os_unix.c

sqlite最新源码

/*
** 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 the VFS implementation for unix-like operating systems
** include Linux, MacOSX, *BSD, QNX, VxWorks, AIX, HPUX, and others.
**
** There are actually several different VFS implementations in this file.
** The differences are in the way that file locking is done.  The default
** implementation uses Posix Advisory Locks.  Alternative implementations
** use flock(), dot-files, various proprietary locking schemas, or simply
** skip locking all together.
**
** This source file is organized into divisions where the logic for various
** subfunctions is contained within the appropriate division.  PLEASE
** KEEP THE STRUCTURE OF THIS FILE INTACT.  New code should be placed
** in the correct division and should be clearly labeled.
**
** The layout of divisions is as follows:
**
**   *  General-purpose declarations and utility functions.
**   *  Unique file ID logic used by VxWorks.
**   *  Various locking primitive implementations (all except proxy locking):
**      + for Posix Advisory Locks
**      + for no-op locks
**      + for dot-file locks
**      + for flock() locking
**      + for named semaphore locks (VxWorks only)
**      + for AFP filesystem locks (MacOSX only)
**   *  sqlite3_file methods not associated with locking.
**   *  Definitions of sqlite3_io_methods objects for all locking
**      methods plus "finder" functions for each locking method.
**   *  sqlite3_vfs method implementations.
**   *  Locking primitives for the proxy uber-locking-method. (MacOSX only)
**   *  Definitions of sqlite3_vfs objects for all locking methods
**      plus implementations of sqlite3_os_init() and sqlite3_os_end().
**
** $Id: os_unix.c,v 1.241 2009/02/09 17:34:07 drh Exp $
*/
#include "sqliteInt.h"
#if SQLITE_OS_UNIX              /* This file is used on unix only */

/*
** There are various methods for file locking used for concurrency
** control:
**
**   1. POSIX locking (the default),
**   2. No locking,
**   3. Dot-file locking,
**   4. flock() locking,
**   5. AFP locking (OSX only),
**   6. Named POSIX semaphores (VXWorks only),
**   7. proxy locking. (OSX only)
**
** Styles 4, 5, and 7 are only available of SQLITE_ENABLE_LOCKING_STYLE
** is defined to 1.  The SQLITE_ENABLE_LOCKING_STYLE also enables automatic
** selection of the appropriate locking style based on the filesystem
** where the database is located.  
*/
#if !defined(SQLITE_ENABLE_LOCKING_STYLE)
#  if defined(__APPLE__)
#    define SQLITE_ENABLE_LOCKING_STYLE 1
#  else
#    define SQLITE_ENABLE_LOCKING_STYLE 0
#  endif
#endif

/*
** Define the OS_VXWORKS pre-processor macro to 1 if building on 
** vxworks, or 0 otherwise.
*/
#ifndef OS_VXWORKS
#  if defined(__RTP__) || defined(_WRS_KERNEL)
#    define OS_VXWORKS 1
#  else
#    define OS_VXWORKS 0
#  endif
#endif

/*
** 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.
**
** The previous paragraph was written in 2005.  (This paragraph is written
** on 2008-11-28.) These days, all Linux kernels support large files, so
** you should probably leave LFS enabled.  But some embedded platforms might
** lack LFS in which case the SQLITE_DISABLE_LFS macro might still be useful.
*/
#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>

#if SQLITE_ENABLE_LOCKING_STYLE
# include <sys/ioctl.h>
# if OS_VXWORKS
#  include <semaphore.h>
#  include <limits.h>
# else
#  include <sys/file.h>
#  include <sys/param.h>
#  include <sys/mount.h>
# endif
#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

/*
 ** Default permissions when creating auto proxy dir
 */
#ifndef SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
# define SQLITE_DEFAULT_PROXYDIR_PERMISSIONS 0755
#endif

/*
** Maximum supported path-length.
*/
#define MAX_PATHNAME 512

/*
** Only set the lastErrno if the error code is a real error and not 
** a normal expected return code of SQLITE_BUSY or SQLITE_OK
*/
#define IS_LOCK_ERROR(x)  ((x != SQLITE_OK) && (x != SQLITE_BUSY))


/*
** The unixFile structure is subclass of sqlite3_file specific to the unix
** VFS implementations.
*/
typedef struct unixFile unixFile;
struct unixFile {
  sqlite3_io_methods const *pMethod;  /* Always the first entry */
  struct unixOpenCnt *pOpen;       /* Info about all open fd's on this inode */
  struct unixLockInfo *pLock;      /* Info about locks on this inode */
  int h;                           /* The file descriptor */
  int dirfd;                       /* File descriptor for the directory */
  unsigned char locktype;          /* The type of lock held on this fd */
  int lastErrno;                   /* The unix errno from the last I/O error */
  void *lockingContext;            /* Locking style specific state */
#if SQLITE_ENABLE_LOCKING_STYLE
  int openFlags;                   /* The flags specified at open() */
#endif
#if SQLITE_THREADSAFE && defined(__linux__)
  pthread_t tid;                   /* The thread that "owns" this unixFile */
#endif
#if OS_VXWORKS
  int isDelete;                    /* Delete on close if true */
  struct vxworksFileId *pId;       /* Unique file ID */
#endif
#ifndef NDEBUG
  /* The next group of variables are used to track whether or not the
  ** transaction counter in bytes 24-27 of database files are updated
  ** whenever any part of the database changes.  An assertion fault will
  ** occur if a file is updated without also updating the transaction
  ** counter.  This test is made to avoid new problems similar to the
  ** one described by ticket #3584. 
  */
  unsigned char transCntrChng;   /* True if the transaction counter changed */
  unsigned char dbUpdate;        /* True if any part of database file changed */
  unsigned char inNormalWrite;   /* True if in a normal write operation */

  /* If true, that means we are dealing with a database file that has
  ** a range of locking bytes from PENDING_BYTE through PENDING_BYTE+511
  ** which should never be read or written.  Asserts() will verify this */
  unsigned char isLockable;      /* True if file might be locked */
#endif
#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
};

/*
** 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 is 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


/*
** Helper functions to obtain and relinquish the global mutex.
*/
static void unixEnterMutex(void){
  sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER));
}
static void unixLeaveMutex(void){
  sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MASTER));
}


#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

#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 */



/*
** This routine translates a standard POSIX errno code into something
** useful to the clients of the sqlite3 functions.  Specifically, it is
** intended to translate a variety of "try again" errors into SQLITE_BUSY
** and a variety of "please close the file descriptor NOW" errors into 
** SQLITE_IOERR
** 
** Errors during initialization of locks, or file system support for locks,
** should handle ENOLCK, ENOTSUP, EOPNOTSUPP separately.
*/
static int sqliteErrorFromPosixError(int posixError, int sqliteIOErr) {
  switch (posixError) {
  case 0: 
    return SQLITE_OK;
    
  case EAGAIN:
  case ETIMEDOUT:
  case EBUSY:
  case EINTR:
  case ENOLCK:  
    /* random NFS retry error, unless during file system support 
     * introspection, in which it actually means what it says */
    return SQLITE_BUSY;
    
  case EACCES: 
    /* EACCES is like EAGAIN during locking operations, but not any other time*/
    if( (sqliteIOErr == SQLITE_IOERR_LOCK) || 
	(sqliteIOErr == SQLITE_IOERR_UNLOCK) || 
	(sqliteIOErr == SQLITE_IOERR_RDLOCK) ||
	(sqliteIOErr == SQLITE_IOERR_CHECKRESERVEDLOCK) ){
      return SQLITE_BUSY;
    }
    /* else fall through */
  case EPERM: 
    return SQLITE_PERM;
    
  case EDEADLK:
    return SQLITE_IOERR_BLOCKED;
    
#if EOPNOTSUPP!=ENOTSUP
  case EOPNOTSUPP: 
    /* something went terribly awry, unless during file system support 
     * introspection, in which it actually means what it says */