4 ** The author disclaims copyright to this source code. In place of
5 ** a legal notice, here is a blessing:
7 ** May you do good and not evil.
8 ** May you find forgiveness for yourself and forgive others.
9 ** May you share freely, never taking more than you give.
11 ******************************************************************************
13 ** This file contains the VFS implementation for unix-like operating systems
14 ** include Linux, MacOSX, *BSD, QNX, VxWorks, AIX, HPUX, and others.
16 ** There are actually several different VFS implementations in this file.
17 ** The differences are in the way that file locking is done. The default
18 ** implementation uses Posix Advisory Locks. Alternative implementations
19 ** use flock(), dot-files, various proprietary locking schemas, or simply
20 ** skip locking all together.
22 ** This source file is organized into divisions where the logic for various
23 ** subfunctions is contained within the appropriate division. PLEASE
24 ** KEEP THE STRUCTURE OF THIS FILE INTACT. New code should be placed
25 ** in the correct division and should be clearly labeled.
27 ** The layout of divisions is as follows:
29 ** * General-purpose declarations and utility functions.
30 ** * Unique file ID logic used by VxWorks.
31 ** * Various locking primitive implementations (all except proxy locking):
32 ** + for Posix Advisory Locks
34 ** + for dot-file locks
35 ** + for flock() locking
36 ** + for named semaphore locks (VxWorks only)
37 ** + for AFP filesystem locks (MacOSX only)
38 ** * sqlite3_file methods not associated with locking.
39 ** * Definitions of sqlite3_io_methods objects for all locking
40 ** methods plus "finder" functions for each locking method.
41 ** * sqlite3_vfs method implementations.
42 ** * Locking primitives for the proxy uber-locking-method. (MacOSX only)
43 ** * Definitions of sqlite3_vfs objects for all locking methods
44 ** plus implementations of sqlite3_os_init() and sqlite3_os_end().
46 #include "sqliteInt.h"
47 #if SQLITE_OS_UNIX /* This file is used on unix only */
50 ** There are various methods for file locking used for concurrency
53 ** 1. POSIX locking (the default),
55 ** 3. Dot-file locking,
56 ** 4. flock() locking,
57 ** 5. AFP locking (OSX only),
58 ** 6. Named POSIX semaphores (VXWorks only),
59 ** 7. proxy locking. (OSX only)
61 ** Styles 4, 5, and 7 are only available of SQLITE_ENABLE_LOCKING_STYLE
62 ** is defined to 1. The SQLITE_ENABLE_LOCKING_STYLE also enables automatic
63 ** selection of the appropriate locking style based on the filesystem
64 ** where the database is located.
66 #if !defined(SQLITE_ENABLE_LOCKING_STYLE)
67 # if defined(__APPLE__)
68 # define SQLITE_ENABLE_LOCKING_STYLE 1
70 # define SQLITE_ENABLE_LOCKING_STYLE 0
74 /* Use pread() and pwrite() if they are available */
75 #if defined(__APPLE__)
77 # define HAVE_PWRITE 1
79 #if defined(HAVE_PREAD64) && defined(HAVE_PWRITE64)
81 # define USE_PREAD64 1
82 #elif defined(HAVE_PREAD) && defined(HAVE_PWRITE)
88 ** standard include files.
90 #include <sys/types.h>
97 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
98 # include <sys/mman.h>
101 #if SQLITE_ENABLE_LOCKING_STYLE
102 # include <sys/ioctl.h>
103 # include <sys/file.h>
104 # include <sys/param.h>
105 #endif /* SQLITE_ENABLE_LOCKING_STYLE */
107 #if defined(__APPLE__) && ((__MAC_OS_X_VERSION_MIN_REQUIRED > 1050) || \
108 (__IPHONE_OS_VERSION_MIN_REQUIRED > 2000))
109 # if (!defined(TARGET_OS_EMBEDDED) || (TARGET_OS_EMBEDDED==0)) \
110 && (!defined(TARGET_IPHONE_SIMULATOR) || (TARGET_IPHONE_SIMULATOR==0))
111 # define HAVE_GETHOSTUUID 1
113 # warning "gethostuuid() is disabled."
119 # include <sys/ioctl.h>
120 # include <semaphore.h>
122 #endif /* OS_VXWORKS */
124 #if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE
125 # include <sys/mount.h>
133 ** Allowed values of unixFile.fsFlags
135 #define SQLITE_FSFLAGS_IS_MSDOS 0x1
138 ** If we are to be thread-safe, include the pthreads header and define
139 ** the SQLITE_UNIX_THREADS macro.
141 #if SQLITE_THREADSAFE
142 # include <pthread.h>
143 # define SQLITE_UNIX_THREADS 1
147 ** Default permissions when creating a new file
149 #ifndef SQLITE_DEFAULT_FILE_PERMISSIONS
150 # define SQLITE_DEFAULT_FILE_PERMISSIONS 0644
154 ** Default permissions when creating auto proxy dir
156 #ifndef SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
157 # define SQLITE_DEFAULT_PROXYDIR_PERMISSIONS 0755
161 ** Maximum supported path-length.
163 #define MAX_PATHNAME 512
166 ** Maximum supported symbolic links
168 #define SQLITE_MAX_SYMLINKS 100
170 /* Always cast the getpid() return type for compatibility with
171 ** kernel modules in VxWorks. */
172 #define osGetpid(X) (pid_t)getpid()
175 ** Only set the lastErrno if the error code is a real error and not
176 ** a normal expected return code of SQLITE_BUSY or SQLITE_OK
178 #define IS_LOCK_ERROR(x) ((x != SQLITE_OK) && (x != SQLITE_BUSY))
180 /* Forward references */
181 typedef struct unixShm unixShm
; /* Connection shared memory */
182 typedef struct unixShmNode unixShmNode
; /* Shared memory instance */
183 typedef struct unixInodeInfo unixInodeInfo
; /* An i-node */
184 typedef struct UnixUnusedFd UnixUnusedFd
; /* An unused file descriptor */
187 ** Sometimes, after a file handle is closed by SQLite, the file descriptor
188 ** cannot be closed immediately. In these cases, instances of the following
189 ** structure are used to store the file descriptor while waiting for an
190 ** opportunity to either close or reuse it.
192 struct UnixUnusedFd
{
193 int fd
; /* File descriptor to close */
194 int flags
; /* Flags this file descriptor was opened with */
195 UnixUnusedFd
*pNext
; /* Next unused file descriptor on same file */
199 ** The unixFile structure is subclass of sqlite3_file specific to the unix
200 ** VFS implementations.
202 typedef struct unixFile unixFile
;
204 sqlite3_io_methods
const *pMethod
; /* Always the first entry */
205 sqlite3_vfs
*pVfs
; /* The VFS that created this unixFile */
206 unixInodeInfo
*pInode
; /* Info about locks on this inode */
207 int h
; /* The file descriptor */
208 unsigned char eFileLock
; /* The type of lock held on this fd */
209 unsigned short int ctrlFlags
; /* Behavioral bits. UNIXFILE_* flags */
210 int lastErrno
; /* The unix errno from last I/O error */
211 void *lockingContext
; /* Locking style specific state */
212 UnixUnusedFd
*pUnused
; /* Pre-allocated UnixUnusedFd */
213 const char *zPath
; /* Name of the file */
214 unixShm
*pShm
; /* Shared memory segment information */
215 int szChunk
; /* Configured by FCNTL_CHUNK_SIZE */
216 #if SQLITE_MAX_MMAP_SIZE>0
217 int nFetchOut
; /* Number of outstanding xFetch refs */
218 sqlite3_int64 mmapSize
; /* Usable size of mapping at pMapRegion */
219 sqlite3_int64 mmapSizeActual
; /* Actual size of mapping at pMapRegion */
220 sqlite3_int64 mmapSizeMax
; /* Configured FCNTL_MMAP_SIZE value */
221 void *pMapRegion
; /* Memory mapped region */
224 int sectorSize
; /* Device sector size */
225 int deviceCharacteristics
; /* Precomputed device characteristics */
227 #if SQLITE_ENABLE_LOCKING_STYLE
228 int openFlags
; /* The flags specified at open() */
230 #if SQLITE_ENABLE_LOCKING_STYLE || defined(__APPLE__)
231 unsigned fsFlags
; /* cached details from statfs() */
234 struct vxworksFileId
*pId
; /* Unique file ID */
237 /* The next group of variables are used to track whether or not the
238 ** transaction counter in bytes 24-27 of database files are updated
239 ** whenever any part of the database changes. An assertion fault will
240 ** occur if a file is updated without also updating the transaction
241 ** counter. This test is made to avoid new problems similar to the
242 ** one described by ticket #3584.
244 unsigned char transCntrChng
; /* True if the transaction counter changed */
245 unsigned char dbUpdate
; /* True if any part of database file changed */
246 unsigned char inNormalWrite
; /* True if in a normal write operation */
251 /* In test mode, increase the size of this structure a bit so that
252 ** it is larger than the struct CrashFile defined in test6.c.
258 /* This variable holds the process id (pid) from when the xRandomness()
259 ** method was called. If xOpen() is called from a different process id,
260 ** indicating that a fork() has occurred, the PRNG will be reset.
262 static pid_t randomnessPid
= 0;
265 ** Allowed values for the unixFile.ctrlFlags bitmask:
267 #define UNIXFILE_EXCL 0x01 /* Connections from one process only */
268 #define UNIXFILE_RDONLY 0x02 /* Connection is read only */
269 #define UNIXFILE_PERSIST_WAL 0x04 /* Persistent WAL mode */
270 #ifndef SQLITE_DISABLE_DIRSYNC
271 # define UNIXFILE_DIRSYNC 0x08 /* Directory sync needed */
273 # define UNIXFILE_DIRSYNC 0x00
275 #define UNIXFILE_PSOW 0x10 /* SQLITE_IOCAP_POWERSAFE_OVERWRITE */
276 #define UNIXFILE_DELETE 0x20 /* Delete on close */
277 #define UNIXFILE_URI 0x40 /* Filename might have query parameters */
278 #define UNIXFILE_NOLOCK 0x80 /* Do no file locking */
281 ** Include code that is common to all os_*.c files
283 #include "os_common.h"
286 ** Define various macros that are missing from some systems.
289 # define O_LARGEFILE 0
291 #ifdef SQLITE_DISABLE_LFS
293 # define O_LARGEFILE 0
296 # define O_NOFOLLOW 0
303 ** The threadid macro resolves to the thread-id or to 0. Used for
304 ** testing and debugging only.
306 #if SQLITE_THREADSAFE
307 #define threadid pthread_self()
313 ** HAVE_MREMAP defaults to true on Linux and false everywhere else.
315 #if !defined(HAVE_MREMAP)
316 # if defined(__linux__) && defined(_GNU_SOURCE)
317 # define HAVE_MREMAP 1
319 # define HAVE_MREMAP 0
324 ** Explicitly call the 64-bit version of lseek() on Android. Otherwise, lseek()
325 ** is the 32-bit version, even if _FILE_OFFSET_BITS=64 is defined.
328 # define lseek lseek64
332 ** Different Unix systems declare open() in different ways. Same use
333 ** open(const char*,int,mode_t). Others use open(const char*,int,...).
334 ** The difference is important when using a pointer to the function.
336 ** The safest way to deal with the problem is to always use this wrapper
337 ** which always has the same well-defined interface.
339 static int posixOpen(const char *zFile
, int flags
, int mode
){
340 return open(zFile
, flags
, mode
);
343 /* Forward reference */
344 static int openDirectory(const char*, int*);
345 static int unixGetpagesize(void);
348 ** Many system calls are accessed through pointer-to-functions so that
349 ** they may be overridden at runtime to facilitate fault injection during
350 ** testing and sandboxing. The following array holds the names and pointers
351 ** to all overrideable system calls.
353 static struct unix_syscall
{
354 const char *zName
; /* Name of the system call */
355 sqlite3_syscall_ptr pCurrent
; /* Current value of the system call */
356 sqlite3_syscall_ptr pDefault
; /* Default value */
358 { "open", (sqlite3_syscall_ptr
)posixOpen
, 0 },
359 #define osOpen ((int(*)(const char*,int,int))aSyscall[0].pCurrent)
361 { "close", (sqlite3_syscall_ptr
)close
, 0 },
362 #define osClose ((int(*)(int))aSyscall[1].pCurrent)
364 { "access", (sqlite3_syscall_ptr
)access
, 0 },
365 #define osAccess ((int(*)(const char*,int))aSyscall[2].pCurrent)
367 { "getcwd", (sqlite3_syscall_ptr
)getcwd
, 0 },
368 #define osGetcwd ((char*(*)(char*,size_t))aSyscall[3].pCurrent)
370 { "stat", (sqlite3_syscall_ptr
)stat
, 0 },
371 #define osStat ((int(*)(const char*,struct stat*))aSyscall[4].pCurrent)
374 ** The DJGPP compiler environment looks mostly like Unix, but it
375 ** lacks the fcntl() system call. So redefine fcntl() to be something
376 ** that always succeeds. This means that locking does not occur under
377 ** DJGPP. But it is DOS - what did you expect?
381 #define osFstat(a,b,c) 0
383 { "fstat", (sqlite3_syscall_ptr
)fstat
, 0 },
384 #define osFstat ((int(*)(int,struct stat*))aSyscall[5].pCurrent)
387 { "ftruncate", (sqlite3_syscall_ptr
)ftruncate
, 0 },
388 #define osFtruncate ((int(*)(int,off_t))aSyscall[6].pCurrent)
390 { "fcntl", (sqlite3_syscall_ptr
)fcntl
, 0 },
391 #define osFcntl ((int(*)(int,int,...))aSyscall[7].pCurrent)
393 { "read", (sqlite3_syscall_ptr
)read
, 0 },
394 #define osRead ((ssize_t(*)(int,void*,size_t))aSyscall[8].pCurrent)
396 #if defined(USE_PREAD) || SQLITE_ENABLE_LOCKING_STYLE
397 { "pread", (sqlite3_syscall_ptr
)pread
, 0 },
399 { "pread", (sqlite3_syscall_ptr
)0, 0 },
401 #define osPread ((ssize_t(*)(int,void*,size_t,off_t))aSyscall[9].pCurrent)
403 #if defined(USE_PREAD64)
404 { "pread64", (sqlite3_syscall_ptr
)pread64
, 0 },
406 { "pread64", (sqlite3_syscall_ptr
)0, 0 },
408 #define osPread64 ((ssize_t(*)(int,void*,size_t,off64_t))aSyscall[10].pCurrent)
410 { "write", (sqlite3_syscall_ptr
)write
, 0 },
411 #define osWrite ((ssize_t(*)(int,const void*,size_t))aSyscall[11].pCurrent)
413 #if defined(USE_PREAD) || SQLITE_ENABLE_LOCKING_STYLE
414 { "pwrite", (sqlite3_syscall_ptr
)pwrite
, 0 },
416 { "pwrite", (sqlite3_syscall_ptr
)0, 0 },
418 #define osPwrite ((ssize_t(*)(int,const void*,size_t,off_t))\
419 aSyscall[12].pCurrent)
421 #if defined(USE_PREAD64)
422 { "pwrite64", (sqlite3_syscall_ptr
)pwrite64
, 0 },
424 { "pwrite64", (sqlite3_syscall_ptr
)0, 0 },
426 #define osPwrite64 ((ssize_t(*)(int,const void*,size_t,off64_t))\
427 aSyscall[13].pCurrent)
429 { "fchmod", (sqlite3_syscall_ptr
)fchmod
, 0 },
430 #define osFchmod ((int(*)(int,mode_t))aSyscall[14].pCurrent)
432 #if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE
433 { "fallocate", (sqlite3_syscall_ptr
)posix_fallocate
, 0 },
435 { "fallocate", (sqlite3_syscall_ptr
)0, 0 },
437 #define osFallocate ((int(*)(int,off_t,off_t))aSyscall[15].pCurrent)
439 { "unlink", (sqlite3_syscall_ptr
)unlink
, 0 },
440 #define osUnlink ((int(*)(const char*))aSyscall[16].pCurrent)
442 { "openDirectory", (sqlite3_syscall_ptr
)openDirectory
, 0 },
443 #define osOpenDirectory ((int(*)(const char*,int*))aSyscall[17].pCurrent)
445 { "mkdir", (sqlite3_syscall_ptr
)mkdir
, 0 },
446 #define osMkdir ((int(*)(const char*,mode_t))aSyscall[18].pCurrent)
448 { "rmdir", (sqlite3_syscall_ptr
)rmdir
, 0 },
449 #define osRmdir ((int(*)(const char*))aSyscall[19].pCurrent)
451 #if defined(HAVE_FCHOWN)
452 { "fchown", (sqlite3_syscall_ptr
)fchown
, 0 },
454 { "fchown", (sqlite3_syscall_ptr
)0, 0 },
456 #define osFchown ((int(*)(int,uid_t,gid_t))aSyscall[20].pCurrent)
458 { "geteuid", (sqlite3_syscall_ptr
)geteuid
, 0 },
459 #define osGeteuid ((uid_t(*)(void))aSyscall[21].pCurrent)
461 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
462 { "mmap", (sqlite3_syscall_ptr
)mmap
, 0 },
464 { "mmap", (sqlite3_syscall_ptr
)0, 0 },
466 #define osMmap ((void*(*)(void*,size_t,int,int,int,off_t))aSyscall[22].pCurrent)
468 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
469 { "munmap", (sqlite3_syscall_ptr
)munmap
, 0 },
471 { "munmap", (sqlite3_syscall_ptr
)0, 0 },
473 #define osMunmap ((void*(*)(void*,size_t))aSyscall[23].pCurrent)
475 #if HAVE_MREMAP && (!defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0)
476 { "mremap", (sqlite3_syscall_ptr
)mremap
, 0 },
478 { "mremap", (sqlite3_syscall_ptr
)0, 0 },
480 #define osMremap ((void*(*)(void*,size_t,size_t,int,...))aSyscall[24].pCurrent)
482 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
483 { "getpagesize", (sqlite3_syscall_ptr
)unixGetpagesize
, 0 },
485 { "getpagesize", (sqlite3_syscall_ptr
)0, 0 },
487 #define osGetpagesize ((int(*)(void))aSyscall[25].pCurrent)
489 #if defined(HAVE_READLINK)
490 { "readlink", (sqlite3_syscall_ptr
)readlink
, 0 },
492 { "readlink", (sqlite3_syscall_ptr
)0, 0 },
494 #define osReadlink ((ssize_t(*)(const char*,char*,size_t))aSyscall[26].pCurrent)
496 #if defined(HAVE_LSTAT)
497 { "lstat", (sqlite3_syscall_ptr
)lstat
, 0 },
499 { "lstat", (sqlite3_syscall_ptr
)0, 0 },
501 #define osLstat ((int(*)(const char*,struct stat*))aSyscall[27].pCurrent)
503 }; /* End of the overrideable system calls */
507 ** On some systems, calls to fchown() will trigger a message in a security
508 ** log if they come from non-root processes. So avoid calling fchown() if
509 ** we are not running as root.
511 static int robustFchown(int fd
, uid_t uid
, gid_t gid
){
512 #if defined(HAVE_FCHOWN)
513 return osGeteuid() ? 0 : osFchown(fd
,uid
,gid
);
520 ** This is the xSetSystemCall() method of sqlite3_vfs for all of the
521 ** "unix" VFSes. Return SQLITE_OK opon successfully updating the
522 ** system call pointer, or SQLITE_NOTFOUND if there is no configurable
523 ** system call named zName.
525 static int unixSetSystemCall(
526 sqlite3_vfs
*pNotUsed
, /* The VFS pointer. Not used */
527 const char *zName
, /* Name of system call to override */
528 sqlite3_syscall_ptr pNewFunc
/* Pointer to new system call value */
531 int rc
= SQLITE_NOTFOUND
;
533 UNUSED_PARAMETER(pNotUsed
);
535 /* If no zName is given, restore all system calls to their default
536 ** settings and return NULL
539 for(i
=0; i
<sizeof(aSyscall
)/sizeof(aSyscall
[0]); i
++){
540 if( aSyscall
[i
].pDefault
){
541 aSyscall
[i
].pCurrent
= aSyscall
[i
].pDefault
;
545 /* If zName is specified, operate on only the one system call
548 for(i
=0; i
<sizeof(aSyscall
)/sizeof(aSyscall
[0]); i
++){
549 if( strcmp(zName
, aSyscall
[i
].zName
)==0 ){
550 if( aSyscall
[i
].pDefault
==0 ){
551 aSyscall
[i
].pDefault
= aSyscall
[i
].pCurrent
;
554 if( pNewFunc
==0 ) pNewFunc
= aSyscall
[i
].pDefault
;
555 aSyscall
[i
].pCurrent
= pNewFunc
;
564 ** Return the value of a system call. Return NULL if zName is not a
565 ** recognized system call name. NULL is also returned if the system call
566 ** is currently undefined.
568 static sqlite3_syscall_ptr
unixGetSystemCall(
569 sqlite3_vfs
*pNotUsed
,
574 UNUSED_PARAMETER(pNotUsed
);
575 for(i
=0; i
<sizeof(aSyscall
)/sizeof(aSyscall
[0]); i
++){
576 if( strcmp(zName
, aSyscall
[i
].zName
)==0 ) return aSyscall
[i
].pCurrent
;
582 ** Return the name of the first system call after zName. If zName==NULL
583 ** then return the name of the first system call. Return NULL if zName
584 ** is the last system call or if zName is not the name of a valid
587 static const char *unixNextSystemCall(sqlite3_vfs
*p
, const char *zName
){
592 for(i
=0; i
<ArraySize(aSyscall
)-1; i
++){
593 if( strcmp(zName
, aSyscall
[i
].zName
)==0 ) break;
596 for(i
++; i
<ArraySize(aSyscall
); i
++){
597 if( aSyscall
[i
].pCurrent
!=0 ) return aSyscall
[i
].zName
;
603 ** Do not accept any file descriptor less than this value, in order to avoid
604 ** opening database file using file descriptors that are commonly used for
605 ** standard input, output, and error.
607 #ifndef SQLITE_MINIMUM_FILE_DESCRIPTOR
608 # define SQLITE_MINIMUM_FILE_DESCRIPTOR 3
612 ** Invoke open(). Do so multiple times, until it either succeeds or
613 ** fails for some reason other than EINTR.
615 ** If the file creation mode "m" is 0 then set it to the default for
616 ** SQLite. The default is SQLITE_DEFAULT_FILE_PERMISSIONS (normally
617 ** 0644) as modified by the system umask. If m is not 0, then
618 ** make the file creation mode be exactly m ignoring the umask.
620 ** The m parameter will be non-zero only when creating -wal, -journal,
621 ** and -shm files. We want those files to have *exactly* the same
622 ** permissions as their original database, unadulterated by the umask.
623 ** In that way, if a database file is -rw-rw-rw or -rw-rw-r-, and a
624 ** transaction crashes and leaves behind hot journals, then any
625 ** process that is able to write to the database will also be able to
626 ** recover the hot journals.
628 static int robust_open(const char *z
, int f
, mode_t m
){
630 mode_t m2
= m
? m
: SQLITE_DEFAULT_FILE_PERMISSIONS
;
632 #if defined(O_CLOEXEC)
633 fd
= osOpen(z
,f
|O_CLOEXEC
,m2
);
638 if( errno
==EINTR
) continue;
641 if( fd
>=SQLITE_MINIMUM_FILE_DESCRIPTOR
) break;
643 sqlite3_log(SQLITE_WARNING
,
644 "attempt to open \"%s\" as file descriptor %d", z
, fd
);
646 if( osOpen("/dev/null", f
, m
)<0 ) break;
651 if( osFstat(fd
, &statbuf
)==0
652 && statbuf
.st_size
==0
653 && (statbuf
.st_mode
&0777)!=m
658 #if defined(FD_CLOEXEC) && (!defined(O_CLOEXEC) || O_CLOEXEC==0)
659 osFcntl(fd
, F_SETFD
, osFcntl(fd
, F_GETFD
, 0) | FD_CLOEXEC
);
666 ** Helper functions to obtain and relinquish the global mutex. The
667 ** global mutex is used to protect the unixInodeInfo and
668 ** vxworksFileId objects used by this file, all of which may be
669 ** shared by multiple threads.
671 ** Function unixMutexHeld() is used to assert() that the global mutex
672 ** is held when required. This function is only used as part of assert()
676 ** assert( unixMutexHeld() );
679 static void unixEnterMutex(void){
680 sqlite3_mutex_enter(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_VFS1
));
682 static void unixLeaveMutex(void){
683 sqlite3_mutex_leave(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_VFS1
));
686 static int unixMutexHeld(void) {
687 return sqlite3_mutex_held(sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_VFS1
));
692 #ifdef SQLITE_HAVE_OS_TRACE
694 ** Helper function for printing out trace information from debugging
695 ** binaries. This returns the string representation of the supplied
696 ** integer lock-type.
698 static const char *azFileLock(int eFileLock
){
700 case NO_LOCK
: return "NONE";
701 case SHARED_LOCK
: return "SHARED";
702 case RESERVED_LOCK
: return "RESERVED";
703 case PENDING_LOCK
: return "PENDING";
704 case EXCLUSIVE_LOCK
: return "EXCLUSIVE";
710 #ifdef SQLITE_LOCK_TRACE
712 ** Print out information about all locking operations.
714 ** This routine is used for troubleshooting locks on multithreaded
715 ** platforms. Enable by compiling with the -DSQLITE_LOCK_TRACE
716 ** command-line option on the compiler. This code is normally
719 static int lockTrace(int fd
, int op
, struct flock
*p
){
720 char *zOpName
, *zType
;
725 }else if( op
==F_SETLK
){
728 s
= osFcntl(fd
, op
, p
);
729 sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd
, op
, s
);
732 if( p
->l_type
==F_RDLCK
){
734 }else if( p
->l_type
==F_WRLCK
){
736 }else if( p
->l_type
==F_UNLCK
){
741 assert( p
->l_whence
==SEEK_SET
);
742 s
= osFcntl(fd
, op
, p
);
744 sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n",
745 threadid
, fd
, zOpName
, zType
, (int)p
->l_start
, (int)p
->l_len
,
747 if( s
==(-1) && op
==F_SETLK
&& (p
->l_type
==F_RDLCK
|| p
->l_type
==F_WRLCK
) ){
750 osFcntl(fd
, F_GETLK
, &l2
);
751 if( l2
.l_type
==F_RDLCK
){
753 }else if( l2
.l_type
==F_WRLCK
){
755 }else if( l2
.l_type
==F_UNLCK
){
760 sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n",
761 zType
, (int)l2
.l_start
, (int)l2
.l_len
, (int)l2
.l_pid
);
767 #define osFcntl lockTrace
768 #endif /* SQLITE_LOCK_TRACE */
771 ** Retry ftruncate() calls that fail due to EINTR
773 ** All calls to ftruncate() within this file should be made through
774 ** this wrapper. On the Android platform, bypassing the logic below
775 ** could lead to a corrupt database.
777 static int robust_ftruncate(int h
, sqlite3_int64 sz
){
780 /* On Android, ftruncate() always uses 32-bit offsets, even if
781 ** _FILE_OFFSET_BITS=64 is defined. This means it is unsafe to attempt to
782 ** truncate a file to any size larger than 2GiB. Silently ignore any
784 if( sz
>(sqlite3_int64
)0x7FFFFFFF ){
788 do{ rc
= osFtruncate(h
,sz
); }while( rc
<0 && errno
==EINTR
);
793 ** This routine translates a standard POSIX errno code into something
794 ** useful to the clients of the sqlite3 functions. Specifically, it is
795 ** intended to translate a variety of "try again" errors into SQLITE_BUSY
796 ** and a variety of "please close the file descriptor NOW" errors into
799 ** Errors during initialization of locks, or file system support for locks,
800 ** should handle ENOLCK, ENOTSUP, EOPNOTSUPP separately.
802 static int sqliteErrorFromPosixError(int posixError
, int sqliteIOErr
) {
803 assert( (sqliteIOErr
== SQLITE_IOERR_LOCK
) ||
804 (sqliteIOErr
== SQLITE_IOERR_UNLOCK
) ||
805 (sqliteIOErr
== SQLITE_IOERR_RDLOCK
) ||
806 (sqliteIOErr
== SQLITE_IOERR_CHECKRESERVEDLOCK
) );
807 switch (posixError
) {
814 /* random NFS retry error, unless during file system support
815 * introspection, in which it actually means what it says */
827 /******************************************************************************
828 ****************** Begin Unique File ID Utility Used By VxWorks ***************
830 ** On most versions of unix, we can get a unique ID for a file by concatenating
831 ** the device number and the inode number. But this does not work on VxWorks.
832 ** On VxWorks, a unique file id must be based on the canonical filename.
834 ** A pointer to an instance of the following structure can be used as a
835 ** unique file ID in VxWorks. Each instance of this structure contains
836 ** a copy of the canonical filename. There is also a reference count.
837 ** The structure is reclaimed when the number of pointers to it drops to
840 ** There are never very many files open at one time and lookups are not
841 ** a performance-critical path, so it is sufficient to put these
842 ** structures on a linked list.
844 struct vxworksFileId
{
845 struct vxworksFileId
*pNext
; /* Next in a list of them all */
846 int nRef
; /* Number of references to this one */
847 int nName
; /* Length of the zCanonicalName[] string */
848 char *zCanonicalName
; /* Canonical filename */
853 ** All unique filenames are held on a linked list headed by this
856 static struct vxworksFileId
*vxworksFileList
= 0;
859 ** Simplify a filename into its canonical form
860 ** by making the following changes:
862 ** * removing any trailing and duplicate /
863 ** * convert /./ into just /
864 ** * convert /A/../ where A is any simple name into just /
866 ** Changes are made in-place. Return the new name length.
868 ** The original filename is in z[0..n-1]. Return the number of
869 ** characters in the simplified name.
871 static int vxworksSimplifyName(char *z
, int n
){
873 while( n
>1 && z
[n
-1]=='/' ){ n
--; }
874 for(i
=j
=0; i
<n
; i
++){
876 if( z
[i
+1]=='/' ) continue;
877 if( z
[i
+1]=='.' && i
+2<n
&& z
[i
+2]=='/' ){
881 if( z
[i
+1]=='.' && i
+3<n
&& z
[i
+2]=='.' && z
[i
+3]=='/' ){
882 while( j
>0 && z
[j
-1]!='/' ){ j
--; }
895 ** Find a unique file ID for the given absolute pathname. Return
896 ** a pointer to the vxworksFileId object. This pointer is the unique
899 ** The nRef field of the vxworksFileId object is incremented before
900 ** the object is returned. A new vxworksFileId object is created
901 ** and added to the global list if necessary.
903 ** If a memory allocation error occurs, return NULL.
905 static struct vxworksFileId
*vxworksFindFileId(const char *zAbsoluteName
){
906 struct vxworksFileId
*pNew
; /* search key and new file ID */
907 struct vxworksFileId
*pCandidate
; /* For looping over existing file IDs */
908 int n
; /* Length of zAbsoluteName string */
910 assert( zAbsoluteName
[0]=='/' );
911 n
= (int)strlen(zAbsoluteName
);
912 pNew
= sqlite3_malloc64( sizeof(*pNew
) + (n
+1) );
913 if( pNew
==0 ) return 0;
914 pNew
->zCanonicalName
= (char*)&pNew
[1];
915 memcpy(pNew
->zCanonicalName
, zAbsoluteName
, n
+1);
916 n
= vxworksSimplifyName(pNew
->zCanonicalName
, n
);
918 /* Search for an existing entry that matching the canonical name.
919 ** If found, increment the reference count and return a pointer to
920 ** the existing file ID.
923 for(pCandidate
=vxworksFileList
; pCandidate
; pCandidate
=pCandidate
->pNext
){
924 if( pCandidate
->nName
==n
925 && memcmp(pCandidate
->zCanonicalName
, pNew
->zCanonicalName
, n
)==0
934 /* No match was found. We will make a new file ID */
937 pNew
->pNext
= vxworksFileList
;
938 vxworksFileList
= pNew
;
944 ** Decrement the reference count on a vxworksFileId object. Free
945 ** the object when the reference count reaches zero.
947 static void vxworksReleaseFileId(struct vxworksFileId
*pId
){
949 assert( pId
->nRef
>0 );
952 struct vxworksFileId
**pp
;
953 for(pp
=&vxworksFileList
; *pp
&& *pp
!=pId
; pp
= &((*pp
)->pNext
)){}
960 #endif /* OS_VXWORKS */
961 /*************** End of Unique File ID Utility Used By VxWorks ****************
962 ******************************************************************************/
965 /******************************************************************************
966 *************************** Posix Advisory Locking ****************************
968 ** POSIX advisory locks are broken by design. ANSI STD 1003.1 (1996)
969 ** section 6.5.2.2 lines 483 through 490 specify that when a process
970 ** sets or clears a lock, that operation overrides any prior locks set
971 ** by the same process. It does not explicitly say so, but this implies
972 ** that it overrides locks set by the same process using a different
973 ** file descriptor. Consider this test case:
975 ** int fd1 = open("./file1", O_RDWR|O_CREAT, 0644);
976 ** int fd2 = open("./file2", O_RDWR|O_CREAT, 0644);
978 ** Suppose ./file1 and ./file2 are really the same file (because
979 ** one is a hard or symbolic link to the other) then if you set
980 ** an exclusive lock on fd1, then try to get an exclusive lock
981 ** on fd2, it works. I would have expected the second lock to
982 ** fail since there was already a lock on the file due to fd1.
983 ** But not so. Since both locks came from the same process, the
984 ** second overrides the first, even though they were on different
985 ** file descriptors opened on different file names.
987 ** This means that we cannot use POSIX locks to synchronize file access
988 ** among competing threads of the same process. POSIX locks will work fine
989 ** to synchronize access for threads in separate processes, but not
990 ** threads within the same process.
992 ** To work around the problem, SQLite has to manage file locks internally
993 ** on its own. Whenever a new database is opened, we have to find the
994 ** specific inode of the database file (the inode is determined by the
995 ** st_dev and st_ino fields of the stat structure that fstat() fills in)
996 ** and check for locks already existing on that inode. When locks are
997 ** created or removed, we have to look at our own internal record of the
998 ** locks to see if another thread has previously set a lock on that same
1001 ** (Aside: The use of inode numbers as unique IDs does not work on VxWorks.
1002 ** For VxWorks, we have to use the alternative unique ID system based on
1003 ** canonical filename and implemented in the previous division.)
1005 ** The sqlite3_file structure for POSIX is no longer just an integer file
1006 ** descriptor. It is now a structure that holds the integer file
1007 ** descriptor and a pointer to a structure that describes the internal
1008 ** locks on the corresponding inode. There is one locking structure
1009 ** per inode, so if the same inode is opened twice, both unixFile structures
1010 ** point to the same locking structure. The locking structure keeps
1011 ** a reference count (so we will know when to delete it) and a "cnt"
1012 ** field that tells us its internal lock status. cnt==0 means the
1013 ** file is unlocked. cnt==-1 means the file has an exclusive lock.
1014 ** cnt>0 means there are cnt shared locks on the file.
1016 ** Any attempt to lock or unlock a file first checks the locking
1017 ** structure. The fcntl() system call is only invoked to set a
1018 ** POSIX lock if the internal lock structure transitions between
1019 ** a locked and an unlocked state.
1021 ** But wait: there are yet more problems with POSIX advisory locks.
1023 ** If you close a file descriptor that points to a file that has locks,
1024 ** all locks on that file that are owned by the current process are
1025 ** released. To work around this problem, each unixInodeInfo object
1026 ** maintains a count of the number of pending locks on tha inode.
1027 ** When an attempt is made to close an unixFile, if there are
1028 ** other unixFile open on the same inode that are holding locks, the call
1029 ** to close() the file descriptor is deferred until all of the locks clear.
1030 ** The unixInodeInfo structure keeps a list of file descriptors that need to
1031 ** be closed and that list is walked (and cleared) when the last lock
1034 ** Yet another problem: LinuxThreads do not play well with posix locks.
1036 ** Many older versions of linux use the LinuxThreads library which is
1037 ** not posix compliant. Under LinuxThreads, a lock created by thread
1038 ** A cannot be modified or overridden by a different thread B.
1039 ** Only thread A can modify the lock. Locking behavior is correct
1040 ** if the appliation uses the newer Native Posix Thread Library (NPTL)
1041 ** on linux - with NPTL a lock created by thread A can override locks
1042 ** in thread B. But there is no way to know at compile-time which
1043 ** threading library is being used. So there is no way to know at
1044 ** compile-time whether or not thread A can override locks on thread B.
1045 ** One has to do a run-time check to discover the behavior of the
1048 ** SQLite used to support LinuxThreads. But support for LinuxThreads
1049 ** was dropped beginning with version 3.7.0. SQLite will still work with
1050 ** LinuxThreads provided that (1) there is no more than one connection
1051 ** per database file in the same process and (2) database connections
1052 ** do not move across threads.
1056 ** An instance of the following structure serves as the key used
1057 ** to locate a particular unixInodeInfo object.
1060 dev_t dev
; /* Device number */
1062 struct vxworksFileId
*pId
; /* Unique file ID for vxworks. */
1064 /* We are told that some versions of Android contain a bug that
1065 ** sizes ino_t at only 32-bits instead of 64-bits. (See
1066 ** https://android-review.googlesource.com/#/c/115351/3/dist/sqlite3.c)
1067 ** To work around this, always allocate 64-bits for the inode number.
1068 ** On small machines that only have 32-bit inodes, this wastes 4 bytes,
1069 ** but that should not be a big deal. */
1070 /* WAS: ino_t ino; */
1071 u64 ino
; /* Inode number */
1076 ** An instance of the following structure is allocated for each open
1077 ** inode. Or, on LinuxThreads, there is one of these structures for
1078 ** each inode opened by each thread.
1080 ** A single inode can have multiple file descriptors, so each unixFile
1081 ** structure contains a pointer to an instance of this object and this
1082 ** object keeps a count of the number of unixFile pointing to it.
1084 struct unixInodeInfo
{
1085 struct unixFileId fileId
; /* The lookup key */
1086 int nShared
; /* Number of SHARED locks held */
1087 unsigned char eFileLock
; /* One of SHARED_LOCK, RESERVED_LOCK etc. */
1088 unsigned char bProcessLock
; /* An exclusive process lock is held */
1089 int nRef
; /* Number of pointers to this structure */
1090 unixShmNode
*pShmNode
; /* Shared memory associated with this inode */
1091 int nLock
; /* Number of outstanding file locks */
1092 UnixUnusedFd
*pUnused
; /* Unused file descriptors to close */
1093 unixInodeInfo
*pNext
; /* List of all unixInodeInfo objects */
1094 unixInodeInfo
*pPrev
; /* .... doubly linked */
1095 #if SQLITE_ENABLE_LOCKING_STYLE
1096 unsigned long long sharedByte
; /* for AFP simulated shared lock */
1099 sem_t
*pSem
; /* Named POSIX semaphore */
1100 char aSemName
[MAX_PATHNAME
+2]; /* Name of that semaphore */
1105 ** A lists of all unixInodeInfo objects.
1107 static unixInodeInfo
*inodeList
= 0;
1111 ** This function - unixLogErrorAtLine(), is only ever called via the macro
1114 ** It is invoked after an error occurs in an OS function and errno has been
1115 ** set. It logs a message using sqlite3_log() containing the current value of
1116 ** errno and, if possible, the human-readable equivalent from strerror() or
1119 ** The first argument passed to the macro should be the error code that
1120 ** will be returned to SQLite (e.g. SQLITE_IOERR_DELETE, SQLITE_CANTOPEN).
1121 ** The two subsequent arguments should be the name of the OS function that
1122 ** failed (e.g. "unlink", "open") and the associated file-system path,
1125 #define unixLogError(a,b,c) unixLogErrorAtLine(a,b,c,__LINE__)
1126 static int unixLogErrorAtLine(
1127 int errcode
, /* SQLite error code */
1128 const char *zFunc
, /* Name of OS function that failed */
1129 const char *zPath
, /* File path associated with error */
1130 int iLine
/* Source line number where error occurred */
1132 char *zErr
; /* Message from strerror() or equivalent */
1133 int iErrno
= errno
; /* Saved syscall error number */
1135 /* If this is not a threadsafe build (SQLITE_THREADSAFE==0), then use
1136 ** the strerror() function to obtain the human-readable error message
1137 ** equivalent to errno. Otherwise, use strerror_r().
1139 #if SQLITE_THREADSAFE && defined(HAVE_STRERROR_R)
1141 memset(aErr
, 0, sizeof(aErr
));
1144 /* If STRERROR_R_CHAR_P (set by autoconf scripts) or __USE_GNU is defined,
1145 ** assume that the system provides the GNU version of strerror_r() that
1146 ** returns a pointer to a buffer containing the error message. That pointer
1147 ** may point to aErr[], or it may point to some static storage somewhere.
1148 ** Otherwise, assume that the system provides the POSIX version of
1149 ** strerror_r(), which always writes an error message into aErr[].
1151 ** If the code incorrectly assumes that it is the POSIX version that is
1152 ** available, the error message will often be an empty string. Not a
1153 ** huge problem. Incorrectly concluding that the GNU version is available
1154 ** could lead to a segfault though.
1156 #if defined(STRERROR_R_CHAR_P) || defined(__USE_GNU)
1159 strerror_r(iErrno
, aErr
, sizeof(aErr
)-1);
1161 #elif SQLITE_THREADSAFE
1162 /* This is a threadsafe build, but strerror_r() is not available. */
1165 /* Non-threadsafe build, use strerror(). */
1166 zErr
= strerror(iErrno
);
1169 if( zPath
==0 ) zPath
= "";
1170 sqlite3_log(errcode
,
1171 "os_unix.c:%d: (%d) %s(%s) - %s",
1172 iLine
, iErrno
, zFunc
, zPath
, zErr
1179 ** Close a file descriptor.
1181 ** We assume that close() almost always works, since it is only in a
1182 ** very sick application or on a very sick platform that it might fail.
1183 ** If it does fail, simply leak the file descriptor, but do log the
1186 ** Note that it is not safe to retry close() after EINTR since the
1187 ** file descriptor might have already been reused by another thread.
1188 ** So we don't even try to recover from an EINTR. Just log the error
1191 static void robust_close(unixFile
*pFile
, int h
, int lineno
){
1193 unixLogErrorAtLine(SQLITE_IOERR_CLOSE
, "close",
1194 pFile
? pFile
->zPath
: 0, lineno
);
1199 ** Set the pFile->lastErrno. Do this in a subroutine as that provides
1200 ** a convenient place to set a breakpoint.
1202 static void storeLastErrno(unixFile
*pFile
, int error
){
1203 pFile
->lastErrno
= error
;
1207 ** Close all file descriptors accumuated in the unixInodeInfo->pUnused list.
1209 static void closePendingFds(unixFile
*pFile
){
1210 unixInodeInfo
*pInode
= pFile
->pInode
;
1212 UnixUnusedFd
*pNext
;
1213 for(p
=pInode
->pUnused
; p
; p
=pNext
){
1215 robust_close(pFile
, p
->fd
, __LINE__
);
1218 pInode
->pUnused
= 0;
1222 ** Release a unixInodeInfo structure previously allocated by findInodeInfo().
1224 ** The mutex entered using the unixEnterMutex() function must be held
1225 ** when this function is called.
1227 static void releaseInodeInfo(unixFile
*pFile
){
1228 unixInodeInfo
*pInode
= pFile
->pInode
;
1229 assert( unixMutexHeld() );
1230 if( ALWAYS(pInode
) ){
1232 if( pInode
->nRef
==0 ){
1233 assert( pInode
->pShmNode
==0 );
1234 closePendingFds(pFile
);
1235 if( pInode
->pPrev
){
1236 assert( pInode
->pPrev
->pNext
==pInode
);
1237 pInode
->pPrev
->pNext
= pInode
->pNext
;
1239 assert( inodeList
==pInode
);
1240 inodeList
= pInode
->pNext
;
1242 if( pInode
->pNext
){
1243 assert( pInode
->pNext
->pPrev
==pInode
);
1244 pInode
->pNext
->pPrev
= pInode
->pPrev
;
1246 sqlite3_free(pInode
);
1252 ** Given a file descriptor, locate the unixInodeInfo object that
1253 ** describes that file descriptor. Create a new one if necessary. The
1254 ** return value might be uninitialized if an error occurs.
1256 ** The mutex entered using the unixEnterMutex() function must be held
1257 ** when this function is called.
1259 ** Return an appropriate error code.
1261 static int findInodeInfo(
1262 unixFile
*pFile
, /* Unix file with file desc used in the key */
1263 unixInodeInfo
**ppInode
/* Return the unixInodeInfo object here */
1265 int rc
; /* System call return code */
1266 int fd
; /* The file descriptor for pFile */
1267 struct unixFileId fileId
; /* Lookup key for the unixInodeInfo */
1268 struct stat statbuf
; /* Low-level file information */
1269 unixInodeInfo
*pInode
= 0; /* Candidate unixInodeInfo object */
1271 assert( unixMutexHeld() );
1273 /* Get low-level information about the file that we can used to
1274 ** create a unique name for the file.
1277 rc
= osFstat(fd
, &statbuf
);
1279 storeLastErrno(pFile
, errno
);
1280 #if defined(EOVERFLOW) && defined(SQLITE_DISABLE_LFS)
1281 if( pFile
->lastErrno
==EOVERFLOW
) return SQLITE_NOLFS
;
1283 return SQLITE_IOERR
;
1287 /* On OS X on an msdos filesystem, the inode number is reported
1288 ** incorrectly for zero-size files. See ticket #3260. To work
1289 ** around this problem (we consider it a bug in OS X, not SQLite)
1290 ** we always increase the file size to 1 by writing a single byte
1291 ** prior to accessing the inode number. The one byte written is
1292 ** an ASCII 'S' character which also happens to be the first byte
1293 ** in the header of every SQLite database. In this way, if there
1294 ** is a race condition such that another thread has already populated
1295 ** the first page of the database, no damage is done.
1297 if( statbuf
.st_size
==0 && (pFile
->fsFlags
& SQLITE_FSFLAGS_IS_MSDOS
)!=0 ){
1298 do{ rc
= osWrite(fd
, "S", 1); }while( rc
<0 && errno
==EINTR
);
1300 storeLastErrno(pFile
, errno
);
1301 return SQLITE_IOERR
;
1303 rc
= osFstat(fd
, &statbuf
);
1305 storeLastErrno(pFile
, errno
);
1306 return SQLITE_IOERR
;
1311 memset(&fileId
, 0, sizeof(fileId
));
1312 fileId
.dev
= statbuf
.st_dev
;
1314 fileId
.pId
= pFile
->pId
;
1316 fileId
.ino
= (u64
)statbuf
.st_ino
;
1319 while( pInode
&& memcmp(&fileId
, &pInode
->fileId
, sizeof(fileId
)) ){
1320 pInode
= pInode
->pNext
;
1323 pInode
= sqlite3_malloc64( sizeof(*pInode
) );
1325 return SQLITE_NOMEM_BKPT
;
1327 memset(pInode
, 0, sizeof(*pInode
));
1328 memcpy(&pInode
->fileId
, &fileId
, sizeof(fileId
));
1330 pInode
->pNext
= inodeList
;
1332 if( inodeList
) inodeList
->pPrev
= pInode
;
1342 ** Return TRUE if pFile has been renamed or unlinked since it was first opened.
1344 static int fileHasMoved(unixFile
*pFile
){
1346 return pFile
->pInode
!=0 && pFile
->pId
!=pFile
->pInode
->fileId
.pId
;
1349 return pFile
->pInode
!=0 &&
1350 (osStat(pFile
->zPath
, &buf
)!=0
1351 || (u64
)buf
.st_ino
!=pFile
->pInode
->fileId
.ino
);
1357 ** Check a unixFile that is a database. Verify the following:
1359 ** (1) There is exactly one hard link on the file
1360 ** (2) The file is not a symbolic link
1361 ** (3) The file has not been renamed or unlinked
1363 ** Issue sqlite3_log(SQLITE_WARNING,...) messages if anything is not right.
1365 static void verifyDbFile(unixFile
*pFile
){
1369 /* These verifications occurs for the main database only */
1370 if( pFile
->ctrlFlags
& UNIXFILE_NOLOCK
) return;
1372 rc
= osFstat(pFile
->h
, &buf
);
1374 sqlite3_log(SQLITE_WARNING
, "cannot fstat db file %s", pFile
->zPath
);
1377 if( buf
.st_nlink
==0 ){
1378 sqlite3_log(SQLITE_WARNING
, "file unlinked while open: %s", pFile
->zPath
);
1381 if( buf
.st_nlink
>1 ){
1382 sqlite3_log(SQLITE_WARNING
, "multiple links to file: %s", pFile
->zPath
);
1385 if( fileHasMoved(pFile
) ){
1386 sqlite3_log(SQLITE_WARNING
, "file renamed while open: %s", pFile
->zPath
);
1393 ** This routine checks if there is a RESERVED lock held on the specified
1394 ** file by this or any other process. If such a lock is held, set *pResOut
1395 ** to a non-zero value otherwise *pResOut is set to zero. The return value
1396 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
1398 static int unixCheckReservedLock(sqlite3_file
*id
, int *pResOut
){
1401 unixFile
*pFile
= (unixFile
*)id
;
1403 SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK
; );
1406 assert( pFile
->eFileLock
<=SHARED_LOCK
);
1407 unixEnterMutex(); /* Because pFile->pInode is shared across threads */
1409 /* Check if a thread in this process holds such a lock */
1410 if( pFile
->pInode
->eFileLock
>SHARED_LOCK
){
1414 /* Otherwise see if some other process holds it.
1417 if( !reserved
&& !pFile
->pInode
->bProcessLock
){
1419 lock
.l_whence
= SEEK_SET
;
1420 lock
.l_start
= RESERVED_BYTE
;
1422 lock
.l_type
= F_WRLCK
;
1423 if( osFcntl(pFile
->h
, F_GETLK
, &lock
) ){
1424 rc
= SQLITE_IOERR_CHECKRESERVEDLOCK
;
1425 storeLastErrno(pFile
, errno
);
1426 } else if( lock
.l_type
!=F_UNLCK
){
1433 OSTRACE(("TEST WR-LOCK %d %d %d (unix)\n", pFile
->h
, rc
, reserved
));
1435 *pResOut
= reserved
;
1440 ** Attempt to set a system-lock on the file pFile. The lock is
1441 ** described by pLock.
1443 ** If the pFile was opened read/write from unix-excl, then the only lock
1444 ** ever obtained is an exclusive lock, and it is obtained exactly once
1445 ** the first time any lock is attempted. All subsequent system locking
1446 ** operations become no-ops. Locking operations still happen internally,
1447 ** in order to coordinate access between separate database connections
1448 ** within this process, but all of that is handled in memory and the
1449 ** operating system does not participate.
1451 ** This function is a pass-through to fcntl(F_SETLK) if pFile is using
1452 ** any VFS other than "unix-excl" or if pFile is opened on "unix-excl"
1453 ** and is read-only.
1455 ** Zero is returned if the call completes successfully, or -1 if a call
1456 ** to fcntl() fails. In this case, errno is set appropriately (by fcntl()).
1458 static int unixFileLock(unixFile
*pFile
, struct flock
*pLock
){
1460 unixInodeInfo
*pInode
= pFile
->pInode
;
1461 assert( unixMutexHeld() );
1462 assert( pInode
!=0 );
1463 if( (pFile
->ctrlFlags
& (UNIXFILE_EXCL
|UNIXFILE_RDONLY
))==UNIXFILE_EXCL
){
1464 if( pInode
->bProcessLock
==0 ){
1466 assert( pInode
->nLock
==0 );
1467 lock
.l_whence
= SEEK_SET
;
1468 lock
.l_start
= SHARED_FIRST
;
1469 lock
.l_len
= SHARED_SIZE
;
1470 lock
.l_type
= F_WRLCK
;
1471 rc
= osFcntl(pFile
->h
, F_SETLK
, &lock
);
1472 if( rc
<0 ) return rc
;
1473 pInode
->bProcessLock
= 1;
1479 rc
= osFcntl(pFile
->h
, F_SETLK
, pLock
);
1485 ** Lock the file with the lock specified by parameter eFileLock - one
1486 ** of the following:
1489 ** (2) RESERVED_LOCK
1491 ** (4) EXCLUSIVE_LOCK
1493 ** Sometimes when requesting one lock state, additional lock states
1494 ** are inserted in between. The locking might fail on one of the later
1495 ** transitions leaving the lock state different from what it started but
1496 ** still short of its goal. The following chart shows the allowed
1497 ** transitions and the inserted intermediate states:
1499 ** UNLOCKED -> SHARED
1500 ** SHARED -> RESERVED
1501 ** SHARED -> (PENDING) -> EXCLUSIVE
1502 ** RESERVED -> (PENDING) -> EXCLUSIVE
1503 ** PENDING -> EXCLUSIVE
1505 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
1506 ** routine to lower a locking level.
1508 static int unixLock(sqlite3_file
*id
, int eFileLock
){
1509 /* The following describes the implementation of the various locks and
1510 ** lock transitions in terms of the POSIX advisory shared and exclusive
1511 ** lock primitives (called read-locks and write-locks below, to avoid
1512 ** confusion with SQLite lock names). The algorithms are complicated
1513 ** slightly in order to be compatible with Windows95 systems simultaneously
1514 ** accessing the same database file, in case that is ever required.
1516 ** Symbols defined in os.h indentify the 'pending byte' and the 'reserved
1517 ** byte', each single bytes at well known offsets, and the 'shared byte
1518 ** range', a range of 510 bytes at a well known offset.
1520 ** To obtain a SHARED lock, a read-lock is obtained on the 'pending
1521 ** byte'. If this is successful, 'shared byte range' is read-locked
1522 ** and the lock on the 'pending byte' released. (Legacy note: When
1523 ** SQLite was first developed, Windows95 systems were still very common,
1524 ** and Widnows95 lacks a shared-lock capability. So on Windows95, a
1525 ** single randomly selected by from the 'shared byte range' is locked.
1526 ** Windows95 is now pretty much extinct, but this work-around for the
1527 ** lack of shared-locks on Windows95 lives on, for backwards
1530 ** A process may only obtain a RESERVED lock after it has a SHARED lock.
1531 ** A RESERVED lock is implemented by grabbing a write-lock on the
1534 ** A process may only obtain a PENDING lock after it has obtained a
1535 ** SHARED lock. A PENDING lock is implemented by obtaining a write-lock
1536 ** on the 'pending byte'. This ensures that no new SHARED locks can be
1537 ** obtained, but existing SHARED locks are allowed to persist. A process
1538 ** does not have to obtain a RESERVED lock on the way to a PENDING lock.
1539 ** This property is used by the algorithm for rolling back a journal file
1542 ** An EXCLUSIVE lock, obtained after a PENDING lock is held, is
1543 ** implemented by obtaining a write-lock on the entire 'shared byte
1544 ** range'. Since all other locks require a read-lock on one of the bytes
1545 ** within this range, this ensures that no other locks are held on the
1549 unixFile
*pFile
= (unixFile
*)id
;
1550 unixInodeInfo
*pInode
;
1555 OSTRACE(("LOCK %d %s was %s(%s,%d) pid=%d (unix)\n", pFile
->h
,
1556 azFileLock(eFileLock
), azFileLock(pFile
->eFileLock
),
1557 azFileLock(pFile
->pInode
->eFileLock
), pFile
->pInode
->nShared
,
1560 /* If there is already a lock of this type or more restrictive on the
1561 ** unixFile, do nothing. Don't use the end_lock: exit path, as
1562 ** unixEnterMutex() hasn't been called yet.
1564 if( pFile
->eFileLock
>=eFileLock
){
1565 OSTRACE(("LOCK %d %s ok (already held) (unix)\n", pFile
->h
,
1566 azFileLock(eFileLock
)));
1570 /* Make sure the locking sequence is correct.
1571 ** (1) We never move from unlocked to anything higher than shared lock.
1572 ** (2) SQLite never explicitly requests a pendig lock.
1573 ** (3) A shared lock is always held when a reserve lock is requested.
1575 assert( pFile
->eFileLock
!=NO_LOCK
|| eFileLock
==SHARED_LOCK
);
1576 assert( eFileLock
!=PENDING_LOCK
);
1577 assert( eFileLock
!=RESERVED_LOCK
|| pFile
->eFileLock
==SHARED_LOCK
);
1579 /* This mutex is needed because pFile->pInode is shared across threads
1582 pInode
= pFile
->pInode
;
1584 /* If some thread using this PID has a lock via a different unixFile*
1585 ** handle that precludes the requested lock, return BUSY.
1587 if( (pFile
->eFileLock
!=pInode
->eFileLock
&&
1588 (pInode
->eFileLock
>=PENDING_LOCK
|| eFileLock
>SHARED_LOCK
))
1594 /* If a SHARED lock is requested, and some thread using this PID already
1595 ** has a SHARED or RESERVED lock, then increment reference counts and
1596 ** return SQLITE_OK.
1598 if( eFileLock
==SHARED_LOCK
&&
1599 (pInode
->eFileLock
==SHARED_LOCK
|| pInode
->eFileLock
==RESERVED_LOCK
) ){
1600 assert( eFileLock
==SHARED_LOCK
);
1601 assert( pFile
->eFileLock
==0 );
1602 assert( pInode
->nShared
>0 );
1603 pFile
->eFileLock
= SHARED_LOCK
;
1610 /* A PENDING lock is needed before acquiring a SHARED lock and before
1611 ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will
1615 lock
.l_whence
= SEEK_SET
;
1616 if( eFileLock
==SHARED_LOCK
1617 || (eFileLock
==EXCLUSIVE_LOCK
&& pFile
->eFileLock
<PENDING_LOCK
)
1619 lock
.l_type
= (eFileLock
==SHARED_LOCK
?F_RDLCK
:F_WRLCK
);
1620 lock
.l_start
= PENDING_BYTE
;
1621 if( unixFileLock(pFile
, &lock
) ){
1623 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
1624 if( rc
!=SQLITE_BUSY
){
1625 storeLastErrno(pFile
, tErrno
);
1632 /* If control gets to this point, then actually go ahead and make
1633 ** operating system calls for the specified lock.
1635 if( eFileLock
==SHARED_LOCK
){
1636 assert( pInode
->nShared
==0 );
1637 assert( pInode
->eFileLock
==0 );
1638 assert( rc
==SQLITE_OK
);
1640 /* Now get the read-lock */
1641 lock
.l_start
= SHARED_FIRST
;
1642 lock
.l_len
= SHARED_SIZE
;
1643 if( unixFileLock(pFile
, &lock
) ){
1645 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
1648 /* Drop the temporary PENDING lock */
1649 lock
.l_start
= PENDING_BYTE
;
1651 lock
.l_type
= F_UNLCK
;
1652 if( unixFileLock(pFile
, &lock
) && rc
==SQLITE_OK
){
1653 /* This could happen with a network mount */
1655 rc
= SQLITE_IOERR_UNLOCK
;
1659 if( rc
!=SQLITE_BUSY
){
1660 storeLastErrno(pFile
, tErrno
);
1664 pFile
->eFileLock
= SHARED_LOCK
;
1666 pInode
->nShared
= 1;
1668 }else if( eFileLock
==EXCLUSIVE_LOCK
&& pInode
->nShared
>1 ){
1669 /* We are trying for an exclusive lock but another thread in this
1670 ** same process is still holding a shared lock. */
1673 /* The request was for a RESERVED or EXCLUSIVE lock. It is
1674 ** assumed that there is a SHARED or greater lock on the file
1677 assert( 0!=pFile
->eFileLock
);
1678 lock
.l_type
= F_WRLCK
;
1680 assert( eFileLock
==RESERVED_LOCK
|| eFileLock
==EXCLUSIVE_LOCK
);
1681 if( eFileLock
==RESERVED_LOCK
){
1682 lock
.l_start
= RESERVED_BYTE
;
1685 lock
.l_start
= SHARED_FIRST
;
1686 lock
.l_len
= SHARED_SIZE
;
1689 if( unixFileLock(pFile
, &lock
) ){
1691 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
1692 if( rc
!=SQLITE_BUSY
){
1693 storeLastErrno(pFile
, tErrno
);
1700 /* Set up the transaction-counter change checking flags when
1701 ** transitioning from a SHARED to a RESERVED lock. The change
1702 ** from SHARED to RESERVED marks the beginning of a normal
1703 ** write operation (not a hot journal rollback).
1706 && pFile
->eFileLock
<=SHARED_LOCK
1707 && eFileLock
==RESERVED_LOCK
1709 pFile
->transCntrChng
= 0;
1710 pFile
->dbUpdate
= 0;
1711 pFile
->inNormalWrite
= 1;
1716 if( rc
==SQLITE_OK
){
1717 pFile
->eFileLock
= eFileLock
;
1718 pInode
->eFileLock
= eFileLock
;
1719 }else if( eFileLock
==EXCLUSIVE_LOCK
){
1720 pFile
->eFileLock
= PENDING_LOCK
;
1721 pInode
->eFileLock
= PENDING_LOCK
;
1726 OSTRACE(("LOCK %d %s %s (unix)\n", pFile
->h
, azFileLock(eFileLock
),
1727 rc
==SQLITE_OK
? "ok" : "failed"));
1732 ** Add the file descriptor used by file handle pFile to the corresponding
1735 static void setPendingFd(unixFile
*pFile
){
1736 unixInodeInfo
*pInode
= pFile
->pInode
;
1737 UnixUnusedFd
*p
= pFile
->pUnused
;
1738 p
->pNext
= pInode
->pUnused
;
1739 pInode
->pUnused
= p
;
1745 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
1746 ** must be either NO_LOCK or SHARED_LOCK.
1748 ** If the locking level of the file descriptor is already at or below
1749 ** the requested locking level, this routine is a no-op.
1751 ** If handleNFSUnlock is true, then on downgrading an EXCLUSIVE_LOCK to SHARED
1752 ** the byte range is divided into 2 parts and the first part is unlocked then
1753 ** set to a read lock, then the other part is simply unlocked. This works
1754 ** around a bug in BSD NFS lockd (also seen on MacOSX 10.3+) that fails to
1755 ** remove the write lock on a region when a read lock is set.
1757 static int posixUnlock(sqlite3_file
*id
, int eFileLock
, int handleNFSUnlock
){
1758 unixFile
*pFile
= (unixFile
*)id
;
1759 unixInodeInfo
*pInode
;
1764 OSTRACE(("UNLOCK %d %d was %d(%d,%d) pid=%d (unix)\n", pFile
->h
, eFileLock
,
1765 pFile
->eFileLock
, pFile
->pInode
->eFileLock
, pFile
->pInode
->nShared
,
1768 assert( eFileLock
<=SHARED_LOCK
);
1769 if( pFile
->eFileLock
<=eFileLock
){
1773 pInode
= pFile
->pInode
;
1774 assert( pInode
->nShared
!=0 );
1775 if( pFile
->eFileLock
>SHARED_LOCK
){
1776 assert( pInode
->eFileLock
==pFile
->eFileLock
);
1779 /* When reducing a lock such that other processes can start
1780 ** reading the database file again, make sure that the
1781 ** transaction counter was updated if any part of the database
1782 ** file changed. If the transaction counter is not updated,
1783 ** other connections to the same file might not realize that
1784 ** the file has changed and hence might not know to flush their
1785 ** cache. The use of a stale cache can lead to database corruption.
1787 pFile
->inNormalWrite
= 0;
1790 /* downgrading to a shared lock on NFS involves clearing the write lock
1791 ** before establishing the readlock - to avoid a race condition we downgrade
1792 ** the lock in 2 blocks, so that part of the range will be covered by a
1793 ** write lock until the rest is covered by a read lock:
1799 if( eFileLock
==SHARED_LOCK
){
1800 #if !defined(__APPLE__) || !SQLITE_ENABLE_LOCKING_STYLE
1801 (void)handleNFSUnlock
;
1802 assert( handleNFSUnlock
==0 );
1804 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
1805 if( handleNFSUnlock
){
1806 int tErrno
; /* Error code from system call errors */
1807 off_t divSize
= SHARED_SIZE
- 1;
1809 lock
.l_type
= F_UNLCK
;
1810 lock
.l_whence
= SEEK_SET
;
1811 lock
.l_start
= SHARED_FIRST
;
1812 lock
.l_len
= divSize
;
1813 if( unixFileLock(pFile
, &lock
)==(-1) ){
1815 rc
= SQLITE_IOERR_UNLOCK
;
1816 storeLastErrno(pFile
, tErrno
);
1819 lock
.l_type
= F_RDLCK
;
1820 lock
.l_whence
= SEEK_SET
;
1821 lock
.l_start
= SHARED_FIRST
;
1822 lock
.l_len
= divSize
;
1823 if( unixFileLock(pFile
, &lock
)==(-1) ){
1825 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_RDLOCK
);
1826 if( IS_LOCK_ERROR(rc
) ){
1827 storeLastErrno(pFile
, tErrno
);
1831 lock
.l_type
= F_UNLCK
;
1832 lock
.l_whence
= SEEK_SET
;
1833 lock
.l_start
= SHARED_FIRST
+divSize
;
1834 lock
.l_len
= SHARED_SIZE
-divSize
;
1835 if( unixFileLock(pFile
, &lock
)==(-1) ){
1837 rc
= SQLITE_IOERR_UNLOCK
;
1838 storeLastErrno(pFile
, tErrno
);
1842 #endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
1844 lock
.l_type
= F_RDLCK
;
1845 lock
.l_whence
= SEEK_SET
;
1846 lock
.l_start
= SHARED_FIRST
;
1847 lock
.l_len
= SHARED_SIZE
;
1848 if( unixFileLock(pFile
, &lock
) ){
1849 /* In theory, the call to unixFileLock() cannot fail because another
1850 ** process is holding an incompatible lock. If it does, this
1851 ** indicates that the other process is not following the locking
1852 ** protocol. If this happens, return SQLITE_IOERR_RDLOCK. Returning
1853 ** SQLITE_BUSY would confuse the upper layer (in practice it causes
1854 ** an assert to fail). */
1855 rc
= SQLITE_IOERR_RDLOCK
;
1856 storeLastErrno(pFile
, errno
);
1861 lock
.l_type
= F_UNLCK
;
1862 lock
.l_whence
= SEEK_SET
;
1863 lock
.l_start
= PENDING_BYTE
;
1864 lock
.l_len
= 2L; assert( PENDING_BYTE
+1==RESERVED_BYTE
);
1865 if( unixFileLock(pFile
, &lock
)==0 ){
1866 pInode
->eFileLock
= SHARED_LOCK
;
1868 rc
= SQLITE_IOERR_UNLOCK
;
1869 storeLastErrno(pFile
, errno
);
1873 if( eFileLock
==NO_LOCK
){
1874 /* Decrement the shared lock counter. Release the lock using an
1875 ** OS call only when all threads in this same process have released
1879 if( pInode
->nShared
==0 ){
1880 lock
.l_type
= F_UNLCK
;
1881 lock
.l_whence
= SEEK_SET
;
1882 lock
.l_start
= lock
.l_len
= 0L;
1883 if( unixFileLock(pFile
, &lock
)==0 ){
1884 pInode
->eFileLock
= NO_LOCK
;
1886 rc
= SQLITE_IOERR_UNLOCK
;
1887 storeLastErrno(pFile
, errno
);
1888 pInode
->eFileLock
= NO_LOCK
;
1889 pFile
->eFileLock
= NO_LOCK
;
1893 /* Decrement the count of locks against this same file. When the
1894 ** count reaches zero, close any other file descriptors whose close
1895 ** was deferred because of outstanding locks.
1898 assert( pInode
->nLock
>=0 );
1899 if( pInode
->nLock
==0 ){
1900 closePendingFds(pFile
);
1906 if( rc
==SQLITE_OK
) pFile
->eFileLock
= eFileLock
;
1911 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
1912 ** must be either NO_LOCK or SHARED_LOCK.
1914 ** If the locking level of the file descriptor is already at or below
1915 ** the requested locking level, this routine is a no-op.
1917 static int unixUnlock(sqlite3_file
*id
, int eFileLock
){
1918 #if SQLITE_MAX_MMAP_SIZE>0
1919 assert( eFileLock
==SHARED_LOCK
|| ((unixFile
*)id
)->nFetchOut
==0 );
1921 return posixUnlock(id
, eFileLock
, 0);
1924 #if SQLITE_MAX_MMAP_SIZE>0
1925 static int unixMapfile(unixFile
*pFd
, i64 nByte
);
1926 static void unixUnmapfile(unixFile
*pFd
);
1930 ** This function performs the parts of the "close file" operation
1931 ** common to all locking schemes. It closes the directory and file
1932 ** handles, if they are valid, and sets all fields of the unixFile
1935 ** It is *not* necessary to hold the mutex when this routine is called,
1936 ** even on VxWorks. A mutex will be acquired on VxWorks by the
1937 ** vxworksReleaseFileId() routine.
1939 static int closeUnixFile(sqlite3_file
*id
){
1940 unixFile
*pFile
= (unixFile
*)id
;
1941 #if SQLITE_MAX_MMAP_SIZE>0
1942 unixUnmapfile(pFile
);
1945 robust_close(pFile
, pFile
->h
, __LINE__
);
1950 if( pFile
->ctrlFlags
& UNIXFILE_DELETE
){
1951 osUnlink(pFile
->pId
->zCanonicalName
);
1953 vxworksReleaseFileId(pFile
->pId
);
1957 #ifdef SQLITE_UNLINK_AFTER_CLOSE
1958 if( pFile
->ctrlFlags
& UNIXFILE_DELETE
){
1959 osUnlink(pFile
->zPath
);
1960 sqlite3_free(*(char**)&pFile
->zPath
);
1964 OSTRACE(("CLOSE %-3d\n", pFile
->h
));
1966 sqlite3_free(pFile
->pUnused
);
1967 memset(pFile
, 0, sizeof(unixFile
));
1974 static int unixClose(sqlite3_file
*id
){
1976 unixFile
*pFile
= (unixFile
*)id
;
1977 verifyDbFile(pFile
);
1978 unixUnlock(id
, NO_LOCK
);
1981 /* unixFile.pInode is always valid here. Otherwise, a different close
1982 ** routine (e.g. nolockClose()) would be called instead.
1984 assert( pFile
->pInode
->nLock
>0 || pFile
->pInode
->bProcessLock
==0 );
1985 if( ALWAYS(pFile
->pInode
) && pFile
->pInode
->nLock
){
1986 /* If there are outstanding locks, do not actually close the file just
1987 ** yet because that would clear those locks. Instead, add the file
1988 ** descriptor to pInode->pUnused list. It will be automatically closed
1989 ** when the last lock is cleared.
1991 setPendingFd(pFile
);
1993 releaseInodeInfo(pFile
);
1994 rc
= closeUnixFile(id
);
1999 /************** End of the posix advisory lock implementation *****************
2000 ******************************************************************************/
2002 /******************************************************************************
2003 ****************************** No-op Locking **********************************
2005 ** Of the various locking implementations available, this is by far the
2006 ** simplest: locking is ignored. No attempt is made to lock the database
2007 ** file for reading or writing.
2009 ** This locking mode is appropriate for use on read-only databases
2010 ** (ex: databases that are burned into CD-ROM, for example.) It can
2011 ** also be used if the application employs some external mechanism to
2012 ** prevent simultaneous access of the same database by two or more
2013 ** database connections. But there is a serious risk of database
2014 ** corruption if this locking mode is used in situations where multiple
2015 ** database connections are accessing the same database file at the same
2016 ** time and one or more of those connections are writing.
2019 static int nolockCheckReservedLock(sqlite3_file
*NotUsed
, int *pResOut
){
2020 UNUSED_PARAMETER(NotUsed
);
2024 static int nolockLock(sqlite3_file
*NotUsed
, int NotUsed2
){
2025 UNUSED_PARAMETER2(NotUsed
, NotUsed2
);
2028 static int nolockUnlock(sqlite3_file
*NotUsed
, int NotUsed2
){
2029 UNUSED_PARAMETER2(NotUsed
, NotUsed2
);
2036 static int nolockClose(sqlite3_file
*id
) {
2037 return closeUnixFile(id
);
2040 /******************* End of the no-op lock implementation *********************
2041 ******************************************************************************/
2043 /******************************************************************************
2044 ************************* Begin dot-file Locking ******************************
2046 ** The dotfile locking implementation uses the existence of separate lock
2047 ** files (really a directory) to control access to the database. This works
2048 ** on just about every filesystem imaginable. But there are serious downsides:
2050 ** (1) There is zero concurrency. A single reader blocks all other
2051 ** connections from reading or writing the database.
2053 ** (2) An application crash or power loss can leave stale lock files
2054 ** sitting around that need to be cleared manually.
2056 ** Nevertheless, a dotlock is an appropriate locking mode for use if no
2057 ** other locking strategy is available.
2059 ** Dotfile locking works by creating a subdirectory in the same directory as
2060 ** the database and with the same name but with a ".lock" extension added.
2061 ** The existence of a lock directory implies an EXCLUSIVE lock. All other
2062 ** lock types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE.
2066 ** The file suffix added to the data base filename in order to create the
2069 #define DOTLOCK_SUFFIX ".lock"
2072 ** This routine checks if there is a RESERVED lock held on the specified
2073 ** file by this or any other process. If such a lock is held, set *pResOut
2074 ** to a non-zero value otherwise *pResOut is set to zero. The return value
2075 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
2077 ** In dotfile locking, either a lock exists or it does not. So in this
2078 ** variation of CheckReservedLock(), *pResOut is set to true if any lock
2079 ** is held on the file and false if the file is unlocked.
2081 static int dotlockCheckReservedLock(sqlite3_file
*id
, int *pResOut
) {
2084 unixFile
*pFile
= (unixFile
*)id
;
2086 SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK
; );
2089 reserved
= osAccess((const char*)pFile
->lockingContext
, 0)==0;
2090 OSTRACE(("TEST WR-LOCK %d %d %d (dotlock)\n", pFile
->h
, rc
, reserved
));
2091 *pResOut
= reserved
;
2096 ** Lock the file with the lock specified by parameter eFileLock - one
2097 ** of the following:
2100 ** (2) RESERVED_LOCK
2102 ** (4) EXCLUSIVE_LOCK
2104 ** Sometimes when requesting one lock state, additional lock states
2105 ** are inserted in between. The locking might fail on one of the later
2106 ** transitions leaving the lock state different from what it started but
2107 ** still short of its goal. The following chart shows the allowed
2108 ** transitions and the inserted intermediate states:
2110 ** UNLOCKED -> SHARED
2111 ** SHARED -> RESERVED
2112 ** SHARED -> (PENDING) -> EXCLUSIVE
2113 ** RESERVED -> (PENDING) -> EXCLUSIVE
2114 ** PENDING -> EXCLUSIVE
2116 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
2117 ** routine to lower a locking level.
2119 ** With dotfile locking, we really only support state (4): EXCLUSIVE.
2120 ** But we track the other locking levels internally.
2122 static int dotlockLock(sqlite3_file
*id
, int eFileLock
) {
2123 unixFile
*pFile
= (unixFile
*)id
;
2124 char *zLockFile
= (char *)pFile
->lockingContext
;
2128 /* If we have any lock, then the lock file already exists. All we have
2129 ** to do is adjust our internal record of the lock level.
2131 if( pFile
->eFileLock
> NO_LOCK
){
2132 pFile
->eFileLock
= eFileLock
;
2133 /* Always update the timestamp on the old file */
2135 utime(zLockFile
, NULL
);
2137 utimes(zLockFile
, NULL
);
2142 /* grab an exclusive lock */
2143 rc
= osMkdir(zLockFile
, 0777);
2145 /* failed to open/create the lock directory */
2147 if( EEXIST
== tErrno
){
2150 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
2151 if( rc
!=SQLITE_BUSY
){
2152 storeLastErrno(pFile
, tErrno
);
2158 /* got it, set the type and return ok */
2159 pFile
->eFileLock
= eFileLock
;
2164 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
2165 ** must be either NO_LOCK or SHARED_LOCK.
2167 ** If the locking level of the file descriptor is already at or below
2168 ** the requested locking level, this routine is a no-op.
2170 ** When the locking level reaches NO_LOCK, delete the lock file.
2172 static int dotlockUnlock(sqlite3_file
*id
, int eFileLock
) {
2173 unixFile
*pFile
= (unixFile
*)id
;
2174 char *zLockFile
= (char *)pFile
->lockingContext
;
2178 OSTRACE(("UNLOCK %d %d was %d pid=%d (dotlock)\n", pFile
->h
, eFileLock
,
2179 pFile
->eFileLock
, osGetpid(0)));
2180 assert( eFileLock
<=SHARED_LOCK
);
2182 /* no-op if possible */
2183 if( pFile
->eFileLock
==eFileLock
){
2187 /* To downgrade to shared, simply update our internal notion of the
2188 ** lock state. No need to mess with the file on disk.
2190 if( eFileLock
==SHARED_LOCK
){
2191 pFile
->eFileLock
= SHARED_LOCK
;
2195 /* To fully unlock the database, delete the lock file */
2196 assert( eFileLock
==NO_LOCK
);
2197 rc
= osRmdir(zLockFile
);
2200 if( tErrno
==ENOENT
){
2203 rc
= SQLITE_IOERR_UNLOCK
;
2204 storeLastErrno(pFile
, tErrno
);
2208 pFile
->eFileLock
= NO_LOCK
;
2213 ** Close a file. Make sure the lock has been released before closing.
2215 static int dotlockClose(sqlite3_file
*id
) {
2216 unixFile
*pFile
= (unixFile
*)id
;
2218 dotlockUnlock(id
, NO_LOCK
);
2219 sqlite3_free(pFile
->lockingContext
);
2220 return closeUnixFile(id
);
2222 /****************** End of the dot-file lock implementation *******************
2223 ******************************************************************************/
2225 /******************************************************************************
2226 ************************** Begin flock Locking ********************************
2228 ** Use the flock() system call to do file locking.
2230 ** flock() locking is like dot-file locking in that the various
2231 ** fine-grain locking levels supported by SQLite are collapsed into
2232 ** a single exclusive lock. In other words, SHARED, RESERVED, and
2233 ** PENDING locks are the same thing as an EXCLUSIVE lock. SQLite
2234 ** still works when you do this, but concurrency is reduced since
2235 ** only a single process can be reading the database at a time.
2237 ** Omit this section if SQLITE_ENABLE_LOCKING_STYLE is turned off
2239 #if SQLITE_ENABLE_LOCKING_STYLE
2242 ** Retry flock() calls that fail with EINTR
2245 static int robust_flock(int fd
, int op
){
2247 do{ rc
= flock(fd
,op
); }while( rc
<0 && errno
==EINTR
);
2251 # define robust_flock(a,b) flock(a,b)
2256 ** This routine checks if there is a RESERVED lock held on the specified
2257 ** file by this or any other process. If such a lock is held, set *pResOut
2258 ** to a non-zero value otherwise *pResOut is set to zero. The return value
2259 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
2261 static int flockCheckReservedLock(sqlite3_file
*id
, int *pResOut
){
2264 unixFile
*pFile
= (unixFile
*)id
;
2266 SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK
; );
2270 /* Check if a thread in this process holds such a lock */
2271 if( pFile
->eFileLock
>SHARED_LOCK
){
2275 /* Otherwise see if some other process holds it. */
2277 /* attempt to get the lock */
2278 int lrc
= robust_flock(pFile
->h
, LOCK_EX
| LOCK_NB
);
2280 /* got the lock, unlock it */
2281 lrc
= robust_flock(pFile
->h
, LOCK_UN
);
2284 /* unlock failed with an error */
2285 lrc
= SQLITE_IOERR_UNLOCK
;
2286 storeLastErrno(pFile
, tErrno
);
2292 /* someone else might have it reserved */
2293 lrc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
2294 if( IS_LOCK_ERROR(lrc
) ){
2295 storeLastErrno(pFile
, tErrno
);
2300 OSTRACE(("TEST WR-LOCK %d %d %d (flock)\n", pFile
->h
, rc
, reserved
));
2302 #ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS
2303 if( (rc
& SQLITE_IOERR
) == SQLITE_IOERR
){
2307 #endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */
2308 *pResOut
= reserved
;
2313 ** Lock the file with the lock specified by parameter eFileLock - one
2314 ** of the following:
2317 ** (2) RESERVED_LOCK
2319 ** (4) EXCLUSIVE_LOCK
2321 ** Sometimes when requesting one lock state, additional lock states
2322 ** are inserted in between. The locking might fail on one of the later
2323 ** transitions leaving the lock state different from what it started but
2324 ** still short of its goal. The following chart shows the allowed
2325 ** transitions and the inserted intermediate states:
2327 ** UNLOCKED -> SHARED
2328 ** SHARED -> RESERVED
2329 ** SHARED -> (PENDING) -> EXCLUSIVE
2330 ** RESERVED -> (PENDING) -> EXCLUSIVE
2331 ** PENDING -> EXCLUSIVE
2333 ** flock() only really support EXCLUSIVE locks. We track intermediate
2334 ** lock states in the sqlite3_file structure, but all locks SHARED or
2335 ** above are really EXCLUSIVE locks and exclude all other processes from
2338 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
2339 ** routine to lower a locking level.
2341 static int flockLock(sqlite3_file
*id
, int eFileLock
) {
2343 unixFile
*pFile
= (unixFile
*)id
;
2347 /* if we already have a lock, it is exclusive.
2348 ** Just adjust level and punt on outta here. */
2349 if (pFile
->eFileLock
> NO_LOCK
) {
2350 pFile
->eFileLock
= eFileLock
;
2354 /* grab an exclusive lock */
2356 if (robust_flock(pFile
->h
, LOCK_EX
| LOCK_NB
)) {
2358 /* didn't get, must be busy */
2359 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_LOCK
);
2360 if( IS_LOCK_ERROR(rc
) ){
2361 storeLastErrno(pFile
, tErrno
);
2364 /* got it, set the type and return ok */
2365 pFile
->eFileLock
= eFileLock
;
2367 OSTRACE(("LOCK %d %s %s (flock)\n", pFile
->h
, azFileLock(eFileLock
),
2368 rc
==SQLITE_OK
? "ok" : "failed"));
2369 #ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS
2370 if( (rc
& SQLITE_IOERR
) == SQLITE_IOERR
){
2373 #endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */
2379 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
2380 ** must be either NO_LOCK or SHARED_LOCK.
2382 ** If the locking level of the file descriptor is already at or below
2383 ** the requested locking level, this routine is a no-op.
2385 static int flockUnlock(sqlite3_file
*id
, int eFileLock
) {
2386 unixFile
*pFile
= (unixFile
*)id
;
2389 OSTRACE(("UNLOCK %d %d was %d pid=%d (flock)\n", pFile
->h
, eFileLock
,
2390 pFile
->eFileLock
, osGetpid(0)));
2391 assert( eFileLock
<=SHARED_LOCK
);
2393 /* no-op if possible */
2394 if( pFile
->eFileLock
==eFileLock
){
2398 /* shared can just be set because we always have an exclusive */
2399 if (eFileLock
==SHARED_LOCK
) {
2400 pFile
->eFileLock
= eFileLock
;
2404 /* no, really, unlock. */
2405 if( robust_flock(pFile
->h
, LOCK_UN
) ){
2406 #ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS
2408 #endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */
2409 return SQLITE_IOERR_UNLOCK
;
2411 pFile
->eFileLock
= NO_LOCK
;
2419 static int flockClose(sqlite3_file
*id
) {
2421 flockUnlock(id
, NO_LOCK
);
2422 return closeUnixFile(id
);
2425 #endif /* SQLITE_ENABLE_LOCKING_STYLE && !OS_VXWORK */
2427 /******************* End of the flock lock implementation *********************
2428 ******************************************************************************/
2430 /******************************************************************************
2431 ************************ Begin Named Semaphore Locking ************************
2433 ** Named semaphore locking is only supported on VxWorks.
2435 ** Semaphore locking is like dot-lock and flock in that it really only
2436 ** supports EXCLUSIVE locking. Only a single process can read or write
2437 ** the database file at a time. This reduces potential concurrency, but
2438 ** makes the lock implementation much easier.
2443 ** This routine checks if there is a RESERVED lock held on the specified
2444 ** file by this or any other process. If such a lock is held, set *pResOut
2445 ** to a non-zero value otherwise *pResOut is set to zero. The return value
2446 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
2448 static int semXCheckReservedLock(sqlite3_file
*id
, int *pResOut
) {
2451 unixFile
*pFile
= (unixFile
*)id
;
2453 SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK
; );
2457 /* Check if a thread in this process holds such a lock */
2458 if( pFile
->eFileLock
>SHARED_LOCK
){
2462 /* Otherwise see if some other process holds it. */
2464 sem_t
*pSem
= pFile
->pInode
->pSem
;
2466 if( sem_trywait(pSem
)==-1 ){
2468 if( EAGAIN
!= tErrno
){
2469 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_CHECKRESERVEDLOCK
);
2470 storeLastErrno(pFile
, tErrno
);
2472 /* someone else has the lock when we are in NO_LOCK */
2473 reserved
= (pFile
->eFileLock
< SHARED_LOCK
);
2476 /* we could have it if we want it */
2480 OSTRACE(("TEST WR-LOCK %d %d %d (sem)\n", pFile
->h
, rc
, reserved
));
2482 *pResOut
= reserved
;
2487 ** Lock the file with the lock specified by parameter eFileLock - one
2488 ** of the following:
2491 ** (2) RESERVED_LOCK
2493 ** (4) EXCLUSIVE_LOCK
2495 ** Sometimes when requesting one lock state, additional lock states
2496 ** are inserted in between. The locking might fail on one of the later
2497 ** transitions leaving the lock state different from what it started but
2498 ** still short of its goal. The following chart shows the allowed
2499 ** transitions and the inserted intermediate states:
2501 ** UNLOCKED -> SHARED
2502 ** SHARED -> RESERVED
2503 ** SHARED -> (PENDING) -> EXCLUSIVE
2504 ** RESERVED -> (PENDING) -> EXCLUSIVE
2505 ** PENDING -> EXCLUSIVE
2507 ** Semaphore locks only really support EXCLUSIVE locks. We track intermediate
2508 ** lock states in the sqlite3_file structure, but all locks SHARED or
2509 ** above are really EXCLUSIVE locks and exclude all other processes from
2512 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
2513 ** routine to lower a locking level.
2515 static int semXLock(sqlite3_file
*id
, int eFileLock
) {
2516 unixFile
*pFile
= (unixFile
*)id
;
2517 sem_t
*pSem
= pFile
->pInode
->pSem
;
2520 /* if we already have a lock, it is exclusive.
2521 ** Just adjust level and punt on outta here. */
2522 if (pFile
->eFileLock
> NO_LOCK
) {
2523 pFile
->eFileLock
= eFileLock
;
2528 /* lock semaphore now but bail out when already locked. */
2529 if( sem_trywait(pSem
)==-1 ){
2534 /* got it, set the type and return ok */
2535 pFile
->eFileLock
= eFileLock
;
2542 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
2543 ** must be either NO_LOCK or SHARED_LOCK.
2545 ** If the locking level of the file descriptor is already at or below
2546 ** the requested locking level, this routine is a no-op.
2548 static int semXUnlock(sqlite3_file
*id
, int eFileLock
) {
2549 unixFile
*pFile
= (unixFile
*)id
;
2550 sem_t
*pSem
= pFile
->pInode
->pSem
;
2554 OSTRACE(("UNLOCK %d %d was %d pid=%d (sem)\n", pFile
->h
, eFileLock
,
2555 pFile
->eFileLock
, osGetpid(0)));
2556 assert( eFileLock
<=SHARED_LOCK
);
2558 /* no-op if possible */
2559 if( pFile
->eFileLock
==eFileLock
){
2563 /* shared can just be set because we always have an exclusive */
2564 if (eFileLock
==SHARED_LOCK
) {
2565 pFile
->eFileLock
= eFileLock
;
2569 /* no, really unlock. */
2570 if ( sem_post(pSem
)==-1 ) {
2571 int rc
, tErrno
= errno
;
2572 rc
= sqliteErrorFromPosixError(tErrno
, SQLITE_IOERR_UNLOCK
);
2573 if( IS_LOCK_ERROR(rc
) ){
2574 storeLastErrno(pFile
, tErrno
);
2578 pFile
->eFileLock
= NO_LOCK
;
2585 static int semXClose(sqlite3_file
*id
) {
2587 unixFile
*pFile
= (unixFile
*)id
;
2588 semXUnlock(id
, NO_LOCK
);
2591 releaseInodeInfo(pFile
);
2598 #endif /* OS_VXWORKS */
2600 ** Named semaphore locking is only available on VxWorks.
2602 *************** End of the named semaphore lock implementation ****************
2603 ******************************************************************************/
2606 /******************************************************************************
2607 *************************** Begin AFP Locking *********************************
2609 ** AFP is the Apple Filing Protocol. AFP is a network filesystem found
2610 ** on Apple Macintosh computers - both OS9 and OSX.
2612 ** Third-party implementations of AFP are available. But this code here
2613 ** only works on OSX.
2616 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
2618 ** The afpLockingContext structure contains all afp lock specific state
2620 typedef struct afpLockingContext afpLockingContext
;
2621 struct afpLockingContext
{
2623 const char *dbPath
; /* Name of the open file */
2626 struct ByteRangeLockPB2
2628 unsigned long long offset
; /* offset to first byte to lock */
2629 unsigned long long length
; /* nbr of bytes to lock */
2630 unsigned long long retRangeStart
; /* nbr of 1st byte locked if successful */
2631 unsigned char unLockFlag
; /* 1 = unlock, 0 = lock */
2632 unsigned char startEndFlag
; /* 1=rel to end of fork, 0=rel to start */
2633 int fd
; /* file desc to assoc this lock with */
2636 #define afpfsByteRangeLock2FSCTL _IOWR('z', 23, struct ByteRangeLockPB2)
2639 ** This is a utility for setting or clearing a bit-range lock on an
2642 ** Return SQLITE_OK on success, SQLITE_BUSY on failure.
2644 static int afpSetLock(
2645 const char *path
, /* Name of the file to be locked or unlocked */
2646 unixFile
*pFile
, /* Open file descriptor on path */
2647 unsigned long long offset
, /* First byte to be locked */
2648 unsigned long long length
, /* Number of bytes to lock */
2649 int setLockFlag
/* True to set lock. False to clear lock */
2651 struct ByteRangeLockPB2 pb
;
2654 pb
.unLockFlag
= setLockFlag
? 0 : 1;
2655 pb
.startEndFlag
= 0;
2660 OSTRACE(("AFPSETLOCK [%s] for %d%s in range %llx:%llx\n",
2661 (setLockFlag
?"ON":"OFF"), pFile
->h
, (pb
.fd
==-1?"[testval-1]":""),
2663 err
= fsctl(path
, afpfsByteRangeLock2FSCTL
, &pb
, 0);
2667 OSTRACE(("AFPSETLOCK failed to fsctl() '%s' %d %s\n",
2668 path
, tErrno
, strerror(tErrno
)));
2669 #ifdef SQLITE_IGNORE_AFP_LOCK_ERRORS
2672 rc
= sqliteErrorFromPosixError(tErrno
,
2673 setLockFlag
? SQLITE_IOERR_LOCK
: SQLITE_IOERR_UNLOCK
);
2674 #endif /* SQLITE_IGNORE_AFP_LOCK_ERRORS */
2675 if( IS_LOCK_ERROR(rc
) ){
2676 storeLastErrno(pFile
, tErrno
);
2685 ** This routine checks if there is a RESERVED lock held on the specified
2686 ** file by this or any other process. If such a lock is held, set *pResOut
2687 ** to a non-zero value otherwise *pResOut is set to zero. The return value
2688 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
2690 static int afpCheckReservedLock(sqlite3_file
*id
, int *pResOut
){
2693 unixFile
*pFile
= (unixFile
*)id
;
2694 afpLockingContext
*context
;
2696 SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK
; );
2699 context
= (afpLockingContext
*) pFile
->lockingContext
;
2700 if( context
->reserved
){
2704 unixEnterMutex(); /* Because pFile->pInode is shared across threads */
2706 /* Check if a thread in this process holds such a lock */
2707 if( pFile
->pInode
->eFileLock
>SHARED_LOCK
){
2711 /* Otherwise see if some other process holds it.
2714 /* lock the RESERVED byte */
2715 int lrc
= afpSetLock(context
->dbPath
, pFile
, RESERVED_BYTE
, 1,1);
2716 if( SQLITE_OK
==lrc
){
2717 /* if we succeeded in taking the reserved lock, unlock it to restore
2718 ** the original state */
2719 lrc
= afpSetLock(context
->dbPath
, pFile
, RESERVED_BYTE
, 1, 0);
2721 /* if we failed to get the lock then someone else must have it */
2724 if( IS_LOCK_ERROR(lrc
) ){
2730 OSTRACE(("TEST WR-LOCK %d %d %d (afp)\n", pFile
->h
, rc
, reserved
));
2732 *pResOut
= reserved
;
2737 ** Lock the file with the lock specified by parameter eFileLock - one
2738 ** of the following:
2741 ** (2) RESERVED_LOCK
2743 ** (4) EXCLUSIVE_LOCK
2745 ** Sometimes when requesting one lock state, additional lock states
2746 ** are inserted in between. The locking might fail on one of the later
2747 ** transitions leaving the lock state different from what it started but
2748 ** still short of its goal. The following chart shows the allowed
2749 ** transitions and the inserted intermediate states:
2751 ** UNLOCKED -> SHARED
2752 ** SHARED -> RESERVED
2753 ** SHARED -> (PENDING) -> EXCLUSIVE
2754 ** RESERVED -> (PENDING) -> EXCLUSIVE
2755 ** PENDING -> EXCLUSIVE
2757 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
2758 ** routine to lower a locking level.
2760 static int afpLock(sqlite3_file
*id
, int eFileLock
){
2762 unixFile
*pFile
= (unixFile
*)id
;
2763 unixInodeInfo
*pInode
= pFile
->pInode
;
2764 afpLockingContext
*context
= (afpLockingContext
*) pFile
->lockingContext
;
2767 OSTRACE(("LOCK %d %s was %s(%s,%d) pid=%d (afp)\n", pFile
->h
,
2768 azFileLock(eFileLock
), azFileLock(pFile
->eFileLock
),
2769 azFileLock(pInode
->eFileLock
), pInode
->nShared
, osGetpid(0)));
2771 /* If there is already a lock of this type or more restrictive on the
2772 ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as
2773 ** unixEnterMutex() hasn't been called yet.
2775 if( pFile
->eFileLock
>=eFileLock
){
2776 OSTRACE(("LOCK %d %s ok (already held) (afp)\n", pFile
->h
,
2777 azFileLock(eFileLock
)));
2781 /* Make sure the locking sequence is correct
2782 ** (1) We never move from unlocked to anything higher than shared lock.
2783 ** (2) SQLite never explicitly requests a pendig lock.
2784 ** (3) A shared lock is always held when a reserve lock is requested.
2786 assert( pFile
->eFileLock
!=NO_LOCK
|| eFileLock
==SHARED_LOCK
);
2787 assert( eFileLock
!=PENDING_LOCK
);
2788 assert( eFileLock
!=RESERVED_LOCK
|| pFile
->eFileLock
==SHARED_LOCK
);
2790 /* This mutex is needed because pFile->pInode is shared across threads
2793 pInode
= pFile
->pInode
;
2795 /* If some thread using this PID has a lock via a different unixFile*
2796 ** handle that precludes the requested lock, return BUSY.
2798 if( (pFile
->eFileLock
!=pInode
->eFileLock
&&
2799 (pInode
->eFileLock
>=PENDING_LOCK
|| eFileLock
>SHARED_LOCK
))
2805 /* If a SHARED lock is requested, and some thread using this PID already
2806 ** has a SHARED or RESERVED lock, then increment reference counts and
2807 ** return SQLITE_OK.
2809 if( eFileLock
==SHARED_LOCK
&&
2810 (pInode
->eFileLock
==SHARED_LOCK
|| pInode
->eFileLock
==RESERVED_LOCK
) ){
2811 assert( eFileLock
==SHARED_LOCK
);
2812 assert( pFile
->eFileLock
==0 );
2813 assert( pInode
->nShared
>0 );
2814 pFile
->eFileLock
= SHARED_LOCK
;
2820 /* A PENDING lock is needed before acquiring a SHARED lock and before
2821 ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will
2824 if( eFileLock
==SHARED_LOCK
2825 || (eFileLock
==EXCLUSIVE_LOCK
&& pFile
->eFileLock
<PENDING_LOCK
)
2828 failed
= afpSetLock(context
->dbPath
, pFile
, PENDING_BYTE
, 1, 1);
2835 /* If control gets to this point, then actually go ahead and make
2836 ** operating system calls for the specified lock.
2838 if( eFileLock
==SHARED_LOCK
){
2839 int lrc1
, lrc2
, lrc1Errno
= 0;
2842 assert( pInode
->nShared
==0 );
2843 assert( pInode
->eFileLock
==0 );
2845 mask
= (sizeof(long)==8) ? LARGEST_INT64
: 0x7fffffff;
2846 /* Now get the read-lock SHARED_LOCK */
2847 /* note that the quality of the randomness doesn't matter that much */
2849 pInode
->sharedByte
= (lk
& mask
)%(SHARED_SIZE
- 1);
2850 lrc1
= afpSetLock(context
->dbPath
, pFile
,
2851 SHARED_FIRST
+pInode
->sharedByte
, 1, 1);
2852 if( IS_LOCK_ERROR(lrc1
) ){
2853 lrc1Errno
= pFile
->lastErrno
;
2855 /* Drop the temporary PENDING lock */
2856 lrc2
= afpSetLock(context
->dbPath
, pFile
, PENDING_BYTE
, 1, 0);
2858 if( IS_LOCK_ERROR(lrc1
) ) {
2859 storeLastErrno(pFile
, lrc1Errno
);
2862 } else if( IS_LOCK_ERROR(lrc2
) ){
2865 } else if( lrc1
!= SQLITE_OK
) {
2868 pFile
->eFileLock
= SHARED_LOCK
;
2870 pInode
->nShared
= 1;
2872 }else if( eFileLock
==EXCLUSIVE_LOCK
&& pInode
->nShared
>1 ){
2873 /* We are trying for an exclusive lock but another thread in this
2874 ** same process is still holding a shared lock. */
2877 /* The request was for a RESERVED or EXCLUSIVE lock. It is
2878 ** assumed that there is a SHARED or greater lock on the file
2882 assert( 0!=pFile
->eFileLock
);
2883 if (eFileLock
>= RESERVED_LOCK
&& pFile
->eFileLock
< RESERVED_LOCK
) {
2884 /* Acquire a RESERVED lock */
2885 failed
= afpSetLock(context
->dbPath
, pFile
, RESERVED_BYTE
, 1,1);
2887 context
->reserved
= 1;
2890 if (!failed
&& eFileLock
== EXCLUSIVE_LOCK
) {
2891 /* Acquire an EXCLUSIVE lock */
2893 /* Remove the shared lock before trying the range. we'll need to
2894 ** reestablish the shared lock if we can't get the afpUnlock
2896 if( !(failed
= afpSetLock(context
->dbPath
, pFile
, SHARED_FIRST
+
2897 pInode
->sharedByte
, 1, 0)) ){
2898 int failed2
= SQLITE_OK
;
2899 /* now attemmpt to get the exclusive lock range */
2900 failed
= afpSetLock(context
->dbPath
, pFile
, SHARED_FIRST
,
2902 if( failed
&& (failed2
= afpSetLock(context
->dbPath
, pFile
,
2903 SHARED_FIRST
+ pInode
->sharedByte
, 1, 1)) ){
2904 /* Can't reestablish the shared lock. Sqlite can't deal, this is
2905 ** a critical I/O error
2907 rc
= ((failed
& SQLITE_IOERR
) == SQLITE_IOERR
) ? failed2
:
2920 if( rc
==SQLITE_OK
){
2921 pFile
->eFileLock
= eFileLock
;
2922 pInode
->eFileLock
= eFileLock
;
2923 }else if( eFileLock
==EXCLUSIVE_LOCK
){
2924 pFile
->eFileLock
= PENDING_LOCK
;
2925 pInode
->eFileLock
= PENDING_LOCK
;
2930 OSTRACE(("LOCK %d %s %s (afp)\n", pFile
->h
, azFileLock(eFileLock
),
2931 rc
==SQLITE_OK
? "ok" : "failed"));
2936 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
2937 ** must be either NO_LOCK or SHARED_LOCK.
2939 ** If the locking level of the file descriptor is already at or below
2940 ** the requested locking level, this routine is a no-op.
2942 static int afpUnlock(sqlite3_file
*id
, int eFileLock
) {
2944 unixFile
*pFile
= (unixFile
*)id
;
2945 unixInodeInfo
*pInode
;
2946 afpLockingContext
*context
= (afpLockingContext
*) pFile
->lockingContext
;
2953 OSTRACE(("UNLOCK %d %d was %d(%d,%d) pid=%d (afp)\n", pFile
->h
, eFileLock
,
2954 pFile
->eFileLock
, pFile
->pInode
->eFileLock
, pFile
->pInode
->nShared
,
2957 assert( eFileLock
<=SHARED_LOCK
);
2958 if( pFile
->eFileLock
<=eFileLock
){
2962 pInode
= pFile
->pInode
;
2963 assert( pInode
->nShared
!=0 );
2964 if( pFile
->eFileLock
>SHARED_LOCK
){
2965 assert( pInode
->eFileLock
==pFile
->eFileLock
);
2966 SimulateIOErrorBenign(1);
2967 SimulateIOError( h
=(-1) )
2968 SimulateIOErrorBenign(0);
2971 /* When reducing a lock such that other processes can start
2972 ** reading the database file again, make sure that the
2973 ** transaction counter was updated if any part of the database
2974 ** file changed. If the transaction counter is not updated,
2975 ** other connections to the same file might not realize that
2976 ** the file has changed and hence might not know to flush their
2977 ** cache. The use of a stale cache can lead to database corruption.
2979 assert( pFile
->inNormalWrite
==0
2980 || pFile
->dbUpdate
==0
2981 || pFile
->transCntrChng
==1 );
2982 pFile
->inNormalWrite
= 0;
2985 if( pFile
->eFileLock
==EXCLUSIVE_LOCK
){
2986 rc
= afpSetLock(context
->dbPath
, pFile
, SHARED_FIRST
, SHARED_SIZE
, 0);
2987 if( rc
==SQLITE_OK
&& (eFileLock
==SHARED_LOCK
|| pInode
->nShared
>1) ){
2988 /* only re-establish the shared lock if necessary */
2989 int sharedLockByte
= SHARED_FIRST
+pInode
->sharedByte
;
2990 rc
= afpSetLock(context
->dbPath
, pFile
, sharedLockByte
, 1, 1);
2995 if( rc
==SQLITE_OK
&& pFile
->eFileLock
>=PENDING_LOCK
){
2996 rc
= afpSetLock(context
->dbPath
, pFile
, PENDING_BYTE
, 1, 0);
2998 if( rc
==SQLITE_OK
&& pFile
->eFileLock
>=RESERVED_LOCK
&& context
->reserved
){
2999 rc
= afpSetLock(context
->dbPath
, pFile
, RESERVED_BYTE
, 1, 0);
3001 context
->reserved
= 0;
3004 if( rc
==SQLITE_OK
&& (eFileLock
==SHARED_LOCK
|| pInode
->nShared
>1)){
3005 pInode
->eFileLock
= SHARED_LOCK
;
3008 if( rc
==SQLITE_OK
&& eFileLock
==NO_LOCK
){
3010 /* Decrement the shared lock counter. Release the lock using an
3011 ** OS call only when all threads in this same process have released
3014 unsigned long long sharedLockByte
= SHARED_FIRST
+pInode
->sharedByte
;
3016 if( pInode
->nShared
==0 ){
3017 SimulateIOErrorBenign(1);
3018 SimulateIOError( h
=(-1) )
3019 SimulateIOErrorBenign(0);
3021 rc
= afpSetLock(context
->dbPath
, pFile
, sharedLockByte
, 1, 0);
3024 pInode
->eFileLock
= NO_LOCK
;
3025 pFile
->eFileLock
= NO_LOCK
;
3028 if( rc
==SQLITE_OK
){
3030 assert( pInode
->nLock
>=0 );
3031 if( pInode
->nLock
==0 ){
3032 closePendingFds(pFile
);
3038 if( rc
==SQLITE_OK
) pFile
->eFileLock
= eFileLock
;
3043 ** Close a file & cleanup AFP specific locking context
3045 static int afpClose(sqlite3_file
*id
) {
3047 unixFile
*pFile
= (unixFile
*)id
;
3049 afpUnlock(id
, NO_LOCK
);
3051 if( pFile
->pInode
&& pFile
->pInode
->nLock
){
3052 /* If there are outstanding locks, do not actually close the file just
3053 ** yet because that would clear those locks. Instead, add the file
3054 ** descriptor to pInode->aPending. It will be automatically closed when
3055 ** the last lock is cleared.
3057 setPendingFd(pFile
);
3059 releaseInodeInfo(pFile
);
3060 sqlite3_free(pFile
->lockingContext
);
3061 rc
= closeUnixFile(id
);
3066 #endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
3068 ** The code above is the AFP lock implementation. The code is specific
3069 ** to MacOSX and does not work on other unix platforms. No alternative
3070 ** is available. If you don't compile for a mac, then the "unix-afp"
3071 ** VFS is not available.
3073 ********************* End of the AFP lock implementation **********************
3074 ******************************************************************************/
3076 /******************************************************************************
3077 *************************** Begin NFS Locking ********************************/
3079 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
3081 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
3082 ** must be either NO_LOCK or SHARED_LOCK.
3084 ** If the locking level of the file descriptor is already at or below
3085 ** the requested locking level, this routine is a no-op.
3087 static int nfsUnlock(sqlite3_file
*id
, int eFileLock
){
3088 return posixUnlock(id
, eFileLock
, 1);
3091 #endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
3093 ** The code above is the NFS lock implementation. The code is specific
3094 ** to MacOSX and does not work on other unix platforms. No alternative
3097 ********************* End of the NFS lock implementation **********************
3098 ******************************************************************************/
3100 /******************************************************************************
3101 **************** Non-locking sqlite3_file methods *****************************
3103 ** The next division contains implementations for all methods of the
3104 ** sqlite3_file object other than the locking methods. The locking
3105 ** methods were defined in divisions above (one locking method per
3106 ** division). Those methods that are common to all locking modes
3107 ** are gather together into this division.
3111 ** Seek to the offset passed as the second argument, then read cnt
3112 ** bytes into pBuf. Return the number of bytes actually read.
3114 ** NB: If you define USE_PREAD or USE_PREAD64, then it might also
3115 ** be necessary to define _XOPEN_SOURCE to be 500. This varies from
3116 ** one system to another. Since SQLite does not define USE_PREAD
3117 ** in any form by default, we will not attempt to define _XOPEN_SOURCE.
3118 ** See tickets #2741 and #2681.
3120 ** To avoid stomping the errno value on a failed read the lastErrno value
3121 ** is set before returning.
3123 static int seekAndRead(unixFile
*id
, sqlite3_int64 offset
, void *pBuf
, int cnt
){
3126 #if (!defined(USE_PREAD) && !defined(USE_PREAD64))
3130 assert( cnt
==(cnt
&0x1ffff) );
3133 #if defined(USE_PREAD)
3134 got
= osPread(id
->h
, pBuf
, cnt
, offset
);
3135 SimulateIOError( got
= -1 );
3136 #elif defined(USE_PREAD64)
3137 got
= osPread64(id
->h
, pBuf
, cnt
, offset
);
3138 SimulateIOError( got
= -1 );
3140 newOffset
= lseek(id
->h
, offset
, SEEK_SET
);
3141 SimulateIOError( newOffset
= -1 );
3143 storeLastErrno((unixFile
*)id
, errno
);
3146 got
= osRead(id
->h
, pBuf
, cnt
);
3148 if( got
==cnt
) break;
3150 if( errno
==EINTR
){ got
= 1; continue; }
3152 storeLastErrno((unixFile
*)id
, errno
);
3158 pBuf
= (void*)(got
+ (char*)pBuf
);
3162 OSTRACE(("READ %-3d %5d %7lld %llu\n",
3163 id
->h
, got
+prior
, offset
-prior
, TIMER_ELAPSED
));
3168 ** Read data from a file into a buffer. Return SQLITE_OK if all
3169 ** bytes were read successfully and SQLITE_IOERR if anything goes
3172 static int unixRead(
3176 sqlite3_int64 offset
3178 unixFile
*pFile
= (unixFile
*)id
;
3181 assert( offset
>=0 );
3184 /* If this is a database file (not a journal, master-journal or temp
3185 ** file), the bytes in the locking range should never be read or written. */
3187 assert( pFile
->pUnused
==0
3188 || offset
>=PENDING_BYTE
+512
3189 || offset
+amt
<=PENDING_BYTE
3193 #if SQLITE_MAX_MMAP_SIZE>0
3194 /* Deal with as much of this read request as possible by transfering
3195 ** data from the memory mapping using memcpy(). */
3196 if( offset
<pFile
->mmapSize
){
3197 if( offset
+amt
<= pFile
->mmapSize
){
3198 memcpy(pBuf
, &((u8
*)(pFile
->pMapRegion
))[offset
], amt
);
3201 int nCopy
= pFile
->mmapSize
- offset
;
3202 memcpy(pBuf
, &((u8
*)(pFile
->pMapRegion
))[offset
], nCopy
);
3203 pBuf
= &((u8
*)pBuf
)[nCopy
];
3210 got
= seekAndRead(pFile
, offset
, pBuf
, amt
);
3214 /* lastErrno set by seekAndRead */
3215 return SQLITE_IOERR_READ
;
3217 storeLastErrno(pFile
, 0); /* not a system error */
3218 /* Unread parts of the buffer must be zero-filled */
3219 memset(&((char*)pBuf
)[got
], 0, amt
-got
);
3220 return SQLITE_IOERR_SHORT_READ
;
3225 ** Attempt to seek the file-descriptor passed as the first argument to
3226 ** absolute offset iOff, then attempt to write nBuf bytes of data from
3227 ** pBuf to it. If an error occurs, return -1 and set *piErrno. Otherwise,
3228 ** return the actual number of bytes written (which may be less than
3231 static int seekAndWriteFd(
3232 int fd
, /* File descriptor to write to */
3233 i64 iOff
, /* File offset to begin writing at */
3234 const void *pBuf
, /* Copy data from this buffer to the file */
3235 int nBuf
, /* Size of buffer pBuf in bytes */
3236 int *piErrno
/* OUT: Error number if error occurs */
3238 int rc
= 0; /* Value returned by system call */
3240 assert( nBuf
==(nBuf
&0x1ffff) );
3242 assert( piErrno
!=0 );
3246 #if defined(USE_PREAD)
3247 do{ rc
= (int)osPwrite(fd
, pBuf
, nBuf
, iOff
); }while( rc
<0 && errno
==EINTR
);
3248 #elif defined(USE_PREAD64)
3249 do{ rc
= (int)osPwrite64(fd
, pBuf
, nBuf
, iOff
);}while( rc
<0 && errno
==EINTR
);
3252 i64 iSeek
= lseek(fd
, iOff
, SEEK_SET
);
3253 SimulateIOError( iSeek
= -1 );
3258 rc
= osWrite(fd
, pBuf
, nBuf
);
3259 }while( rc
<0 && errno
==EINTR
);
3263 OSTRACE(("WRITE %-3d %5d %7lld %llu\n", fd
, rc
, iOff
, TIMER_ELAPSED
));
3265 if( rc
<0 ) *piErrno
= errno
;
3271 ** Seek to the offset in id->offset then read cnt bytes into pBuf.
3272 ** Return the number of bytes actually read. Update the offset.
3274 ** To avoid stomping the errno value on a failed write the lastErrno value
3275 ** is set before returning.
3277 static int seekAndWrite(unixFile
*id
, i64 offset
, const void *pBuf
, int cnt
){
3278 return seekAndWriteFd(id
->h
, offset
, pBuf
, cnt
, &id
->lastErrno
);
3283 ** Write data from a buffer into a file. Return SQLITE_OK on success
3284 ** or some other error code on failure.
3286 static int unixWrite(
3290 sqlite3_int64 offset
3292 unixFile
*pFile
= (unixFile
*)id
;
3297 /* If this is a database file (not a journal, master-journal or temp
3298 ** file), the bytes in the locking range should never be read or written. */
3300 assert( pFile
->pUnused
==0
3301 || offset
>=PENDING_BYTE
+512
3302 || offset
+amt
<=PENDING_BYTE
3307 /* If we are doing a normal write to a database file (as opposed to
3308 ** doing a hot-journal rollback or a write to some file other than a
3309 ** normal database file) then record the fact that the database
3310 ** has changed. If the transaction counter is modified, record that
3313 if( pFile
->inNormalWrite
){
3314 pFile
->dbUpdate
= 1; /* The database has been modified */
3315 if( offset
<=24 && offset
+amt
>=27 ){
3318 SimulateIOErrorBenign(1);
3319 rc
= seekAndRead(pFile
, 24, oldCntr
, 4);
3320 SimulateIOErrorBenign(0);
3321 if( rc
!=4 || memcmp(oldCntr
, &((char*)pBuf
)[24-offset
], 4)!=0 ){
3322 pFile
->transCntrChng
= 1; /* The transaction counter has changed */
3328 #if defined(SQLITE_MMAP_READWRITE) && SQLITE_MAX_MMAP_SIZE>0
3329 /* Deal with as much of this write request as possible by transfering
3330 ** data from the memory mapping using memcpy(). */
3331 if( offset
<pFile
->mmapSize
){
3332 if( offset
+amt
<= pFile
->mmapSize
){
3333 memcpy(&((u8
*)(pFile
->pMapRegion
))[offset
], pBuf
, amt
);
3336 int nCopy
= pFile
->mmapSize
- offset
;
3337 memcpy(&((u8
*)(pFile
->pMapRegion
))[offset
], pBuf
, nCopy
);
3338 pBuf
= &((u8
*)pBuf
)[nCopy
];
3345 while( (wrote
= seekAndWrite(pFile
, offset
, pBuf
, amt
))<amt
&& wrote
>0 ){
3348 pBuf
= &((char*)pBuf
)[wrote
];
3350 SimulateIOError(( wrote
=(-1), amt
=1 ));
3351 SimulateDiskfullError(( wrote
=0, amt
=1 ));
3354 if( wrote
<0 && pFile
->lastErrno
!=ENOSPC
){
3355 /* lastErrno set by seekAndWrite */
3356 return SQLITE_IOERR_WRITE
;
3358 storeLastErrno(pFile
, 0); /* not a system error */
3368 ** Count the number of fullsyncs and normal syncs. This is used to test
3369 ** that syncs and fullsyncs are occurring at the right times.
3371 int sqlite3_sync_count
= 0;
3372 int sqlite3_fullsync_count
= 0;
3376 ** We do not trust systems to provide a working fdatasync(). Some do.
3377 ** Others do no. To be safe, we will stick with the (slightly slower)
3378 ** fsync(). If you know that your system does support fdatasync() correctly,
3379 ** then simply compile with -Dfdatasync=fdatasync or -DHAVE_FDATASYNC
3381 #if !defined(fdatasync) && !HAVE_FDATASYNC
3382 # define fdatasync fsync
3386 ** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
3387 ** the F_FULLFSYNC macro is defined. F_FULLFSYNC is currently
3388 ** only available on Mac OS X. But that could change.
3391 # define HAVE_FULLFSYNC 1
3393 # define HAVE_FULLFSYNC 0
3398 ** The fsync() system call does not work as advertised on many
3399 ** unix systems. The following procedure is an attempt to make
3402 ** The SQLITE_NO_SYNC macro disables all fsync()s. This is useful
3403 ** for testing when we want to run through the test suite quickly.
3404 ** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
3405 ** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
3406 ** or power failure will likely corrupt the database file.
3408 ** SQLite sets the dataOnly flag if the size of the file is unchanged.
3409 ** The idea behind dataOnly is that it should only write the file content
3410 ** to disk, not the inode. We only set dataOnly if the file size is
3411 ** unchanged since the file size is part of the inode. However,
3412 ** Ted Ts'o tells us that fdatasync() will also write the inode if the
3413 ** file size has changed. The only real difference between fdatasync()
3414 ** and fsync(), Ted tells us, is that fdatasync() will not flush the
3415 ** inode if the mtime or owner or other inode attributes have changed.
3416 ** We only care about the file size, not the other file attributes, so
3417 ** as far as SQLite is concerned, an fdatasync() is always adequate.
3418 ** So, we always use fdatasync() if it is available, regardless of
3419 ** the value of the dataOnly flag.
3421 static int full_fsync(int fd
, int fullSync
, int dataOnly
){
3424 /* The following "ifdef/elif/else/" block has the same structure as
3425 ** the one below. It is replicated here solely to avoid cluttering
3426 ** up the real code with the UNUSED_PARAMETER() macros.
3428 #ifdef SQLITE_NO_SYNC
3429 UNUSED_PARAMETER(fd
);
3430 UNUSED_PARAMETER(fullSync
);
3431 UNUSED_PARAMETER(dataOnly
);
3432 #elif HAVE_FULLFSYNC
3433 UNUSED_PARAMETER(dataOnly
);
3435 UNUSED_PARAMETER(fullSync
);
3436 UNUSED_PARAMETER(dataOnly
);
3439 /* Record the number of times that we do a normal fsync() and
3440 ** FULLSYNC. This is used during testing to verify that this procedure
3441 ** gets called with the correct arguments.
3444 if( fullSync
) sqlite3_fullsync_count
++;
3445 sqlite3_sync_count
++;
3448 /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
3449 ** no-op. But go ahead and call fstat() to validate the file
3450 ** descriptor as we need a method to provoke a failure during
3451 ** coverate testing.
3453 #ifdef SQLITE_NO_SYNC
3456 rc
= osFstat(fd
, &buf
);
3458 #elif HAVE_FULLFSYNC
3460 rc
= osFcntl(fd
, F_FULLFSYNC
, 0);
3464 /* If the FULLFSYNC failed, fall back to attempting an fsync().
3465 ** It shouldn't be possible for fullfsync to fail on the local
3466 ** file system (on OSX), so failure indicates that FULLFSYNC
3467 ** isn't supported for this file system. So, attempt an fsync
3468 ** and (for now) ignore the overhead of a superfluous fcntl call.
3469 ** It'd be better to detect fullfsync support once and avoid
3470 ** the fcntl call every time sync is called.
3472 if( rc
) rc
= fsync(fd
);
3474 #elif defined(__APPLE__)
3475 /* fdatasync() on HFS+ doesn't yet flush the file size if it changed correctly
3476 ** so currently we default to the macro that redefines fdatasync to fsync
3482 if( rc
==-1 && errno
==ENOTSUP
){
3485 #endif /* OS_VXWORKS */
3486 #endif /* ifdef SQLITE_NO_SYNC elif HAVE_FULLFSYNC */
3488 if( OS_VXWORKS
&& rc
!= -1 ){
3495 ** Open a file descriptor to the directory containing file zFilename.
3496 ** If successful, *pFd is set to the opened file descriptor and
3497 ** SQLITE_OK is returned. If an error occurs, either SQLITE_NOMEM
3498 ** or SQLITE_CANTOPEN is returned and *pFd is set to an undefined
3501 ** The directory file descriptor is used for only one thing - to
3502 ** fsync() a directory to make sure file creation and deletion events
3503 ** are flushed to disk. Such fsyncs are not needed on newer
3504 ** journaling filesystems, but are required on older filesystems.
3506 ** This routine can be overridden using the xSetSysCall interface.
3507 ** The ability to override this routine was added in support of the
3508 ** chromium sandbox. Opening a directory is a security risk (we are
3509 ** told) so making it overrideable allows the chromium sandbox to
3510 ** replace this routine with a harmless no-op. To make this routine
3511 ** a no-op, replace it with a stub that returns SQLITE_OK but leaves
3512 ** *pFd set to a negative number.
3514 ** If SQLITE_OK is returned, the caller is responsible for closing
3515 ** the file descriptor *pFd using close().
3517 static int openDirectory(const char *zFilename
, int *pFd
){
3520 char zDirname
[MAX_PATHNAME
+1];
3522 sqlite3_snprintf(MAX_PATHNAME
, zDirname
, "%s", zFilename
);
3523 for(ii
=(int)strlen(zDirname
); ii
>0 && zDirname
[ii
]!='/'; ii
--);
3525 zDirname
[ii
] = '\0';
3527 if( zDirname
[0]!='/' ) zDirname
[0] = '.';
3530 fd
= robust_open(zDirname
, O_RDONLY
|O_BINARY
, 0);
3532 OSTRACE(("OPENDIR %-3d %s\n", fd
, zDirname
));
3535 if( fd
>=0 ) return SQLITE_OK
;
3536 return unixLogError(SQLITE_CANTOPEN_BKPT
, "openDirectory", zDirname
);
3540 ** Make sure all writes to a particular file are committed to disk.
3542 ** If dataOnly==0 then both the file itself and its metadata (file
3543 ** size, access time, etc) are synced. If dataOnly!=0 then only the
3544 ** file data is synced.
3546 ** Under Unix, also make sure that the directory entry for the file
3547 ** has been created by fsync-ing the directory that contains the file.
3548 ** If we do not do this and we encounter a power failure, the directory
3549 ** entry for the journal might not exist after we reboot. The next
3550 ** SQLite to access the file will not know that the journal exists (because
3551 ** the directory entry for the journal was never created) and the transaction
3552 ** will not roll back - possibly leading to database corruption.
3554 static int unixSync(sqlite3_file
*id
, int flags
){
3556 unixFile
*pFile
= (unixFile
*)id
;
3558 int isDataOnly
= (flags
&SQLITE_SYNC_DATAONLY
);
3559 int isFullsync
= (flags
&0x0F)==SQLITE_SYNC_FULL
;
3561 /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */
3562 assert((flags
&0x0F)==SQLITE_SYNC_NORMAL
3563 || (flags
&0x0F)==SQLITE_SYNC_FULL
3566 /* Unix cannot, but some systems may return SQLITE_FULL from here. This
3567 ** line is to test that doing so does not cause any problems.
3569 SimulateDiskfullError( return SQLITE_FULL
);
3572 OSTRACE(("SYNC %-3d\n", pFile
->h
));
3573 rc
= full_fsync(pFile
->h
, isFullsync
, isDataOnly
);
3574 SimulateIOError( rc
=1 );
3576 storeLastErrno(pFile
, errno
);
3577 return unixLogError(SQLITE_IOERR_FSYNC
, "full_fsync", pFile
->zPath
);
3580 /* Also fsync the directory containing the file if the DIRSYNC flag
3581 ** is set. This is a one-time occurrence. Many systems (examples: AIX)
3582 ** are unable to fsync a directory, so ignore errors on the fsync.
3584 if( pFile
->ctrlFlags
& UNIXFILE_DIRSYNC
){
3586 OSTRACE(("DIRSYNC %s (have_fullfsync=%d fullsync=%d)\n", pFile
->zPath
,
3587 HAVE_FULLFSYNC
, isFullsync
));
3588 rc
= osOpenDirectory(pFile
->zPath
, &dirfd
);
3589 if( rc
==SQLITE_OK
){
3590 full_fsync(dirfd
, 0, 0);
3591 robust_close(pFile
, dirfd
, __LINE__
);
3593 assert( rc
==SQLITE_CANTOPEN
);
3596 pFile
->ctrlFlags
&= ~UNIXFILE_DIRSYNC
;
3602 ** Truncate an open file to a specified size
3604 static int unixTruncate(sqlite3_file
*id
, i64 nByte
){
3605 unixFile
*pFile
= (unixFile
*)id
;
3608 SimulateIOError( return SQLITE_IOERR_TRUNCATE
);
3610 /* If the user has configured a chunk-size for this file, truncate the
3611 ** file so that it consists of an integer number of chunks (i.e. the
3612 ** actual file size after the operation may be larger than the requested
3615 if( pFile
->szChunk
>0 ){
3616 nByte
= ((nByte
+ pFile
->szChunk
- 1)/pFile
->szChunk
) * pFile
->szChunk
;
3619 rc
= robust_ftruncate(pFile
->h
, nByte
);
3621 storeLastErrno(pFile
, errno
);
3622 return unixLogError(SQLITE_IOERR_TRUNCATE
, "ftruncate", pFile
->zPath
);
3625 /* If we are doing a normal write to a database file (as opposed to
3626 ** doing a hot-journal rollback or a write to some file other than a
3627 ** normal database file) and we truncate the file to zero length,
3628 ** that effectively updates the change counter. This might happen
3629 ** when restoring a database using the backup API from a zero-length
3632 if( pFile
->inNormalWrite
&& nByte
==0 ){
3633 pFile
->transCntrChng
= 1;
3637 #if SQLITE_MAX_MMAP_SIZE>0
3638 /* If the file was just truncated to a size smaller than the currently
3639 ** mapped region, reduce the effective mapping size as well. SQLite will
3640 ** use read() and write() to access data beyond this point from now on.
3642 if( nByte
<pFile
->mmapSize
){
3643 pFile
->mmapSize
= nByte
;
3652 ** Determine the current size of a file in bytes
3654 static int unixFileSize(sqlite3_file
*id
, i64
*pSize
){
3658 rc
= osFstat(((unixFile
*)id
)->h
, &buf
);
3659 SimulateIOError( rc
=1 );
3661 storeLastErrno((unixFile
*)id
, errno
);
3662 return SQLITE_IOERR_FSTAT
;
3664 *pSize
= buf
.st_size
;
3666 /* When opening a zero-size database, the findInodeInfo() procedure
3667 ** writes a single byte into that file in order to work around a bug
3668 ** in the OS-X msdos filesystem. In order to avoid problems with upper
3669 ** layers, we need to report this file size as zero even though it is
3670 ** really 1. Ticket #3260.
3672 if( *pSize
==1 ) *pSize
= 0;
3678 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
3680 ** Handler for proxy-locking file-control verbs. Defined below in the
3681 ** proxying locking division.
3683 static int proxyFileControl(sqlite3_file
*,int,void*);
3687 ** This function is called to handle the SQLITE_FCNTL_SIZE_HINT
3688 ** file-control operation. Enlarge the database to nBytes in size
3689 ** (rounded up to the next chunk-size). If the database is already
3690 ** nBytes or larger, this routine is a no-op.
3692 static int fcntlSizeHint(unixFile
*pFile
, i64 nByte
){
3693 if( pFile
->szChunk
>0 ){
3694 i64 nSize
; /* Required file size */
3695 struct stat buf
; /* Used to hold return values of fstat() */
3697 if( osFstat(pFile
->h
, &buf
) ){
3698 return SQLITE_IOERR_FSTAT
;
3701 nSize
= ((nByte
+pFile
->szChunk
-1) / pFile
->szChunk
) * pFile
->szChunk
;
3702 if( nSize
>(i64
)buf
.st_size
){
3704 #if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE
3705 /* The code below is handling the return value of osFallocate()
3706 ** correctly. posix_fallocate() is defined to "returns zero on success,
3707 ** or an error number on failure". See the manpage for details. */
3710 err
= osFallocate(pFile
->h
, buf
.st_size
, nSize
-buf
.st_size
);
3711 }while( err
==EINTR
);
3712 if( err
) return SQLITE_IOERR_WRITE
;
3714 /* If the OS does not have posix_fallocate(), fake it. Write a
3715 ** single byte to the last byte in each block that falls entirely
3716 ** within the extended region. Then, if required, a single byte
3717 ** at offset (nSize-1), to set the size of the file correctly.
3718 ** This is a similar technique to that used by glibc on systems
3719 ** that do not have a real fallocate() call.
3721 int nBlk
= buf
.st_blksize
; /* File-system block size */
3722 int nWrite
= 0; /* Number of bytes written by seekAndWrite */
3723 i64 iWrite
; /* Next offset to write to */
3725 iWrite
= (buf
.st_size
/nBlk
)*nBlk
+ nBlk
- 1;
3726 assert( iWrite
>=buf
.st_size
);
3727 assert( ((iWrite
+1)%nBlk
)==0 );
3728 for(/*no-op*/; iWrite
<nSize
+nBlk
-1; iWrite
+=nBlk
){
3729 if( iWrite
>=nSize
) iWrite
= nSize
- 1;
3730 nWrite
= seekAndWrite(pFile
, iWrite
, "", 1);
3731 if( nWrite
!=1 ) return SQLITE_IOERR_WRITE
;
3737 #if SQLITE_MAX_MMAP_SIZE>0
3738 if( pFile
->mmapSizeMax
>0 && nByte
>pFile
->mmapSize
){
3740 if( pFile
->szChunk
<=0 ){
3741 if( robust_ftruncate(pFile
->h
, nByte
) ){
3742 storeLastErrno(pFile
, errno
);
3743 return unixLogError(SQLITE_IOERR_TRUNCATE
, "ftruncate", pFile
->zPath
);
3747 rc
= unixMapfile(pFile
, nByte
);
3756 ** If *pArg is initially negative then this is a query. Set *pArg to
3757 ** 1 or 0 depending on whether or not bit mask of pFile->ctrlFlags is set.
3759 ** If *pArg is 0 or 1, then clear or set the mask bit of pFile->ctrlFlags.
3761 static void unixModeBit(unixFile
*pFile
, unsigned char mask
, int *pArg
){
3763 *pArg
= (pFile
->ctrlFlags
& mask
)!=0;
3764 }else if( (*pArg
)==0 ){
3765 pFile
->ctrlFlags
&= ~mask
;
3767 pFile
->ctrlFlags
|= mask
;
3771 /* Forward declaration */
3772 static int unixGetTempname(int nBuf
, char *zBuf
);
3775 ** Information and control of an open file handle.
3777 static int unixFileControl(sqlite3_file
*id
, int op
, void *pArg
){
3778 unixFile
*pFile
= (unixFile
*)id
;
3780 case SQLITE_FCNTL_LOCKSTATE
: {
3781 *(int*)pArg
= pFile
->eFileLock
;
3784 case SQLITE_FCNTL_LAST_ERRNO
: {
3785 *(int*)pArg
= pFile
->lastErrno
;
3788 case SQLITE_FCNTL_CHUNK_SIZE
: {
3789 pFile
->szChunk
= *(int *)pArg
;
3792 case SQLITE_FCNTL_SIZE_HINT
: {
3794 SimulateIOErrorBenign(1);
3795 rc
= fcntlSizeHint(pFile
, *(i64
*)pArg
);
3796 SimulateIOErrorBenign(0);
3799 case SQLITE_FCNTL_PERSIST_WAL
: {
3800 unixModeBit(pFile
, UNIXFILE_PERSIST_WAL
, (int*)pArg
);
3803 case SQLITE_FCNTL_POWERSAFE_OVERWRITE
: {
3804 unixModeBit(pFile
, UNIXFILE_PSOW
, (int*)pArg
);
3807 case SQLITE_FCNTL_VFSNAME
: {
3808 *(char**)pArg
= sqlite3_mprintf("%s", pFile
->pVfs
->zName
);
3811 case SQLITE_FCNTL_TEMPFILENAME
: {
3812 char *zTFile
= sqlite3_malloc64( pFile
->pVfs
->mxPathname
);
3814 unixGetTempname(pFile
->pVfs
->mxPathname
, zTFile
);
3815 *(char**)pArg
= zTFile
;
3819 case SQLITE_FCNTL_HAS_MOVED
: {
3820 *(int*)pArg
= fileHasMoved(pFile
);
3823 #if SQLITE_MAX_MMAP_SIZE>0
3824 case SQLITE_FCNTL_MMAP_SIZE
: {
3825 i64 newLimit
= *(i64
*)pArg
;
3827 if( newLimit
>sqlite3GlobalConfig
.mxMmap
){
3828 newLimit
= sqlite3GlobalConfig
.mxMmap
;
3830 *(i64
*)pArg
= pFile
->mmapSizeMax
;
3831 if( newLimit
>=0 && newLimit
!=pFile
->mmapSizeMax
&& pFile
->nFetchOut
==0 ){
3832 pFile
->mmapSizeMax
= newLimit
;
3833 if( pFile
->mmapSize
>0 ){
3834 unixUnmapfile(pFile
);
3835 rc
= unixMapfile(pFile
, -1);
3842 /* The pager calls this method to signal that it has done
3843 ** a rollback and that the database is therefore unchanged and
3844 ** it hence it is OK for the transaction change counter to be
3847 case SQLITE_FCNTL_DB_UNCHANGED
: {
3848 ((unixFile
*)id
)->dbUpdate
= 0;
3852 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
3853 case SQLITE_FCNTL_SET_LOCKPROXYFILE
:
3854 case SQLITE_FCNTL_GET_LOCKPROXYFILE
: {
3855 return proxyFileControl(id
,op
,pArg
);
3857 #endif /* SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__) */
3859 return SQLITE_NOTFOUND
;
3863 ** Return the sector size in bytes of the underlying block device for
3864 ** the specified file. This is almost always 512 bytes, but may be
3865 ** larger for some devices.
3867 ** SQLite code assumes this function cannot fail. It also assumes that
3868 ** if two files are created in the same file-system directory (i.e.
3869 ** a database and its journal file) that the sector size will be the
3873 static int unixSectorSize(sqlite3_file
*NotUsed
){
3874 UNUSED_PARAMETER(NotUsed
);
3875 return SQLITE_DEFAULT_SECTOR_SIZE
;
3880 ** The following version of unixSectorSize() is optimized for QNX.
3883 #include <sys/dcmd_blk.h>
3884 #include <sys/statvfs.h>
3885 static int unixSectorSize(sqlite3_file
*id
){
3886 unixFile
*pFile
= (unixFile
*)id
;
3887 if( pFile
->sectorSize
== 0 ){
3888 struct statvfs fsInfo
;
3890 /* Set defaults for non-supported filesystems */
3891 pFile
->sectorSize
= SQLITE_DEFAULT_SECTOR_SIZE
;
3892 pFile
->deviceCharacteristics
= 0;
3893 if( fstatvfs(pFile
->h
, &fsInfo
) == -1 ) {
3894 return pFile
->sectorSize
;
3897 if( !strcmp(fsInfo
.f_basetype
, "tmp") ) {
3898 pFile
->sectorSize
= fsInfo
.f_bsize
;
3899 pFile
->deviceCharacteristics
=
3900 SQLITE_IOCAP_ATOMIC4K
| /* All ram filesystem writes are atomic */
3901 SQLITE_IOCAP_SAFE_APPEND
| /* growing the file does not occur until
3902 ** the write succeeds */
3903 SQLITE_IOCAP_SEQUENTIAL
| /* The ram filesystem has no write behind
3904 ** so it is ordered */
3906 }else if( strstr(fsInfo
.f_basetype
, "etfs") ){
3907 pFile
->sectorSize
= fsInfo
.f_bsize
;
3908 pFile
->deviceCharacteristics
=
3909 /* etfs cluster size writes are atomic */
3910 (pFile
->sectorSize
/ 512 * SQLITE_IOCAP_ATOMIC512
) |
3911 SQLITE_IOCAP_SAFE_APPEND
| /* growing the file does not occur until
3912 ** the write succeeds */
3913 SQLITE_IOCAP_SEQUENTIAL
| /* The ram filesystem has no write behind
3914 ** so it is ordered */
3916 }else if( !strcmp(fsInfo
.f_basetype
, "qnx6") ){
3917 pFile
->sectorSize
= fsInfo
.f_bsize
;
3918 pFile
->deviceCharacteristics
=
3919 SQLITE_IOCAP_ATOMIC
| /* All filesystem writes are atomic */
3920 SQLITE_IOCAP_SAFE_APPEND
| /* growing the file does not occur until
3921 ** the write succeeds */
3922 SQLITE_IOCAP_SEQUENTIAL
| /* The ram filesystem has no write behind
3923 ** so it is ordered */
3925 }else if( !strcmp(fsInfo
.f_basetype
, "qnx4") ){
3926 pFile
->sectorSize
= fsInfo
.f_bsize
;
3927 pFile
->deviceCharacteristics
=
3928 /* full bitset of atomics from max sector size and smaller */
3929 ((pFile
->sectorSize
/ 512 * SQLITE_IOCAP_ATOMIC512
) << 1) - 2 |
3930 SQLITE_IOCAP_SEQUENTIAL
| /* The ram filesystem has no write behind
3931 ** so it is ordered */
3933 }else if( strstr(fsInfo
.f_basetype
, "dos") ){
3934 pFile
->sectorSize
= fsInfo
.f_bsize
;
3935 pFile
->deviceCharacteristics
=
3936 /* full bitset of atomics from max sector size and smaller */
3937 ((pFile
->sectorSize
/ 512 * SQLITE_IOCAP_ATOMIC512
) << 1) - 2 |
3938 SQLITE_IOCAP_SEQUENTIAL
| /* The ram filesystem has no write behind
3939 ** so it is ordered */
3942 pFile
->deviceCharacteristics
=
3943 SQLITE_IOCAP_ATOMIC512
| /* blocks are atomic */
3944 SQLITE_IOCAP_SAFE_APPEND
| /* growing the file does not occur until
3945 ** the write succeeds */
3949 /* Last chance verification. If the sector size isn't a multiple of 512
3950 ** then it isn't valid.*/
3951 if( pFile
->sectorSize
% 512 != 0 ){
3952 pFile
->deviceCharacteristics
= 0;
3953 pFile
->sectorSize
= SQLITE_DEFAULT_SECTOR_SIZE
;
3955 return pFile
->sectorSize
;
3957 #endif /* __QNXNTO__ */
3960 ** Return the device characteristics for the file.
3962 ** This VFS is set up to return SQLITE_IOCAP_POWERSAFE_OVERWRITE by default.
3963 ** However, that choice is controversial since technically the underlying
3964 ** file system does not always provide powersafe overwrites. (In other
3965 ** words, after a power-loss event, parts of the file that were never
3966 ** written might end up being altered.) However, non-PSOW behavior is very,
3967 ** very rare. And asserting PSOW makes a large reduction in the amount
3968 ** of required I/O for journaling, since a lot of padding is eliminated.
3969 ** Hence, while POWERSAFE_OVERWRITE is on by default, there is a file-control
3970 ** available to turn it off and URI query parameter available to turn it off.
3972 static int unixDeviceCharacteristics(sqlite3_file
*id
){
3973 unixFile
*p
= (unixFile
*)id
;
3976 if( p
->sectorSize
==0 ) unixSectorSize(id
);
3977 rc
= p
->deviceCharacteristics
;
3979 if( p
->ctrlFlags
& UNIXFILE_PSOW
){
3980 rc
|= SQLITE_IOCAP_POWERSAFE_OVERWRITE
;
3985 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
3988 ** Return the system page size.
3990 ** This function should not be called directly by other code in this file.
3991 ** Instead, it should be called via macro osGetpagesize().
3993 static int unixGetpagesize(void){
3996 #elif defined(_BSD_SOURCE)
3997 return getpagesize();
3999 return (int)sysconf(_SC_PAGESIZE
);
4003 #endif /* !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0 */
4005 #ifndef SQLITE_OMIT_WAL
4008 ** Object used to represent an shared memory buffer.
4010 ** When multiple threads all reference the same wal-index, each thread
4011 ** has its own unixShm object, but they all point to a single instance
4012 ** of this unixShmNode object. In other words, each wal-index is opened
4013 ** only once per process.
4015 ** Each unixShmNode object is connected to a single unixInodeInfo object.
4016 ** We could coalesce this object into unixInodeInfo, but that would mean
4017 ** every open file that does not use shared memory (in other words, most
4018 ** open files) would have to carry around this extra information. So
4019 ** the unixInodeInfo object contains a pointer to this unixShmNode object
4020 ** and the unixShmNode object is created only when needed.
4022 ** unixMutexHeld() must be true when creating or destroying
4023 ** this object or while reading or writing the following fields:
4027 ** The following fields are read-only after the object is created:
4032 ** Either unixShmNode.mutex must be held or unixShmNode.nRef==0 and
4033 ** unixMutexHeld() is true when reading or writing any other field
4034 ** in this structure.
4036 struct unixShmNode
{
4037 unixInodeInfo
*pInode
; /* unixInodeInfo that owns this SHM node */
4038 sqlite3_mutex
*mutex
; /* Mutex to access this object */
4039 char *zFilename
; /* Name of the mmapped file */
4040 int h
; /* Open file descriptor */
4041 int szRegion
; /* Size of shared-memory regions */
4042 u16 nRegion
; /* Size of array apRegion */
4043 u8 isReadonly
; /* True if read-only */
4044 char **apRegion
; /* Array of mapped shared-memory regions */
4045 int nRef
; /* Number of unixShm objects pointing to this */
4046 unixShm
*pFirst
; /* All unixShm objects pointing to this */
4048 u8 exclMask
; /* Mask of exclusive locks held */
4049 u8 sharedMask
; /* Mask of shared locks held */
4050 u8 nextShmId
; /* Next available unixShm.id value */
4055 ** Structure used internally by this VFS to record the state of an
4056 ** open shared memory connection.
4058 ** The following fields are initialized when this object is created and
4059 ** are read-only thereafter:
4064 ** All other fields are read/write. The unixShm.pFile->mutex must be held
4065 ** while accessing any read/write fields.
4068 unixShmNode
*pShmNode
; /* The underlying unixShmNode object */
4069 unixShm
*pNext
; /* Next unixShm with the same unixShmNode */
4070 u8 hasMutex
; /* True if holding the unixShmNode mutex */
4071 u8 id
; /* Id of this connection within its unixShmNode */
4072 u16 sharedMask
; /* Mask of shared locks held */
4073 u16 exclMask
; /* Mask of exclusive locks held */
4077 ** Constants used for locking
4079 #define UNIX_SHM_BASE ((22+SQLITE_SHM_NLOCK)*4) /* first lock byte */
4080 #define UNIX_SHM_DMS (UNIX_SHM_BASE+SQLITE_SHM_NLOCK) /* deadman switch */
4083 ** Apply posix advisory locks for all bytes from ofst through ofst+n-1.
4085 ** Locks block if the mask is exactly UNIX_SHM_C and are non-blocking
4088 static int unixShmSystemLock(
4089 unixFile
*pFile
, /* Open connection to the WAL file */
4090 int lockType
, /* F_UNLCK, F_RDLCK, or F_WRLCK */
4091 int ofst
, /* First byte of the locking range */
4092 int n
/* Number of bytes to lock */
4094 unixShmNode
*pShmNode
; /* Apply locks to this open shared-memory segment */
4095 struct flock f
; /* The posix advisory locking structure */
4096 int rc
= SQLITE_OK
; /* Result code form fcntl() */
4098 /* Access to the unixShmNode object is serialized by the caller */
4099 pShmNode
= pFile
->pInode
->pShmNode
;
4100 assert( sqlite3_mutex_held(pShmNode
->mutex
) || pShmNode
->nRef
==0 );
4102 /* Shared locks never span more than one byte */
4103 assert( n
==1 || lockType
!=F_RDLCK
);
4105 /* Locks are within range */
4106 assert( n
>=1 && n
<=SQLITE_SHM_NLOCK
);
4108 if( pShmNode
->h
>=0 ){
4109 /* Initialize the locking parameters */
4110 memset(&f
, 0, sizeof(f
));
4111 f
.l_type
= lockType
;
4112 f
.l_whence
= SEEK_SET
;
4116 rc
= osFcntl(pShmNode
->h
, F_SETLK
, &f
);
4117 rc
= (rc
!=(-1)) ? SQLITE_OK
: SQLITE_BUSY
;
4120 /* Update the global lock state and do debug tracing */
4123 OSTRACE(("SHM-LOCK "));
4124 mask
= ofst
>31 ? 0xffff : (1<<(ofst
+n
)) - (1<<ofst
);
4125 if( rc
==SQLITE_OK
){
4126 if( lockType
==F_UNLCK
){
4127 OSTRACE(("unlock %d ok", ofst
));
4128 pShmNode
->exclMask
&= ~mask
;
4129 pShmNode
->sharedMask
&= ~mask
;
4130 }else if( lockType
==F_RDLCK
){
4131 OSTRACE(("read-lock %d ok", ofst
));
4132 pShmNode
->exclMask
&= ~mask
;
4133 pShmNode
->sharedMask
|= mask
;
4135 assert( lockType
==F_WRLCK
);
4136 OSTRACE(("write-lock %d ok", ofst
));
4137 pShmNode
->exclMask
|= mask
;
4138 pShmNode
->sharedMask
&= ~mask
;
4141 if( lockType
==F_UNLCK
){
4142 OSTRACE(("unlock %d failed", ofst
));
4143 }else if( lockType
==F_RDLCK
){
4144 OSTRACE(("read-lock failed"));
4146 assert( lockType
==F_WRLCK
);
4147 OSTRACE(("write-lock %d failed", ofst
));
4150 OSTRACE((" - afterwards %03x,%03x\n",
4151 pShmNode
->sharedMask
, pShmNode
->exclMask
));
4159 ** Return the minimum number of 32KB shm regions that should be mapped at
4160 ** a time, assuming that each mapping must be an integer multiple of the
4161 ** current system page-size.
4163 ** Usually, this is 1. The exception seems to be systems that are configured
4164 ** to use 64KB pages - in this case each mapping must cover at least two
4167 static int unixShmRegionPerMap(void){
4168 int shmsz
= 32*1024; /* SHM region size */
4169 int pgsz
= osGetpagesize(); /* System page size */
4170 assert( ((pgsz
-1)&pgsz
)==0 ); /* Page size must be a power of 2 */
4171 if( pgsz
<shmsz
) return 1;
4176 ** Purge the unixShmNodeList list of all entries with unixShmNode.nRef==0.
4178 ** This is not a VFS shared-memory method; it is a utility function called
4179 ** by VFS shared-memory methods.
4181 static void unixShmPurge(unixFile
*pFd
){
4182 unixShmNode
*p
= pFd
->pInode
->pShmNode
;
4183 assert( unixMutexHeld() );
4184 if( p
&& ALWAYS(p
->nRef
==0) ){
4185 int nShmPerMap
= unixShmRegionPerMap();
4187 assert( p
->pInode
==pFd
->pInode
);
4188 sqlite3_mutex_free(p
->mutex
);
4189 for(i
=0; i
<p
->nRegion
; i
+=nShmPerMap
){
4191 osMunmap(p
->apRegion
[i
], p
->szRegion
);
4193 sqlite3_free(p
->apRegion
[i
]);
4196 sqlite3_free(p
->apRegion
);
4198 robust_close(pFd
, p
->h
, __LINE__
);
4201 p
->pInode
->pShmNode
= 0;
4207 ** Open a shared-memory area associated with open database file pDbFd.
4208 ** This particular implementation uses mmapped files.
4210 ** The file used to implement shared-memory is in the same directory
4211 ** as the open database file and has the same name as the open database
4212 ** file with the "-shm" suffix added. For example, if the database file
4213 ** is "/home/user1/config.db" then the file that is created and mmapped
4214 ** for shared memory will be called "/home/user1/config.db-shm".
4216 ** Another approach to is to use files in /dev/shm or /dev/tmp or an
4217 ** some other tmpfs mount. But if a file in a different directory
4218 ** from the database file is used, then differing access permissions
4219 ** or a chroot() might cause two different processes on the same
4220 ** database to end up using different files for shared memory -
4221 ** meaning that their memory would not really be shared - resulting
4222 ** in database corruption. Nevertheless, this tmpfs file usage
4223 ** can be enabled at compile-time using -DSQLITE_SHM_DIRECTORY="/dev/shm"
4224 ** or the equivalent. The use of the SQLITE_SHM_DIRECTORY compile-time
4225 ** option results in an incompatible build of SQLite; builds of SQLite
4226 ** that with differing SQLITE_SHM_DIRECTORY settings attempt to use the
4227 ** same database file at the same time, database corruption will likely
4228 ** result. The SQLITE_SHM_DIRECTORY compile-time option is considered
4229 ** "unsupported" and may go away in a future SQLite release.
4231 ** When opening a new shared-memory file, if no other instances of that
4232 ** file are currently open, in this process or in other processes, then
4233 ** the file must be truncated to zero length or have its header cleared.
4235 ** If the original database file (pDbFd) is using the "unix-excl" VFS
4236 ** that means that an exclusive lock is held on the database file and
4237 ** that no other processes are able to read or write the database. In
4238 ** that case, we do not really need shared memory. No shared memory
4239 ** file is created. The shared memory will be simulated with heap memory.
4241 static int unixOpenSharedMemory(unixFile
*pDbFd
){
4242 struct unixShm
*p
= 0; /* The connection to be opened */
4243 struct unixShmNode
*pShmNode
; /* The underlying mmapped file */
4244 int rc
; /* Result code */
4245 unixInodeInfo
*pInode
; /* The inode of fd */
4246 char *zShmFilename
; /* Name of the file used for SHM */
4247 int nShmFilename
; /* Size of the SHM filename in bytes */
4249 /* Allocate space for the new unixShm object. */
4250 p
= sqlite3_malloc64( sizeof(*p
) );
4251 if( p
==0 ) return SQLITE_NOMEM_BKPT
;
4252 memset(p
, 0, sizeof(*p
));
4253 assert( pDbFd
->pShm
==0 );
4255 /* Check to see if a unixShmNode object already exists. Reuse an existing
4256 ** one if present. Create a new one if necessary.
4259 pInode
= pDbFd
->pInode
;
4260 pShmNode
= pInode
->pShmNode
;
4262 struct stat sStat
; /* fstat() info for database file */
4263 #ifndef SQLITE_SHM_DIRECTORY
4264 const char *zBasePath
= pDbFd
->zPath
;
4267 /* Call fstat() to figure out the permissions on the database file. If
4268 ** a new *-shm file is created, an attempt will be made to create it
4269 ** with the same permissions.
4271 if( osFstat(pDbFd
->h
, &sStat
) ){
4272 rc
= SQLITE_IOERR_FSTAT
;
4276 #ifdef SQLITE_SHM_DIRECTORY
4277 nShmFilename
= sizeof(SQLITE_SHM_DIRECTORY
) + 31;
4279 nShmFilename
= 6 + (int)strlen(zBasePath
);
4281 pShmNode
= sqlite3_malloc64( sizeof(*pShmNode
) + nShmFilename
);
4283 rc
= SQLITE_NOMEM_BKPT
;
4286 memset(pShmNode
, 0, sizeof(*pShmNode
)+nShmFilename
);
4287 zShmFilename
= pShmNode
->zFilename
= (char*)&pShmNode
[1];
4288 #ifdef SQLITE_SHM_DIRECTORY
4289 sqlite3_snprintf(nShmFilename
, zShmFilename
,
4290 SQLITE_SHM_DIRECTORY
"/sqlite-shm-%x-%x",
4291 (u32
)sStat
.st_ino
, (u32
)sStat
.st_dev
);
4293 sqlite3_snprintf(nShmFilename
, zShmFilename
, "%s-shm", zBasePath
);
4294 sqlite3FileSuffix3(pDbFd
->zPath
, zShmFilename
);
4297 pDbFd
->pInode
->pShmNode
= pShmNode
;
4298 pShmNode
->pInode
= pDbFd
->pInode
;
4299 if( sqlite3GlobalConfig
.bCoreMutex
){
4300 pShmNode
->mutex
= sqlite3_mutex_alloc(SQLITE_MUTEX_FAST
);
4301 if( pShmNode
->mutex
==0 ){
4302 rc
= SQLITE_NOMEM_BKPT
;
4307 if( pInode
->bProcessLock
==0 ){
4308 int openFlags
= O_RDWR
| O_CREAT
;
4309 if( sqlite3_uri_boolean(pDbFd
->zPath
, "readonly_shm", 0) ){
4310 openFlags
= O_RDONLY
;
4311 pShmNode
->isReadonly
= 1;
4313 pShmNode
->h
= robust_open(zShmFilename
, openFlags
, (sStat
.st_mode
&0777));
4314 if( pShmNode
->h
<0 ){
4315 rc
= unixLogError(SQLITE_CANTOPEN_BKPT
, "open", zShmFilename
);
4319 /* If this process is running as root, make sure that the SHM file
4320 ** is owned by the same user that owns the original database. Otherwise,
4321 ** the original owner will not be able to connect.
4323 robustFchown(pShmNode
->h
, sStat
.st_uid
, sStat
.st_gid
);
4325 /* Check to see if another process is holding the dead-man switch.
4326 ** If not, truncate the file to zero length.
4329 if( unixShmSystemLock(pDbFd
, F_WRLCK
, UNIX_SHM_DMS
, 1)==SQLITE_OK
){
4330 if( robust_ftruncate(pShmNode
->h
, 0) ){
4331 rc
= unixLogError(SQLITE_IOERR_SHMOPEN
, "ftruncate", zShmFilename
);
4334 if( rc
==SQLITE_OK
){
4335 rc
= unixShmSystemLock(pDbFd
, F_RDLCK
, UNIX_SHM_DMS
, 1);
4337 if( rc
) goto shm_open_err
;
4341 /* Make the new connection a child of the unixShmNode */
4342 p
->pShmNode
= pShmNode
;
4344 p
->id
= pShmNode
->nextShmId
++;
4350 /* The reference count on pShmNode has already been incremented under
4351 ** the cover of the unixEnterMutex() mutex and the pointer from the
4352 ** new (struct unixShm) object to the pShmNode has been set. All that is
4353 ** left to do is to link the new object into the linked list starting
4354 ** at pShmNode->pFirst. This must be done while holding the pShmNode->mutex
4357 sqlite3_mutex_enter(pShmNode
->mutex
);
4358 p
->pNext
= pShmNode
->pFirst
;
4359 pShmNode
->pFirst
= p
;
4360 sqlite3_mutex_leave(pShmNode
->mutex
);
4363 /* Jump here on any error */
4365 unixShmPurge(pDbFd
); /* This call frees pShmNode if required */
4372 ** This function is called to obtain a pointer to region iRegion of the
4373 ** shared-memory associated with the database file fd. Shared-memory regions
4374 ** are numbered starting from zero. Each shared-memory region is szRegion
4377 ** If an error occurs, an error code is returned and *pp is set to NULL.
4379 ** Otherwise, if the bExtend parameter is 0 and the requested shared-memory
4380 ** region has not been allocated (by any client, including one running in a
4381 ** separate process), then *pp is set to NULL and SQLITE_OK returned. If
4382 ** bExtend is non-zero and the requested shared-memory region has not yet
4383 ** been allocated, it is allocated by this function.
4385 ** If the shared-memory region has already been allocated or is allocated by
4386 ** this call as described above, then it is mapped into this processes
4387 ** address space (if it is not already), *pp is set to point to the mapped
4388 ** memory and SQLITE_OK returned.
4390 static int unixShmMap(
4391 sqlite3_file
*fd
, /* Handle open on database file */
4392 int iRegion
, /* Region to retrieve */
4393 int szRegion
, /* Size of regions */
4394 int bExtend
, /* True to extend file if necessary */
4395 void volatile **pp
/* OUT: Mapped memory */
4397 unixFile
*pDbFd
= (unixFile
*)fd
;
4399 unixShmNode
*pShmNode
;
4401 int nShmPerMap
= unixShmRegionPerMap();
4404 /* If the shared-memory file has not yet been opened, open it now. */
4405 if( pDbFd
->pShm
==0 ){
4406 rc
= unixOpenSharedMemory(pDbFd
);
4407 if( rc
!=SQLITE_OK
) return rc
;
4411 pShmNode
= p
->pShmNode
;
4412 sqlite3_mutex_enter(pShmNode
->mutex
);
4413 assert( szRegion
==pShmNode
->szRegion
|| pShmNode
->nRegion
==0 );
4414 assert( pShmNode
->pInode
==pDbFd
->pInode
);
4415 assert( pShmNode
->h
>=0 || pDbFd
->pInode
->bProcessLock
==1 );
4416 assert( pShmNode
->h
<0 || pDbFd
->pInode
->bProcessLock
==0 );
4418 /* Minimum number of regions required to be mapped. */
4419 nReqRegion
= ((iRegion
+nShmPerMap
) / nShmPerMap
) * nShmPerMap
;
4421 if( pShmNode
->nRegion
<nReqRegion
){
4422 char **apNew
; /* New apRegion[] array */
4423 int nByte
= nReqRegion
*szRegion
; /* Minimum required file size */
4424 struct stat sStat
; /* Used by fstat() */
4426 pShmNode
->szRegion
= szRegion
;
4428 if( pShmNode
->h
>=0 ){
4429 /* The requested region is not mapped into this processes address space.
4430 ** Check to see if it has been allocated (i.e. if the wal-index file is
4431 ** large enough to contain the requested region).
4433 if( osFstat(pShmNode
->h
, &sStat
) ){
4434 rc
= SQLITE_IOERR_SHMSIZE
;
4438 if( sStat
.st_size
<nByte
){
4439 /* The requested memory region does not exist. If bExtend is set to
4440 ** false, exit early. *pp will be set to NULL and SQLITE_OK returned.
4446 /* Alternatively, if bExtend is true, extend the file. Do this by
4447 ** writing a single byte to the end of each (OS) page being
4448 ** allocated or extended. Technically, we need only write to the
4449 ** last page in order to extend the file. But writing to all new
4450 ** pages forces the OS to allocate them immediately, which reduces
4451 ** the chances of SIGBUS while accessing the mapped region later on.
4454 static const int pgsz
= 4096;
4457 /* Write to the last byte of each newly allocated or extended page */
4458 assert( (nByte
% pgsz
)==0 );
4459 for(iPg
=(sStat
.st_size
/pgsz
); iPg
<(nByte
/pgsz
); iPg
++){
4461 if( seekAndWriteFd(pShmNode
->h
, iPg
*pgsz
+ pgsz
-1, "", 1, &x
)!=1 ){
4462 const char *zFile
= pShmNode
->zFilename
;
4463 rc
= unixLogError(SQLITE_IOERR_SHMSIZE
, "write", zFile
);
4471 /* Map the requested memory region into this processes address space. */
4472 apNew
= (char **)sqlite3_realloc(
4473 pShmNode
->apRegion
, nReqRegion
*sizeof(char *)
4476 rc
= SQLITE_IOERR_NOMEM_BKPT
;
4479 pShmNode
->apRegion
= apNew
;
4480 while( pShmNode
->nRegion
<nReqRegion
){
4481 int nMap
= szRegion
*nShmPerMap
;
4484 if( pShmNode
->h
>=0 ){
4485 pMem
= osMmap(0, nMap
,
4486 pShmNode
->isReadonly
? PROT_READ
: PROT_READ
|PROT_WRITE
,
4487 MAP_SHARED
, pShmNode
->h
, szRegion
*(i64
)pShmNode
->nRegion
4489 if( pMem
==MAP_FAILED
){
4490 rc
= unixLogError(SQLITE_IOERR_SHMMAP
, "mmap", pShmNode
->zFilename
);
4494 pMem
= sqlite3_malloc64(szRegion
);
4496 rc
= SQLITE_NOMEM_BKPT
;
4499 memset(pMem
, 0, szRegion
);
4502 for(i
=0; i
<nShmPerMap
; i
++){
4503 pShmNode
->apRegion
[pShmNode
->nRegion
+i
] = &((char*)pMem
)[szRegion
*i
];
4505 pShmNode
->nRegion
+= nShmPerMap
;
4510 if( pShmNode
->nRegion
>iRegion
){
4511 *pp
= pShmNode
->apRegion
[iRegion
];
4515 if( pShmNode
->isReadonly
&& rc
==SQLITE_OK
) rc
= SQLITE_READONLY
;
4516 sqlite3_mutex_leave(pShmNode
->mutex
);
4521 ** Change the lock state for a shared-memory segment.
4523 ** Note that the relationship between SHAREd and EXCLUSIVE locks is a little
4524 ** different here than in posix. In xShmLock(), one can go from unlocked
4525 ** to shared and back or from unlocked to exclusive and back. But one may
4526 ** not go from shared to exclusive or from exclusive to shared.
4528 static int unixShmLock(
4529 sqlite3_file
*fd
, /* Database file holding the shared memory */
4530 int ofst
, /* First lock to acquire or release */
4531 int n
, /* Number of locks to acquire or release */
4532 int flags
/* What to do with the lock */
4534 unixFile
*pDbFd
= (unixFile
*)fd
; /* Connection holding shared memory */
4535 unixShm
*p
= pDbFd
->pShm
; /* The shared memory being locked */
4536 unixShm
*pX
; /* For looping over all siblings */
4537 unixShmNode
*pShmNode
= p
->pShmNode
; /* The underlying file iNode */
4538 int rc
= SQLITE_OK
; /* Result code */
4539 u16 mask
; /* Mask of locks to take or release */
4541 assert( pShmNode
==pDbFd
->pInode
->pShmNode
);
4542 assert( pShmNode
->pInode
==pDbFd
->pInode
);
4543 assert( ofst
>=0 && ofst
+n
<=SQLITE_SHM_NLOCK
);
4545 assert( flags
==(SQLITE_SHM_LOCK
| SQLITE_SHM_SHARED
)
4546 || flags
==(SQLITE_SHM_LOCK
| SQLITE_SHM_EXCLUSIVE
)
4547 || flags
==(SQLITE_SHM_UNLOCK
| SQLITE_SHM_SHARED
)
4548 || flags
==(SQLITE_SHM_UNLOCK
| SQLITE_SHM_EXCLUSIVE
) );
4549 assert( n
==1 || (flags
& SQLITE_SHM_EXCLUSIVE
)!=0 );
4550 assert( pShmNode
->h
>=0 || pDbFd
->pInode
->bProcessLock
==1 );
4551 assert( pShmNode
->h
<0 || pDbFd
->pInode
->bProcessLock
==0 );
4553 mask
= (1<<(ofst
+n
)) - (1<<ofst
);
4554 assert( n
>1 || mask
==(1<<ofst
) );
4555 sqlite3_mutex_enter(pShmNode
->mutex
);
4556 if( flags
& SQLITE_SHM_UNLOCK
){
4557 u16 allMask
= 0; /* Mask of locks held by siblings */
4559 /* See if any siblings hold this same lock */
4560 for(pX
=pShmNode
->pFirst
; pX
; pX
=pX
->pNext
){
4561 if( pX
==p
) continue;
4562 assert( (pX
->exclMask
& (p
->exclMask
|p
->sharedMask
))==0 );
4563 allMask
|= pX
->sharedMask
;
4566 /* Unlock the system-level locks */
4567 if( (mask
& allMask
)==0 ){
4568 rc
= unixShmSystemLock(pDbFd
, F_UNLCK
, ofst
+UNIX_SHM_BASE
, n
);
4573 /* Undo the local locks */
4574 if( rc
==SQLITE_OK
){
4575 p
->exclMask
&= ~mask
;
4576 p
->sharedMask
&= ~mask
;
4578 }else if( flags
& SQLITE_SHM_SHARED
){
4579 u16 allShared
= 0; /* Union of locks held by connections other than "p" */
4581 /* Find out which shared locks are already held by sibling connections.
4582 ** If any sibling already holds an exclusive lock, go ahead and return
4585 for(pX
=pShmNode
->pFirst
; pX
; pX
=pX
->pNext
){
4586 if( (pX
->exclMask
& mask
)!=0 ){
4590 allShared
|= pX
->sharedMask
;
4593 /* Get shared locks at the system level, if necessary */
4594 if( rc
==SQLITE_OK
){
4595 if( (allShared
& mask
)==0 ){
4596 rc
= unixShmSystemLock(pDbFd
, F_RDLCK
, ofst
+UNIX_SHM_BASE
, n
);
4602 /* Get the local shared locks */
4603 if( rc
==SQLITE_OK
){
4604 p
->sharedMask
|= mask
;
4607 /* Make sure no sibling connections hold locks that will block this
4608 ** lock. If any do, return SQLITE_BUSY right away.
4610 for(pX
=pShmNode
->pFirst
; pX
; pX
=pX
->pNext
){
4611 if( (pX
->exclMask
& mask
)!=0 || (pX
->sharedMask
& mask
)!=0 ){
4617 /* Get the exclusive locks at the system level. Then if successful
4618 ** also mark the local connection as being locked.
4620 if( rc
==SQLITE_OK
){
4621 rc
= unixShmSystemLock(pDbFd
, F_WRLCK
, ofst
+UNIX_SHM_BASE
, n
);
4622 if( rc
==SQLITE_OK
){
4623 assert( (p
->sharedMask
& mask
)==0 );
4624 p
->exclMask
|= mask
;
4628 sqlite3_mutex_leave(pShmNode
->mutex
);
4629 OSTRACE(("SHM-LOCK shmid-%d, pid-%d got %03x,%03x\n",
4630 p
->id
, osGetpid(0), p
->sharedMask
, p
->exclMask
));
4635 ** Implement a memory barrier or memory fence on shared memory.
4637 ** All loads and stores begun before the barrier must complete before
4638 ** any load or store begun after the barrier.
4640 static void unixShmBarrier(
4641 sqlite3_file
*fd
/* Database file holding the shared memory */
4643 UNUSED_PARAMETER(fd
);
4644 sqlite3MemoryBarrier(); /* compiler-defined memory barrier */
4645 unixEnterMutex(); /* Also mutex, for redundancy */
4650 ** Close a connection to shared-memory. Delete the underlying
4651 ** storage if deleteFlag is true.
4653 ** If there is no shared memory associated with the connection then this
4654 ** routine is a harmless no-op.
4656 static int unixShmUnmap(
4657 sqlite3_file
*fd
, /* The underlying database file */
4658 int deleteFlag
/* Delete shared-memory if true */
4660 unixShm
*p
; /* The connection to be closed */
4661 unixShmNode
*pShmNode
; /* The underlying shared-memory file */
4662 unixShm
**pp
; /* For looping over sibling connections */
4663 unixFile
*pDbFd
; /* The underlying database file */
4665 pDbFd
= (unixFile
*)fd
;
4667 if( p
==0 ) return SQLITE_OK
;
4668 pShmNode
= p
->pShmNode
;
4670 assert( pShmNode
==pDbFd
->pInode
->pShmNode
);
4671 assert( pShmNode
->pInode
==pDbFd
->pInode
);
4673 /* Remove connection p from the set of connections associated
4675 sqlite3_mutex_enter(pShmNode
->mutex
);
4676 for(pp
=&pShmNode
->pFirst
; (*pp
)!=p
; pp
= &(*pp
)->pNext
){}
4679 /* Free the connection p */
4682 sqlite3_mutex_leave(pShmNode
->mutex
);
4684 /* If pShmNode->nRef has reached 0, then close the underlying
4685 ** shared-memory file, too */
4687 assert( pShmNode
->nRef
>0 );
4689 if( pShmNode
->nRef
==0 ){
4690 if( deleteFlag
&& pShmNode
->h
>=0 ){
4691 osUnlink(pShmNode
->zFilename
);
4693 unixShmPurge(pDbFd
);
4702 # define unixShmMap 0
4703 # define unixShmLock 0
4704 # define unixShmBarrier 0
4705 # define unixShmUnmap 0
4706 #endif /* #ifndef SQLITE_OMIT_WAL */
4708 #if SQLITE_MAX_MMAP_SIZE>0
4710 ** If it is currently memory mapped, unmap file pFd.
4712 static void unixUnmapfile(unixFile
*pFd
){
4713 assert( pFd
->nFetchOut
==0 );
4714 if( pFd
->pMapRegion
){
4715 osMunmap(pFd
->pMapRegion
, pFd
->mmapSizeActual
);
4716 pFd
->pMapRegion
= 0;
4718 pFd
->mmapSizeActual
= 0;
4723 ** Attempt to set the size of the memory mapping maintained by file
4724 ** descriptor pFd to nNew bytes. Any existing mapping is discarded.
4726 ** If successful, this function sets the following variables:
4728 ** unixFile.pMapRegion
4729 ** unixFile.mmapSize
4730 ** unixFile.mmapSizeActual
4732 ** If unsuccessful, an error message is logged via sqlite3_log() and
4733 ** the three variables above are zeroed. In this case SQLite should
4734 ** continue accessing the database using the xRead() and xWrite()
4737 static void unixRemapfile(
4738 unixFile
*pFd
, /* File descriptor object */
4739 i64 nNew
/* Required mapping size */
4741 const char *zErr
= "mmap";
4742 int h
= pFd
->h
; /* File descriptor open on db file */
4743 u8
*pOrig
= (u8
*)pFd
->pMapRegion
; /* Pointer to current file mapping */
4744 i64 nOrig
= pFd
->mmapSizeActual
; /* Size of pOrig region in bytes */
4745 u8
*pNew
= 0; /* Location of new mapping */
4746 int flags
= PROT_READ
; /* Flags to pass to mmap() */
4748 assert( pFd
->nFetchOut
==0 );
4749 assert( nNew
>pFd
->mmapSize
);
4750 assert( nNew
<=pFd
->mmapSizeMax
);
4752 assert( pFd
->mmapSizeActual
>=pFd
->mmapSize
);
4753 assert( MAP_FAILED
!=0 );
4755 #ifdef SQLITE_MMAP_READWRITE
4756 if( (pFd
->ctrlFlags
& UNIXFILE_RDONLY
)==0 ) flags
|= PROT_WRITE
;
4761 i64 nReuse
= pFd
->mmapSize
;
4763 const int szSyspage
= osGetpagesize();
4764 i64 nReuse
= (pFd
->mmapSize
& ~(szSyspage
-1));
4766 u8
*pReq
= &pOrig
[nReuse
];
4768 /* Unmap any pages of the existing mapping that cannot be reused. */
4769 if( nReuse
!=nOrig
){
4770 osMunmap(pReq
, nOrig
-nReuse
);
4774 pNew
= osMremap(pOrig
, nReuse
, nNew
, MREMAP_MAYMOVE
);
4777 pNew
= osMmap(pReq
, nNew
-nReuse
, flags
, MAP_SHARED
, h
, nReuse
);
4778 if( pNew
!=MAP_FAILED
){
4780 osMunmap(pNew
, nNew
- nReuse
);
4788 /* The attempt to extend the existing mapping failed. Free it. */
4789 if( pNew
==MAP_FAILED
|| pNew
==0 ){
4790 osMunmap(pOrig
, nReuse
);
4794 /* If pNew is still NULL, try to create an entirely new mapping. */
4796 pNew
= osMmap(0, nNew
, flags
, MAP_SHARED
, h
, 0);
4799 if( pNew
==MAP_FAILED
){
4802 unixLogError(SQLITE_OK
, zErr
, pFd
->zPath
);
4804 /* If the mmap() above failed, assume that all subsequent mmap() calls
4805 ** will probably fail too. Fall back to using xRead/xWrite exclusively
4807 pFd
->mmapSizeMax
= 0;
4809 pFd
->pMapRegion
= (void *)pNew
;
4810 pFd
->mmapSize
= pFd
->mmapSizeActual
= nNew
;
4814 ** Memory map or remap the file opened by file-descriptor pFd (if the file
4815 ** is already mapped, the existing mapping is replaced by the new). Or, if
4816 ** there already exists a mapping for this file, and there are still
4817 ** outstanding xFetch() references to it, this function is a no-op.
4819 ** If parameter nByte is non-negative, then it is the requested size of
4820 ** the mapping to create. Otherwise, if nByte is less than zero, then the
4821 ** requested size is the size of the file on disk. The actual size of the
4822 ** created mapping is either the requested size or the value configured
4823 ** using SQLITE_FCNTL_MMAP_LIMIT, whichever is smaller.
4825 ** SQLITE_OK is returned if no error occurs (even if the mapping is not
4826 ** recreated as a result of outstanding references) or an SQLite error
4829 static int unixMapfile(unixFile
*pFd
, i64 nMap
){
4830 assert( nMap
>=0 || pFd
->nFetchOut
==0 );
4831 assert( nMap
>0 || (pFd
->mmapSize
==0 && pFd
->pMapRegion
==0) );
4832 if( pFd
->nFetchOut
>0 ) return SQLITE_OK
;
4835 struct stat statbuf
; /* Low-level file information */
4836 if( osFstat(pFd
->h
, &statbuf
) ){
4837 return SQLITE_IOERR_FSTAT
;
4839 nMap
= statbuf
.st_size
;
4841 if( nMap
>pFd
->mmapSizeMax
){
4842 nMap
= pFd
->mmapSizeMax
;
4845 assert( nMap
>0 || (pFd
->mmapSize
==0 && pFd
->pMapRegion
==0) );
4846 if( nMap
!=pFd
->mmapSize
){
4847 unixRemapfile(pFd
, nMap
);
4852 #endif /* SQLITE_MAX_MMAP_SIZE>0 */
4855 ** If possible, return a pointer to a mapping of file fd starting at offset
4856 ** iOff. The mapping must be valid for at least nAmt bytes.
4858 ** If such a pointer can be obtained, store it in *pp and return SQLITE_OK.
4859 ** Or, if one cannot but no error occurs, set *pp to 0 and return SQLITE_OK.
4860 ** Finally, if an error does occur, return an SQLite error code. The final
4861 ** value of *pp is undefined in this case.
4863 ** If this function does return a pointer, the caller must eventually
4864 ** release the reference by calling unixUnfetch().
4866 static int unixFetch(sqlite3_file
*fd
, i64 iOff
, int nAmt
, void **pp
){
4867 #if SQLITE_MAX_MMAP_SIZE>0
4868 unixFile
*pFd
= (unixFile
*)fd
; /* The underlying database file */
4872 #if SQLITE_MAX_MMAP_SIZE>0
4873 if( pFd
->mmapSizeMax
>0 ){
4874 if( pFd
->pMapRegion
==0 ){
4875 int rc
= unixMapfile(pFd
, -1);
4876 if( rc
!=SQLITE_OK
) return rc
;
4878 if( pFd
->mmapSize
>= iOff
+nAmt
){
4879 *pp
= &((u8
*)pFd
->pMapRegion
)[iOff
];
4888 ** If the third argument is non-NULL, then this function releases a
4889 ** reference obtained by an earlier call to unixFetch(). The second
4890 ** argument passed to this function must be the same as the corresponding
4891 ** argument that was passed to the unixFetch() invocation.
4893 ** Or, if the third argument is NULL, then this function is being called
4894 ** to inform the VFS layer that, according to POSIX, any existing mapping
4895 ** may now be invalid and should be unmapped.
4897 static int unixUnfetch(sqlite3_file
*fd
, i64 iOff
, void *p
){
4898 #if SQLITE_MAX_MMAP_SIZE>0
4899 unixFile
*pFd
= (unixFile
*)fd
; /* The underlying database file */
4900 UNUSED_PARAMETER(iOff
);
4902 /* If p==0 (unmap the entire file) then there must be no outstanding
4903 ** xFetch references. Or, if p!=0 (meaning it is an xFetch reference),
4904 ** then there must be at least one outstanding. */
4905 assert( (p
==0)==(pFd
->nFetchOut
==0) );
4907 /* If p!=0, it must match the iOff value. */
4908 assert( p
==0 || p
==&((u8
*)pFd
->pMapRegion
)[iOff
] );
4916 assert( pFd
->nFetchOut
>=0 );
4918 UNUSED_PARAMETER(fd
);
4919 UNUSED_PARAMETER(p
);
4920 UNUSED_PARAMETER(iOff
);
4926 ** Here ends the implementation of all sqlite3_file methods.
4928 ********************** End sqlite3_file Methods *******************************
4929 ******************************************************************************/
4932 ** This division contains definitions of sqlite3_io_methods objects that
4933 ** implement various file locking strategies. It also contains definitions
4934 ** of "finder" functions. A finder-function is used to locate the appropriate
4935 ** sqlite3_io_methods object for a particular database file. The pAppData
4936 ** field of the sqlite3_vfs VFS objects are initialized to be pointers to
4937 ** the correct finder-function for that VFS.
4939 ** Most finder functions return a pointer to a fixed sqlite3_io_methods
4940 ** object. The only interesting finder-function is autolockIoFinder, which
4941 ** looks at the filesystem type and tries to guess the best locking
4942 ** strategy from that.
4944 ** For finder-function F, two objects are created:
4946 ** (1) The real finder-function named "FImpt()".
4948 ** (2) A constant pointer to this function named just "F".
4951 ** A pointer to the F pointer is used as the pAppData value for VFS
4952 ** objects. We have to do this instead of letting pAppData point
4953 ** directly at the finder-function since C90 rules prevent a void*
4954 ** from be cast into a function pointer.
4957 ** Each instance of this macro generates two objects:
4959 ** * A constant sqlite3_io_methods object call METHOD that has locking
4960 ** methods CLOSE, LOCK, UNLOCK, CKRESLOCK.
4962 ** * An I/O method finder function called FINDER that returns a pointer
4963 ** to the METHOD object in the previous bullet.
4965 #define IOMETHODS(FINDER,METHOD,VERSION,CLOSE,LOCK,UNLOCK,CKLOCK,SHMMAP) \
4966 static const sqlite3_io_methods METHOD = { \
4967 VERSION, /* iVersion */ \
4968 CLOSE, /* xClose */ \
4969 unixRead, /* xRead */ \
4970 unixWrite, /* xWrite */ \
4971 unixTruncate, /* xTruncate */ \
4972 unixSync, /* xSync */ \
4973 unixFileSize, /* xFileSize */ \
4975 UNLOCK, /* xUnlock */ \
4976 CKLOCK, /* xCheckReservedLock */ \
4977 unixFileControl, /* xFileControl */ \
4978 unixSectorSize, /* xSectorSize */ \
4979 unixDeviceCharacteristics, /* xDeviceCapabilities */ \
4980 SHMMAP, /* xShmMap */ \
4981 unixShmLock, /* xShmLock */ \
4982 unixShmBarrier, /* xShmBarrier */ \
4983 unixShmUnmap, /* xShmUnmap */ \
4984 unixFetch, /* xFetch */ \
4985 unixUnfetch, /* xUnfetch */ \
4987 static const sqlite3_io_methods *FINDER##Impl(const char *z, unixFile *p){ \
4988 UNUSED_PARAMETER(z); UNUSED_PARAMETER(p); \
4991 static const sqlite3_io_methods *(*const FINDER)(const char*,unixFile *p) \
4995 ** Here are all of the sqlite3_io_methods objects for each of the
4996 ** locking strategies. Functions that return pointers to these methods
4997 ** are also created.
5000 posixIoFinder
, /* Finder function name */
5001 posixIoMethods
, /* sqlite3_io_methods object name */
5002 3, /* shared memory and mmap are enabled */
5003 unixClose
, /* xClose method */
5004 unixLock
, /* xLock method */
5005 unixUnlock
, /* xUnlock method */
5006 unixCheckReservedLock
, /* xCheckReservedLock method */
5007 unixShmMap
/* xShmMap method */
5010 nolockIoFinder
, /* Finder function name */
5011 nolockIoMethods
, /* sqlite3_io_methods object name */
5012 3, /* shared memory is disabled */
5013 nolockClose
, /* xClose method */
5014 nolockLock
, /* xLock method */
5015 nolockUnlock
, /* xUnlock method */
5016 nolockCheckReservedLock
, /* xCheckReservedLock method */
5017 0 /* xShmMap method */
5020 dotlockIoFinder
, /* Finder function name */
5021 dotlockIoMethods
, /* sqlite3_io_methods object name */
5022 1, /* shared memory is disabled */
5023 dotlockClose
, /* xClose method */
5024 dotlockLock
, /* xLock method */
5025 dotlockUnlock
, /* xUnlock method */
5026 dotlockCheckReservedLock
, /* xCheckReservedLock method */
5027 0 /* xShmMap method */
5030 #if SQLITE_ENABLE_LOCKING_STYLE
5032 flockIoFinder
, /* Finder function name */
5033 flockIoMethods
, /* sqlite3_io_methods object name */
5034 1, /* shared memory is disabled */
5035 flockClose
, /* xClose method */
5036 flockLock
, /* xLock method */
5037 flockUnlock
, /* xUnlock method */
5038 flockCheckReservedLock
, /* xCheckReservedLock method */
5039 0 /* xShmMap method */
5045 semIoFinder
, /* Finder function name */
5046 semIoMethods
, /* sqlite3_io_methods object name */
5047 1, /* shared memory is disabled */
5048 semXClose
, /* xClose method */
5049 semXLock
, /* xLock method */
5050 semXUnlock
, /* xUnlock method */
5051 semXCheckReservedLock
, /* xCheckReservedLock method */
5052 0 /* xShmMap method */
5056 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5058 afpIoFinder
, /* Finder function name */
5059 afpIoMethods
, /* sqlite3_io_methods object name */
5060 1, /* shared memory is disabled */
5061 afpClose
, /* xClose method */
5062 afpLock
, /* xLock method */
5063 afpUnlock
, /* xUnlock method */
5064 afpCheckReservedLock
, /* xCheckReservedLock method */
5065 0 /* xShmMap method */
5070 ** The proxy locking method is a "super-method" in the sense that it
5071 ** opens secondary file descriptors for the conch and lock files and
5072 ** it uses proxy, dot-file, AFP, and flock() locking methods on those
5073 ** secondary files. For this reason, the division that implements
5074 ** proxy locking is located much further down in the file. But we need
5075 ** to go ahead and define the sqlite3_io_methods and finder function
5076 ** for proxy locking here. So we forward declare the I/O methods.
5078 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5079 static int proxyClose(sqlite3_file
*);
5080 static int proxyLock(sqlite3_file
*, int);
5081 static int proxyUnlock(sqlite3_file
*, int);
5082 static int proxyCheckReservedLock(sqlite3_file
*, int*);
5084 proxyIoFinder
, /* Finder function name */
5085 proxyIoMethods
, /* sqlite3_io_methods object name */
5086 1, /* shared memory is disabled */
5087 proxyClose
, /* xClose method */
5088 proxyLock
, /* xLock method */
5089 proxyUnlock
, /* xUnlock method */
5090 proxyCheckReservedLock
, /* xCheckReservedLock method */
5091 0 /* xShmMap method */
5095 /* nfs lockd on OSX 10.3+ doesn't clear write locks when a read lock is set */
5096 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5098 nfsIoFinder
, /* Finder function name */
5099 nfsIoMethods
, /* sqlite3_io_methods object name */
5100 1, /* shared memory is disabled */
5101 unixClose
, /* xClose method */
5102 unixLock
, /* xLock method */
5103 nfsUnlock
, /* xUnlock method */
5104 unixCheckReservedLock
, /* xCheckReservedLock method */
5105 0 /* xShmMap method */
5109 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5111 ** This "finder" function attempts to determine the best locking strategy
5112 ** for the database file "filePath". It then returns the sqlite3_io_methods
5113 ** object that implements that strategy.
5115 ** This is for MacOSX only.
5117 static const sqlite3_io_methods
*autolockIoFinderImpl(
5118 const char *filePath
, /* name of the database file */
5119 unixFile
*pNew
/* open file object for the database file */
5121 static const struct Mapping
{
5122 const char *zFilesystem
; /* Filesystem type name */
5123 const sqlite3_io_methods
*pMethods
; /* Appropriate locking method */
5125 { "hfs", &posixIoMethods
},
5126 { "ufs", &posixIoMethods
},
5127 { "afpfs", &afpIoMethods
},
5128 { "smbfs", &afpIoMethods
},
5129 { "webdav", &nolockIoMethods
},
5133 struct statfs fsInfo
;
5134 struct flock lockInfo
;
5137 /* If filePath==NULL that means we are dealing with a transient file
5138 ** that does not need to be locked. */
5139 return &nolockIoMethods
;
5141 if( statfs(filePath
, &fsInfo
) != -1 ){
5142 if( fsInfo
.f_flags
& MNT_RDONLY
){
5143 return &nolockIoMethods
;
5145 for(i
=0; aMap
[i
].zFilesystem
; i
++){
5146 if( strcmp(fsInfo
.f_fstypename
, aMap
[i
].zFilesystem
)==0 ){
5147 return aMap
[i
].pMethods
;
5152 /* Default case. Handles, amongst others, "nfs".
5153 ** Test byte-range lock using fcntl(). If the call succeeds,
5154 ** assume that the file-system supports POSIX style locks.
5157 lockInfo
.l_start
= 0;
5158 lockInfo
.l_whence
= SEEK_SET
;
5159 lockInfo
.l_type
= F_RDLCK
;
5160 if( osFcntl(pNew
->h
, F_GETLK
, &lockInfo
)!=-1 ) {
5161 if( strcmp(fsInfo
.f_fstypename
, "nfs")==0 ){
5162 return &nfsIoMethods
;
5164 return &posixIoMethods
;
5167 return &dotlockIoMethods
;
5170 static const sqlite3_io_methods
5171 *(*const autolockIoFinder
)(const char*,unixFile
*) = autolockIoFinderImpl
;
5173 #endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
5177 ** This "finder" function for VxWorks checks to see if posix advisory
5178 ** locking works. If it does, then that is what is used. If it does not
5179 ** work, then fallback to named semaphore locking.
5181 static const sqlite3_io_methods
*vxworksIoFinderImpl(
5182 const char *filePath
, /* name of the database file */
5183 unixFile
*pNew
/* the open file object */
5185 struct flock lockInfo
;
5188 /* If filePath==NULL that means we are dealing with a transient file
5189 ** that does not need to be locked. */
5190 return &nolockIoMethods
;
5193 /* Test if fcntl() is supported and use POSIX style locks.
5194 ** Otherwise fall back to the named semaphore method.
5197 lockInfo
.l_start
= 0;
5198 lockInfo
.l_whence
= SEEK_SET
;
5199 lockInfo
.l_type
= F_RDLCK
;
5200 if( osFcntl(pNew
->h
, F_GETLK
, &lockInfo
)!=-1 ) {
5201 return &posixIoMethods
;
5203 return &semIoMethods
;
5206 static const sqlite3_io_methods
5207 *(*const vxworksIoFinder
)(const char*,unixFile
*) = vxworksIoFinderImpl
;
5209 #endif /* OS_VXWORKS */
5212 ** An abstract type for a pointer to an IO method finder function:
5214 typedef const sqlite3_io_methods
*(*finder_type
)(const char*,unixFile
*);
5217 /****************************************************************************
5218 **************************** sqlite3_vfs methods ****************************
5220 ** This division contains the implementation of methods on the
5221 ** sqlite3_vfs object.
5225 ** Initialize the contents of the unixFile structure pointed to by pId.
5227 static int fillInUnixFile(
5228 sqlite3_vfs
*pVfs
, /* Pointer to vfs object */
5229 int h
, /* Open file descriptor of file being opened */
5230 sqlite3_file
*pId
, /* Write to the unixFile structure here */
5231 const char *zFilename
, /* Name of the file being opened */
5232 int ctrlFlags
/* Zero or more UNIXFILE_* values */
5234 const sqlite3_io_methods
*pLockingStyle
;
5235 unixFile
*pNew
= (unixFile
*)pId
;
5238 assert( pNew
->pInode
==NULL
);
5240 /* Usually the path zFilename should not be a relative pathname. The
5241 ** exception is when opening the proxy "conch" file in builds that
5242 ** include the special Apple locking styles.
5244 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5245 assert( zFilename
==0 || zFilename
[0]=='/'
5246 || pVfs
->pAppData
==(void*)&autolockIoFinder
);
5248 assert( zFilename
==0 || zFilename
[0]=='/' );
5251 /* No locking occurs in temporary files */
5252 assert( zFilename
!=0 || (ctrlFlags
& UNIXFILE_NOLOCK
)!=0 );
5254 OSTRACE(("OPEN %-3d %s\n", h
, zFilename
));
5257 pNew
->zPath
= zFilename
;
5258 pNew
->ctrlFlags
= (u8
)ctrlFlags
;
5259 #if SQLITE_MAX_MMAP_SIZE>0
5260 pNew
->mmapSizeMax
= sqlite3GlobalConfig
.szMmap
;
5262 if( sqlite3_uri_boolean(((ctrlFlags
& UNIXFILE_URI
) ? zFilename
: 0),
5263 "psow", SQLITE_POWERSAFE_OVERWRITE
) ){
5264 pNew
->ctrlFlags
|= UNIXFILE_PSOW
;
5266 if( strcmp(pVfs
->zName
,"unix-excl")==0 ){
5267 pNew
->ctrlFlags
|= UNIXFILE_EXCL
;
5271 pNew
->pId
= vxworksFindFileId(zFilename
);
5273 ctrlFlags
|= UNIXFILE_NOLOCK
;
5274 rc
= SQLITE_NOMEM_BKPT
;
5278 if( ctrlFlags
& UNIXFILE_NOLOCK
){
5279 pLockingStyle
= &nolockIoMethods
;
5281 pLockingStyle
= (**(finder_type
*)pVfs
->pAppData
)(zFilename
, pNew
);
5282 #if SQLITE_ENABLE_LOCKING_STYLE
5283 /* Cache zFilename in the locking context (AFP and dotlock override) for
5284 ** proxyLock activation is possible (remote proxy is based on db name)
5285 ** zFilename remains valid until file is closed, to support */
5286 pNew
->lockingContext
= (void*)zFilename
;
5290 if( pLockingStyle
== &posixIoMethods
5291 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
5292 || pLockingStyle
== &nfsIoMethods
5296 rc
= findInodeInfo(pNew
, &pNew
->pInode
);
5297 if( rc
!=SQLITE_OK
){
5298 /* If an error occurred in findInodeInfo(), close the file descriptor
5299 ** immediately, before releasing the mutex. findInodeInfo() may fail
5300 ** in two scenarios:
5302 ** (a) A call to fstat() failed.
5303 ** (b) A malloc failed.
5305 ** Scenario (b) may only occur if the process is holding no other
5306 ** file descriptors open on the same file. If there were other file
5307 ** descriptors on this file, then no malloc would be required by
5308 ** findInodeInfo(). If this is the case, it is quite safe to close
5309 ** handle h - as it is guaranteed that no posix locks will be released
5312 ** If scenario (a) caused the error then things are not so safe. The
5313 ** implicit assumption here is that if fstat() fails, things are in
5314 ** such bad shape that dropping a lock or two doesn't matter much.
5316 robust_close(pNew
, h
, __LINE__
);
5322 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
5323 else if( pLockingStyle
== &afpIoMethods
){
5324 /* AFP locking uses the file path so it needs to be included in
5325 ** the afpLockingContext.
5327 afpLockingContext
*pCtx
;
5328 pNew
->lockingContext
= pCtx
= sqlite3_malloc64( sizeof(*pCtx
) );
5330 rc
= SQLITE_NOMEM_BKPT
;
5332 /* NB: zFilename exists and remains valid until the file is closed
5333 ** according to requirement F11141. So we do not need to make a
5334 ** copy of the filename. */
5335 pCtx
->dbPath
= zFilename
;
5339 rc
= findInodeInfo(pNew
, &pNew
->pInode
);
5340 if( rc
!=SQLITE_OK
){
5341 sqlite3_free(pNew
->lockingContext
);
5342 robust_close(pNew
, h
, __LINE__
);
5350 else if( pLockingStyle
== &dotlockIoMethods
){
5351 /* Dotfile locking uses the file path so it needs to be included in
5352 ** the dotlockLockingContext
5356 assert( zFilename
!=0 );
5357 nFilename
= (int)strlen(zFilename
) + 6;
5358 zLockFile
= (char *)sqlite3_malloc64(nFilename
);
5360 rc
= SQLITE_NOMEM_BKPT
;
5362 sqlite3_snprintf(nFilename
, zLockFile
, "%s" DOTLOCK_SUFFIX
, zFilename
);
5364 pNew
->lockingContext
= zLockFile
;
5368 else if( pLockingStyle
== &semIoMethods
){
5369 /* Named semaphore locking uses the file path so it needs to be
5370 ** included in the semLockingContext
5373 rc
= findInodeInfo(pNew
, &pNew
->pInode
);
5374 if( (rc
==SQLITE_OK
) && (pNew
->pInode
->pSem
==NULL
) ){
5375 char *zSemName
= pNew
->pInode
->aSemName
;
5377 sqlite3_snprintf(MAX_PATHNAME
, zSemName
, "/%s.sem",
5378 pNew
->pId
->zCanonicalName
);
5379 for( n
=1; zSemName
[n
]; n
++ )
5380 if( zSemName
[n
]=='/' ) zSemName
[n
] = '_';
5381 pNew
->pInode
->pSem
= sem_open(zSemName
, O_CREAT
, 0666, 1);
5382 if( pNew
->pInode
->pSem
== SEM_FAILED
){
5383 rc
= SQLITE_NOMEM_BKPT
;
5384 pNew
->pInode
->aSemName
[0] = '\0';
5391 storeLastErrno(pNew
, 0);
5393 if( rc
!=SQLITE_OK
){
5394 if( h
>=0 ) robust_close(pNew
, h
, __LINE__
);
5396 osUnlink(zFilename
);
5397 pNew
->ctrlFlags
|= UNIXFILE_DELETE
;
5400 if( rc
!=SQLITE_OK
){
5401 if( h
>=0 ) robust_close(pNew
, h
, __LINE__
);
5403 pNew
->pMethod
= pLockingStyle
;
5411 ** Return the name of a directory in which to put temporary files.
5412 ** If no suitable temporary file directory can be found, return NULL.
5414 static const char *unixTempFileDir(void){
5415 static const char *azDirs
[] = {
5425 const char *zDir
= sqlite3_temp_directory
;
5427 if( !azDirs
[0] ) azDirs
[0] = getenv("SQLITE_TMPDIR");
5428 if( !azDirs
[1] ) azDirs
[1] = getenv("TMPDIR");
5431 && osStat(zDir
, &buf
)==0
5432 && S_ISDIR(buf
.st_mode
)
5433 && osAccess(zDir
, 03)==0
5437 if( i
>=sizeof(azDirs
)/sizeof(azDirs
[0]) ) break;
5444 ** Create a temporary file name in zBuf. zBuf must be allocated
5445 ** by the calling process and must be big enough to hold at least
5446 ** pVfs->mxPathname bytes.
5448 static int unixGetTempname(int nBuf
, char *zBuf
){
5452 /* It's odd to simulate an io-error here, but really this is just
5453 ** using the io-error infrastructure to test that SQLite handles this
5454 ** function failing.
5457 SimulateIOError( return SQLITE_IOERR
);
5459 zDir
= unixTempFileDir();
5460 if( zDir
==0 ) return SQLITE_IOERR_GETTEMPPATH
;
5463 sqlite3_randomness(sizeof(r
), &r
);
5466 sqlite3_snprintf(nBuf
, zBuf
, "%s/"SQLITE_TEMP_FILE_PREFIX
"%llx%c",
5468 if( zBuf
[nBuf
-2]!=0 || (iLimit
++)>10 ) return SQLITE_ERROR
;
5469 }while( osAccess(zBuf
,0)==0 );
5473 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
5475 ** Routine to transform a unixFile into a proxy-locking unixFile.
5476 ** Implementation in the proxy-lock division, but used by unixOpen()
5477 ** if SQLITE_PREFER_PROXY_LOCKING is defined.
5479 static int proxyTransformUnixFile(unixFile
*, const char*);
5483 ** Search for an unused file descriptor that was opened on the database
5484 ** file (not a journal or master-journal file) identified by pathname
5485 ** zPath with SQLITE_OPEN_XXX flags matching those passed as the second
5486 ** argument to this function.
5488 ** Such a file descriptor may exist if a database connection was closed
5489 ** but the associated file descriptor could not be closed because some
5490 ** other file descriptor open on the same file is holding a file-lock.
5491 ** Refer to comments in the unixClose() function and the lengthy comment
5492 ** describing "Posix Advisory Locking" at the start of this file for
5493 ** further details. Also, ticket #4018.
5495 ** If a suitable file descriptor is found, then it is returned. If no
5496 ** such file descriptor is located, -1 is returned.
5498 static UnixUnusedFd
*findReusableFd(const char *zPath
, int flags
){
5499 UnixUnusedFd
*pUnused
= 0;
5501 /* Do not search for an unused file descriptor on vxworks. Not because
5502 ** vxworks would not benefit from the change (it might, we're not sure),
5503 ** but because no way to test it is currently available. It is better
5504 ** not to risk breaking vxworks support for the sake of such an obscure
5507 struct stat sStat
; /* Results of stat() call */
5509 /* A stat() call may fail for various reasons. If this happens, it is
5510 ** almost certain that an open() call on the same path will also fail.
5511 ** For this reason, if an error occurs in the stat() call here, it is
5512 ** ignored and -1 is returned. The caller will try to open a new file
5513 ** descriptor on the same path, fail, and return an error to SQLite.
5515 ** Even if a subsequent open() call does succeed, the consequences of
5516 ** not searching for a reusable file descriptor are not dire. */
5517 if( 0==osStat(zPath
, &sStat
) ){
5518 unixInodeInfo
*pInode
;
5522 while( pInode
&& (pInode
->fileId
.dev
!=sStat
.st_dev
5523 || pInode
->fileId
.ino
!=(u64
)sStat
.st_ino
) ){
5524 pInode
= pInode
->pNext
;
5528 for(pp
=&pInode
->pUnused
; *pp
&& (*pp
)->flags
!=flags
; pp
=&((*pp
)->pNext
));
5531 *pp
= pUnused
->pNext
;
5536 #endif /* if !OS_VXWORKS */
5541 ** Find the mode, uid and gid of file zFile.
5543 static int getFileMode(
5544 const char *zFile
, /* File name */
5545 mode_t
*pMode
, /* OUT: Permissions of zFile */
5546 uid_t
*pUid
, /* OUT: uid of zFile. */
5547 gid_t
*pGid
/* OUT: gid of zFile. */
5549 struct stat sStat
; /* Output of stat() on database file */
5551 if( 0==osStat(zFile
, &sStat
) ){
5552 *pMode
= sStat
.st_mode
& 0777;
5553 *pUid
= sStat
.st_uid
;
5554 *pGid
= sStat
.st_gid
;
5556 rc
= SQLITE_IOERR_FSTAT
;
5562 ** This function is called by unixOpen() to determine the unix permissions
5563 ** to create new files with. If no error occurs, then SQLITE_OK is returned
5564 ** and a value suitable for passing as the third argument to open(2) is
5565 ** written to *pMode. If an IO error occurs, an SQLite error code is
5566 ** returned and the value of *pMode is not modified.
5568 ** In most cases, this routine sets *pMode to 0, which will become
5569 ** an indication to robust_open() to create the file using
5570 ** SQLITE_DEFAULT_FILE_PERMISSIONS adjusted by the umask.
5571 ** But if the file being opened is a WAL or regular journal file, then
5572 ** this function queries the file-system for the permissions on the
5573 ** corresponding database file and sets *pMode to this value. Whenever
5574 ** possible, WAL and journal files are created using the same permissions
5575 ** as the associated database file.
5577 ** If the SQLITE_ENABLE_8_3_NAMES option is enabled, then the
5578 ** original filename is unavailable. But 8_3_NAMES is only used for
5579 ** FAT filesystems and permissions do not matter there, so just use
5580 ** the default permissions.
5582 static int findCreateFileMode(
5583 const char *zPath
, /* Path of file (possibly) being created */
5584 int flags
, /* Flags passed as 4th argument to xOpen() */
5585 mode_t
*pMode
, /* OUT: Permissions to open file with */
5586 uid_t
*pUid
, /* OUT: uid to set on the file */
5587 gid_t
*pGid
/* OUT: gid to set on the file */
5589 int rc
= SQLITE_OK
; /* Return Code */
5593 if( flags
& (SQLITE_OPEN_WAL
|SQLITE_OPEN_MAIN_JOURNAL
) ){
5594 char zDb
[MAX_PATHNAME
+1]; /* Database file path */
5595 int nDb
; /* Number of valid bytes in zDb */
5597 /* zPath is a path to a WAL or journal file. The following block derives
5598 ** the path to the associated database file from zPath. This block handles
5599 ** the following naming conventions:
5601 ** "<path to db>-journal"
5602 ** "<path to db>-wal"
5603 ** "<path to db>-journalNN"
5604 ** "<path to db>-walNN"
5606 ** where NN is a decimal number. The NN naming schemes are
5607 ** used by the test_multiplex.c module.
5609 nDb
= sqlite3Strlen30(zPath
) - 1;
5610 while( zPath
[nDb
]!='-' ){
5611 #ifndef SQLITE_ENABLE_8_3_NAMES
5612 /* In the normal case (8+3 filenames disabled) the journal filename
5613 ** is guaranteed to contain a '-' character. */
5615 assert( sqlite3Isalnum(zPath
[nDb
]) );
5617 /* If 8+3 names are possible, then the journal file might not contain
5618 ** a '-' character. So check for that case and return early. */
5619 if( nDb
==0 || zPath
[nDb
]=='.' ) return SQLITE_OK
;
5623 memcpy(zDb
, zPath
, nDb
);
5626 rc
= getFileMode(zDb
, pMode
, pUid
, pGid
);
5627 }else if( flags
& SQLITE_OPEN_DELETEONCLOSE
){
5629 }else if( flags
& SQLITE_OPEN_URI
){
5630 /* If this is a main database file and the file was opened using a URI
5631 ** filename, check for the "modeof" parameter. If present, interpret
5632 ** its value as a filename and try to copy the mode, uid and gid from
5634 const char *z
= sqlite3_uri_parameter(zPath
, "modeof");
5636 rc
= getFileMode(z
, pMode
, pUid
, pGid
);
5643 ** Open the file zPath.
5645 ** Previously, the SQLite OS layer used three functions in place of this
5648 ** sqlite3OsOpenReadWrite();
5649 ** sqlite3OsOpenReadOnly();
5650 ** sqlite3OsOpenExclusive();
5652 ** These calls correspond to the following combinations of flags:
5654 ** ReadWrite() -> (READWRITE | CREATE)
5655 ** ReadOnly() -> (READONLY)
5656 ** OpenExclusive() -> (READWRITE | CREATE | EXCLUSIVE)
5658 ** The old OpenExclusive() accepted a boolean argument - "delFlag". If
5659 ** true, the file was configured to be automatically deleted when the
5660 ** file handle closed. To achieve the same effect using this new
5661 ** interface, add the DELETEONCLOSE flag to those specified above for
5664 static int unixOpen(
5665 sqlite3_vfs
*pVfs
, /* The VFS for which this is the xOpen method */
5666 const char *zPath
, /* Pathname of file to be opened */
5667 sqlite3_file
*pFile
, /* The file descriptor to be filled in */
5668 int flags
, /* Input flags to control the opening */
5669 int *pOutFlags
/* Output flags returned to SQLite core */
5671 unixFile
*p
= (unixFile
*)pFile
;
5672 int fd
= -1; /* File descriptor returned by open() */
5673 int openFlags
= 0; /* Flags to pass to open() */
5674 int eType
= flags
&0xFFFFFF00; /* Type of file to open */
5675 int noLock
; /* True to omit locking primitives */
5676 int rc
= SQLITE_OK
; /* Function Return Code */
5677 int ctrlFlags
= 0; /* UNIXFILE_* flags */
5679 int isExclusive
= (flags
& SQLITE_OPEN_EXCLUSIVE
);
5680 int isDelete
= (flags
& SQLITE_OPEN_DELETEONCLOSE
);
5681 int isCreate
= (flags
& SQLITE_OPEN_CREATE
);
5682 int isReadonly
= (flags
& SQLITE_OPEN_READONLY
);
5683 int isReadWrite
= (flags
& SQLITE_OPEN_READWRITE
);
5684 #if SQLITE_ENABLE_LOCKING_STYLE
5685 int isAutoProxy
= (flags
& SQLITE_OPEN_AUTOPROXY
);
5687 #if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE
5688 struct statfs fsInfo
;
5691 /* If creating a master or main-file journal, this function will open
5692 ** a file-descriptor on the directory too. The first time unixSync()
5693 ** is called the directory file descriptor will be fsync()ed and close()d.
5695 int syncDir
= (isCreate
&& (
5696 eType
==SQLITE_OPEN_MASTER_JOURNAL
5697 || eType
==SQLITE_OPEN_MAIN_JOURNAL
5698 || eType
==SQLITE_OPEN_WAL
5701 /* If argument zPath is a NULL pointer, this function is required to open
5702 ** a temporary file. Use this buffer to store the file name in.
5704 char zTmpname
[MAX_PATHNAME
+2];
5705 const char *zName
= zPath
;
5707 /* Check the following statements are true:
5709 ** (a) Exactly one of the READWRITE and READONLY flags must be set, and
5710 ** (b) if CREATE is set, then READWRITE must also be set, and
5711 ** (c) if EXCLUSIVE is set, then CREATE must also be set.
5712 ** (d) if DELETEONCLOSE is set, then CREATE must also be set.
5714 assert((isReadonly
==0 || isReadWrite
==0) && (isReadWrite
|| isReadonly
));
5715 assert(isCreate
==0 || isReadWrite
);
5716 assert(isExclusive
==0 || isCreate
);
5717 assert(isDelete
==0 || isCreate
);
5719 /* The main DB, main journal, WAL file and master journal are never
5720 ** automatically deleted. Nor are they ever temporary files. */
5721 assert( (!isDelete
&& zName
) || eType
!=SQLITE_OPEN_MAIN_DB
);
5722 assert( (!isDelete
&& zName
) || eType
!=SQLITE_OPEN_MAIN_JOURNAL
);
5723 assert( (!isDelete
&& zName
) || eType
!=SQLITE_OPEN_MASTER_JOURNAL
);
5724 assert( (!isDelete
&& zName
) || eType
!=SQLITE_OPEN_WAL
);
5726 /* Assert that the upper layer has set one of the "file-type" flags. */
5727 assert( eType
==SQLITE_OPEN_MAIN_DB
|| eType
==SQLITE_OPEN_TEMP_DB
5728 || eType
==SQLITE_OPEN_MAIN_JOURNAL
|| eType
==SQLITE_OPEN_TEMP_JOURNAL
5729 || eType
==SQLITE_OPEN_SUBJOURNAL
|| eType
==SQLITE_OPEN_MASTER_JOURNAL
5730 || eType
==SQLITE_OPEN_TRANSIENT_DB
|| eType
==SQLITE_OPEN_WAL
5733 /* Detect a pid change and reset the PRNG. There is a race condition
5734 ** here such that two or more threads all trying to open databases at
5735 ** the same instant might all reset the PRNG. But multiple resets
5738 if( randomnessPid
!=osGetpid(0) ){
5739 randomnessPid
= osGetpid(0);
5740 sqlite3_randomness(0,0);
5743 memset(p
, 0, sizeof(unixFile
));
5745 if( eType
==SQLITE_OPEN_MAIN_DB
){
5746 UnixUnusedFd
*pUnused
;
5747 pUnused
= findReusableFd(zName
, flags
);
5751 pUnused
= sqlite3_malloc64(sizeof(*pUnused
));
5753 return SQLITE_NOMEM_BKPT
;
5756 p
->pUnused
= pUnused
;
5758 /* Database filenames are double-zero terminated if they are not
5759 ** URIs with parameters. Hence, they can always be passed into
5760 ** sqlite3_uri_parameter(). */
5761 assert( (flags
& SQLITE_OPEN_URI
) || zName
[strlen(zName
)+1]==0 );
5764 /* If zName is NULL, the upper layer is requesting a temp file. */
5765 assert(isDelete
&& !syncDir
);
5766 rc
= unixGetTempname(pVfs
->mxPathname
, zTmpname
);
5767 if( rc
!=SQLITE_OK
){
5772 /* Generated temporary filenames are always double-zero terminated
5773 ** for use by sqlite3_uri_parameter(). */
5774 assert( zName
[strlen(zName
)+1]==0 );
5777 /* Determine the value of the flags parameter passed to POSIX function
5778 ** open(). These must be calculated even if open() is not called, as
5779 ** they may be stored as part of the file handle and used by the
5780 ** 'conch file' locking functions later on. */
5781 if( isReadonly
) openFlags
|= O_RDONLY
;
5782 if( isReadWrite
) openFlags
|= O_RDWR
;
5783 if( isCreate
) openFlags
|= O_CREAT
;
5784 if( isExclusive
) openFlags
|= (O_EXCL
|O_NOFOLLOW
);
5785 openFlags
|= (O_LARGEFILE
|O_BINARY
);
5788 mode_t openMode
; /* Permissions to create file with */
5789 uid_t uid
; /* Userid for the file */
5790 gid_t gid
; /* Groupid for the file */
5791 rc
= findCreateFileMode(zName
, flags
, &openMode
, &uid
, &gid
);
5792 if( rc
!=SQLITE_OK
){
5793 assert( !p
->pUnused
);
5794 assert( eType
==SQLITE_OPEN_WAL
|| eType
==SQLITE_OPEN_MAIN_JOURNAL
);
5797 fd
= robust_open(zName
, openFlags
, openMode
);
5798 OSTRACE(("OPENX %-3d %s 0%o\n", fd
, zName
, openFlags
));
5799 assert( !isExclusive
|| (openFlags
& O_CREAT
)!=0 );
5800 if( fd
<0 && errno
!=EISDIR
&& isReadWrite
){
5801 /* Failed to open the file for read/write access. Try read-only. */
5802 flags
&= ~(SQLITE_OPEN_READWRITE
|SQLITE_OPEN_CREATE
);
5803 openFlags
&= ~(O_RDWR
|O_CREAT
);
5804 flags
|= SQLITE_OPEN_READONLY
;
5805 openFlags
|= O_RDONLY
;
5807 fd
= robust_open(zName
, openFlags
, openMode
);
5810 rc
= unixLogError(SQLITE_CANTOPEN_BKPT
, "open", zName
);
5814 /* If this process is running as root and if creating a new rollback
5815 ** journal or WAL file, set the ownership of the journal or WAL to be
5816 ** the same as the original database.
5818 if( flags
& (SQLITE_OPEN_WAL
|SQLITE_OPEN_MAIN_JOURNAL
) ){
5819 robustFchown(fd
, uid
, gid
);
5828 p
->pUnused
->fd
= fd
;
5829 p
->pUnused
->flags
= flags
;
5835 #elif defined(SQLITE_UNLINK_AFTER_CLOSE)
5836 zPath
= sqlite3_mprintf("%s", zName
);
5838 robust_close(p
, fd
, __LINE__
);
5839 return SQLITE_NOMEM_BKPT
;
5845 #if SQLITE_ENABLE_LOCKING_STYLE
5847 p
->openFlags
= openFlags
;
5851 #if defined(__APPLE__) || SQLITE_ENABLE_LOCKING_STYLE
5852 if( fstatfs(fd
, &fsInfo
) == -1 ){
5853 storeLastErrno(p
, errno
);
5854 robust_close(p
, fd
, __LINE__
);
5855 return SQLITE_IOERR_ACCESS
;
5857 if (0 == strncmp("msdos", fsInfo
.f_fstypename
, 5)) {
5858 ((unixFile
*)pFile
)->fsFlags
|= SQLITE_FSFLAGS_IS_MSDOS
;
5860 if (0 == strncmp("exfat", fsInfo
.f_fstypename
, 5)) {
5861 ((unixFile
*)pFile
)->fsFlags
|= SQLITE_FSFLAGS_IS_MSDOS
;
5865 /* Set up appropriate ctrlFlags */
5866 if( isDelete
) ctrlFlags
|= UNIXFILE_DELETE
;
5867 if( isReadonly
) ctrlFlags
|= UNIXFILE_RDONLY
;
5868 noLock
= eType
!=SQLITE_OPEN_MAIN_DB
;
5869 if( noLock
) ctrlFlags
|= UNIXFILE_NOLOCK
;
5870 if( syncDir
) ctrlFlags
|= UNIXFILE_DIRSYNC
;
5871 if( flags
& SQLITE_OPEN_URI
) ctrlFlags
|= UNIXFILE_URI
;
5873 #if SQLITE_ENABLE_LOCKING_STYLE
5874 #if SQLITE_PREFER_PROXY_LOCKING
5877 if( isAutoProxy
&& (zPath
!=NULL
) && (!noLock
) && pVfs
->xOpen
){
5878 char *envforce
= getenv("SQLITE_FORCE_PROXY_LOCKING");
5881 /* SQLITE_FORCE_PROXY_LOCKING==1 means force always use proxy, 0 means
5882 ** never use proxy, NULL means use proxy for non-local files only. */
5883 if( envforce
!=NULL
){
5884 useProxy
= atoi(envforce
)>0;
5886 useProxy
= !(fsInfo
.f_flags
&MNT_LOCAL
);
5889 rc
= fillInUnixFile(pVfs
, fd
, pFile
, zPath
, ctrlFlags
);
5890 if( rc
==SQLITE_OK
){
5891 rc
= proxyTransformUnixFile((unixFile
*)pFile
, ":auto:");
5892 if( rc
!=SQLITE_OK
){
5893 /* Use unixClose to clean up the resources added in fillInUnixFile
5894 ** and clear all the structure's references. Specifically,
5895 ** pFile->pMethods will be NULL so sqlite3OsClose will be a no-op
5906 rc
= fillInUnixFile(pVfs
, fd
, pFile
, zPath
, ctrlFlags
);
5909 if( rc
!=SQLITE_OK
){
5910 sqlite3_free(p
->pUnused
);
5917 ** Delete the file at zPath. If the dirSync argument is true, fsync()
5918 ** the directory after deleting the file.
5920 static int unixDelete(
5921 sqlite3_vfs
*NotUsed
, /* VFS containing this as the xDelete method */
5922 const char *zPath
, /* Name of file to be deleted */
5923 int dirSync
/* If true, fsync() directory after deleting file */
5926 UNUSED_PARAMETER(NotUsed
);
5927 SimulateIOError(return SQLITE_IOERR_DELETE
);
5928 if( osUnlink(zPath
)==(-1) ){
5931 || osAccess(zPath
,0)!=0
5934 rc
= SQLITE_IOERR_DELETE_NOENT
;
5936 rc
= unixLogError(SQLITE_IOERR_DELETE
, "unlink", zPath
);
5940 #ifndef SQLITE_DISABLE_DIRSYNC
5941 if( (dirSync
& 1)!=0 ){
5943 rc
= osOpenDirectory(zPath
, &fd
);
5944 if( rc
==SQLITE_OK
){
5945 if( full_fsync(fd
,0,0) ){
5946 rc
= unixLogError(SQLITE_IOERR_DIR_FSYNC
, "fsync", zPath
);
5948 robust_close(0, fd
, __LINE__
);
5950 assert( rc
==SQLITE_CANTOPEN
);
5959 ** Test the existence of or access permissions of file zPath. The
5960 ** test performed depends on the value of flags:
5962 ** SQLITE_ACCESS_EXISTS: Return 1 if the file exists
5963 ** SQLITE_ACCESS_READWRITE: Return 1 if the file is read and writable.
5964 ** SQLITE_ACCESS_READONLY: Return 1 if the file is readable.
5966 ** Otherwise return 0.
5968 static int unixAccess(
5969 sqlite3_vfs
*NotUsed
, /* The VFS containing this xAccess method */
5970 const char *zPath
, /* Path of the file to examine */
5971 int flags
, /* What do we want to learn about the zPath file? */
5972 int *pResOut
/* Write result boolean here */
5974 UNUSED_PARAMETER(NotUsed
);
5975 SimulateIOError( return SQLITE_IOERR_ACCESS
; );
5976 assert( pResOut
!=0 );
5978 /* The spec says there are three possible values for flags. But only
5979 ** two of them are actually used */
5980 assert( flags
==SQLITE_ACCESS_EXISTS
|| flags
==SQLITE_ACCESS_READWRITE
);
5982 if( flags
==SQLITE_ACCESS_EXISTS
){
5984 *pResOut
= (0==osStat(zPath
, &buf
) && buf
.st_size
>0);
5986 *pResOut
= osAccess(zPath
, W_OK
|R_OK
)==0;
5994 static int mkFullPathname(
5995 const char *zPath
, /* Input path */
5996 char *zOut
, /* Output buffer */
5997 int nOut
/* Allocated size of buffer zOut */
5999 int nPath
= sqlite3Strlen30(zPath
);
6001 if( zPath
[0]!='/' ){
6002 if( osGetcwd(zOut
, nOut
-2)==0 ){
6003 return unixLogError(SQLITE_CANTOPEN_BKPT
, "getcwd", zPath
);
6005 iOff
= sqlite3Strlen30(zOut
);
6008 if( (iOff
+nPath
+1)>nOut
){
6009 /* SQLite assumes that xFullPathname() nul-terminates the output buffer
6010 ** even if it returns an error. */
6012 return SQLITE_CANTOPEN_BKPT
;
6014 sqlite3_snprintf(nOut
-iOff
, &zOut
[iOff
], "%s", zPath
);
6019 ** Turn a relative pathname into a full pathname. The relative path
6020 ** is stored as a nul-terminated string in the buffer pointed to by
6023 ** zOut points to a buffer of at least sqlite3_vfs.mxPathname bytes
6024 ** (in this case, MAX_PATHNAME bytes). The full-path is written to
6025 ** this buffer before returning.
6027 static int unixFullPathname(
6028 sqlite3_vfs
*pVfs
, /* Pointer to vfs object */
6029 const char *zPath
, /* Possibly relative input path */
6030 int nOut
, /* Size of output buffer in bytes */
6031 char *zOut
/* Output buffer */
6033 #if !defined(HAVE_READLINK) || !defined(HAVE_LSTAT)
6034 return mkFullPathname(zPath
, zOut
, nOut
);
6038 int nLink
= 1; /* Number of symbolic links followed so far */
6039 const char *zIn
= zPath
; /* Input path for each iteration of loop */
6042 assert( pVfs
->mxPathname
==MAX_PATHNAME
);
6043 UNUSED_PARAMETER(pVfs
);
6045 /* It's odd to simulate an io-error here, but really this is just
6046 ** using the io-error infrastructure to test that SQLite handles this
6047 ** function failing. This function could fail if, for example, the
6048 ** current working directory has been unlinked.
6050 SimulateIOError( return SQLITE_ERROR
);
6054 /* Call stat() on path zIn. Set bLink to true if the path is a symbolic
6055 ** link, or false otherwise. */
6058 if( osLstat(zIn
, &buf
)!=0 ){
6059 if( errno
!=ENOENT
){
6060 rc
= unixLogError(SQLITE_CANTOPEN_BKPT
, "lstat", zIn
);
6063 bLink
= S_ISLNK(buf
.st_mode
);
6068 zDel
= sqlite3_malloc(nOut
);
6069 if( zDel
==0 ) rc
= SQLITE_NOMEM_BKPT
;
6070 }else if( ++nLink
>SQLITE_MAX_SYMLINKS
){
6071 rc
= SQLITE_CANTOPEN_BKPT
;
6074 if( rc
==SQLITE_OK
){
6075 nByte
= osReadlink(zIn
, zDel
, nOut
-1);
6077 rc
= unixLogError(SQLITE_CANTOPEN_BKPT
, "readlink", zIn
);
6081 for(n
= sqlite3Strlen30(zIn
); n
>0 && zIn
[n
-1]!='/'; n
--);
6082 if( nByte
+n
+1>nOut
){
6083 rc
= SQLITE_CANTOPEN_BKPT
;
6085 memmove(&zDel
[n
], zDel
, nByte
+1);
6086 memcpy(zDel
, zIn
, n
);
6097 assert( rc
!=SQLITE_OK
|| zIn
!=zOut
|| zIn
[0]=='/' );
6098 if( rc
==SQLITE_OK
&& zIn
!=zOut
){
6099 rc
= mkFullPathname(zIn
, zOut
, nOut
);
6101 if( bLink
==0 ) break;
6103 }while( rc
==SQLITE_OK
);
6107 #endif /* HAVE_READLINK && HAVE_LSTAT */
6111 #ifndef SQLITE_OMIT_LOAD_EXTENSION
6113 ** Interfaces for opening a shared library, finding entry points
6114 ** within the shared library, and closing the shared library.
6117 static void *unixDlOpen(sqlite3_vfs
*NotUsed
, const char *zFilename
){
6118 UNUSED_PARAMETER(NotUsed
);
6119 return dlopen(zFilename
, RTLD_NOW
| RTLD_GLOBAL
);
6123 ** SQLite calls this function immediately after a call to unixDlSym() or
6124 ** unixDlOpen() fails (returns a null pointer). If a more detailed error
6125 ** message is available, it is written to zBufOut. If no error message
6126 ** is available, zBufOut is left unmodified and SQLite uses a default
6129 static void unixDlError(sqlite3_vfs
*NotUsed
, int nBuf
, char *zBufOut
){
6131 UNUSED_PARAMETER(NotUsed
);
6135 sqlite3_snprintf(nBuf
, zBufOut
, "%s", zErr
);
6139 static void (*unixDlSym(sqlite3_vfs
*NotUsed
, void *p
, const char*zSym
))(void){
6141 ** GCC with -pedantic-errors says that C90 does not allow a void* to be
6142 ** cast into a pointer to a function. And yet the library dlsym() routine
6143 ** returns a void* which is really a pointer to a function. So how do we
6144 ** use dlsym() with -pedantic-errors?
6146 ** Variable x below is defined to be a pointer to a function taking
6147 ** parameters void* and const char* and returning a pointer to a function.
6148 ** We initialize x by assigning it a pointer to the dlsym() function.
6149 ** (That assignment requires a cast.) Then we call the function that
6152 ** This work-around is unlikely to work correctly on any system where
6153 ** you really cannot cast a function pointer into void*. But then, on the
6154 ** other hand, dlsym() will not work on such a system either, so we have
6155 ** not really lost anything.
6157 void (*(*x
)(void*,const char*))(void);
6158 UNUSED_PARAMETER(NotUsed
);
6159 x
= (void(*(*)(void*,const char*))(void))dlsym
;
6160 return (*x
)(p
, zSym
);
6162 static void unixDlClose(sqlite3_vfs
*NotUsed
, void *pHandle
){
6163 UNUSED_PARAMETER(NotUsed
);
6166 #else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */
6167 #define unixDlOpen 0
6168 #define unixDlError 0
6170 #define unixDlClose 0
6174 ** Write nBuf bytes of random data to the supplied buffer zBuf.
6176 static int unixRandomness(sqlite3_vfs
*NotUsed
, int nBuf
, char *zBuf
){
6177 UNUSED_PARAMETER(NotUsed
);
6178 assert((size_t)nBuf
>=(sizeof(time_t)+sizeof(int)));
6180 /* We have to initialize zBuf to prevent valgrind from reporting
6181 ** errors. The reports issued by valgrind are incorrect - we would
6182 ** prefer that the randomness be increased by making use of the
6183 ** uninitialized space in zBuf - but valgrind errors tend to worry
6184 ** some users. Rather than argue, it seems easier just to initialize
6185 ** the whole array and silence valgrind, even if that means less randomness
6186 ** in the random seed.
6188 ** When testing, initializing zBuf[] to zero is all we do. That means
6189 ** that we always use the same random number sequence. This makes the
6190 ** tests repeatable.
6192 memset(zBuf
, 0, nBuf
);
6193 randomnessPid
= osGetpid(0);
6194 #if !defined(SQLITE_TEST) && !defined(SQLITE_OMIT_RANDOMNESS)
6197 fd
= robust_open("/dev/urandom", O_RDONLY
, 0);
6201 memcpy(zBuf
, &t
, sizeof(t
));
6202 memcpy(&zBuf
[sizeof(t
)], &randomnessPid
, sizeof(randomnessPid
));
6203 assert( sizeof(t
)+sizeof(randomnessPid
)<=(size_t)nBuf
);
6204 nBuf
= sizeof(t
) + sizeof(randomnessPid
);
6206 do{ got
= osRead(fd
, zBuf
, nBuf
); }while( got
<0 && errno
==EINTR
);
6207 robust_close(0, fd
, __LINE__
);
6216 ** Sleep for a little while. Return the amount of time slept.
6217 ** The argument is the number of microseconds we want to sleep.
6218 ** The return value is the number of microseconds of sleep actually
6219 ** requested from the underlying operating system, a number which
6220 ** might be greater than or equal to the argument, but not less
6221 ** than the argument.
6223 static int unixSleep(sqlite3_vfs
*NotUsed
, int microseconds
){
6227 sp
.tv_sec
= microseconds
/ 1000000;
6228 sp
.tv_nsec
= (microseconds
% 1000000) * 1000;
6229 nanosleep(&sp
, NULL
);
6230 UNUSED_PARAMETER(NotUsed
);
6231 return microseconds
;
6232 #elif defined(HAVE_USLEEP) && HAVE_USLEEP
6233 usleep(microseconds
);
6234 UNUSED_PARAMETER(NotUsed
);
6235 return microseconds
;
6237 int seconds
= (microseconds
+999999)/1000000;
6239 UNUSED_PARAMETER(NotUsed
);
6240 return seconds
*1000000;
6245 ** The following variable, if set to a non-zero value, is interpreted as
6246 ** the number of seconds since 1970 and is used to set the result of
6247 ** sqlite3OsCurrentTime() during testing.
6250 int sqlite3_current_time
= 0; /* Fake system time in seconds since 1970. */
6254 ** Find the current time (in Universal Coordinated Time). Write into *piNow
6255 ** the current time and date as a Julian Day number times 86_400_000. In
6256 ** other words, write into *piNow the number of milliseconds since the Julian
6257 ** epoch of noon in Greenwich on November 24, 4714 B.C according to the
6258 ** proleptic Gregorian calendar.
6260 ** On success, return SQLITE_OK. Return SQLITE_ERROR if the time and date
6263 static int unixCurrentTimeInt64(sqlite3_vfs
*NotUsed
, sqlite3_int64
*piNow
){
6264 static const sqlite3_int64 unixEpoch
= 24405875*(sqlite3_int64
)8640000;
6266 #if defined(NO_GETTOD)
6269 *piNow
= ((sqlite3_int64
)t
)*1000 + unixEpoch
;
6271 struct timespec sNow
;
6272 clock_gettime(CLOCK_REALTIME
, &sNow
);
6273 *piNow
= unixEpoch
+ 1000*(sqlite3_int64
)sNow
.tv_sec
+ sNow
.tv_nsec
/1000000;
6275 struct timeval sNow
;
6276 (void)gettimeofday(&sNow
, 0); /* Cannot fail given valid arguments */
6277 *piNow
= unixEpoch
+ 1000*(sqlite3_int64
)sNow
.tv_sec
+ sNow
.tv_usec
/1000;
6281 if( sqlite3_current_time
){
6282 *piNow
= 1000*(sqlite3_int64
)sqlite3_current_time
+ unixEpoch
;
6285 UNUSED_PARAMETER(NotUsed
);
6289 #ifndef SQLITE_OMIT_DEPRECATED
6291 ** Find the current time (in Universal Coordinated Time). Write the
6292 ** current time and date as a Julian Day number into *prNow and
6293 ** return 0. Return 1 if the time and date cannot be found.
6295 static int unixCurrentTime(sqlite3_vfs
*NotUsed
, double *prNow
){
6296 sqlite3_int64 i
= 0;
6298 UNUSED_PARAMETER(NotUsed
);
6299 rc
= unixCurrentTimeInt64(0, &i
);
6300 *prNow
= i
/86400000.0;
6304 # define unixCurrentTime 0
6308 ** The xGetLastError() method is designed to return a better
6309 ** low-level error message when operating-system problems come up
6310 ** during SQLite operation. Only the integer return code is currently
6313 static int unixGetLastError(sqlite3_vfs
*NotUsed
, int NotUsed2
, char *NotUsed3
){
6314 UNUSED_PARAMETER(NotUsed
);
6315 UNUSED_PARAMETER(NotUsed2
);
6316 UNUSED_PARAMETER(NotUsed3
);
6322 ************************ End of sqlite3_vfs methods ***************************
6323 ******************************************************************************/
6325 /******************************************************************************
6326 ************************** Begin Proxy Locking ********************************
6328 ** Proxy locking is a "uber-locking-method" in this sense: It uses the
6329 ** other locking methods on secondary lock files. Proxy locking is a
6330 ** meta-layer over top of the primitive locking implemented above. For
6331 ** this reason, the division that implements of proxy locking is deferred
6332 ** until late in the file (here) after all of the other I/O methods have
6333 ** been defined - so that the primitive locking methods are available
6334 ** as services to help with the implementation of proxy locking.
6338 ** The default locking schemes in SQLite use byte-range locks on the
6339 ** database file to coordinate safe, concurrent access by multiple readers
6340 ** and writers [http://sqlite.org/lockingv3.html]. The five file locking
6341 ** states (UNLOCKED, PENDING, SHARED, RESERVED, EXCLUSIVE) are implemented
6342 ** as POSIX read & write locks over fixed set of locations (via fsctl),
6343 ** on AFP and SMB only exclusive byte-range locks are available via fsctl
6344 ** with _IOWR('z', 23, struct ByteRangeLockPB2) to track the same 5 states.
6345 ** To simulate a F_RDLCK on the shared range, on AFP a randomly selected
6346 ** address in the shared range is taken for a SHARED lock, the entire
6347 ** shared range is taken for an EXCLUSIVE lock):
6349 ** PENDING_BYTE 0x40000000
6350 ** RESERVED_BYTE 0x40000001
6351 ** SHARED_RANGE 0x40000002 -> 0x40000200
6353 ** This works well on the local file system, but shows a nearly 100x
6354 ** slowdown in read performance on AFP because the AFP client disables
6355 ** the read cache when byte-range locks are present. Enabling the read
6356 ** cache exposes a cache coherency problem that is present on all OS X
6357 ** supported network file systems. NFS and AFP both observe the
6358 ** close-to-open semantics for ensuring cache coherency
6359 ** [http://nfs.sourceforge.net/#faq_a8], which does not effectively
6360 ** address the requirements for concurrent database access by multiple
6361 ** readers and writers
6362 ** [http://www.nabble.com/SQLite-on-NFS-cache-coherency-td15655701.html].
6364 ** To address the performance and cache coherency issues, proxy file locking
6365 ** changes the way database access is controlled by limiting access to a
6366 ** single host at a time and moving file locks off of the database file
6367 ** and onto a proxy file on the local file system.
6370 ** Using proxy locks
6371 ** -----------------
6375 ** sqlite3_file_control(db, dbname, SQLITE_FCNTL_SET_LOCKPROXYFILE,
6376 ** <proxy_path> | ":auto:");
6377 ** sqlite3_file_control(db, dbname, SQLITE_FCNTL_GET_LOCKPROXYFILE,
6383 ** PRAGMA [database.]lock_proxy_file=<proxy_path> | :auto:
6384 ** PRAGMA [database.]lock_proxy_file
6386 ** Specifying ":auto:" means that if there is a conch file with a matching
6387 ** host ID in it, the proxy path in the conch file will be used, otherwise
6388 ** a proxy path based on the user's temp dir
6389 ** (via confstr(_CS_DARWIN_USER_TEMP_DIR,...)) will be used and the
6390 ** actual proxy file name is generated from the name and path of the
6391 ** database file. For example:
6393 ** For database path "/Users/me/foo.db"
6394 ** The lock path will be "<tmpdir>/sqliteplocks/_Users_me_foo.db:auto:")
6396 ** Once a lock proxy is configured for a database connection, it can not
6397 ** be removed, however it may be switched to a different proxy path via
6398 ** the above APIs (assuming the conch file is not being held by another
6399 ** connection or process).
6402 ** How proxy locking works
6403 ** -----------------------
6405 ** Proxy file locking relies primarily on two new supporting files:
6407 ** * conch file to limit access to the database file to a single host
6410 ** * proxy file to act as a proxy for the advisory locks normally
6411 ** taken on the database
6413 ** The conch file - to use a proxy file, sqlite must first "hold the conch"
6414 ** by taking an sqlite-style shared lock on the conch file, reading the
6415 ** contents and comparing the host's unique host ID (see below) and lock
6416 ** proxy path against the values stored in the conch. The conch file is
6417 ** stored in the same directory as the database file and the file name
6418 ** is patterned after the database file name as ".<databasename>-conch".
6419 ** If the conch file does not exist, or its contents do not match the
6420 ** host ID and/or proxy path, then the lock is escalated to an exclusive
6421 ** lock and the conch file contents is updated with the host ID and proxy
6422 ** path and the lock is downgraded to a shared lock again. If the conch
6423 ** is held by another process (with a shared lock), the exclusive lock
6424 ** will fail and SQLITE_BUSY is returned.
6426 ** The proxy file - a single-byte file used for all advisory file locks
6427 ** normally taken on the database file. This allows for safe sharing
6428 ** of the database file for multiple readers and writers on the same
6429 ** host (the conch ensures that they all use the same local lock file).
6431 ** Requesting the lock proxy does not immediately take the conch, it is
6432 ** only taken when the first request to lock database file is made.
6433 ** This matches the semantics of the traditional locking behavior, where
6434 ** opening a connection to a database file does not take a lock on it.
6435 ** The shared lock and an open file descriptor are maintained until
6436 ** the connection to the database is closed.
6438 ** The proxy file and the lock file are never deleted so they only need
6439 ** to be created the first time they are used.
6441 ** Configuration options
6442 ** ---------------------
6444 ** SQLITE_PREFER_PROXY_LOCKING
6446 ** Database files accessed on non-local file systems are
6447 ** automatically configured for proxy locking, lock files are
6448 ** named automatically using the same logic as
6449 ** PRAGMA lock_proxy_file=":auto:"
6451 ** SQLITE_PROXY_DEBUG
6453 ** Enables the logging of error messages during host id file
6454 ** retrieval and creation
6458 ** Overrides the default directory used for lock proxy files that
6459 ** are named automatically via the ":auto:" setting
6461 ** SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
6463 ** Permissions to use when creating a directory for storing the
6464 ** lock proxy files, only used when LOCKPROXYDIR is not set.
6467 ** As mentioned above, when compiled with SQLITE_PREFER_PROXY_LOCKING,
6468 ** setting the environment variable SQLITE_FORCE_PROXY_LOCKING to 1 will
6469 ** force proxy locking to be used for every database file opened, and 0
6470 ** will force automatic proxy locking to be disabled for all database
6471 ** files (explicitly calling the SQLITE_FCNTL_SET_LOCKPROXYFILE pragma or
6472 ** sqlite_file_control API is not affected by SQLITE_FORCE_PROXY_LOCKING).
6476 ** Proxy locking is only available on MacOSX
6478 #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
6481 ** The proxyLockingContext has the path and file structures for the remote
6482 ** and local proxy files in it
6484 typedef struct proxyLockingContext proxyLockingContext
;
6485 struct proxyLockingContext
{
6486 unixFile
*conchFile
; /* Open conch file */
6487 char *conchFilePath
; /* Name of the conch file */
6488 unixFile
*lockProxy
; /* Open proxy lock file */
6489 char *lockProxyPath
; /* Name of the proxy lock file */
6490 char *dbPath
; /* Name of the open file */
6491 int conchHeld
; /* 1 if the conch is held, -1 if lockless */
6492 int nFails
; /* Number of conch taking failures */
6493 void *oldLockingContext
; /* Original lockingcontext to restore on close */
6494 sqlite3_io_methods
const *pOldMethod
; /* Original I/O methods for close */
6498 ** The proxy lock file path for the database at dbPath is written into lPath,
6499 ** which must point to valid, writable memory large enough for a maxLen length
6502 static int proxyGetLockPath(const char *dbPath
, char *lPath
, size_t maxLen
){
6508 len
= strlcpy(lPath
, LOCKPROXYDIR
, maxLen
);
6510 # ifdef _CS_DARWIN_USER_TEMP_DIR
6512 if( !confstr(_CS_DARWIN_USER_TEMP_DIR
, lPath
, maxLen
) ){
6513 OSTRACE(("GETLOCKPATH failed %s errno=%d pid=%d\n",
6514 lPath
, errno
, osGetpid(0)));
6515 return SQLITE_IOERR_LOCK
;
6517 len
= strlcat(lPath
, "sqliteplocks", maxLen
);
6520 len
= strlcpy(lPath
, "/tmp/", maxLen
);
6524 if( lPath
[len
-1]!='/' ){
6525 len
= strlcat(lPath
, "/", maxLen
);
6528 /* transform the db path to a unique cache name */
6529 dbLen
= (int)strlen(dbPath
);
6530 for( i
=0; i
<dbLen
&& (i
+len
+7)<(int)maxLen
; i
++){
6532 lPath
[i
+len
] = (c
=='/')?'_':c
;
6535 strlcat(lPath
, ":auto:", maxLen
);
6536 OSTRACE(("GETLOCKPATH proxy lock path=%s pid=%d\n", lPath
, osGetpid(0)));
6541 ** Creates the lock file and any missing directories in lockPath
6543 static int proxyCreateLockPath(const char *lockPath
){
6545 char buf
[MAXPATHLEN
];
6548 assert(lockPath
!=NULL
);
6549 /* try to create all the intermediate directories */
6550 len
= (int)strlen(lockPath
);
6551 buf
[0] = lockPath
[0];
6552 for( i
=1; i
<len
; i
++ ){
6553 if( lockPath
[i
] == '/' && (i
- start
> 0) ){
6554 /* only mkdir if leaf dir != "." or "/" or ".." */
6555 if( i
-start
>2 || (i
-start
==1 && buf
[start
] != '.' && buf
[start
] != '/')
6556 || (i
-start
==2 && buf
[start
] != '.' && buf
[start
+1] != '.') ){
6558 if( osMkdir(buf
, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
) ){
6561 OSTRACE(("CREATELOCKPATH FAILED creating %s, "
6562 "'%s' proxy lock path=%s pid=%d\n",
6563 buf
, strerror(err
), lockPath
, osGetpid(0)));
6570 buf
[i
] = lockPath
[i
];
6572 OSTRACE(("CREATELOCKPATH proxy lock path=%s pid=%d\n",lockPath
,osGetpid(0)));
6577 ** Create a new VFS file descriptor (stored in memory obtained from
6578 ** sqlite3_malloc) and open the file named "path" in the file descriptor.
6580 ** The caller is responsible not only for closing the file descriptor
6581 ** but also for freeing the memory associated with the file descriptor.
6583 static int proxyCreateUnixFile(
6584 const char *path
, /* path for the new unixFile */
6585 unixFile
**ppFile
, /* unixFile created and returned by ref */
6586 int islockfile
/* if non zero missing dirs will be created */
6591 int openFlags
= O_RDWR
| O_CREAT
;
6592 sqlite3_vfs dummyVfs
;
6594 UnixUnusedFd
*pUnused
= NULL
;
6596 /* 1. first try to open/create the file
6597 ** 2. if that fails, and this is a lock file (not-conch), try creating
6598 ** the parent directories and then try again.
6599 ** 3. if that fails, try to open the file read-only
6600 ** otherwise return BUSY (if lock file) or CANTOPEN for the conch file
6602 pUnused
= findReusableFd(path
, openFlags
);
6606 pUnused
= sqlite3_malloc64(sizeof(*pUnused
));
6608 return SQLITE_NOMEM_BKPT
;
6612 fd
= robust_open(path
, openFlags
, 0);
6614 if( fd
<0 && errno
==ENOENT
&& islockfile
){
6615 if( proxyCreateLockPath(path
) == SQLITE_OK
){
6616 fd
= robust_open(path
, openFlags
, 0);
6621 openFlags
= O_RDONLY
;
6622 fd
= robust_open(path
, openFlags
, 0);
6633 return SQLITE_IOERR_LOCK
; /* even though it is the conch */
6635 return SQLITE_CANTOPEN_BKPT
;
6639 pNew
= (unixFile
*)sqlite3_malloc64(sizeof(*pNew
));
6641 rc
= SQLITE_NOMEM_BKPT
;
6642 goto end_create_proxy
;
6644 memset(pNew
, 0, sizeof(unixFile
));
6645 pNew
->openFlags
= openFlags
;
6646 memset(&dummyVfs
, 0, sizeof(dummyVfs
));
6647 dummyVfs
.pAppData
= (void*)&autolockIoFinder
;
6648 dummyVfs
.zName
= "dummy";
6650 pUnused
->flags
= openFlags
;
6651 pNew
->pUnused
= pUnused
;
6653 rc
= fillInUnixFile(&dummyVfs
, fd
, (sqlite3_file
*)pNew
, path
, 0);
6654 if( rc
==SQLITE_OK
){
6659 robust_close(pNew
, fd
, __LINE__
);
6661 sqlite3_free(pUnused
);
6666 /* simulate multiple hosts by creating unique hostid file paths */
6667 int sqlite3_hostid_num
= 0;
6670 #define PROXY_HOSTIDLEN 16 /* conch file host id length */
6672 #ifdef HAVE_GETHOSTUUID
6673 /* Not always defined in the headers as it ought to be */
6674 extern int gethostuuid(uuid_t id
, const struct timespec
*wait
);
6677 /* get the host ID via gethostuuid(), pHostID must point to PROXY_HOSTIDLEN
6678 ** bytes of writable memory.
6680 static int proxyGetHostID(unsigned char *pHostID
, int *pError
){
6681 assert(PROXY_HOSTIDLEN
== sizeof(uuid_t
));
6682 memset(pHostID
, 0, PROXY_HOSTIDLEN
);
6683 #ifdef HAVE_GETHOSTUUID
6685 struct timespec timeout
= {1, 0}; /* 1 sec timeout */
6686 if( gethostuuid(pHostID
, &timeout
) ){
6691 return SQLITE_IOERR
;
6695 UNUSED_PARAMETER(pError
);
6698 /* simulate multiple hosts by creating unique hostid file paths */
6699 if( sqlite3_hostid_num
!= 0){
6700 pHostID
[0] = (char)(pHostID
[0] + (char)(sqlite3_hostid_num
& 0xFF));
6707 /* The conch file contains the header, host id and lock file path
6709 #define PROXY_CONCHVERSION 2 /* 1-byte header, 16-byte host id, path */
6710 #define PROXY_HEADERLEN 1 /* conch file header length */
6711 #define PROXY_PATHINDEX (PROXY_HEADERLEN+PROXY_HOSTIDLEN)
6712 #define PROXY_MAXCONCHLEN (PROXY_HEADERLEN+PROXY_HOSTIDLEN+MAXPATHLEN)
6715 ** Takes an open conch file, copies the contents to a new path and then moves
6716 ** it back. The newly created file's file descriptor is assigned to the
6717 ** conch file structure and finally the original conch file descriptor is
6718 ** closed. Returns zero if successful.
6720 static int proxyBreakConchLock(unixFile
*pFile
, uuid_t myHostID
){
6721 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
6722 unixFile
*conchFile
= pCtx
->conchFile
;
6723 char tPath
[MAXPATHLEN
];
6724 char buf
[PROXY_MAXCONCHLEN
];
6725 char *cPath
= pCtx
->conchFilePath
;
6728 char errmsg
[64] = "";
6731 UNUSED_PARAMETER(myHostID
);
6733 /* create a new path by replace the trailing '-conch' with '-break' */
6734 pathLen
= strlcpy(tPath
, cPath
, MAXPATHLEN
);
6735 if( pathLen
>MAXPATHLEN
|| pathLen
<6 ||
6736 (strlcpy(&tPath
[pathLen
-5], "break", 6) != 5) ){
6737 sqlite3_snprintf(sizeof(errmsg
),errmsg
,"path error (len %d)",(int)pathLen
);
6740 /* read the conch content */
6741 readLen
= osPread(conchFile
->h
, buf
, PROXY_MAXCONCHLEN
, 0);
6742 if( readLen
<PROXY_PATHINDEX
){
6743 sqlite3_snprintf(sizeof(errmsg
),errmsg
,"read error (len %d)",(int)readLen
);
6746 /* write it out to the temporary break file */
6747 fd
= robust_open(tPath
, (O_RDWR
|O_CREAT
|O_EXCL
), 0);
6749 sqlite3_snprintf(sizeof(errmsg
), errmsg
, "create failed (%d)", errno
);
6752 if( osPwrite(fd
, buf
, readLen
, 0) != (ssize_t
)readLen
){
6753 sqlite3_snprintf(sizeof(errmsg
), errmsg
, "write failed (%d)", errno
);
6756 if( rename(tPath
, cPath
) ){
6757 sqlite3_snprintf(sizeof(errmsg
), errmsg
, "rename failed (%d)", errno
);
6761 fprintf(stderr
, "broke stale lock on %s\n", cPath
);
6762 robust_close(pFile
, conchFile
->h
, __LINE__
);
6764 conchFile
->openFlags
= O_RDWR
| O_CREAT
;
6770 robust_close(pFile
, fd
, __LINE__
);
6772 fprintf(stderr
, "failed to break stale lock on %s, %s\n", cPath
, errmsg
);
6777 /* Take the requested lock on the conch file and break a stale lock if the
6780 static int proxyConchLock(unixFile
*pFile
, uuid_t myHostID
, int lockType
){
6781 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
6782 unixFile
*conchFile
= pCtx
->conchFile
;
6785 struct timespec conchModTime
;
6787 memset(&conchModTime
, 0, sizeof(conchModTime
));
6789 rc
= conchFile
->pMethod
->xLock((sqlite3_file
*)conchFile
, lockType
);
6791 if( rc
==SQLITE_BUSY
){
6792 /* If the lock failed (busy):
6793 * 1st try: get the mod time of the conch, wait 0.5s and try again.
6794 * 2nd try: fail if the mod time changed or host id is different, wait
6795 * 10 sec and try again
6796 * 3rd try: break the lock unless the mod time has changed.
6799 if( osFstat(conchFile
->h
, &buf
) ){
6800 storeLastErrno(pFile
, errno
);
6801 return SQLITE_IOERR_LOCK
;
6805 conchModTime
= buf
.st_mtimespec
;
6806 usleep(500000); /* wait 0.5 sec and try the lock again*/
6811 if( conchModTime
.tv_sec
!= buf
.st_mtimespec
.tv_sec
||
6812 conchModTime
.tv_nsec
!= buf
.st_mtimespec
.tv_nsec
){
6817 char tBuf
[PROXY_MAXCONCHLEN
];
6818 int len
= osPread(conchFile
->h
, tBuf
, PROXY_MAXCONCHLEN
, 0);
6820 storeLastErrno(pFile
, errno
);
6821 return SQLITE_IOERR_LOCK
;
6823 if( len
>PROXY_PATHINDEX
&& tBuf
[0]==(char)PROXY_CONCHVERSION
){
6824 /* don't break the lock if the host id doesn't match */
6825 if( 0!=memcmp(&tBuf
[PROXY_HEADERLEN
], myHostID
, PROXY_HOSTIDLEN
) ){
6829 /* don't break the lock on short read or a version mismatch */
6832 usleep(10000000); /* wait 10 sec and try the lock again */
6836 assert( nTries
==3 );
6837 if( 0==proxyBreakConchLock(pFile
, myHostID
) ){
6839 if( lockType
==EXCLUSIVE_LOCK
){
6840 rc
= conchFile
->pMethod
->xLock((sqlite3_file
*)conchFile
, SHARED_LOCK
);
6843 rc
= conchFile
->pMethod
->xLock((sqlite3_file
*)conchFile
, lockType
);
6847 } while( rc
==SQLITE_BUSY
&& nTries
<3 );
6852 /* Takes the conch by taking a shared lock and read the contents conch, if
6853 ** lockPath is non-NULL, the host ID and lock file path must match. A NULL
6854 ** lockPath means that the lockPath in the conch file will be used if the
6855 ** host IDs match, or a new lock path will be generated automatically
6856 ** and written to the conch file.
6858 static int proxyTakeConch(unixFile
*pFile
){
6859 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
6861 if( pCtx
->conchHeld
!=0 ){
6864 unixFile
*conchFile
= pCtx
->conchFile
;
6867 char readBuf
[PROXY_MAXCONCHLEN
];
6868 char lockPath
[MAXPATHLEN
];
6869 char *tempLockPath
= NULL
;
6871 int createConch
= 0;
6872 int hostIdMatch
= 0;
6874 int tryOldLockPath
= 0;
6875 int forceNewLockPath
= 0;
6877 OSTRACE(("TAKECONCH %d for %s pid=%d\n", conchFile
->h
,
6878 (pCtx
->lockProxyPath
? pCtx
->lockProxyPath
: ":auto:"),
6881 rc
= proxyGetHostID(myHostID
, &pError
);
6882 if( (rc
&0xff)==SQLITE_IOERR
){
6883 storeLastErrno(pFile
, pError
);
6886 rc
= proxyConchLock(pFile
, myHostID
, SHARED_LOCK
);
6887 if( rc
!=SQLITE_OK
){
6890 /* read the existing conch file */
6891 readLen
= seekAndRead((unixFile
*)conchFile
, 0, readBuf
, PROXY_MAXCONCHLEN
);
6893 /* I/O error: lastErrno set by seekAndRead */
6894 storeLastErrno(pFile
, conchFile
->lastErrno
);
6895 rc
= SQLITE_IOERR_READ
;
6897 }else if( readLen
<=(PROXY_HEADERLEN
+PROXY_HOSTIDLEN
) ||
6898 readBuf
[0]!=(char)PROXY_CONCHVERSION
){
6899 /* a short read or version format mismatch means we need to create a new
6904 /* if the host id matches and the lock path already exists in the conch
6905 ** we'll try to use the path there, if we can't open that path, we'll
6906 ** retry with a new auto-generated path
6908 do { /* in case we need to try again for an :auto: named lock file */
6910 if( !createConch
&& !forceNewLockPath
){
6911 hostIdMatch
= !memcmp(&readBuf
[PROXY_HEADERLEN
], myHostID
,
6913 /* if the conch has data compare the contents */
6914 if( !pCtx
->lockProxyPath
){
6915 /* for auto-named local lock file, just check the host ID and we'll
6916 ** use the local lock file path that's already in there
6919 size_t pathLen
= (readLen
- PROXY_PATHINDEX
);
6921 if( pathLen
>=MAXPATHLEN
){
6922 pathLen
=MAXPATHLEN
-1;
6924 memcpy(lockPath
, &readBuf
[PROXY_PATHINDEX
], pathLen
);
6925 lockPath
[pathLen
] = 0;
6926 tempLockPath
= lockPath
;
6928 /* create a copy of the lock path if the conch is taken */
6931 }else if( hostIdMatch
6932 && !strncmp(pCtx
->lockProxyPath
, &readBuf
[PROXY_PATHINDEX
],
6933 readLen
-PROXY_PATHINDEX
)
6935 /* conch host and lock path match */
6940 /* if the conch isn't writable and doesn't match, we can't take it */
6941 if( (conchFile
->openFlags
&O_RDWR
) == 0 ){
6946 /* either the conch didn't match or we need to create a new one */
6947 if( !pCtx
->lockProxyPath
){
6948 proxyGetLockPath(pCtx
->dbPath
, lockPath
, MAXPATHLEN
);
6949 tempLockPath
= lockPath
;
6950 /* create a copy of the lock path _only_ if the conch is taken */
6953 /* update conch with host and path (this will fail if other process
6954 ** has a shared lock already), if the host id matches, use the big
6957 futimes(conchFile
->h
, NULL
);
6958 if( hostIdMatch
&& !createConch
){
6959 if( conchFile
->pInode
&& conchFile
->pInode
->nShared
>1 ){
6960 /* We are trying for an exclusive lock but another thread in this
6961 ** same process is still holding a shared lock. */
6964 rc
= proxyConchLock(pFile
, myHostID
, EXCLUSIVE_LOCK
);
6967 rc
= proxyConchLock(pFile
, myHostID
, EXCLUSIVE_LOCK
);
6969 if( rc
==SQLITE_OK
){
6970 char writeBuffer
[PROXY_MAXCONCHLEN
];
6973 writeBuffer
[0] = (char)PROXY_CONCHVERSION
;
6974 memcpy(&writeBuffer
[PROXY_HEADERLEN
], myHostID
, PROXY_HOSTIDLEN
);
6975 if( pCtx
->lockProxyPath
!=NULL
){
6976 strlcpy(&writeBuffer
[PROXY_PATHINDEX
], pCtx
->lockProxyPath
,
6979 strlcpy(&writeBuffer
[PROXY_PATHINDEX
], tempLockPath
, MAXPATHLEN
);
6981 writeSize
= PROXY_PATHINDEX
+ strlen(&writeBuffer
[PROXY_PATHINDEX
]);
6982 robust_ftruncate(conchFile
->h
, writeSize
);
6983 rc
= unixWrite((sqlite3_file
*)conchFile
, writeBuffer
, writeSize
, 0);
6984 full_fsync(conchFile
->h
,0,0);
6985 /* If we created a new conch file (not just updated the contents of a
6986 ** valid conch file), try to match the permissions of the database
6988 if( rc
==SQLITE_OK
&& createConch
){
6990 int err
= osFstat(pFile
->h
, &buf
);
6992 mode_t cmode
= buf
.st_mode
&(S_IRUSR
|S_IWUSR
| S_IRGRP
|S_IWGRP
|
6994 /* try to match the database file R/W permissions, ignore failure */
6995 #ifndef SQLITE_PROXY_DEBUG
6996 osFchmod(conchFile
->h
, cmode
);
6999 rc
= osFchmod(conchFile
->h
, cmode
);
7000 }while( rc
==(-1) && errno
==EINTR
);
7003 fprintf(stderr
, "fchmod %o FAILED with %d %s\n",
7004 cmode
, code
, strerror(code
));
7006 fprintf(stderr
, "fchmod %o SUCCEDED\n",cmode
);
7010 fprintf(stderr
, "STAT FAILED[%d] with %d %s\n",
7011 err
, code
, strerror(code
));
7016 conchFile
->pMethod
->xUnlock((sqlite3_file
*)conchFile
, SHARED_LOCK
);
7019 OSTRACE(("TRANSPROXY: CLOSE %d\n", pFile
->h
));
7020 if( rc
==SQLITE_OK
&& pFile
->openFlags
){
7023 robust_close(pFile
, pFile
->h
, __LINE__
);
7026 fd
= robust_open(pCtx
->dbPath
, pFile
->openFlags
, 0);
7027 OSTRACE(("TRANSPROXY: OPEN %d\n", fd
));
7031 rc
=SQLITE_CANTOPEN_BKPT
; /* SQLITE_BUSY? proxyTakeConch called
7035 if( rc
==SQLITE_OK
&& !pCtx
->lockProxy
){
7036 char *path
= tempLockPath
? tempLockPath
: pCtx
->lockProxyPath
;
7037 rc
= proxyCreateUnixFile(path
, &pCtx
->lockProxy
, 1);
7038 if( rc
!=SQLITE_OK
&& rc
!=SQLITE_NOMEM
&& tryOldLockPath
){
7039 /* we couldn't create the proxy lock file with the old lock file path
7040 ** so try again via auto-naming
7042 forceNewLockPath
= 1;
7044 continue; /* go back to the do {} while start point, try again */
7047 if( rc
==SQLITE_OK
){
7048 /* Need to make a copy of path if we extracted the value
7049 ** from the conch file or the path was allocated on the stack
7052 pCtx
->lockProxyPath
= sqlite3DbStrDup(0, tempLockPath
);
7053 if( !pCtx
->lockProxyPath
){
7054 rc
= SQLITE_NOMEM_BKPT
;
7058 if( rc
==SQLITE_OK
){
7059 pCtx
->conchHeld
= 1;
7061 if( pCtx
->lockProxy
->pMethod
== &afpIoMethods
){
7062 afpLockingContext
*afpCtx
;
7063 afpCtx
= (afpLockingContext
*)pCtx
->lockProxy
->lockingContext
;
7064 afpCtx
->dbPath
= pCtx
->lockProxyPath
;
7067 conchFile
->pMethod
->xUnlock((sqlite3_file
*)conchFile
, NO_LOCK
);
7069 OSTRACE(("TAKECONCH %d %s\n", conchFile
->h
,
7070 rc
==SQLITE_OK
?"ok":"failed"));
7072 } while (1); /* in case we need to retry the :auto: lock file -
7073 ** we should never get here except via the 'continue' call. */
7078 ** If pFile holds a lock on a conch file, then release that lock.
7080 static int proxyReleaseConch(unixFile
*pFile
){
7081 int rc
= SQLITE_OK
; /* Subroutine return code */
7082 proxyLockingContext
*pCtx
; /* The locking context for the proxy lock */
7083 unixFile
*conchFile
; /* Name of the conch file */
7085 pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7086 conchFile
= pCtx
->conchFile
;
7087 OSTRACE(("RELEASECONCH %d for %s pid=%d\n", conchFile
->h
,
7088 (pCtx
->lockProxyPath
? pCtx
->lockProxyPath
: ":auto:"),
7090 if( pCtx
->conchHeld
>0 ){
7091 rc
= conchFile
->pMethod
->xUnlock((sqlite3_file
*)conchFile
, NO_LOCK
);
7093 pCtx
->conchHeld
= 0;
7094 OSTRACE(("RELEASECONCH %d %s\n", conchFile
->h
,
7095 (rc
==SQLITE_OK
? "ok" : "failed")));
7100 ** Given the name of a database file, compute the name of its conch file.
7101 ** Store the conch filename in memory obtained from sqlite3_malloc64().
7102 ** Make *pConchPath point to the new name. Return SQLITE_OK on success
7103 ** or SQLITE_NOMEM if unable to obtain memory.
7105 ** The caller is responsible for ensuring that the allocated memory
7106 ** space is eventually freed.
7108 ** *pConchPath is set to NULL if a memory allocation error occurs.
7110 static int proxyCreateConchPathname(char *dbPath
, char **pConchPath
){
7111 int i
; /* Loop counter */
7112 int len
= (int)strlen(dbPath
); /* Length of database filename - dbPath */
7113 char *conchPath
; /* buffer in which to construct conch name */
7115 /* Allocate space for the conch filename and initialize the name to
7116 ** the name of the original database file. */
7117 *pConchPath
= conchPath
= (char *)sqlite3_malloc64(len
+ 8);
7119 return SQLITE_NOMEM_BKPT
;
7121 memcpy(conchPath
, dbPath
, len
+1);
7123 /* now insert a "." before the last / character */
7124 for( i
=(len
-1); i
>=0; i
-- ){
7125 if( conchPath
[i
]=='/' ){
7132 conchPath
[i
+1]=dbPath
[i
];
7136 /* append the "-conch" suffix to the file */
7137 memcpy(&conchPath
[i
+1], "-conch", 7);
7138 assert( (int)strlen(conchPath
) == len
+7 );
7144 /* Takes a fully configured proxy locking-style unix file and switches
7145 ** the local lock file path
7147 static int switchLockProxyPath(unixFile
*pFile
, const char *path
) {
7148 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7149 char *oldPath
= pCtx
->lockProxyPath
;
7152 if( pFile
->eFileLock
!=NO_LOCK
){
7156 /* nothing to do if the path is NULL, :auto: or matches the existing path */
7157 if( !path
|| path
[0]=='\0' || !strcmp(path
, ":auto:") ||
7158 (oldPath
&& !strncmp(oldPath
, path
, MAXPATHLEN
)) ){
7161 unixFile
*lockProxy
= pCtx
->lockProxy
;
7162 pCtx
->lockProxy
=NULL
;
7163 pCtx
->conchHeld
= 0;
7164 if( lockProxy
!=NULL
){
7165 rc
=lockProxy
->pMethod
->xClose((sqlite3_file
*)lockProxy
);
7167 sqlite3_free(lockProxy
);
7169 sqlite3_free(oldPath
);
7170 pCtx
->lockProxyPath
= sqlite3DbStrDup(0, path
);
7177 ** pFile is a file that has been opened by a prior xOpen call. dbPath
7178 ** is a string buffer at least MAXPATHLEN+1 characters in size.
7180 ** This routine find the filename associated with pFile and writes it
7183 static int proxyGetDbPathForUnixFile(unixFile
*pFile
, char *dbPath
){
7184 #if defined(__APPLE__)
7185 if( pFile
->pMethod
== &afpIoMethods
){
7186 /* afp style keeps a reference to the db path in the filePath field
7188 assert( (int)strlen((char*)pFile
->lockingContext
)<=MAXPATHLEN
);
7189 strlcpy(dbPath
, ((afpLockingContext
*)pFile
->lockingContext
)->dbPath
,
7193 if( pFile
->pMethod
== &dotlockIoMethods
){
7194 /* dot lock style uses the locking context to store the dot lock
7196 int len
= strlen((char *)pFile
->lockingContext
) - strlen(DOTLOCK_SUFFIX
);
7197 memcpy(dbPath
, (char *)pFile
->lockingContext
, len
+ 1);
7199 /* all other styles use the locking context to store the db file path */
7200 assert( strlen((char*)pFile
->lockingContext
)<=MAXPATHLEN
);
7201 strlcpy(dbPath
, (char *)pFile
->lockingContext
, MAXPATHLEN
);
7207 ** Takes an already filled in unix file and alters it so all file locking
7208 ** will be performed on the local proxy lock file. The following fields
7209 ** are preserved in the locking context so that they can be restored and
7210 ** the unix structure properly cleaned up at close time:
7214 static int proxyTransformUnixFile(unixFile
*pFile
, const char *path
) {
7215 proxyLockingContext
*pCtx
;
7216 char dbPath
[MAXPATHLEN
+1]; /* Name of the database file */
7217 char *lockPath
=NULL
;
7220 if( pFile
->eFileLock
!=NO_LOCK
){
7223 proxyGetDbPathForUnixFile(pFile
, dbPath
);
7224 if( !path
|| path
[0]=='\0' || !strcmp(path
, ":auto:") ){
7227 lockPath
=(char *)path
;
7230 OSTRACE(("TRANSPROXY %d for %s pid=%d\n", pFile
->h
,
7231 (lockPath
? lockPath
: ":auto:"), osGetpid(0)));
7233 pCtx
= sqlite3_malloc64( sizeof(*pCtx
) );
7235 return SQLITE_NOMEM_BKPT
;
7237 memset(pCtx
, 0, sizeof(*pCtx
));
7239 rc
= proxyCreateConchPathname(dbPath
, &pCtx
->conchFilePath
);
7240 if( rc
==SQLITE_OK
){
7241 rc
= proxyCreateUnixFile(pCtx
->conchFilePath
, &pCtx
->conchFile
, 0);
7242 if( rc
==SQLITE_CANTOPEN
&& ((pFile
->openFlags
&O_RDWR
) == 0) ){
7243 /* if (a) the open flags are not O_RDWR, (b) the conch isn't there, and
7244 ** (c) the file system is read-only, then enable no-locking access.
7245 ** Ugh, since O_RDONLY==0x0000 we test for !O_RDWR since unixOpen asserts
7246 ** that openFlags will have only one of O_RDONLY or O_RDWR.
7248 struct statfs fsInfo
;
7249 struct stat conchInfo
;
7252 if( osStat(pCtx
->conchFilePath
, &conchInfo
) == -1 ) {
7254 if( (err
==ENOENT
) && (statfs(dbPath
, &fsInfo
) != -1) ){
7255 goLockless
= (fsInfo
.f_flags
&MNT_RDONLY
) == MNT_RDONLY
;
7259 pCtx
->conchHeld
= -1; /* read only FS/ lockless */
7264 if( rc
==SQLITE_OK
&& lockPath
){
7265 pCtx
->lockProxyPath
= sqlite3DbStrDup(0, lockPath
);
7268 if( rc
==SQLITE_OK
){
7269 pCtx
->dbPath
= sqlite3DbStrDup(0, dbPath
);
7270 if( pCtx
->dbPath
==NULL
){
7271 rc
= SQLITE_NOMEM_BKPT
;
7274 if( rc
==SQLITE_OK
){
7275 /* all memory is allocated, proxys are created and assigned,
7276 ** switch the locking context and pMethod then return.
7278 pCtx
->oldLockingContext
= pFile
->lockingContext
;
7279 pFile
->lockingContext
= pCtx
;
7280 pCtx
->pOldMethod
= pFile
->pMethod
;
7281 pFile
->pMethod
= &proxyIoMethods
;
7283 if( pCtx
->conchFile
){
7284 pCtx
->conchFile
->pMethod
->xClose((sqlite3_file
*)pCtx
->conchFile
);
7285 sqlite3_free(pCtx
->conchFile
);
7287 sqlite3DbFree(0, pCtx
->lockProxyPath
);
7288 sqlite3_free(pCtx
->conchFilePath
);
7291 OSTRACE(("TRANSPROXY %d %s\n", pFile
->h
,
7292 (rc
==SQLITE_OK
? "ok" : "failed")));
7298 ** This routine handles sqlite3_file_control() calls that are specific
7299 ** to proxy locking.
7301 static int proxyFileControl(sqlite3_file
*id
, int op
, void *pArg
){
7303 case SQLITE_FCNTL_GET_LOCKPROXYFILE
: {
7304 unixFile
*pFile
= (unixFile
*)id
;
7305 if( pFile
->pMethod
== &proxyIoMethods
){
7306 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7307 proxyTakeConch(pFile
);
7308 if( pCtx
->lockProxyPath
){
7309 *(const char **)pArg
= pCtx
->lockProxyPath
;
7311 *(const char **)pArg
= ":auto: (not held)";
7314 *(const char **)pArg
= NULL
;
7318 case SQLITE_FCNTL_SET_LOCKPROXYFILE
: {
7319 unixFile
*pFile
= (unixFile
*)id
;
7321 int isProxyStyle
= (pFile
->pMethod
== &proxyIoMethods
);
7322 if( pArg
==NULL
|| (const char *)pArg
==0 ){
7324 /* turn off proxy locking - not supported. If support is added for
7325 ** switching proxy locking mode off then it will need to fail if
7326 ** the journal mode is WAL mode.
7328 rc
= SQLITE_ERROR
/*SQLITE_PROTOCOL? SQLITE_MISUSE?*/;
7330 /* turn off proxy locking - already off - NOOP */
7334 const char *proxyPath
= (const char *)pArg
;
7336 proxyLockingContext
*pCtx
=
7337 (proxyLockingContext
*)pFile
->lockingContext
;
7338 if( !strcmp(pArg
, ":auto:")
7339 || (pCtx
->lockProxyPath
&&
7340 !strncmp(pCtx
->lockProxyPath
, proxyPath
, MAXPATHLEN
))
7344 rc
= switchLockProxyPath(pFile
, proxyPath
);
7347 /* turn on proxy file locking */
7348 rc
= proxyTransformUnixFile(pFile
, proxyPath
);
7354 assert( 0 ); /* The call assures that only valid opcodes are sent */
7358 return SQLITE_ERROR
;
7362 ** Within this division (the proxying locking implementation) the procedures
7363 ** above this point are all utilities. The lock-related methods of the
7364 ** proxy-locking sqlite3_io_method object follow.
7369 ** This routine checks if there is a RESERVED lock held on the specified
7370 ** file by this or any other process. If such a lock is held, set *pResOut
7371 ** to a non-zero value otherwise *pResOut is set to zero. The return value
7372 ** is set to SQLITE_OK unless an I/O error occurs during lock checking.
7374 static int proxyCheckReservedLock(sqlite3_file
*id
, int *pResOut
) {
7375 unixFile
*pFile
= (unixFile
*)id
;
7376 int rc
= proxyTakeConch(pFile
);
7377 if( rc
==SQLITE_OK
){
7378 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7379 if( pCtx
->conchHeld
>0 ){
7380 unixFile
*proxy
= pCtx
->lockProxy
;
7381 return proxy
->pMethod
->xCheckReservedLock((sqlite3_file
*)proxy
, pResOut
);
7382 }else{ /* conchHeld < 0 is lockless */
7390 ** Lock the file with the lock specified by parameter eFileLock - one
7391 ** of the following:
7394 ** (2) RESERVED_LOCK
7396 ** (4) EXCLUSIVE_LOCK
7398 ** Sometimes when requesting one lock state, additional lock states
7399 ** are inserted in between. The locking might fail on one of the later
7400 ** transitions leaving the lock state different from what it started but
7401 ** still short of its goal. The following chart shows the allowed
7402 ** transitions and the inserted intermediate states:
7404 ** UNLOCKED -> SHARED
7405 ** SHARED -> RESERVED
7406 ** SHARED -> (PENDING) -> EXCLUSIVE
7407 ** RESERVED -> (PENDING) -> EXCLUSIVE
7408 ** PENDING -> EXCLUSIVE
7410 ** This routine will only increase a lock. Use the sqlite3OsUnlock()
7411 ** routine to lower a locking level.
7413 static int proxyLock(sqlite3_file
*id
, int eFileLock
) {
7414 unixFile
*pFile
= (unixFile
*)id
;
7415 int rc
= proxyTakeConch(pFile
);
7416 if( rc
==SQLITE_OK
){
7417 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7418 if( pCtx
->conchHeld
>0 ){
7419 unixFile
*proxy
= pCtx
->lockProxy
;
7420 rc
= proxy
->pMethod
->xLock((sqlite3_file
*)proxy
, eFileLock
);
7421 pFile
->eFileLock
= proxy
->eFileLock
;
7423 /* conchHeld < 0 is lockless */
7431 ** Lower the locking level on file descriptor pFile to eFileLock. eFileLock
7432 ** must be either NO_LOCK or SHARED_LOCK.
7434 ** If the locking level of the file descriptor is already at or below
7435 ** the requested locking level, this routine is a no-op.
7437 static int proxyUnlock(sqlite3_file
*id
, int eFileLock
) {
7438 unixFile
*pFile
= (unixFile
*)id
;
7439 int rc
= proxyTakeConch(pFile
);
7440 if( rc
==SQLITE_OK
){
7441 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7442 if( pCtx
->conchHeld
>0 ){
7443 unixFile
*proxy
= pCtx
->lockProxy
;
7444 rc
= proxy
->pMethod
->xUnlock((sqlite3_file
*)proxy
, eFileLock
);
7445 pFile
->eFileLock
= proxy
->eFileLock
;
7447 /* conchHeld < 0 is lockless */
7454 ** Close a file that uses proxy locks.
7456 static int proxyClose(sqlite3_file
*id
) {
7458 unixFile
*pFile
= (unixFile
*)id
;
7459 proxyLockingContext
*pCtx
= (proxyLockingContext
*)pFile
->lockingContext
;
7460 unixFile
*lockProxy
= pCtx
->lockProxy
;
7461 unixFile
*conchFile
= pCtx
->conchFile
;
7465 rc
= lockProxy
->pMethod
->xUnlock((sqlite3_file
*)lockProxy
, NO_LOCK
);
7467 rc
= lockProxy
->pMethod
->xClose((sqlite3_file
*)lockProxy
);
7469 sqlite3_free(lockProxy
);
7470 pCtx
->lockProxy
= 0;
7473 if( pCtx
->conchHeld
){
7474 rc
= proxyReleaseConch(pFile
);
7477 rc
= conchFile
->pMethod
->xClose((sqlite3_file
*)conchFile
);
7479 sqlite3_free(conchFile
);
7481 sqlite3DbFree(0, pCtx
->lockProxyPath
);
7482 sqlite3_free(pCtx
->conchFilePath
);
7483 sqlite3DbFree(0, pCtx
->dbPath
);
7484 /* restore the original locking context and pMethod then close it */
7485 pFile
->lockingContext
= pCtx
->oldLockingContext
;
7486 pFile
->pMethod
= pCtx
->pOldMethod
;
7488 return pFile
->pMethod
->xClose(id
);
7495 #endif /* defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE */
7497 ** The proxy locking style is intended for use with AFP filesystems.
7498 ** And since AFP is only supported on MacOSX, the proxy locking is also
7499 ** restricted to MacOSX.
7502 ******************* End of the proxy lock implementation **********************
7503 ******************************************************************************/
7506 ** Initialize the operating system interface.
7508 ** This routine registers all VFS implementations for unix-like operating
7509 ** systems. This routine, and the sqlite3_os_end() routine that follows,
7510 ** should be the only routines in this file that are visible from other
7513 ** This routine is called once during SQLite initialization and by a
7514 ** single thread. The memory allocation and mutex subsystems have not
7515 ** necessarily been initialized when this routine is called, and so they
7516 ** should not be used.
7518 int sqlite3_os_init(void){
7520 ** The following macro defines an initializer for an sqlite3_vfs object.
7521 ** The name of the VFS is NAME. The pAppData is a pointer to a pointer
7522 ** to the "finder" function. (pAppData is a pointer to a pointer because
7523 ** silly C90 rules prohibit a void* from being cast to a function pointer
7524 ** and so we have to go through the intermediate pointer to avoid problems
7525 ** when compiling with -pedantic-errors on GCC.)
7527 ** The FINDER parameter to this macro is the name of the pointer to the
7528 ** finder-function. The finder-function returns a pointer to the
7529 ** sqlite_io_methods object that implements the desired locking
7530 ** behaviors. See the division above that contains the IOMETHODS
7531 ** macro for addition information on finder-functions.
7533 ** Most finders simply return a pointer to a fixed sqlite3_io_methods
7534 ** object. But the "autolockIoFinder" available on MacOSX does a little
7535 ** more than that; it looks at the filesystem type that hosts the
7536 ** database file and tries to choose an locking method appropriate for
7537 ** that filesystem time.
7539 #define UNIXVFS(VFSNAME, FINDER) { \
7541 sizeof(unixFile), /* szOsFile */ \
7542 MAX_PATHNAME, /* mxPathname */ \
7544 VFSNAME, /* zName */ \
7545 (void*)&FINDER, /* pAppData */ \
7546 unixOpen, /* xOpen */ \
7547 unixDelete, /* xDelete */ \
7548 unixAccess, /* xAccess */ \
7549 unixFullPathname, /* xFullPathname */ \
7550 unixDlOpen, /* xDlOpen */ \
7551 unixDlError, /* xDlError */ \
7552 unixDlSym, /* xDlSym */ \
7553 unixDlClose, /* xDlClose */ \
7554 unixRandomness, /* xRandomness */ \
7555 unixSleep, /* xSleep */ \
7556 unixCurrentTime, /* xCurrentTime */ \
7557 unixGetLastError, /* xGetLastError */ \
7558 unixCurrentTimeInt64, /* xCurrentTimeInt64 */ \
7559 unixSetSystemCall, /* xSetSystemCall */ \
7560 unixGetSystemCall, /* xGetSystemCall */ \
7561 unixNextSystemCall, /* xNextSystemCall */ \
7565 ** All default VFSes for unix are contained in the following array.
7567 ** Note that the sqlite3_vfs.pNext field of the VFS object is modified
7568 ** by the SQLite core when the VFS is registered. So the following
7569 ** array cannot be const.
7571 static sqlite3_vfs aVfs
[] = {
7572 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
7573 UNIXVFS("unix", autolockIoFinder
),
7575 UNIXVFS("unix", vxworksIoFinder
),
7577 UNIXVFS("unix", posixIoFinder
),
7579 UNIXVFS("unix-none", nolockIoFinder
),
7580 UNIXVFS("unix-dotfile", dotlockIoFinder
),
7581 UNIXVFS("unix-excl", posixIoFinder
),
7583 UNIXVFS("unix-namedsem", semIoFinder
),
7585 #if SQLITE_ENABLE_LOCKING_STYLE || OS_VXWORKS
7586 UNIXVFS("unix-posix", posixIoFinder
),
7588 #if SQLITE_ENABLE_LOCKING_STYLE
7589 UNIXVFS("unix-flock", flockIoFinder
),
7591 #if SQLITE_ENABLE_LOCKING_STYLE && defined(__APPLE__)
7592 UNIXVFS("unix-afp", afpIoFinder
),
7593 UNIXVFS("unix-nfs", nfsIoFinder
),
7594 UNIXVFS("unix-proxy", proxyIoFinder
),
7597 unsigned int i
; /* Loop counter */
7599 /* Double-check that the aSyscall[] array has been constructed
7600 ** correctly. See ticket [bb3a86e890c8e96ab] */
7601 assert( ArraySize(aSyscall
)==28 );
7603 /* Register all VFSes defined in the aVfs[] array */
7604 for(i
=0; i
<(sizeof(aVfs
)/sizeof(sqlite3_vfs
)); i
++){
7605 sqlite3_vfs_register(&aVfs
[i
], i
==0);
7611 ** Shutdown the operating system interface.
7613 ** Some operating systems might need to do some cleanup in this routine,
7614 ** to release dynamically allocated objects. But not on unix.
7615 ** This routine is a no-op for unix.
7617 int sqlite3_os_end(void){
7621 #endif /* SQLITE_OS_UNIX */