ** descriptor to pInode->aPending.  It will be automatically closed when
    ** the last lock is cleared.
    */
    setPendingFd(pFile);


  }
  releaseInodeInfo(pFile);
  sqlite3_free(pFile->lockingContext);
  rc = closeUnixFile(id);
  unixLeaveMutex();
  return rc;
}







|





<
<
<
<
|













|







 







<
|
<






<







 







>

>







 







<
|
<


<







 







<







 







|
<
<






<
<
<
<
<







 







>
>
>









>
|







>







 







<







 







|






<
<
<
<
<







 







>
>
>
|
|
|
|
|
|
|
>
>







1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134




1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
....
1280
1281
1282
1283
1284
1285
1286

1287

1288
1289
1290
1291
1292
1293

1294
1295
1296
1297
1298
1299
1300
....
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
....
1854
1855
1856
1857
1858
1859
1860

1861

1862
1863

1864
1865
1866
1867
1868
1869
1870
....
1879
1880
1881
1882
1883
1884
1885

1886
1887
1888
1889
1890
1891
1892
....
2016
2017
2018
2019
2020
2021
2022
2023


2024
2025
2026
2027
2028
2029





2030
2031
2032
2033
2034
2035
2036
....
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
....
3071
3072
3073
3074
3075
3076
3077

3078
3079
3080
3081
3082
3083
3084
....
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167





3168
3169
3170
3171
3172
3173
3174
....
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
** structure contains a pointer to an instance of this object and this
** object keeps a count of the number of unixFile pointing to it.
**
** Mutex rules:
**
**  (1) Only the pLockMutex mutex must be held in order to read or write
**      any of the locking fields:
**          nShared, nLock, eFileLock, bProcessLock, pUnused
**
**  (2) When nRef>0, then the following fields are unchanging and can
**      be read (but not written) without holding any mutex:
**          fileId, pLockMutex
**




**  (3) With the exceptions above, all the fields may only be read
**      or written while holding the global unixBigLock mutex.
**
** Deadlock prevention:  The global unixBigLock mutex may not
** be acquired while holding the pLockMutex mutex.  If both unixBigLock
** and pLockMutex are needed, then unixBigLock must be acquired first.
*/
struct unixInodeInfo {
  struct unixFileId fileId;       /* The lookup key */
  sqlite3_mutex *pLockMutex;      /* Hold this mutex for... */
  int nShared;                      /* Number of SHARED locks held */
  int nLock;                        /* Number of outstanding file locks */
  unsigned char eFileLock;          /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  unsigned char bProcessLock;       /* An exclusive process lock is held */
  UnixUnusedFd *pUnused;            /* Unused file descriptors to close */
  int nRef;                       /* Number of pointers to this structure */
  unixShmNode *pShmNode;          /* Shared memory associated with this inode */
  unixInodeInfo *pNext;           /* List of all unixInodeInfo objects */
  unixInodeInfo *pPrev;           /*    .... doubly linked */
#if SQLITE_ENABLE_LOCKING_STYLE
  unsigned long long sharedByte;  /* for AFP simulated shared lock */
#endif
................................................................................
/*
** Close all file descriptors accumuated in the unixInodeInfo->pUnused list.
*/ 
static void closePendingFds(unixFile *pFile){
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p;
  UnixUnusedFd *pNext;

  assert( unixFileMutexHeld(pFile) );

  for(p=pInode->pUnused; p; p=pNext){
    pNext = p->pNext;
    robust_close(pFile, p->fd, __LINE__);
    sqlite3_free(p);
  }
  pInode->pUnused = 0;

}

/*
** Release a unixInodeInfo structure previously allocated by findInodeInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
................................................................................
  unixInodeInfo *pInode = pFile->pInode;
  assert( unixMutexHeld() );
  assert( unixFileMutexNotheld(pFile) );
  if( ALWAYS(pInode) ){
    pInode->nRef--;
    if( pInode->nRef==0 ){
      assert( pInode->pShmNode==0 );
      sqlite3_mutex_enter(pInode->pLockMutex);
      closePendingFds(pFile);
      sqlite3_mutex_leave(pInode->pLockMutex);
      if( pInode->pPrev ){
        assert( pInode->pPrev->pNext==pInode );
        pInode->pPrev->pNext = pInode->pNext;
      }else{
        assert( inodeList==pInode );
        inodeList = pInode->pNext;
      }
................................................................................
/*
** Add the file descriptor used by file handle pFile to the corresponding
** pUnused list.
*/
static void setPendingFd(unixFile *pFile){
  unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p = pFile->pPreallocatedUnused;

  assert( unixFileMutexHeld(pFile) );

  p->pNext = pInode->pUnused;
  pInode->pUnused = p;

  pFile->h = -1;
  pFile->pPreallocatedUnused = 0;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................
** remove the write lock on a region when a read lock is set.
*/
static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  struct flock lock;
  int rc = SQLITE_OK;


  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock,
      pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
      osGetpid(0)));

  assert( eFileLock<=SHARED_LOCK );
................................................................................

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */
    pInode->nLock--;
    assert( pInode->nLock>=0 );
    if( pInode->nLock==0 ) closePendingFds(pFile);


  }

end_unlock:
  sqlite3_mutex_leave(pInode->pLockMutex);
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;





  }
  return rc;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................

/*
** Close a file.
*/
static int unixClose(sqlite3_file *id){
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile *)id;
  unixInodeInfo *pInode = pFile->pInode;

  assert( pInode!=0 );
  verifyDbFile(pFile);
  unixUnlock(id, NO_LOCK);
  assert( unixFileMutexNotheld(pFile) );
  unixEnterMutex();

  /* unixFile.pInode is always valid here. Otherwise, a different close
  ** routine (e.g. nolockClose()) would be called instead.
  */
  assert( pFile->pInode->nLock>0 || pFile->pInode->bProcessLock==0 );
  sqlite3_mutex_enter(pInode->pLockMutex);
  if( pFile->pInode->nLock ){
    /* If there are outstanding locks, do not actually close the file just
    ** yet because that would clear those locks.  Instead, add the file
    ** descriptor to pInode->pUnused list.  It will be automatically closed 
    ** when the last lock is cleared.
    */
    setPendingFd(pFile);
  }
  sqlite3_mutex_leave(pInode->pLockMutex);
  releaseInodeInfo(pFile);
  rc = closeUnixFile(id);
  unixLeaveMutex();
  return rc;
}

/************** End of the posix advisory lock implementation *****************
................................................................................
*/
static int afpUnlock(sqlite3_file *id, int eFileLock) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  unixInodeInfo *pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int skipShared = 0;

#ifdef SQLITE_TEST
  int h = pFile->h;
#endif

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (afp)\n", pFile->h, eFileLock,
           pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
................................................................................
        pInode->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }
    if( rc==SQLITE_OK ){
      pInode->nLock--;
      assert( pInode->nLock>=0 );
      if( pInode->nLock==0 ) closePendingFds(pFile);
    }
  }
  
  sqlite3_mutex_leave(pInode->pLockMutex);
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;





  }
  return rc;
}

/*
** Close a file & cleanup AFP specific locking context 
*/
................................................................................
static int afpClose(sqlite3_file *id) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  assert( id!=0 );
  afpUnlock(id, NO_LOCK);
  assert( unixFileMutexNotheld(pFile) );
  unixEnterMutex();
  if( pFile->pInode ){
    unixInodeInfo *pInode = pFile->pInode;
    sqlite3_mutex_enter(pInode->pLockMutex);
    if( pFile->pInode->nLock ){
      /* If there are outstanding locks, do not actually close the file just
      ** yet because that would clear those locks.  Instead, add the file
      ** descriptor to pInode->aPending.  It will be automatically closed when
      ** the last lock is cleared.
      */
      setPendingFd(pFile);
    }
    sqlite3_mutex_leave(pInode->pLockMutex);
  }
  releaseInodeInfo(pFile);
  sqlite3_free(pFile->lockingContext);
  rc = closeUnixFile(id);
  unixLeaveMutex();
  return rc;
}