/*

 ** 2004 May 22

 **

 ** 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 contains code that is specific to Unix systems.

 */

 #include "sqliteInt.h"

 #if OS_UNIX              /* This file is used on unix only */

 /* #define SQLITE_ENABLE_LOCKING_STYLE 0 */

 /*

 ** These #defines should enable >2GB file support on Posix if the

 ** underlying operating system supports it.  If the OS lacks

 ** large file support, these should be no-ops.

 **

 ** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch

 ** on the compiler command line.  This is necessary if you are compiling

 ** on a recent machine (ex: RedHat 7.2) but you want your code to work

 ** on an older machine (ex: RedHat 6.0).  If you compile on RedHat 7.2

 ** without this option, LFS is enable.  But LFS does not exist in the kernel

 ** in RedHat 6.0, so the code won't work.  Hence, for maximum binary

 ** portability you should omit LFS.

 */

 #ifndef SQLITE_DISABLE_LFS

 # define _LARGE_FILE       1

 # ifndef _FILE_OFFSET_BITS

 #   define _FILE_OFFSET_BITS 64

 # endif

 # define _LARGEFILE_SOURCE 1

 #endif

 /*

 ** standard include files.

 */

 #include <sys/types.h>

 #include <sys/stat.h>

 #include <fcntl.h>

 #include <unistd.h>

 #include <time.h>

 #include <sys/time.h>

 #include <errno.h>

 #ifdef SQLITE_ENABLE_LOCKING_STYLE

 #include <sys/ioctl.h>

 #include <sys/param.h>

 #include <sys/mount.h>

 #endif /* SQLITE_ENABLE_LOCKING_STYLE */

 /*

 ** If we are to be thread-safe, include the pthreads header and define

 ** the SQLITE_UNIX_THREADS macro.

 */

 #if SQLITE_THREADSAFE

 # include <pthread.h>

 # define SQLITE_UNIX_THREADS 1

 #endif

 /*

 ** Default permissions when creating a new file

 */

 #ifndef SQLITE_DEFAULT_FILE_PERMISSIONS

 # define SQLITE_DEFAULT_FILE_PERMISSIONS 0644

 #endif

 /*

 ** Maximum supported path-length.

 */

 #define MAX_PATHNAME 512

 /*

 ** The unixFile structure is subclass of sqlite3_file specific for the unix

 ** protability layer.

 */

 typedef struct unixFile unixFile;

 struct unixFile {

   sqlite3_io_methods const *pMethod;  /* Always the first entry */

 #ifdef SQLITE_TEST

   /* In test mode, increase the size of this structure a bit so that 

   ** it is larger than the struct CrashFile defined in test6.c.

   */

   char aPadding[32];

 #endif

   struct openCnt *pOpen;    /* Info about all open fd's on this inode */

   struct lockInfo *pLock;   /* Info about locks on this inode */

 #ifdef SQLITE_ENABLE_LOCKING_STYLE

   void *lockingContext;     /* Locking style specific state */

 #endif /* SQLITE_ENABLE_LOCKING_STYLE */

   int h;                    /* The file descriptor */

   unsigned char locktype;   /* The type of lock held on this fd */

   int dirfd;                /* File descriptor for the directory */

 #if SQLITE_THREADSAFE

   pthread_t tid;            /* The thread that "owns" this unixFile */

 #endif

 };

 /*

 ** Include code that is common to all os_*.c files

 */

 #include "os_common.h"

 /*

 ** Define various macros that are missing from some systems.

 */

 #ifndef O_LARGEFILE

 # define O_LARGEFILE 0

 #endif

 #ifdef SQLITE_DISABLE_LFS

 # undef O_LARGEFILE

 # define O_LARGEFILE 0

 #endif

 #ifndef O_NOFOLLOW

 # define O_NOFOLLOW 0

 #endif

 #ifndef O_BINARY

 # define O_BINARY 0

 #endif

 /*

 ** The DJGPP compiler environment looks mostly like Unix, but it

 ** lacks the fcntl() system call.  So redefine fcntl() to be something

 ** that always succeeds.  This means that locking does not occur under

 ** DJGPP.  But it's DOS - what did you expect?

 */

 #ifdef __DJGPP__

 # define fcntl(A,B,C) 0

 #endif

 /*

 ** The threadid macro resolves to the thread-id or to 0.  Used for

 ** testing and debugging only.

 */

 #if SQLITE_THREADSAFE

 #define threadid pthread_self()

 #else

 #define threadid 0

 #endif

 /*

 ** Set or check the unixFile.tid field.  This field is set when an unixFile

 ** is first opened.  All subsequent uses of the unixFile verify that the

 ** same thread is operating on the unixFile.  Some operating systems do

 ** not allow locks to be overridden by other threads and that restriction

 ** means that sqlite3* database handles cannot be moved from one thread

 ** to another.  This logic makes sure a user does not try to do that

 ** by mistake.

 **

 ** Version 3.3.1 (2006-01-15):  unixFile can be moved from one thread to

 ** another as long as we are running on a system that supports threads

 ** overriding each others locks (which now the most common behavior)

 ** or if no locks are held.  But the unixFile.pLock field needs to be

 ** recomputed because its key includes the thread-id.  See the 

 ** transferOwnership() function below for additional information

 */

 #if SQLITE_THREADSAFE

 # define SET_THREADID(X)   (X)->tid = pthread_self()

 # define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \

                             !pthread_equal((X)->tid, pthread_self()))

 #else

 # define SET_THREADID(X)

 # define CHECK_THREADID(X) 0

 #endif

 /*

 ** Here is the dirt on POSIX advisory locks:  ANSI STD 1003.1 (1996)

 ** section 6.5.2.2 lines 483 through 490 specify that when a process

 ** sets or clears a lock, that operation overrides any prior locks set

 ** by the same process.  It does not explicitly say so, but this implies

 ** that it overrides locks set by the same process using a different

 ** file descriptor.  Consider this test case:

 **

 **       int fd1 = open("./file1", O_RDWR|O_CREAT, 0644);

 **       int fd2 = open("./file2", O_RDWR|O_CREAT, 0644);

 **

 ** Suppose ./file1 and ./file2 are really the same file (because

 ** one is a hard or symbolic link to the other) then if you set

 ** an exclusive lock on fd1, then try to get an exclusive lock

 ** on fd2, it works.  I would have expected the second lock to

 ** fail since there was already a lock on the file due to fd1.

 ** But not so.  Since both locks came from the same process, the

 ** second overrides the first, even though they were on different

 ** file descriptors opened on different file names.

 **

 ** Bummer.  If you ask me, this is broken.  Badly broken.  It means

 ** that we cannot use POSIX locks to synchronize file access among

 ** competing threads of the same process.  POSIX locks will work fine

 ** to synchronize access for threads in separate processes, but not

 ** threads within the same process.

 **

 ** To work around the problem, SQLite has to manage file locks internally

 ** on its own.  Whenever a new database is opened, we have to find the

 ** specific inode of the database file (the inode is determined by the

 ** st_dev and st_ino fields of the stat structure that fstat() fills in)

 ** and check for locks already existing on that inode.  When locks are

 ** created or removed, we have to look at our own internal record of the

 ** locks to see if another thread has previously set a lock on that same

 ** inode.

 **

 ** The sqlite3_file structure for POSIX is no longer just an integer file

 ** descriptor.  It is now a structure that holds the integer file

 ** descriptor and a pointer to a structure that describes the internal

 ** locks on the corresponding inode.  There is one locking structure

 ** per inode, so if the same inode is opened twice, both unixFile structures

 ** point to the same locking structure.  The locking structure keeps

 ** a reference count (so we will know when to delete it) and a "cnt"

 ** field that tells us its internal lock status.  cnt==0 means the

 ** file is unlocked.  cnt==-1 means the file has an exclusive lock.

 ** cnt>0 means there are cnt shared locks on the file.

 **

 ** Any attempt to lock or unlock a file first checks the locking

 ** structure.  The fcntl() system call is only invoked to set a 

 ** POSIX lock if the internal lock structure transitions between

 ** a locked and an unlocked state.

 **

 ** 2004-Jan-11:

 ** More recent discoveries about POSIX advisory locks.  (The more

 ** I discover, the more I realize the a POSIX advisory locks are

 ** an abomination.)

 **

 ** If you close a file descriptor that points to a file that has locks,

 ** all locks on that file that are owned by the current process are

 ** released.  To work around this problem, each unixFile structure contains

 ** a pointer to an openCnt structure.  There is one openCnt structure

 ** per open inode, which means that multiple unixFile can point to a single

 ** openCnt.  When an attempt is made to close an unixFile, if there are

 ** other unixFile open on the same inode that are holding locks, the call

 ** to close() the file descriptor is deferred until all of the locks clear.

 ** The openCnt structure keeps a list of file descriptors that need to

 ** be closed and that list is walked (and cleared) when the last lock

 ** clears.

 **

 ** First, under Linux threads, because each thread has a separate

 ** process ID, lock operations in one thread do not override locks

 ** to the same file in other threads.  Linux threads behave like

 ** separate processes in this respect.  But, if you close a file

 ** descriptor in linux threads, all locks are cleared, even locks

 ** on other threads and even though the other threads have different

 ** process IDs.  Linux threads is inconsistent in this respect.

 ** (I'm beginning to think that linux threads is an abomination too.)

 ** The consequence of this all is that the hash table for the lockInfo

 ** structure has to include the process id as part of its key because

 ** locks in different threads are treated as distinct.  But the 

 ** openCnt structure should not include the process id in its

 ** key because close() clears lock on all threads, not just the current

 ** thread.  Were it not for this goofiness in linux threads, we could

 ** combine the lockInfo and openCnt structures into a single structure.

 **

 ** 2004-Jun-28:

 ** On some versions of linux, threads can override each others locks.

 ** On others not.  Sometimes you can change the behavior on the same

 ** system by setting the LD_ASSUME_KERNEL environment variable.  The

 ** POSIX standard is silent as to which behavior is correct, as far

 ** as I can tell, so other versions of unix might show the same

 ** inconsistency.  There is no little doubt in my mind that posix

 ** advisory locks and linux threads are profoundly broken.

 **

 ** To work around the inconsistencies, we have to test at runtime 

 ** whether or not threads can override each others locks.  This test

 ** is run once, the first time any lock is attempted.  A static 

 ** variable is set to record the results of this test for future

 ** use.

 */

 /*

 ** An instance of the following structure serves as the key used

 ** to locate a particular lockInfo structure given its inode.

 **

 ** If threads cannot override each others locks, then we set the

 ** lockKey.tid field to the thread ID.  If threads can override

 ** each others locks then tid is always set to zero.  tid is omitted

 ** if we compile without threading support.

 */

 struct lockKey {

   dev_t dev;       /* Device number */

   ino_t ino;       /* Inode number */

 #if SQLITE_THREADSAFE

   pthread_t tid;   /* Thread ID or zero if threads can override each other */

 #endif

 };

 /*

 ** An instance of the following structure is allocated for each open

 ** inode on each thread with a different process ID.  (Threads have

 ** different process IDs on linux, but not on most other unixes.)

 **

 ** A single inode can have multiple file descriptors, so each unixFile

 ** structure contains a pointer to an instance of this object and this

 ** object keeps a count of the number of unixFile pointing to it.

 */

 struct lockInfo {

   struct lockKey key;  /* The lookup key */

   int cnt;             /* Number of SHARED locks held */

   int locktype;        /* One of SHARED_LOCK, RESERVED_LOCK etc. */

   int nRef;            /* Number of pointers to this structure */

 };

 /*

 ** An instance of the following structure serves as the key used

 ** to locate a particular openCnt structure given its inode.  This

 ** is the same as the lockKey except that the thread ID is omitted.

 */

 struct openKey {

   dev_t dev;   /* Device number */

   ino_t ino;   /* Inode number */

 };

 /*

 ** An instance of the following structure is allocated for each open

 ** inode.  This structure keeps track of the number of locks on that

 ** inode.  If a close is attempted against an inode that is holding

 ** locks, the close is deferred until all locks clear by adding the

 ** file descriptor to be closed to the pending list.

 */

 struct openCnt {

   struct openKey key;   /* The lookup key */

   int nRef;             /* Number of pointers to this structure */

   int nLock;            /* Number of outstanding locks */

   int nPending;         /* Number of pending close() operations */

   int *aPending;        /* Malloced space holding fd's awaiting a close() */

 };

 /* 

 ** These hash tables map inodes and file descriptors (really, lockKey and

 ** openKey structures) into lockInfo and openCnt structures.  Access to 

 ** these hash tables must be protected by a mutex.

 */

 static Hash lockHash = {SQLITE_HASH_BINARY, 0, 0, 0, 0, 0};

 static Hash openHash = {SQLITE_HASH_BINARY, 0, 0, 0, 0, 0};

 #ifdef SQLITE_ENABLE_LOCKING_STYLE

 /*

 ** The locking styles are associated with the different file locking

 ** capabilities supported by different file systems.  

 **

 ** POSIX locking style fully supports shared and exclusive byte-range locks 

 ** ADP locking only supports exclusive byte-range locks

 ** FLOCK only supports a single file-global exclusive lock

 ** DOTLOCK isn't a true locking style, it refers to the use of a special

 **   file named the same as the database file with a '.lock' extension, this

 **   can be used on file systems that do not offer any reliable file locking

 ** NO locking means that no locking will be attempted, this is only used for

 **   read-only file systems currently

 ** UNSUPPORTED means that no locking will be attempted, this is only used for

 **   file systems that are known to be unsupported

 */

 typedef enum {

         posixLockingStyle = 0,       /* standard posix-advisory locks */

         afpLockingStyle,             /* use afp locks */

         flockLockingStyle,           /* use flock() */

         dotlockLockingStyle,         /* use <file>.lock files */

         noLockingStyle,              /* useful for read-only file system */

         unsupportedLockingStyle      /* indicates unsupported file system */

 } sqlite3LockingStyle;

 #endif /* SQLITE_ENABLE_LOCKING_STYLE */

 /*

 ** Helper functions to obtain and relinquish the global mutex.

 */

 static void enterMutex(){

   sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MASTER));

 }

 static void leaveMutex(){

   sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MASTER));

 }

 #if SQLITE_THREADSAFE

 /*

 ** This variable records whether or not threads can override each others

 ** locks.

 **

 **    0:  No.  Threads cannot override each others locks.

 **    1:  Yes.  Threads can override each others locks.

 **   -1:  We don't know yet.

 **

 ** On some systems, we know at compile-time if threads can override each

 ** others locks.  On those systems, the SQLITE_THREAD_OVERRIDE_LOCK macro

 ** will be set appropriately.  On other systems, we have to check at

 ** runtime.  On these latter systems, SQLTIE_THREAD_OVERRIDE_LOCK is

 ** undefined.

 **

 ** This variable normally has file scope only.  But during testing, we make

 ** it a global so that the test code can change its value in order to verify

 ** that the right stuff happens in either case.

 */

 #ifndef SQLITE_THREAD_OVERRIDE_LOCK

 # define SQLITE_THREAD_OVERRIDE_LOCK -1

 #endif

 #ifdef SQLITE_TEST

 int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;

 #else

 static int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;

 #endif

 /*

 ** This structure holds information passed into individual test

 ** threads by the testThreadLockingBehavior() routine.

 */

 struct threadTestData {

   int fd;                /* File to be locked */

   struct flock lock;     /* The locking operation */

   int result;            /* Result of the locking operation */

 };

 #ifdef SQLITE_LOCK_TRACE

 /*

 ** Print out information about all locking operations.

 **

 ** This routine is used for troubleshooting locks on multithreaded

 ** platforms.  Enable by compiling with the -DSQLITE_LOCK_TRACE

 ** command-line option on the compiler.  This code is normally

 ** turned off.

 */

 static int lockTrace(int fd, int op, struct flock *p){

   char *zOpName, *zType;

   int s;

   int savedErrno;

   if( op==F_GETLK ){

     zOpName = "GETLK";

   }else if( op==F_SETLK ){

     zOpName = "SETLK";

   }else{

     s = fcntl(fd, op, p);

     sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd, op, s);

     return s;

   }

   if( p->l_type==F_RDLCK ){

     zType = "RDLCK";

   }else if( p->l_type==F_WRLCK ){

     zType = "WRLCK";

   }else if( p->l_type==F_UNLCK ){

     zType = "UNLCK";

   }else{

     assert( 0 );

   }

   assert( p->l_whence==SEEK_SET );

   s = fcntl(fd, op, p);

   savedErrno = errno;

   sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n",

      threadid, fd, zOpName, zType, (int)p->l_start, (int)p->l_len,

      (int)p->l_pid, s);

   if( s==(-1) && op==F_SETLK && (p->l_type==F_RDLCK || p->l_type==F_WRLCK) ){

     struct flock l2;

     l2 = *p;

     fcntl(fd, F_GETLK, &l2);

     if( l2.l_type==F_RDLCK ){

       zType = "RDLCK";

     }else if( l2.l_type==F_WRLCK ){

       zType = "WRLCK";

     }else if( l2.l_type==F_UNLCK ){

       zType = "UNLCK";

     }else{

       assert( 0 );

     }

     sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n",

        zType, (int)l2.l_start, (int)l2.l_len, (int)l2.l_pid);

   }

   errno = savedErrno;

   return s;

 }

 #define fcntl lockTrace

 #endif /* SQLITE_LOCK_TRACE */

 /*

 ** The testThreadLockingBehavior() routine launches two separate

 ** threads on this routine.  This routine attempts to lock a file

 ** descriptor then returns.  The success or failure of that attempt

 ** allows the testThreadLockingBehavior() procedure to determine

 ** whether or not threads can override each others locks.

 */

 static void *threadLockingTest(void *pArg){

   struct threadTestData *pData = (struct threadTestData*)pArg;

   pData->result = fcntl(pData->fd, F_SETLK, &pData->lock);

   return pArg;

 }

 /*

 ** This procedure attempts to determine whether or not threads

 ** can override each others locks then sets the 

 ** threadsOverrideEachOthersLocks variable appropriately.

 */

 static void testThreadLockingBehavior(int fd_orig){

   int fd;

   struct threadTestData d[2];

   pthread_t t[2];

   fd = dup(fd_orig);

   if( fd<0 ) return;

   memset(d, 0, sizeof(d));

   d[0].fd = fd;

   d[0].lock.l_type = F_RDLCK;

   d[0].lock.l_len = 1;

   d[0].lock.l_start = 0;

   d[0].lock.l_whence = SEEK_SET;

   d[1] = d[0];

   d[1].lock.l_type = F_WRLCK;

   pthread_create(&t[0], 0, threadLockingTest, &d[0]);

   pthread_create(&t[1], 0, threadLockingTest, &d[1]);

   pthread_join(t[0], 0);

   pthread_join(t[1], 0);

   close(fd);

   threadsOverrideEachOthersLocks =  d[0].result==0 && d[1].result==0;

 }

 #endif /* SQLITE_THREADSAFE */

 /*

 ** Release a lockInfo structure previously allocated by findLockInfo().

 */

 static void releaseLockInfo(struct lockInfo *pLock){

   if (pLock == NULL)

     return;

   pLock->nRef--;

   if( pLock->nRef==0 ){

     sqlite3HashInsert(&lockHash, &pLock->key, sizeof(pLock->key), 0);

     sqlite3_free(pLock);

   }

 }

 /*

 ** Release a openCnt structure previously allocated by findLockInfo().

 */

 static void releaseOpenCnt(struct openCnt *pOpen){

   if (pOpen == NULL)

     return;

   pOpen->nRef--;

   if( pOpen->nRef==0 ){

     sqlite3HashInsert(&openHash, &pOpen->key, sizeof(pOpen->key), 0);

     free(pOpen->aPending);

     sqlite3_free(pOpen);

   }

 }

 #ifdef SQLITE_ENABLE_LOCKING_STYLE

 /*

 ** Tests a byte-range locking query to see if byte range locks are 

 ** supported, if not we fall back to dotlockLockingStyle.

 */

 static sqlite3LockingStyle sqlite3TestLockingStyle(

   const char *filePath, 

   int fd

 ){

   /* test byte-range lock using fcntl */

   struct flock lockInfo;

   lockInfo.l_len = 1;

   lockInfo.l_start = 0;

   lockInfo.l_whence = SEEK_SET;

   lockInfo.l_type = F_RDLCK;

   if( fcntl(fd, F_GETLK, &lockInfo)!=-1 ) {

     return posixLockingStyle;

   } 

   /* testing for flock can give false positives.  So if if the above test

   ** fails, then we fall back to using dot-lock style locking.

   */  

   return dotlockLockingStyle;

 }

 /* 

 ** Examines the f_fstypename entry in the statfs structure as returned by 

 ** stat() for the file system hosting the database file, assigns the 

 ** appropriate locking style based on it's value.  These values and 

 ** assignments are based on Darwin/OSX behavior and have not been tested on 

 ** other systems.

 */

 static sqlite3LockingStyle sqlite3DetectLockingStyle(

   const char *filePath, 

   int fd

 ){

 #ifdef SQLITE_FIXED_LOCKING_STYLE

   return (sqlite3LockingStyle)SQLITE_FIXED_LOCKING_STYLE;

 #else

   struct statfs fsInfo;

   if (statfs(filePath, &fsInfo) == -1)

     return sqlite3TestLockingStyle(filePath, fd);

   if (fsInfo.f_flags & MNT_RDONLY)

     return noLockingStyle;

   if( (!strcmp(fsInfo.f_fstypename, "hfs")) ||

     (!strcmp(fsInfo.f_fstypename, "ufs")) )

                 return posixLockingStyle;

   if(!strcmp(fsInfo.f_fstypename, "afpfs"))

     return afpLockingStyle;

   if(!strcmp(fsInfo.f_fstypename, "nfs")) 

     return sqlite3TestLockingStyle(filePath, fd);

   if(!strcmp(fsInfo.f_fstypename, "smbfs"))

     return flockLockingStyle;

   if(!strcmp(fsInfo.f_fstypename, "msdos"))

     return dotlockLockingStyle;

   if(!strcmp(fsInfo.f_fstypename, "webdav"))

     return unsupportedLockingStyle;

   return sqlite3TestLockingStyle(filePath, fd);  

 #endif /* SQLITE_FIXED_LOCKING_STYLE */

 }

 #endif /* SQLITE_ENABLE_LOCKING_STYLE */

 /*

 ** Given a file descriptor, locate lockInfo and openCnt structures that

 ** describes that file descriptor.  Create new ones if necessary.  The

 ** return values might be uninitialized if an error occurs.

 **

 ** Return the number of errors.

 */

 static int findLockInfo(

   int fd,                      /* The file descriptor used in the key */

   struct lockInfo **ppLock,    /* Return the lockInfo structure here */

   struct openCnt **ppOpen      /* Return the openCnt structure here */

 ){

   int rc;

   struct lockKey key1;

   struct openKey key2;

   struct stat statbuf;

   struct lockInfo *pLock;

   struct openCnt *pOpen;

   rc = fstat(fd, &statbuf);

   if( rc!=0 ) return 1;

   memset(&key1, 0, sizeof(key1));

   key1.dev = statbuf.st_dev;

   key1.ino = statbuf.st_ino;

 #if SQLITE_THREADSAFE

   if( threadsOverrideEachOthersLocks<0 ){

     testThreadLockingBehavior(fd);

   }

   key1.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();

 #endif

   memset(&key2, 0, sizeof(key2));

   key2.dev = statbuf.st_dev;

   key2.ino = statbuf.st_ino;

   pLock = (struct lockInfo*)sqlite3HashFind(&lockHash, &key1, sizeof(key1));

   if( pLock==0 ){

     struct lockInfo *pOld;

     pLock = sqlite3_malloc( sizeof(*pLock) );

     if( pLock==0 ){

       rc = 1;

       goto exit_findlockinfo;

     }

     pLock->key = key1;

     pLock->nRef = 1;

     pLock->cnt = 0;

     pLock->locktype = 0;

     pOld = sqlite3HashInsert(&lockHash, &pLock->key, sizeof(key1), pLock);

     if( pOld!=0 ){

       assert( pOld==pLock );

       sqlite3_free(pLock);

       rc = 1;

       goto exit_findlockinfo;

     }

   }else{

     pLock->nRef++;

   }

   *ppLock = pLock;

   if( ppOpen!=0 ){

     pOpen = (struct openCnt*)sqlite3HashFind(&openHash, &key2, sizeof(key2));

     if( pOpen==0 ){

       struct openCnt *pOld;

       pOpen = sqlite3_malloc( sizeof(*pOpen) );

       if( pOpen==0 ){

         releaseLockInfo(pLock);

         rc = 1;

         goto exit_findlockinfo;

       }

       pOpen->key = key2;

       pOpen->nRef = 1;

       pOpen->nLock = 0;

       pOpen->nPending = 0;

       pOpen->aPending = 0;

       pOld = sqlite3HashInsert(&openHash, &pOpen->key, sizeof(key2), pOpen);

       if( pOld!=0 ){

         assert( pOld==pOpen );

         sqlite3_free(pOpen);

         releaseLockInfo(pLock);

         rc = 1;

         goto exit_findlockinfo;

       }

     }else{

       pOpen->nRef++;

     }

     *ppOpen = pOpen;

   }

 exit_findlockinfo:

   return rc;

 }

 #ifdef SQLITE_DEBUG

 /*

 ** Helper function for printing out trace information from debugging

 ** binaries. This returns the string represetation of the supplied

 ** integer lock-type.

 */

 static const char *locktypeName(int locktype){

   switch( locktype ){

   case NO_LOCK: return "NONE";

   case SHARED_LOCK: return "SHARED";

   case RESERVED_LOCK: return "RESERVED";

   case PENDING_LOCK: return "PENDING";

   case EXCLUSIVE_LOCK: return "EXCLUSIVE";

   }

   return "ERROR";

 }

 #endif

 /*

 ** If we are currently in a different thread than the thread that the

 ** unixFile argument belongs to, then transfer ownership of the unixFile

 ** over to the current thread.

 **

 ** A unixFile is only owned by a thread on systems where one thread is

 ** unable to override locks created by a different thread.  RedHat9 is

 ** an example of such a system.

 **

 ** Ownership transfer is only allowed if the unixFile is currently unlocked.

 ** If the unixFile is locked and an ownership is wrong, then return

 ** SQLITE_MISUSE.  SQLITE_OK is returned if everything works.

 */

 #if SQLITE_THREADSAFE

 static int transferOwnership(unixFile *pFile){

   int rc;

   pthread_t hSelf;

   if( threadsOverrideEachOthersLocks ){

     /* Ownership transfers not needed on this system */

     return SQLITE_OK;

   }

   hSelf = pthread_self();

   if( pthread_equal(pFile->tid, hSelf) ){

     /* We are still in the same thread */

     OSTRACE1("No-transfer, same thread\n");

     return SQLITE_OK;

   }

   if( pFile->locktype!=NO_LOCK ){

     /* We cannot change ownership while we are holding a lock! */

     return SQLITE_MISUSE;

   }

   OSTRACE4("Transfer ownership of %d from %d to %d\n",

             pFile->h, pFile->tid, hSelf);

   pFile->tid = hSelf;

   if (pFile->pLock != NULL) {

     releaseLockInfo(pFile->pLock);

     rc = findLockInfo(pFile->h, &pFile->pLock, 0);

     OSTRACE5("LOCK    %d is now %s(%s,%d)\n", pFile->h,

            locktypeName(pFile->locktype),

            locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);

     return rc;

   } else {

     return SQLITE_OK;

   }

 }

 #else

   /* On single-threaded builds, ownership transfer is a no-op */

 # define transferOwnership(X) SQLITE_OK

 #endif

 /*

 ** Seek to the offset passed as the second argument, then read cnt 

 ** bytes into pBuf. Return the number of bytes actually read.

 */

 static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){

   int got;

   i64 newOffset;

   TIMER_START;

 #if defined(USE_PREAD)

   got = pread(id->h, pBuf, cnt, offset);

   SimulateIOError( got = -1 );

 #elif defined(USE_PREAD64)

   got = pread64(id->h, pBuf, cnt, offset);

   SimulateIOError( got = -1 );

 #else

   newOffset = lseek(id->h, offset, SEEK_SET);

   SimulateIOError( newOffset-- );

   if( newOffset!=offset ){

     return -1;

   }

   got = read(id->h, pBuf, cnt);

 #endif

   TIMER_END;

   OSTRACE5("READ    %-3d %5d %7lld %d\n", id->h, got, offset, TIMER_ELAPSED);

   return got;

 }

 /*

 ** Read data from a file into a buffer.  Return SQLITE_OK if all

 ** bytes were read successfully and SQLITE_IOERR if anything goes

 ** wrong.

 */

 static int unixRead(

   sqlite3_file *id, 

   void *pBuf, 

   int amt,

   sqlite3_int64 offset

 ){

   int got;

   assert( id );

   got = seekAndRead((unixFile*)id, offset, pBuf, amt);

   if( got==amt ){

     return SQLITE_OK;

   }else if( got<0 ){

     return SQLITE_IOERR_READ;

   }else{

     memset(&((char*)pBuf)[got], 0, amt-got);

     return SQLITE_IOERR_SHORT_READ;

   }

 }

 /*

 ** Seek to the offset in id->offset then read cnt bytes into pBuf.

 ** Return the number of bytes actually read.  Update the offset.

 */

 static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){

   int got;

   i64 newOffset;

   TIMER_START;

 #if defined(USE_PREAD)

   got = pwrite(id->h, pBuf, cnt, offset);

 #elif defined(USE_PREAD64)

   got = pwrite64(id->h, pBuf, cnt, offset);

 #else

   newOffset = lseek(id->h, offset, SEEK_SET);

   if( newOffset!=offset ){

     return -1;

   }

   got = write(id->h, pBuf, cnt);

 #endif

   TIMER_END;

   OSTRACE5("WRITE   %-3d %5d %7lld %d\n", id->h, got, offset, TIMER_ELAPSED);

   return got;

 }

 /*

 ** Write data from a buffer into a file.  Return SQLITE_OK on success

 ** or some other error code on failure.

 */

 static int unixWrite(

   sqlite3_file *id, 

   const void *pBuf, 

   int amt,

   sqlite3_int64 offset 

 ){

   int wrote = 0;

   assert( id );

   assert( amt>0 );

   while( amt>0 && (wrote = seekAndWrite((unixFile*)id, offset, pBuf, amt))>0 ){

     amt -= wrote;

     offset += wrote;

     pBuf = &((char*)pBuf)[wrote];

   }

   SimulateIOError(( wrote=(-1), amt=1 ));

   SimulateDiskfullError(( wrote=0, amt=1 ));

   if( amt>0 ){

     if( wrote<0 ){

       return SQLITE_IOERR_WRITE;

     }else{

       return SQLITE_FULL;

     }

   }

   return SQLITE_OK;

 }

 #ifdef SQLITE_TEST

 /*

 ** Count the number of fullsyncs and normal syncs.  This is used to test

 ** that syncs and fullsyncs are occuring at the right times.

 */

 int sqlite3_sync_count = 0;

 int sqlite3_fullsync_count = 0;

 #endif

 /*

 ** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.

 ** Otherwise use fsync() in its place.

 */

 #ifndef HAVE_FDATASYNC

 # define fdatasync fsync

 #endif

 /*

 ** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not

 ** the F_FULLFSYNC macro is defined.  F_FULLFSYNC is currently

 ** only available on Mac OS X.  But that could change.

 */

 #ifdef F_FULLFSYNC

 # define HAVE_FULLFSYNC 1

 #else

 # define HAVE_FULLFSYNC 0

 #endif

 /*

 ** The fsync() system call does not work as advertised on many

 ** unix systems.  The following procedure is an attempt to make

 ** it work better.

 **

 ** The SQLITE_NO_SYNC macro disables all fsync()s.  This is useful

 ** for testing when we want to run through the test suite quickly.

 ** You are strongly advised *not* to deploy with SQLITE_NO_SYNC

 ** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash

 ** or power failure will likely corrupt the database file.

 */

 static int full_fsync(int fd, int fullSync, int dataOnly){

   int rc;

   /* Record the number of times that we do a normal fsync() and 

   ** FULLSYNC.  This is used during testing to verify that this procedure

   ** gets called with the correct arguments.

   */

 #ifdef SQLITE_TEST

   if( fullSync ) sqlite3_fullsync_count++;

   sqlite3_sync_count++;

 #endif

   /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a

   ** no-op

   */

 #ifdef SQLITE_NO_SYNC

   rc = SQLITE_OK;

 #else

 #if HAVE_FULLFSYNC

   if( fullSync ){

     rc = fcntl(fd, F_FULLFSYNC, 0);

   }else{

     rc = 1;

   }

   /* If the FULLFSYNC failed, fall back to attempting an fsync().

    * It shouldn't be possible for fullfsync to fail on the local 

    * file system (on OSX), so failure indicates that FULLFSYNC

    * isn't supported for this file system. So, attempt an fsync 

    * and (for now) ignore the overhead of a superfluous fcntl call.  

    * It'd be better to detect fullfsync support once and avoid 

    * the fcntl call every time sync is called.

    */

   if( rc ) rc = fsync(fd);

 #else 

   if( dataOnly ){

     rc = fdatasync(fd);

   }else{

     rc = fsync(fd);

   }

 #endif /* HAVE_FULLFSYNC */

 #endif /* defined(SQLITE_NO_SYNC) */

   return rc;

 }

 /*

 ** Make sure all writes to a particular file are committed to disk.

 **

 ** If dataOnly==0 then both the file itself and its metadata (file

 ** size, access time, etc) are synced.  If dataOnly!=0 then only the

 ** file data is synced.

 **

 ** Under Unix, also make sure that the directory entry for the file

 ** has been created by fsync-ing the directory that contains the file.

 ** If we do not do this and we encounter a power failure, the directory

 ** entry for the journal might not exist after we reboot.  The next

 ** SQLite to access the file will not know that the journal exists (because

 ** the directory entry for the journal was never created) and the transaction

 ** will not roll back - possibly leading to database corruption.

 */

 static int unixSync(sqlite3_file *id, int flags){

   int rc;

   unixFile *pFile = (unixFile*)id;

   int isDataOnly = (flags&SQLITE_SYNC_DATAONLY);

   int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL;

   /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */

   assert((flags&0x0F)==SQLITE_SYNC_NORMAL

       || (flags&0x0F)==SQLITE_SYNC_FULL

   );

   assert( pFile );

   OSTRACE2("SYNC    %-3d\n", pFile->h);

   rc = full_fsync(pFile->h, isFullsync, isDataOnly);

   SimulateIOError( rc=1 );

   if( rc ){

     return SQLITE_IOERR_FSYNC;

   }

   if( pFile->dirfd>=0 ){

     OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,

             HAVE_FULLFSYNC, isFullsync);

 #ifndef SQLITE_DISABLE_DIRSYNC

     /* The directory sync is only attempted if full_fsync is

     ** turned off or unavailable.  If a full_fsync occurred above,

     ** then the directory sync is superfluous.

     */

     if( (!HAVE_FULLFSYNC || !isFullsync) && full_fsync(pFile->dirfd,0,0) ){

        /*

        ** We have received multiple reports of fsync() returning

        ** errors when applied to directories on certain file systems.

        ** A failed directory sync is not a big deal.  So it seems

        ** better to ignore the error.  Ticket #1657

        */

        /* return SQLITE_IOERR; */

     }

 #endif

     close(pFile->dirfd);  /* Only need to sync once, so close the directory */

     pFile->dirfd = -1;    /* when we are done. */

   }

   return SQLITE_OK;

 }

 /*

 ** Truncate an open file to a specified size

 */

 static int unixTruncate(sqlite3_file *id, i64 nByte){

   int rc;

   assert( id );

   rc = ftruncate(((unixFile*)id)->h, (off_t)nByte);

   SimulateIOError( rc=1 );

   if( rc ){

     return SQLITE_IOERR_TRUNCATE;

   }else{

     return SQLITE_OK;

   }

 }

 /*

 ** Determine the current size of a file in bytes

 */

 static int unixFileSize(sqlite3_file *id, i64 *pSize){

   int rc;

   struct stat buf;

   assert( id );

   rc = fstat(((unixFile*)id)->h, &buf);

   SimulateIOError( rc=1 );

   if( rc!=0 ){

     return SQLITE_IOERR_FSTAT;

   }

   *pSize = buf.st_size;

   return SQLITE_OK;

 }

 /*

 ** This routine checks if there is a RESERVED lock held on the specified

 ** file by this or any other process. If such a lock is held, return

 ** non-zero.  If the file is unlocked or holds only SHARED locks, then

 ** return zero.

 */

 static int unixCheckReservedLock(sqlite3_file *id){

   int r = 0;

   unixFile *pFile = (unixFile*)id;

   assert( pFile );

   enterMutex(); /* Because pFile->pLock is shared across threads */

   /* Check if a thread in this process holds such a lock */

   if( pFile->pLock->locktype>SHARED_LOCK ){

     r = 1;

   }

   /* Otherwise see if some other process holds it.

   */

   if( !r ){

     struct flock lock;

     lock.l_whence = SEEK_SET;

     lock.l_start = RESERVED_BYTE;

     lock.l_len = 1;

     lock.l_type = F_WRLCK;

     fcntl(pFile->h, F_GETLK, &lock);

     if( lock.l_type!=F_UNLCK ){

       r = 1;

     }

   }

   leaveMutex();

   OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);

   return r;

 }

 /*

 ** Lock the file with the lock specified by parameter locktype - one

 ** of the following:

 **

 **     (1) SHARED_LOCK

 **     (2) RESERVED_LOCK

 **     (3) PENDING_LOCK

 **     (4) EXCLUSIVE_LOCK

 **

 ** Sometimes when requesting one lock state, additional lock states

 ** are inserted in between.  The locking might fail on one of the later

 ** transitions leaving the lock state different from what it started but

 ** still short of its goal.  The following chart shows the allowed

 ** transitions and the inserted intermediate states:

 **

 **    UNLOCKED -> SHARED

 **    SHARED -> RESERVED

 **    SHARED -> (PENDING) -> EXCLUSIVE

 **    RESERVED -> (PENDING) -> EXCLUSIVE

 **    PENDING -> EXCLUSIVE

 **

 ** This routine will only increase a lock.  Use the sqlite3OsUnlock()

 ** routine to lower a locking level.

 */

 static int unixLock(sqlite3_file *id, int locktype){

   /* The following describes the implementation of the various locks and