Merge all of the latest trunk changes into the sessions branch.

FossilOrigin-Name: a9bcb432f58b96f079a73c456efd4851c582221e

1 files changed, 113 insertions, 61 deletions

diff --git a/src/os_unix.c b/src/os_unix.c
index 0ea6daf27..4f263bcdd 100644
--- a/src/os_unix.c
+++ b/src/os_unix.c
@@ -123,6 +123,7 @@
 #include <sys/mman.h>
 #endif
 
+
 #if SQLITE_ENABLE_LOCKING_STYLE
 # include <sys/ioctl.h>
 # if OS_VXWORKS
@@ -206,6 +207,7 @@ struct UnixUnusedFd {
 typedef struct unixFile unixFile;
 struct unixFile {
   sqlite3_io_methods const *pMethod;  /* Always the first entry */
+  sqlite3_vfs *pVfs;                  /* The VFS that created this unixFile */
   unixInodeInfo *pInode;              /* Info about locks on this inode */
   int h;                              /* The file descriptor */
   unsigned char eFileLock;            /* The type of lock held on this fd */
@@ -257,6 +259,7 @@ struct unixFile {
 #else
 # define UNIXFILE_DIRSYNC    0x00
 #endif
+#define UNIXFILE_PSOW        0x10     /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */
 
 /*
 ** Include code that is common to all os_*.c files
@@ -407,6 +410,12 @@ static struct unix_syscall {
   { "openDirectory",    (sqlite3_syscall_ptr)openDirectory,      0 },
 #define osOpenDirectory ((int(*)(const char*,int*))aSyscall[17].pCurrent)
 
+  { "mkdir",        (sqlite3_syscall_ptr)mkdir,           0 },
+#define osMkdir     ((int(*)(const char*,mode_t))aSyscall[18].pCurrent)
+
+  { "rmdir",        (sqlite3_syscall_ptr)rmdir,           0 },
+#define osRmdir     ((int(*)(const char*))aSyscall[19].pCurrent)
+
 }; /* End of the overrideable system calls */
 
 /*
@@ -1845,8 +1854,8 @@ static int nolockClose(sqlite3_file *id) {
 ************************* Begin dot-file Locking ******************************
 **
 ** The dotfile locking implementation uses the existance of separate lock
-** files in order to control access to the database.  This works on just
-** about every filesystem imaginable.  But there are serious downsides:
+** files (really a directory) to control access to the database.  This works
+** on just about every filesystem imaginable.  But there are serious downsides:
 **
 **    (1)  There is zero concurrency.  A single reader blocks all other
 **         connections from reading or writing the database.
@@ -1857,15 +1866,15 @@ static int nolockClose(sqlite3_file *id) {
 ** Nevertheless, a dotlock is an appropriate locking mode for use if no
 ** other locking strategy is available.
 **
-** Dotfile locking works by creating a file in the same directory as the
-** database and with the same name but with a ".lock" extension added.
-** The existance of a lock file implies an EXCLUSIVE lock.  All other lock
-** types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE.
+** Dotfile locking works by creating a subdirectory in the same directory as
+** the database and with the same name but with a ".lock" extension added.
+** The existance of a lock directory implies an EXCLUSIVE lock.  All other
+** lock types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE.
 */
 
 /*
 ** The file suffix added to the data base filename in order to create the
-** lock file.
+** lock directory.
 */
 #define DOTLOCK_SUFFIX ".lock"
 
@@ -1932,7 +1941,6 @@ static int dotlockCheckReservedLock(sqlite3_file *id, int *pResOut) {
 */
 static int dotlockLock(sqlite3_file *id, int eFileLock) {
   unixFile *pFile = (unixFile*)id;
-  int fd;
   char *zLockFile = (char *)pFile->lockingContext;
   int rc = SQLITE_OK;
 
@@ -1952,9 +1960,9 @@ static int dotlockLock(sqlite3_file *id, int eFileLock) {
   }
   
   /* grab an exclusive lock */
-  fd = robust_open(zLockFile,O_RDONLY|O_CREAT|O_EXCL,0600);
-  if( fd<0 ){
-    /* failed to open/create the file, someone else may have stolen the lock */
+  rc = osMkdir(zLockFile, 0777);
+  if( rc<0 ){
+    /* failed to open/create the lock directory */
     int tErrno = errno;
     if( EEXIST == tErrno ){
       rc = SQLITE_BUSY;
@@ -1966,7 +1974,6 @@ static int dotlockLock(sqlite3_file *id, int eFileLock) {
     }
     return rc;
   } 
-  robust_close(pFile, fd, __LINE__);
   
   /* got it, set the type and return ok */
   pFile->eFileLock = eFileLock;
@@ -1985,6 +1992,7 @@ static int dotlockLock(sqlite3_file *id, int eFileLock) {
 static int dotlockUnlock(sqlite3_file *id, int eFileLock) {
   unixFile *pFile = (unixFile*)id;
   char *zLockFile = (char *)pFile->lockingContext;
+  int rc;
 
   assert( pFile );
   OSTRACE(("UNLOCK  %d %d was %d pid=%d (dotlock)\n", pFile->h, eFileLock,
@@ -2006,9 +2014,11 @@ static int dotlockUnlock(sqlite3_file *id, int eFileLock) {
   
   /* To fully unlock the database, delete the lock file */
   assert( eFileLock==NO_LOCK );
-  if( osUnlink(zLockFile) ){
-    int rc = 0;
+  rc = osRmdir(zLockFile);
+  if( rc<0 && errno==ENOTDIR ) rc = osUnlink(zLockFile);
+  if( rc<0 ){
     int tErrno = errno;
+    rc = 0;
     if( ENOENT != tErrno ){
       rc = SQLITE_IOERR_UNLOCK;
     }
@@ -2944,35 +2954,48 @@ static int nfsUnlock(sqlite3_file *id, int eFileLock){
 */
 static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
   int got;
+  int prior = 0;
 #if (!defined(USE_PREAD) && !defined(USE_PREAD64))
   i64 newOffset;
 #endif
   TIMER_START;
+  do{
 #if defined(USE_PREAD)
-  do{ got = osPread(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR );
-  SimulateIOError( got = -1 );
+    got = osPread(id->h, pBuf, cnt, offset);
+    SimulateIOError( got = -1 );
 #elif defined(USE_PREAD64)
-  do{ got = osPread64(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR);
-  SimulateIOError( got = -1 );
+    got = osPread64(id->h, pBuf, cnt, offset);
+    SimulateIOError( got = -1 );
 #else
-  newOffset = lseek(id->h, offset, SEEK_SET);
-  SimulateIOError( newOffset-- );
-  if( newOffset!=offset ){
-    if( newOffset == -1 ){
-      ((unixFile*)id)->lastErrno = errno;
-    }else{
-      ((unixFile*)id)->lastErrno = 0;			
+    newOffset = lseek(id->h, offset, SEEK_SET);
+    SimulateIOError( newOffset-- );
+    if( newOffset!=offset ){
+      if( newOffset == -1 ){
+        ((unixFile*)id)->lastErrno = errno;
+      }else{
+        ((unixFile*)id)->lastErrno = 0;			
+      }
+      return -1;
     }
-    return -1;
-  }
-  do{ got = osRead(id->h, pBuf, cnt); }while( got<0 && errno==EINTR );
+    got = osRead(id->h, pBuf, cnt);
 #endif
+    if( got==cnt ) break;
+    if( got<0 ){
+      if( errno==EINTR ){ got = 1; continue; }
+      prior = 0;
+      ((unixFile*)id)->lastErrno = errno;
+      break;
+    }else if( got>0 ){
+      cnt -= got;
+      offset += got;
+      prior += got;
+      pBuf = (void*)(got + (char*)pBuf);
+    }
+  }while( got>0 );
   TIMER_END;
-  if( got<0 ){
-    ((unixFile*)id)->lastErrno = errno;
-  }
-  OSTRACE(("READ    %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED));
-  return got;
+  OSTRACE(("READ    %-3d %5d %7lld %llu\n",
+            id->h, got+prior, offset-prior, TIMER_ELAPSED));
+  return got+prior;
 }
 
 /*
@@ -3478,6 +3501,22 @@ static int fcntlSizeHint(unixFile *pFile, i64 nByte){
 }
 
 /*
+** If *pArg is inititially negative then this is a query.  Set *pArg to
+** 1 or 0 depending on whether or not bit mask of pFile->ctrlFlags is set.
+**
+** If *pArg is 0 or 1, then clear or set the mask bit of pFile->ctrlFlags.
+*/
+static void unixModeBit(unixFile *pFile, unsigned char mask, int *pArg){
+  if( *pArg<0 ){
+    *pArg = (pFile->ctrlFlags & mask)!=0;
+  }else if( (*pArg)==0 ){
+    pFile->ctrlFlags &= ~mask;
+  }else{
+    pFile->ctrlFlags |= mask;
+  }
+}
+
+/*
 ** Information and control of an open file handle.
 */
 static int unixFileControl(sqlite3_file *id, int op, void *pArg){
@@ -3503,14 +3542,15 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
       return rc;
     }
     case SQLITE_FCNTL_PERSIST_WAL: {
-      int bPersist = *(int*)pArg;
-      if( bPersist<0 ){
-        *(int*)pArg = (pFile->ctrlFlags & UNIXFILE_PERSIST_WAL)!=0;
-      }else if( bPersist==0 ){
-        pFile->ctrlFlags &= ~UNIXFILE_PERSIST_WAL;
-      }else{
-        pFile->ctrlFlags |= UNIXFILE_PERSIST_WAL;
-      }
+      unixModeBit(pFile, UNIXFILE_PERSIST_WAL, (int*)pArg);
+      return SQLITE_OK;
+    }
+    case SQLITE_FCNTL_POWERSAFE_OVERWRITE: {
+      unixModeBit(pFile, UNIXFILE_PSOW, (int*)pArg);
+      return SQLITE_OK;
+    }
+    case SQLITE_FCNTL_VFSNAME: {
+      *(char**)pArg = sqlite3_mprintf("%s", pFile->pVfs->zName);
       return SQLITE_OK;
     }
 #ifndef NDEBUG
@@ -3530,9 +3570,6 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
       return proxyFileControl(id,op,pArg);
     }
 #endif /* SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) */
-    case SQLITE_FCNTL_SYNC_OMITTED: {
-      return SQLITE_OK;  /* A no-op */
-    }
   }
   return SQLITE_NOTFOUND;
 }
@@ -3547,17 +3584,31 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
 ** a database and its journal file) that the sector size will be the
 ** same for both.
 */
-static int unixSectorSize(sqlite3_file *NotUsed){
-  UNUSED_PARAMETER(NotUsed);
+static int unixSectorSize(sqlite3_file *pFile){
+  (void)pFile;
   return SQLITE_DEFAULT_SECTOR_SIZE;
 }
 
 /*
-** Return the device characteristics for the file. This is always 0 for unix.
+** Return the device characteristics for the file.
+**
+** This VFS is set up to return SQLITE_IOCAP_POWERSAFE_OVERWRITE by default.
+** However, that choice is contraversial since technically the underlying
+** file system does not always provide powersafe overwrites.  (In other
+** words, after a power-loss event, parts of the file that were never
+** written might end up being altered.)  However, non-PSOW behavior is very,
+** very rare.  And asserting PSOW makes a large reduction in the amount
+** of required I/O for journaling, since a lot of padding is eliminated.
+**  Hence, while POWERSAFE_OVERWRITE is on by default, there is a file-control
+** available to turn it off and URI query parameter available to turn it off.
 */
-static int unixDeviceCharacteristics(sqlite3_file *NotUsed){
-  UNUSED_PARAMETER(NotUsed);
-  return 0;
+static int unixDeviceCharacteristics(sqlite3_file *id){
+  unixFile *p = (unixFile*)id;
+  if( p->ctrlFlags & UNIXFILE_PSOW ){
+    return SQLITE_IOCAP_POWERSAFE_OVERWRITE;
+  }else{
+    return 0;
+  }
 }
 
 #ifndef SQLITE_OMIT_WAL
@@ -3812,9 +3863,9 @@ static int unixOpenSharedMemory(unixFile *pDbFd){
     }
 
 #ifdef SQLITE_SHM_DIRECTORY
-    nShmFilename = sizeof(SQLITE_SHM_DIRECTORY) + 30;
+    nShmFilename = sizeof(SQLITE_SHM_DIRECTORY) + 31;
 #else
-    nShmFilename = 5 + (int)strlen(pDbFd->zPath);
+    nShmFilename = 6 + (int)strlen(pDbFd->zPath);
 #endif
     pShmNode = sqlite3_malloc( sizeof(*pShmNode) + nShmFilename );
     if( pShmNode==0 ){
@@ -3841,10 +3892,8 @@ static int unixOpenSharedMemory(unixFile *pDbFd){
     }
 
     if( pInode->bProcessLock==0 ){
-      const char *zRO;
       int openFlags = O_RDWR | O_CREAT;
-      zRO = sqlite3_uri_parameter(pDbFd->zPath, "readonly_shm");
-      if( zRO && sqlite3GetBoolean(zRO) ){
+      if( sqlite3_uri_boolean(pDbFd->zPath, "readonly_shm", 0) ){
         openFlags = O_RDONLY;
         pShmNode->isReadonly = 1;
       }
@@ -4540,11 +4589,14 @@ static int fillInUnixFile(
 
   OSTRACE(("OPEN    %-3d %s\n", h, zFilename));
   pNew->h = h;
+  pNew->pVfs = pVfs;
   pNew->zPath = zFilename;
+  pNew->ctrlFlags = 0;
+  if( sqlite3_uri_boolean(zFilename, "psow", SQLITE_POWERSAFE_OVERWRITE) ){
+    pNew->ctrlFlags |= UNIXFILE_PSOW;
+  }
   if( memcmp(pVfs->zName,"unix-excl",10)==0 ){
-    pNew->ctrlFlags = UNIXFILE_EXCL;
-  }else{
-    pNew->ctrlFlags = 0;
+    pNew->ctrlFlags |= UNIXFILE_EXCL;
   }
   if( isReadOnly ){
     pNew->ctrlFlags |= UNIXFILE_RDONLY;
@@ -4879,7 +4931,7 @@ static int findCreateFileMode(
     */
     nDb = sqlite3Strlen30(zPath) - 1; 
 #ifdef SQLITE_ENABLE_8_3_NAMES
-    while( nDb>0 && !sqlite3Isalnum(zPath[nDb]) ) nDb--;
+    while( nDb>0 && sqlite3Isalnum(zPath[nDb]) ) nDb--;
     if( nDb==0 || zPath[nDb]!='-' ) return SQLITE_OK;
 #else
     while( zPath[nDb]!='-' ){
@@ -5715,7 +5767,7 @@ static int proxyCreateLockPath(const char *lockPath){
       if( i-start>2 || (i-start==1 && buf[start] != '.' && buf[start] != '/') 
          || (i-start==2 && buf[start] != '.' && buf[start+1] != '.') ){
         buf[i]='\0';
-        if( mkdir(buf, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
+        if( osMkdir(buf, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
           int err=errno;
           if( err!=EEXIST ) {
             OSTRACE(("CREATELOCKPATH  FAILED creating %s, "
@@ -6751,7 +6803,7 @@ int sqlite3_os_init(void){
 
   /* Double-check that the aSyscall[] array has been constructed
   ** correctly.  See ticket [bb3a86e890c8e96ab] */
-  assert( ArraySize(aSyscall)==18 );
+  assert( ArraySize(aSyscall)==20 );
 
   /* Register all VFSes defined in the aVfs[] array */
   for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){