aboutsummaryrefslogtreecommitdiffstatshomepage
path: root/libraries/sqlite/win32/os_unix.c
diff options
context:
space:
mode:
Diffstat (limited to 'libraries/sqlite/win32/os_unix.c')
-rwxr-xr-xlibraries/sqlite/win32/os_unix.c2749
1 files changed, 2749 insertions, 0 deletions
diff --git a/libraries/sqlite/win32/os_unix.c b/libraries/sqlite/win32/os_unix.c
new file mode 100755
index 0000000..e95435e
--- /dev/null
+++ b/libraries/sqlite/win32/os_unix.c
@@ -0,0 +1,2749 @@
1/*
2** 2004 May 22
3**
4** The author disclaims copyright to this source code. In place of
5** a legal notice, here is a blessing:
6**
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.
10**
11******************************************************************************
12**
13** This file contains code that is specific to Unix systems.
14*/
15#include "sqliteInt.h"
16#if OS_UNIX /* This file is used on unix only */
17
18/* #define SQLITE_ENABLE_LOCKING_STYLE 0 */
19
20/*
21** These #defines should enable >2GB file support on Posix if the
22** underlying operating system supports it. If the OS lacks
23** large file support, these should be no-ops.
24**
25** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch
26** on the compiler command line. This is necessary if you are compiling
27** on a recent machine (ex: RedHat 7.2) but you want your code to work
28** on an older machine (ex: RedHat 6.0). If you compile on RedHat 7.2
29** without this option, LFS is enable. But LFS does not exist in the kernel
30** in RedHat 6.0, so the code won't work. Hence, for maximum binary
31** portability you should omit LFS.
32*/
33#ifndef SQLITE_DISABLE_LFS
34# define _LARGE_FILE 1
35# ifndef _FILE_OFFSET_BITS
36# define _FILE_OFFSET_BITS 64
37# endif
38# define _LARGEFILE_SOURCE 1
39#endif
40
41/*
42** standard include files.
43*/
44#include <sys/types.h>
45#include <sys/stat.h>
46#include <fcntl.h>
47#include <unistd.h>
48#include <time.h>
49#include <sys/time.h>
50#include <errno.h>
51#ifdef SQLITE_ENABLE_LOCKING_STYLE
52#include <sys/ioctl.h>
53#include <sys/param.h>
54#include <sys/mount.h>
55#endif /* SQLITE_ENABLE_LOCKING_STYLE */
56
57/*
58** If we are to be thread-safe, include the pthreads header and define
59** the SQLITE_UNIX_THREADS macro.
60*/
61#if SQLITE_THREADSAFE
62# include <pthread.h>
63# define SQLITE_UNIX_THREADS 1
64#endif
65
66/*
67** Default permissions when creating a new file
68*/
69#ifndef SQLITE_DEFAULT_FILE_PERMISSIONS
70# define SQLITE_DEFAULT_FILE_PERMISSIONS 0644
71#endif
72
73/*
74** Maximum supported path-length.
75*/
76#define MAX_PATHNAME 512
77
78
79/*
80** The unixFile structure is subclass of sqlite3_file specific for the unix
81** protability layer.
82*/
83typedef struct unixFile unixFile;
84struct unixFile {
85 sqlite3_io_methods const *pMethod; /* Always the first entry */
86#ifdef SQLITE_TEST
87 /* In test mode, increase the size of this structure a bit so that
88 ** it is larger than the struct CrashFile defined in test6.c.
89 */
90 char aPadding[32];
91#endif
92 struct openCnt *pOpen; /* Info about all open fd's on this inode */
93 struct lockInfo *pLock; /* Info about locks on this inode */
94#ifdef SQLITE_ENABLE_LOCKING_STYLE
95 void *lockingContext; /* Locking style specific state */
96#endif /* SQLITE_ENABLE_LOCKING_STYLE */
97 int h; /* The file descriptor */
98 unsigned char locktype; /* The type of lock held on this fd */
99 int dirfd; /* File descriptor for the directory */
100#if SQLITE_THREADSAFE
101 pthread_t tid; /* The thread that "owns" this unixFile */
102#endif
103};
104
105/*
106** Include code that is common to all os_*.c files
107*/
108#include "os_common.h"
109
110/*
111** Define various macros that are missing from some systems.
112*/
113#ifndef O_LARGEFILE
114# define O_LARGEFILE 0
115#endif
116#ifdef SQLITE_DISABLE_LFS
117# undef O_LARGEFILE
118# define O_LARGEFILE 0
119#endif
120#ifndef O_NOFOLLOW
121# define O_NOFOLLOW 0
122#endif
123#ifndef O_BINARY
124# define O_BINARY 0
125#endif
126
127/*
128** The DJGPP compiler environment looks mostly like Unix, but it
129** lacks the fcntl() system call. So redefine fcntl() to be something
130** that always succeeds. This means that locking does not occur under
131** DJGPP. But it's DOS - what did you expect?
132*/
133#ifdef __DJGPP__
134# define fcntl(A,B,C) 0
135#endif
136
137/*
138** The threadid macro resolves to the thread-id or to 0. Used for
139** testing and debugging only.
140*/
141#if SQLITE_THREADSAFE
142#define threadid pthread_self()
143#else
144#define threadid 0
145#endif
146
147/*
148** Set or check the unixFile.tid field. This field is set when an unixFile
149** is first opened. All subsequent uses of the unixFile verify that the
150** same thread is operating on the unixFile. Some operating systems do
151** not allow locks to be overridden by other threads and that restriction
152** means that sqlite3* database handles cannot be moved from one thread
153** to another. This logic makes sure a user does not try to do that
154** by mistake.
155**
156** Version 3.3.1 (2006-01-15): unixFile can be moved from one thread to
157** another as long as we are running on a system that supports threads
158** overriding each others locks (which now the most common behavior)
159** or if no locks are held. But the unixFile.pLock field needs to be
160** recomputed because its key includes the thread-id. See the
161** transferOwnership() function below for additional information
162*/
163#if SQLITE_THREADSAFE
164# define SET_THREADID(X) (X)->tid = pthread_self()
165# define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \
166 !pthread_equal((X)->tid, pthread_self()))
167#else
168# define SET_THREADID(X)
169# define CHECK_THREADID(X) 0
170#endif
171
172/*
173** Here is the dirt on POSIX advisory locks: ANSI STD 1003.1 (1996)
174** section 6.5.2.2 lines 483 through 490 specify that when a process
175** sets or clears a lock, that operation overrides any prior locks set
176** by the same process. It does not explicitly say so, but this implies
177** that it overrides locks set by the same process using a different
178** file descriptor. Consider this test case:
179**
180** int fd1 = open("./file1", O_RDWR|O_CREAT, 0644);
181** int fd2 = open("./file2", O_RDWR|O_CREAT, 0644);
182**
183** Suppose ./file1 and ./file2 are really the same file (because
184** one is a hard or symbolic link to the other) then if you set
185** an exclusive lock on fd1, then try to get an exclusive lock
186** on fd2, it works. I would have expected the second lock to
187** fail since there was already a lock on the file due to fd1.
188** But not so. Since both locks came from the same process, the
189** second overrides the first, even though they were on different
190** file descriptors opened on different file names.
191**
192** Bummer. If you ask me, this is broken. Badly broken. It means
193** that we cannot use POSIX locks to synchronize file access among
194** competing threads of the same process. POSIX locks will work fine
195** to synchronize access for threads in separate processes, but not
196** threads within the same process.
197**
198** To work around the problem, SQLite has to manage file locks internally
199** on its own. Whenever a new database is opened, we have to find the
200** specific inode of the database file (the inode is determined by the
201** st_dev and st_ino fields of the stat structure that fstat() fills in)
202** and check for locks already existing on that inode. When locks are
203** created or removed, we have to look at our own internal record of the
204** locks to see if another thread has previously set a lock on that same
205** inode.
206**
207** The sqlite3_file structure for POSIX is no longer just an integer file
208** descriptor. It is now a structure that holds the integer file
209** descriptor and a pointer to a structure that describes the internal
210** locks on the corresponding inode. There is one locking structure
211** per inode, so if the same inode is opened twice, both unixFile structures
212** point to the same locking structure. The locking structure keeps
213** a reference count (so we will know when to delete it) and a "cnt"
214** field that tells us its internal lock status. cnt==0 means the
215** file is unlocked. cnt==-1 means the file has an exclusive lock.
216** cnt>0 means there are cnt shared locks on the file.
217**
218** Any attempt to lock or unlock a file first checks the locking
219** structure. The fcntl() system call is only invoked to set a
220** POSIX lock if the internal lock structure transitions between
221** a locked and an unlocked state.
222**
223** 2004-Jan-11:
224** More recent discoveries about POSIX advisory locks. (The more
225** I discover, the more I realize the a POSIX advisory locks are
226** an abomination.)
227**
228** If you close a file descriptor that points to a file that has locks,
229** all locks on that file that are owned by the current process are
230** released. To work around this problem, each unixFile structure contains
231** a pointer to an openCnt structure. There is one openCnt structure
232** per open inode, which means that multiple unixFile can point to a single
233** openCnt. When an attempt is made to close an unixFile, if there are
234** other unixFile open on the same inode that are holding locks, the call
235** to close() the file descriptor is deferred until all of the locks clear.
236** The openCnt structure keeps a list of file descriptors that need to
237** be closed and that list is walked (and cleared) when the last lock
238** clears.
239**
240** First, under Linux threads, because each thread has a separate
241** process ID, lock operations in one thread do not override locks
242** to the same file in other threads. Linux threads behave like
243** separate processes in this respect. But, if you close a file
244** descriptor in linux threads, all locks are cleared, even locks
245** on other threads and even though the other threads have different
246** process IDs. Linux threads is inconsistent in this respect.
247** (I'm beginning to think that linux threads is an abomination too.)
248** The consequence of this all is that the hash table for the lockInfo
249** structure has to include the process id as part of its key because
250** locks in different threads are treated as distinct. But the
251** openCnt structure should not include the process id in its
252** key because close() clears lock on all threads, not just the current
253** thread. Were it not for this goofiness in linux threads, we could
254** combine the lockInfo and openCnt structures into a single structure.
255**
256** 2004-Jun-28:
257** On some versions of linux, threads can override each others locks.
258** On others not. Sometimes you can change the behavior on the same
259** system by setting the LD_ASSUME_KERNEL environment variable. The
260** POSIX standard is silent as to which behavior is correct, as far
261** as I can tell, so other versions of unix might show the same
262** inconsistency. There is no little doubt in my mind that posix
263** advisory locks and linux threads are profoundly broken.
264**
265** To work around the inconsistencies, we have to test at runtime
266** whether or not threads can override each others locks. This test
267** is run once, the first time any lock is attempted. A static
268** variable is set to record the results of this test for future
269** use.
270*/
271
272/*
273** An instance of the following structure serves as the key used
274** to locate a particular lockInfo structure given its inode.
275**
276** If threads cannot override each others locks, then we set the
277** lockKey.tid field to the thread ID. If threads can override
278** each others locks then tid is always set to zero. tid is omitted
279** if we compile without threading support.
280*/
281struct lockKey {
282 dev_t dev; /* Device number */
283 ino_t ino; /* Inode number */
284#if SQLITE_THREADSAFE
285 pthread_t tid; /* Thread ID or zero if threads can override each other */
286#endif
287};
288
289/*
290** An instance of the following structure is allocated for each open
291** inode on each thread with a different process ID. (Threads have
292** different process IDs on linux, but not on most other unixes.)
293**
294** A single inode can have multiple file descriptors, so each unixFile
295** structure contains a pointer to an instance of this object and this
296** object keeps a count of the number of unixFile pointing to it.
297*/
298struct lockInfo {
299 struct lockKey key; /* The lookup key */
300 int cnt; /* Number of SHARED locks held */
301 int locktype; /* One of SHARED_LOCK, RESERVED_LOCK etc. */
302 int nRef; /* Number of pointers to this structure */
303};
304
305/*
306** An instance of the following structure serves as the key used
307** to locate a particular openCnt structure given its inode. This
308** is the same as the lockKey except that the thread ID is omitted.
309*/
310struct openKey {
311 dev_t dev; /* Device number */
312 ino_t ino; /* Inode number */
313};
314
315/*
316** An instance of the following structure is allocated for each open
317** inode. This structure keeps track of the number of locks on that
318** inode. If a close is attempted against an inode that is holding
319** locks, the close is deferred until all locks clear by adding the
320** file descriptor to be closed to the pending list.
321*/
322struct openCnt {
323 struct openKey key; /* The lookup key */
324 int nRef; /* Number of pointers to this structure */
325 int nLock; /* Number of outstanding locks */
326 int nPending; /* Number of pending close() operations */
327 int *aPending; /* Malloced space holding fd's awaiting a close() */
328};
329
330/*
331** These hash tables map inodes and file descriptors (really, lockKey and
332** openKey structures) into lockInfo and openCnt structures. Access to
333** these hash tables must be protected by a mutex.
334*/
335static Hash lockHash = {SQLITE_HASH_BINARY, 0, 0, 0, 0, 0};
336static Hash openHash = {SQLITE_HASH_BINARY, 0, 0, 0, 0, 0};
337
338#ifdef SQLITE_ENABLE_LOCKING_STYLE
339/*
340** The locking styles are associated with the different file locking
341** capabilities supported by different file systems.
342**
343** POSIX locking style fully supports shared and exclusive byte-range locks
344** ADP locking only supports exclusive byte-range locks
345** FLOCK only supports a single file-global exclusive lock
346** DOTLOCK isn't a true locking style, it refers to the use of a special
347** file named the same as the database file with a '.lock' extension, this
348** can be used on file systems that do not offer any reliable file locking
349** NO locking means that no locking will be attempted, this is only used for
350** read-only file systems currently
351** UNSUPPORTED means that no locking will be attempted, this is only used for
352** file systems that are known to be unsupported
353*/
354typedef enum {
355 posixLockingStyle = 0, /* standard posix-advisory locks */
356 afpLockingStyle, /* use afp locks */
357 flockLockingStyle, /* use flock() */
358 dotlockLockingStyle, /* use <file>.lock files */
359 noLockingStyle, /* useful for read-only file system */
360 unsupportedLockingStyle /* indicates unsupported file system */
361} sqlite3LockingStyle;
362#endif /* SQLITE_ENABLE_LOCKING_STYLE */
363
364/*
365** Helper functions to obtain and relinquish the global mutex.
366*/
367static void enterMutex(){
368 sqlite3_mutex_enter(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MASTER));
369}
370static void leaveMutex(){
371 sqlite3_mutex_leave(sqlite3_mutex_alloc(SQLITE_MUTEX_STATIC_MASTER));
372}
373
374#if SQLITE_THREADSAFE
375/*
376** This variable records whether or not threads can override each others
377** locks.
378**
379** 0: No. Threads cannot override each others locks.
380** 1: Yes. Threads can override each others locks.
381** -1: We don't know yet.
382**
383** On some systems, we know at compile-time if threads can override each
384** others locks. On those systems, the SQLITE_THREAD_OVERRIDE_LOCK macro
385** will be set appropriately. On other systems, we have to check at
386** runtime. On these latter systems, SQLTIE_THREAD_OVERRIDE_LOCK is
387** undefined.
388**
389** This variable normally has file scope only. But during testing, we make
390** it a global so that the test code can change its value in order to verify
391** that the right stuff happens in either case.
392*/
393#ifndef SQLITE_THREAD_OVERRIDE_LOCK
394# define SQLITE_THREAD_OVERRIDE_LOCK -1
395#endif
396#ifdef SQLITE_TEST
397int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
398#else
399static int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
400#endif
401
402/*
403** This structure holds information passed into individual test
404** threads by the testThreadLockingBehavior() routine.
405*/
406struct threadTestData {
407 int fd; /* File to be locked */
408 struct flock lock; /* The locking operation */
409 int result; /* Result of the locking operation */
410};
411
412#ifdef SQLITE_LOCK_TRACE
413/*
414** Print out information about all locking operations.
415**
416** This routine is used for troubleshooting locks on multithreaded
417** platforms. Enable by compiling with the -DSQLITE_LOCK_TRACE
418** command-line option on the compiler. This code is normally
419** turned off.
420*/
421static int lockTrace(int fd, int op, struct flock *p){
422 char *zOpName, *zType;
423 int s;
424 int savedErrno;
425 if( op==F_GETLK ){
426 zOpName = "GETLK";
427 }else if( op==F_SETLK ){
428 zOpName = "SETLK";
429 }else{
430 s = fcntl(fd, op, p);
431 sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd, op, s);
432 return s;
433 }
434 if( p->l_type==F_RDLCK ){
435 zType = "RDLCK";
436 }else if( p->l_type==F_WRLCK ){
437 zType = "WRLCK";
438 }else if( p->l_type==F_UNLCK ){
439 zType = "UNLCK";
440 }else{
441 assert( 0 );
442 }
443 assert( p->l_whence==SEEK_SET );
444 s = fcntl(fd, op, p);
445 savedErrno = errno;
446 sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n",
447 threadid, fd, zOpName, zType, (int)p->l_start, (int)p->l_len,
448 (int)p->l_pid, s);
449 if( s==(-1) && op==F_SETLK && (p->l_type==F_RDLCK || p->l_type==F_WRLCK) ){
450 struct flock l2;
451 l2 = *p;
452 fcntl(fd, F_GETLK, &l2);
453 if( l2.l_type==F_RDLCK ){
454 zType = "RDLCK";
455 }else if( l2.l_type==F_WRLCK ){
456 zType = "WRLCK";
457 }else if( l2.l_type==F_UNLCK ){
458 zType = "UNLCK";
459 }else{
460 assert( 0 );
461 }
462 sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n",
463 zType, (int)l2.l_start, (int)l2.l_len, (int)l2.l_pid);
464 }
465 errno = savedErrno;
466 return s;
467}
468#define fcntl lockTrace
469#endif /* SQLITE_LOCK_TRACE */
470
471/*
472** The testThreadLockingBehavior() routine launches two separate
473** threads on this routine. This routine attempts to lock a file
474** descriptor then returns. The success or failure of that attempt
475** allows the testThreadLockingBehavior() procedure to determine
476** whether or not threads can override each others locks.
477*/
478static void *threadLockingTest(void *pArg){
479 struct threadTestData *pData = (struct threadTestData*)pArg;
480 pData->result = fcntl(pData->fd, F_SETLK, &pData->lock);
481 return pArg;
482}
483
484/*
485** This procedure attempts to determine whether or not threads
486** can override each others locks then sets the
487** threadsOverrideEachOthersLocks variable appropriately.
488*/
489static void testThreadLockingBehavior(int fd_orig){
490 int fd;
491 struct threadTestData d[2];
492 pthread_t t[2];
493
494 fd = dup(fd_orig);
495 if( fd<0 ) return;
496 memset(d, 0, sizeof(d));
497 d[0].fd = fd;
498 d[0].lock.l_type = F_RDLCK;
499 d[0].lock.l_len = 1;
500 d[0].lock.l_start = 0;
501 d[0].lock.l_whence = SEEK_SET;
502 d[1] = d[0];
503 d[1].lock.l_type = F_WRLCK;
504 pthread_create(&t[0], 0, threadLockingTest, &d[0]);
505 pthread_create(&t[1], 0, threadLockingTest, &d[1]);
506 pthread_join(t[0], 0);
507 pthread_join(t[1], 0);
508 close(fd);
509 threadsOverrideEachOthersLocks = d[0].result==0 && d[1].result==0;
510}
511#endif /* SQLITE_THREADSAFE */
512
513/*
514** Release a lockInfo structure previously allocated by findLockInfo().
515*/
516static void releaseLockInfo(struct lockInfo *pLock){
517 if (pLock == NULL)
518 return;
519 pLock->nRef--;
520 if( pLock->nRef==0 ){
521 sqlite3HashInsert(&lockHash, &pLock->key, sizeof(pLock->key), 0);
522 sqlite3_free(pLock);
523 }
524}
525
526/*
527** Release a openCnt structure previously allocated by findLockInfo().
528*/
529static void releaseOpenCnt(struct openCnt *pOpen){
530 if (pOpen == NULL)
531 return;
532 pOpen->nRef--;
533 if( pOpen->nRef==0 ){
534 sqlite3HashInsert(&openHash, &pOpen->key, sizeof(pOpen->key), 0);
535 free(pOpen->aPending);
536 sqlite3_free(pOpen);
537 }
538}
539
540#ifdef SQLITE_ENABLE_LOCKING_STYLE
541/*
542** Tests a byte-range locking query to see if byte range locks are
543** supported, if not we fall back to dotlockLockingStyle.
544*/
545static sqlite3LockingStyle sqlite3TestLockingStyle(
546 const char *filePath,
547 int fd
548){
549 /* test byte-range lock using fcntl */
550 struct flock lockInfo;
551
552 lockInfo.l_len = 1;
553 lockInfo.l_start = 0;
554 lockInfo.l_whence = SEEK_SET;
555 lockInfo.l_type = F_RDLCK;
556
557 if( fcntl(fd, F_GETLK, &lockInfo)!=-1 ) {
558 return posixLockingStyle;
559 }
560
561 /* testing for flock can give false positives. So if if the above test
562 ** fails, then we fall back to using dot-lock style locking.
563 */
564 return dotlockLockingStyle;
565}
566
567/*
568** Examines the f_fstypename entry in the statfs structure as returned by
569** stat() for the file system hosting the database file, assigns the
570** appropriate locking style based on it's value. These values and
571** assignments are based on Darwin/OSX behavior and have not been tested on
572** other systems.
573*/
574static sqlite3LockingStyle sqlite3DetectLockingStyle(
575 const char *filePath,
576 int fd
577){
578
579#ifdef SQLITE_FIXED_LOCKING_STYLE
580 return (sqlite3LockingStyle)SQLITE_FIXED_LOCKING_STYLE;
581#else
582 struct statfs fsInfo;
583
584 if (statfs(filePath, &fsInfo) == -1)
585 return sqlite3TestLockingStyle(filePath, fd);
586
587 if (fsInfo.f_flags & MNT_RDONLY)
588 return noLockingStyle;
589
590 if( (!strcmp(fsInfo.f_fstypename, "hfs")) ||
591 (!strcmp(fsInfo.f_fstypename, "ufs")) )
592 return posixLockingStyle;
593
594 if(!strcmp(fsInfo.f_fstypename, "afpfs"))
595 return afpLockingStyle;
596
597 if(!strcmp(fsInfo.f_fstypename, "nfs"))
598 return sqlite3TestLockingStyle(filePath, fd);
599
600 if(!strcmp(fsInfo.f_fstypename, "smbfs"))
601 return flockLockingStyle;
602
603 if(!strcmp(fsInfo.f_fstypename, "msdos"))
604 return dotlockLockingStyle;
605
606 if(!strcmp(fsInfo.f_fstypename, "webdav"))
607 return unsupportedLockingStyle;
608
609 return sqlite3TestLockingStyle(filePath, fd);
610#endif /* SQLITE_FIXED_LOCKING_STYLE */
611}
612
613#endif /* SQLITE_ENABLE_LOCKING_STYLE */
614
615/*
616** Given a file descriptor, locate lockInfo and openCnt structures that
617** describes that file descriptor. Create new ones if necessary. The
618** return values might be uninitialized if an error occurs.
619**
620** Return the number of errors.
621*/
622static int findLockInfo(
623 int fd, /* The file descriptor used in the key */
624 struct lockInfo **ppLock, /* Return the lockInfo structure here */
625 struct openCnt **ppOpen /* Return the openCnt structure here */
626){
627 int rc;
628 struct lockKey key1;
629 struct openKey key2;
630 struct stat statbuf;
631 struct lockInfo *pLock;
632 struct openCnt *pOpen;
633 rc = fstat(fd, &statbuf);
634 if( rc!=0 ) return 1;
635
636 memset(&key1, 0, sizeof(key1));
637 key1.dev = statbuf.st_dev;
638 key1.ino = statbuf.st_ino;
639#if SQLITE_THREADSAFE
640 if( threadsOverrideEachOthersLocks<0 ){
641 testThreadLockingBehavior(fd);
642 }
643 key1.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();
644#endif
645 memset(&key2, 0, sizeof(key2));
646 key2.dev = statbuf.st_dev;
647 key2.ino = statbuf.st_ino;
648 pLock = (struct lockInfo*)sqlite3HashFind(&lockHash, &key1, sizeof(key1));
649 if( pLock==0 ){
650 struct lockInfo *pOld;
651 pLock = sqlite3_malloc( sizeof(*pLock) );
652 if( pLock==0 ){
653 rc = 1;
654 goto exit_findlockinfo;
655 }
656 pLock->key = key1;
657 pLock->nRef = 1;
658 pLock->cnt = 0;
659 pLock->locktype = 0;
660 pOld = sqlite3HashInsert(&lockHash, &pLock->key, sizeof(key1), pLock);
661 if( pOld!=0 ){
662 assert( pOld==pLock );
663 sqlite3_free(pLock);
664 rc = 1;
665 goto exit_findlockinfo;
666 }
667 }else{
668 pLock->nRef++;
669 }
670 *ppLock = pLock;
671 if( ppOpen!=0 ){
672 pOpen = (struct openCnt*)sqlite3HashFind(&openHash, &key2, sizeof(key2));
673 if( pOpen==0 ){
674 struct openCnt *pOld;
675 pOpen = sqlite3_malloc( sizeof(*pOpen) );
676 if( pOpen==0 ){
677 releaseLockInfo(pLock);
678 rc = 1;
679 goto exit_findlockinfo;
680 }
681 pOpen->key = key2;
682 pOpen->nRef = 1;
683 pOpen->nLock = 0;
684 pOpen->nPending = 0;
685 pOpen->aPending = 0;
686 pOld = sqlite3HashInsert(&openHash, &pOpen->key, sizeof(key2), pOpen);
687 if( pOld!=0 ){
688 assert( pOld==pOpen );
689 sqlite3_free(pOpen);
690 releaseLockInfo(pLock);
691 rc = 1;
692 goto exit_findlockinfo;
693 }
694 }else{
695 pOpen->nRef++;
696 }
697 *ppOpen = pOpen;
698 }
699
700exit_findlockinfo:
701 return rc;
702}
703
704#ifdef SQLITE_DEBUG
705/*
706** Helper function for printing out trace information from debugging
707** binaries. This returns the string represetation of the supplied
708** integer lock-type.
709*/
710static const char *locktypeName(int locktype){
711 switch( locktype ){
712 case NO_LOCK: return "NONE";
713 case SHARED_LOCK: return "SHARED";
714 case RESERVED_LOCK: return "RESERVED";
715 case PENDING_LOCK: return "PENDING";
716 case EXCLUSIVE_LOCK: return "EXCLUSIVE";
717 }
718 return "ERROR";
719}
720#endif
721
722/*
723** If we are currently in a different thread than the thread that the
724** unixFile argument belongs to, then transfer ownership of the unixFile
725** over to the current thread.
726**
727** A unixFile is only owned by a thread on systems where one thread is
728** unable to override locks created by a different thread. RedHat9 is
729** an example of such a system.
730**
731** Ownership transfer is only allowed if the unixFile is currently unlocked.
732** If the unixFile is locked and an ownership is wrong, then return
733** SQLITE_MISUSE. SQLITE_OK is returned if everything works.
734*/
735#if SQLITE_THREADSAFE
736static int transferOwnership(unixFile *pFile){
737 int rc;
738 pthread_t hSelf;
739 if( threadsOverrideEachOthersLocks ){
740 /* Ownership transfers not needed on this system */
741 return SQLITE_OK;
742 }
743 hSelf = pthread_self();
744 if( pthread_equal(pFile->tid, hSelf) ){
745 /* We are still in the same thread */
746 OSTRACE1("No-transfer, same thread\n");
747 return SQLITE_OK;
748 }
749 if( pFile->locktype!=NO_LOCK ){
750 /* We cannot change ownership while we are holding a lock! */
751 return SQLITE_MISUSE;
752 }
753 OSTRACE4("Transfer ownership of %d from %d to %d\n",
754 pFile->h, pFile->tid, hSelf);
755 pFile->tid = hSelf;
756 if (pFile->pLock != NULL) {
757 releaseLockInfo(pFile->pLock);
758 rc = findLockInfo(pFile->h, &pFile->pLock, 0);
759 OSTRACE5("LOCK %d is now %s(%s,%d)\n", pFile->h,
760 locktypeName(pFile->locktype),
761 locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);
762 return rc;
763 } else {
764 return SQLITE_OK;
765 }
766}
767#else
768 /* On single-threaded builds, ownership transfer is a no-op */
769# define transferOwnership(X) SQLITE_OK
770#endif
771
772/*
773** Seek to the offset passed as the second argument, then read cnt
774** bytes into pBuf. Return the number of bytes actually read.
775*/
776static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
777 int got;
778 i64 newOffset;
779 TIMER_START;
780#if defined(USE_PREAD)
781 got = pread(id->h, pBuf, cnt, offset);
782 SimulateIOError( got = -1 );
783#elif defined(USE_PREAD64)
784 got = pread64(id->h, pBuf, cnt, offset);
785 SimulateIOError( got = -1 );
786#else
787 newOffset = lseek(id->h, offset, SEEK_SET);
788 SimulateIOError( newOffset-- );
789 if( newOffset!=offset ){
790 return -1;
791 }
792 got = read(id->h, pBuf, cnt);
793#endif
794 TIMER_END;
795 OSTRACE5("READ %-3d %5d %7lld %d\n", id->h, got, offset, TIMER_ELAPSED);
796 return got;
797}
798
799/*
800** Read data from a file into a buffer. Return SQLITE_OK if all
801** bytes were read successfully and SQLITE_IOERR if anything goes
802** wrong.
803*/
804static int unixRead(
805 sqlite3_file *id,
806 void *pBuf,
807 int amt,
808 sqlite3_int64 offset
809){
810 int got;
811 assert( id );
812 got = seekAndRead((unixFile*)id, offset, pBuf, amt);
813 if( got==amt ){
814 return SQLITE_OK;
815 }else if( got<0 ){
816 return SQLITE_IOERR_READ;
817 }else{
818 memset(&((char*)pBuf)[got], 0, amt-got);
819 return SQLITE_IOERR_SHORT_READ;
820 }
821}
822
823/*
824** Seek to the offset in id->offset then read cnt bytes into pBuf.
825** Return the number of bytes actually read. Update the offset.
826*/
827static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
828 int got;
829 i64 newOffset;
830 TIMER_START;
831#if defined(USE_PREAD)
832 got = pwrite(id->h, pBuf, cnt, offset);
833#elif defined(USE_PREAD64)
834 got = pwrite64(id->h, pBuf, cnt, offset);
835#else
836 newOffset = lseek(id->h, offset, SEEK_SET);
837 if( newOffset!=offset ){
838 return -1;
839 }
840 got = write(id->h, pBuf, cnt);
841#endif
842 TIMER_END;
843 OSTRACE5("WRITE %-3d %5d %7lld %d\n", id->h, got, offset, TIMER_ELAPSED);
844 return got;
845}
846
847
848/*
849** Write data from a buffer into a file. Return SQLITE_OK on success
850** or some other error code on failure.
851*/
852static int unixWrite(
853 sqlite3_file *id,
854 const void *pBuf,
855 int amt,
856 sqlite3_int64 offset
857){
858 int wrote = 0;
859 assert( id );
860 assert( amt>0 );
861 while( amt>0 && (wrote = seekAndWrite((unixFile*)id, offset, pBuf, amt))>0 ){
862 amt -= wrote;
863 offset += wrote;
864 pBuf = &((char*)pBuf)[wrote];
865 }
866 SimulateIOError(( wrote=(-1), amt=1 ));
867 SimulateDiskfullError(( wrote=0, amt=1 ));
868 if( amt>0 ){
869 if( wrote<0 ){
870 return SQLITE_IOERR_WRITE;
871 }else{
872 return SQLITE_FULL;
873 }
874 }
875 return SQLITE_OK;
876}
877
878#ifdef SQLITE_TEST
879/*
880** Count the number of fullsyncs and normal syncs. This is used to test
881** that syncs and fullsyncs are occuring at the right times.
882*/
883int sqlite3_sync_count = 0;
884int sqlite3_fullsync_count = 0;
885#endif
886
887/*
888** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.
889** Otherwise use fsync() in its place.
890*/
891#ifndef HAVE_FDATASYNC
892# define fdatasync fsync
893#endif
894
895/*
896** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
897** the F_FULLFSYNC macro is defined. F_FULLFSYNC is currently
898** only available on Mac OS X. But that could change.
899*/
900#ifdef F_FULLFSYNC
901# define HAVE_FULLFSYNC 1
902#else
903# define HAVE_FULLFSYNC 0
904#endif
905
906
907/*
908** The fsync() system call does not work as advertised on many
909** unix systems. The following procedure is an attempt to make
910** it work better.
911**
912** The SQLITE_NO_SYNC macro disables all fsync()s. This is useful
913** for testing when we want to run through the test suite quickly.
914** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
915** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
916** or power failure will likely corrupt the database file.
917*/
918static int full_fsync(int fd, int fullSync, int dataOnly){
919 int rc;
920
921 /* Record the number of times that we do a normal fsync() and
922 ** FULLSYNC. This is used during testing to verify that this procedure
923 ** gets called with the correct arguments.
924 */
925#ifdef SQLITE_TEST
926 if( fullSync ) sqlite3_fullsync_count++;
927 sqlite3_sync_count++;
928#endif
929
930 /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
931 ** no-op
932 */
933#ifdef SQLITE_NO_SYNC
934 rc = SQLITE_OK;
935#else
936
937#if HAVE_FULLFSYNC
938 if( fullSync ){
939 rc = fcntl(fd, F_FULLFSYNC, 0);
940 }else{
941 rc = 1;
942 }
943 /* If the FULLFSYNC failed, fall back to attempting an fsync().
944 * It shouldn't be possible for fullfsync to fail on the local
945 * file system (on OSX), so failure indicates that FULLFSYNC
946 * isn't supported for this file system. So, attempt an fsync
947 * and (for now) ignore the overhead of a superfluous fcntl call.
948 * It'd be better to detect fullfsync support once and avoid
949 * the fcntl call every time sync is called.
950 */
951 if( rc ) rc = fsync(fd);
952
953#else
954 if( dataOnly ){
955 rc = fdatasync(fd);
956 }else{
957 rc = fsync(fd);
958 }
959#endif /* HAVE_FULLFSYNC */
960#endif /* defined(SQLITE_NO_SYNC) */
961
962 return rc;
963}
964
965/*
966** Make sure all writes to a particular file are committed to disk.
967**
968** If dataOnly==0 then both the file itself and its metadata (file
969** size, access time, etc) are synced. If dataOnly!=0 then only the
970** file data is synced.
971**
972** Under Unix, also make sure that the directory entry for the file
973** has been created by fsync-ing the directory that contains the file.
974** If we do not do this and we encounter a power failure, the directory
975** entry for the journal might not exist after we reboot. The next
976** SQLite to access the file will not know that the journal exists (because
977** the directory entry for the journal was never created) and the transaction
978** will not roll back - possibly leading to database corruption.
979*/
980static int unixSync(sqlite3_file *id, int flags){
981 int rc;
982 unixFile *pFile = (unixFile*)id;
983
984 int isDataOnly = (flags&SQLITE_SYNC_DATAONLY);
985 int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL;
986
987 /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */
988 assert((flags&0x0F)==SQLITE_SYNC_NORMAL
989 || (flags&0x0F)==SQLITE_SYNC_FULL
990 );
991
992 assert( pFile );
993 OSTRACE2("SYNC %-3d\n", pFile->h);
994 rc = full_fsync(pFile->h, isFullsync, isDataOnly);
995 SimulateIOError( rc=1 );
996 if( rc ){
997 return SQLITE_IOERR_FSYNC;
998 }
999 if( pFile->dirfd>=0 ){
1000 OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,
1001 HAVE_FULLFSYNC, isFullsync);
1002#ifndef SQLITE_DISABLE_DIRSYNC
1003 /* The directory sync is only attempted if full_fsync is
1004 ** turned off or unavailable. If a full_fsync occurred above,
1005 ** then the directory sync is superfluous.
1006 */
1007 if( (!HAVE_FULLFSYNC || !isFullsync) && full_fsync(pFile->dirfd,0,0) ){
1008 /*
1009 ** We have received multiple reports of fsync() returning
1010 ** errors when applied to directories on certain file systems.
1011 ** A failed directory sync is not a big deal. So it seems
1012 ** better to ignore the error. Ticket #1657
1013 */
1014 /* return SQLITE_IOERR; */
1015 }
1016#endif
1017 close(pFile->dirfd); /* Only need to sync once, so close the directory */
1018 pFile->dirfd = -1; /* when we are done. */
1019 }
1020 return SQLITE_OK;
1021}
1022
1023/*
1024** Truncate an open file to a specified size
1025*/
1026static int unixTruncate(sqlite3_file *id, i64 nByte){
1027 int rc;
1028 assert( id );
1029 rc = ftruncate(((unixFile*)id)->h, (off_t)nByte);
1030 SimulateIOError( rc=1 );
1031 if( rc ){
1032 return SQLITE_IOERR_TRUNCATE;
1033 }else{
1034 return SQLITE_OK;
1035 }
1036}
1037
1038/*
1039** Determine the current size of a file in bytes
1040*/
1041static int unixFileSize(sqlite3_file *id, i64 *pSize){
1042 int rc;
1043 struct stat buf;
1044 assert( id );
1045 rc = fstat(((unixFile*)id)->h, &buf);
1046 SimulateIOError( rc=1 );
1047 if( rc!=0 ){
1048 return SQLITE_IOERR_FSTAT;
1049 }
1050 *pSize = buf.st_size;
1051 return SQLITE_OK;
1052}
1053
1054/*
1055** This routine checks if there is a RESERVED lock held on the specified
1056** file by this or any other process. If such a lock is held, return
1057** non-zero. If the file is unlocked or holds only SHARED locks, then
1058** return zero.
1059*/
1060static int unixCheckReservedLock(sqlite3_file *id){
1061 int r = 0;
1062 unixFile *pFile = (unixFile*)id;
1063
1064 assert( pFile );
1065 enterMutex(); /* Because pFile->pLock is shared across threads */
1066
1067 /* Check if a thread in this process holds such a lock */
1068 if( pFile->pLock->locktype>SHARED_LOCK ){
1069 r = 1;
1070 }
1071
1072 /* Otherwise see if some other process holds it.
1073 */
1074 if( !r ){
1075 struct flock lock;
1076 lock.l_whence = SEEK_SET;
1077 lock.l_start = RESERVED_BYTE;
1078 lock.l_len = 1;
1079 lock.l_type = F_WRLCK;
1080 fcntl(pFile->h, F_GETLK, &lock);
1081 if( lock.l_type!=F_UNLCK ){
1082 r = 1;
1083 }
1084 }
1085
1086 leaveMutex();
1087 OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);
1088
1089 return r;
1090}
1091
1092/*
1093** Lock the file with the lock specified by parameter locktype - one
1094** of the following:
1095**
1096** (1) SHARED_LOCK
1097** (2) RESERVED_LOCK
1098** (3) PENDING_LOCK
1099** (4) EXCLUSIVE_LOCK
1100**
1101** Sometimes when requesting one lock state, additional lock states
1102** are inserted in between. The locking might fail on one of the later
1103** transitions leaving the lock state different from what it started but
1104** still short of its goal. The following chart shows the allowed
1105** transitions and the inserted intermediate states:
1106**
1107** UNLOCKED -> SHARED
1108** SHARED -> RESERVED
1109** SHARED -> (PENDING) -> EXCLUSIVE
1110** RESERVED -> (PENDING) -> EXCLUSIVE
1111** PENDING -> EXCLUSIVE
1112**
1113** This routine will only increase a lock. Use the sqlite3OsUnlock()
1114** routine to lower a locking level.
1115*/
1116static int unixLock(sqlite3_file *id, int locktype){
1117 /* The following describes the implementation of the various locks and
1118 ** lock transitions in terms of the POSIX advisory shared and exclusive
1119 ** lock primitives (called read-locks and write-locks below, to avoid
1120 ** confusion with SQLite lock names). The algorithms are complicated
1121 ** slightly in order to be compatible with windows systems simultaneously
1122 ** accessing the same database file, in case that is ever required.
1123 **
1124 ** Symbols defined in os.h indentify the 'pending byte' and the 'reserved
1125 ** byte', each single bytes at well known offsets, and the 'shared byte
1126 ** range', a range of 510 bytes at a well known offset.
1127 **
1128 ** To obtain a SHARED lock, a read-lock is obtained on the 'pending
1129 ** byte'. If this is successful, a random byte from the 'shared byte
1130 ** range' is read-locked and the lock on the 'pending byte' released.
1131 **
1132 ** A process may only obtain a RESERVED lock after it has a SHARED lock.
1133 ** A RESERVED lock is implemented by grabbing a write-lock on the
1134 ** 'reserved byte'.
1135 **
1136 ** A process may only obtain a PENDING lock after it has obtained a
1137 ** SHARED lock. A PENDING lock is implemented by obtaining a write-lock
1138 ** on the 'pending byte'. This ensures that no new SHARED locks can be
1139 ** obtained, but existing SHARED locks are allowed to persist. A process
1140 ** does not have to obtain a RESERVED lock on the way to a PENDING lock.
1141 ** This property is used by the algorithm for rolling back a journal file
1142 ** after a crash.
1143 **
1144 ** An EXCLUSIVE lock, obtained after a PENDING lock is held, is
1145 ** implemented by obtaining a write-lock on the entire 'shared byte
1146 ** range'. Since all other locks require a read-lock on one of the bytes
1147 ** within this range, this ensures that no other locks are held on the
1148 ** database.
1149 **
1150 ** The reason a single byte cannot be used instead of the 'shared byte
1151 ** range' is that some versions of windows do not support read-locks. By
1152 ** locking a random byte from a range, concurrent SHARED locks may exist
1153 ** even if the locking primitive used is always a write-lock.
1154 */
1155 int rc = SQLITE_OK;
1156 unixFile *pFile = (unixFile*)id;
1157 struct lockInfo *pLock = pFile->pLock;
1158 struct flock lock;
1159 int s;
1160
1161 assert( pFile );
1162 OSTRACE7("LOCK %d %s was %s(%s,%d) pid=%d\n", pFile->h,
1163 locktypeName(locktype), locktypeName(pFile->locktype),
1164 locktypeName(pLock->locktype), pLock->cnt , getpid());
1165
1166 /* If there is already a lock of this type or more restrictive on the
1167 ** unixFile, do nothing. Don't use the end_lock: exit path, as
1168 ** enterMutex() hasn't been called yet.
1169 */
1170 if( pFile->locktype>=locktype ){
1171 OSTRACE3("LOCK %d %s ok (already held)\n", pFile->h,
1172 locktypeName(locktype));
1173 return SQLITE_OK;
1174 }
1175
1176 /* Make sure the locking sequence is correct
1177 */
1178 assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
1179 assert( locktype!=PENDING_LOCK );
1180 assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );
1181
1182 /* This mutex is needed because pFile->pLock is shared across threads
1183 */
1184 enterMutex();
1185
1186 /* Make sure the current thread owns the pFile.
1187 */
1188 rc = transferOwnership(pFile);
1189 if( rc!=SQLITE_OK ){
1190 leaveMutex();
1191 return rc;
1192 }
1193 pLock = pFile->pLock;
1194
1195 /* If some thread using this PID has a lock via a different unixFile*
1196 ** handle that precludes the requested lock, return BUSY.
1197 */
1198 if( (pFile->locktype!=pLock->locktype &&
1199 (pLock->locktype>=PENDING_LOCK || locktype>SHARED_LOCK))
1200 ){
1201 rc = SQLITE_BUSY;
1202 goto end_lock;
1203 }
1204
1205 /* If a SHARED lock is requested, and some thread using this PID already
1206 ** has a SHARED or RESERVED lock, then increment reference counts and
1207 ** return SQLITE_OK.
1208 */
1209 if( locktype==SHARED_LOCK &&
1210 (pLock->locktype==SHARED_LOCK || pLock->locktype==RESERVED_LOCK) ){
1211 assert( locktype==SHARED_LOCK );
1212 assert( pFile->locktype==0 );
1213 assert( pLock->cnt>0 );
1214 pFile->locktype = SHARED_LOCK;
1215 pLock->cnt++;
1216 pFile->pOpen->nLock++;
1217 goto end_lock;
1218 }
1219
1220 lock.l_len = 1L;
1221
1222 lock.l_whence = SEEK_SET;
1223
1224 /* A PENDING lock is needed before acquiring a SHARED lock and before
1225 ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will
1226 ** be released.
1227 */
1228 if( locktype==SHARED_LOCK
1229 || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
1230 ){
1231 lock.l_type = (locktype==SHARED_LOCK?F_RDLCK:F_WRLCK);
1232 lock.l_start = PENDING_BYTE;
1233 s = fcntl(pFile->h, F_SETLK, &lock);
1234 if( s==(-1) ){
1235 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
1236 goto end_lock;
1237 }
1238 }
1239
1240
1241 /* If control gets to this point, then actually go ahead and make
1242 ** operating system calls for the specified lock.
1243 */
1244 if( locktype==SHARED_LOCK ){
1245 assert( pLock->cnt==0 );
1246 assert( pLock->locktype==0 );
1247
1248 /* Now get the read-lock */
1249 lock.l_start = SHARED_FIRST;
1250 lock.l_len = SHARED_SIZE;
1251 s = fcntl(pFile->h, F_SETLK, &lock);
1252
1253 /* Drop the temporary PENDING lock */
1254 lock.l_start = PENDING_BYTE;
1255 lock.l_len = 1L;
1256 lock.l_type = F_UNLCK;
1257 if( fcntl(pFile->h, F_SETLK, &lock)!=0 ){
1258 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1259 goto end_lock;
1260 }
1261 if( s==(-1) ){
1262 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
1263 }else{
1264 pFile->locktype = SHARED_LOCK;
1265 pFile->pOpen->nLock++;
1266 pLock->cnt = 1;
1267 }
1268 }else if( locktype==EXCLUSIVE_LOCK && pLock->cnt>1 ){
1269 /* We are trying for an exclusive lock but another thread in this
1270 ** same process is still holding a shared lock. */
1271 rc = SQLITE_BUSY;
1272 }else{
1273 /* The request was for a RESERVED or EXCLUSIVE lock. It is
1274 ** assumed that there is a SHARED or greater lock on the file
1275 ** already.
1276 */
1277 assert( 0!=pFile->locktype );
1278 lock.l_type = F_WRLCK;
1279 switch( locktype ){
1280 case RESERVED_LOCK:
1281 lock.l_start = RESERVED_BYTE;
1282 break;
1283 case EXCLUSIVE_LOCK:
1284 lock.l_start = SHARED_FIRST;
1285 lock.l_len = SHARED_SIZE;
1286 break;
1287 default:
1288 assert(0);
1289 }
1290 s = fcntl(pFile->h, F_SETLK, &lock);
1291 if( s==(-1) ){
1292 rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
1293 }
1294 }
1295
1296 if( rc==SQLITE_OK ){
1297 pFile->locktype = locktype;
1298 pLock->locktype = locktype;
1299 }else if( locktype==EXCLUSIVE_LOCK ){
1300 pFile->locktype = PENDING_LOCK;
1301 pLock->locktype = PENDING_LOCK;
1302 }
1303
1304end_lock:
1305 leaveMutex();
1306 OSTRACE4("LOCK %d %s %s\n", pFile->h, locktypeName(locktype),
1307 rc==SQLITE_OK ? "ok" : "failed");
1308 return rc;
1309}
1310
1311/*
1312** Lower the locking level on file descriptor pFile to locktype. locktype
1313** must be either NO_LOCK or SHARED_LOCK.
1314**
1315** If the locking level of the file descriptor is already at or below
1316** the requested locking level, this routine is a no-op.
1317*/
1318static int unixUnlock(sqlite3_file *id, int locktype){
1319 struct lockInfo *pLock;
1320 struct flock lock;
1321 int rc = SQLITE_OK;
1322 unixFile *pFile = (unixFile*)id;
1323
1324 assert( pFile );
1325 OSTRACE7("UNLOCK %d %d was %d(%d,%d) pid=%d\n", pFile->h, locktype,
1326 pFile->locktype, pFile->pLock->locktype, pFile->pLock->cnt, getpid());
1327
1328 assert( locktype<=SHARED_LOCK );
1329 if( pFile->locktype<=locktype ){
1330 return SQLITE_OK;
1331 }
1332 if( CHECK_THREADID(pFile) ){
1333 return SQLITE_MISUSE;
1334 }
1335 enterMutex();
1336 pLock = pFile->pLock;
1337 assert( pLock->cnt!=0 );
1338 if( pFile->locktype>SHARED_LOCK ){
1339 assert( pLock->locktype==pFile->locktype );
1340 if( locktype==SHARED_LOCK ){
1341 lock.l_type = F_RDLCK;
1342 lock.l_whence = SEEK_SET;
1343 lock.l_start = SHARED_FIRST;
1344 lock.l_len = SHARED_SIZE;
1345 if( fcntl(pFile->h, F_SETLK, &lock)==(-1) ){
1346 /* This should never happen */
1347 rc = SQLITE_IOERR_RDLOCK;
1348 }
1349 }
1350 lock.l_type = F_UNLCK;
1351 lock.l_whence = SEEK_SET;
1352 lock.l_start = PENDING_BYTE;
1353 lock.l_len = 2L; assert( PENDING_BYTE+1==RESERVED_BYTE );
1354 if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
1355 pLock->locktype = SHARED_LOCK;
1356 }else{
1357 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1358 }
1359 }
1360 if( locktype==NO_LOCK ){
1361 struct openCnt *pOpen;
1362
1363 /* Decrement the shared lock counter. Release the lock using an
1364 ** OS call only when all threads in this same process have released
1365 ** the lock.
1366 */
1367 pLock->cnt--;
1368 if( pLock->cnt==0 ){
1369 lock.l_type = F_UNLCK;
1370 lock.l_whence = SEEK_SET;
1371 lock.l_start = lock.l_len = 0L;
1372 if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
1373 pLock->locktype = NO_LOCK;
1374 }else{
1375 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1376 }
1377 }
1378
1379 /* Decrement the count of locks against this same file. When the
1380 ** count reaches zero, close any other file descriptors whose close
1381 ** was deferred because of outstanding locks.
1382 */
1383 pOpen = pFile->pOpen;
1384 pOpen->nLock--;
1385 assert( pOpen->nLock>=0 );
1386 if( pOpen->nLock==0 && pOpen->nPending>0 ){
1387 int i;
1388 for(i=0; i<pOpen->nPending; i++){
1389 close(pOpen->aPending[i]);
1390 }
1391 free(pOpen->aPending);
1392 pOpen->nPending = 0;
1393 pOpen->aPending = 0;
1394 }
1395 }
1396 leaveMutex();
1397 pFile->locktype = locktype;
1398 return rc;
1399}
1400
1401/*
1402** Close a file.
1403*/
1404static int unixClose(sqlite3_file *id){
1405 unixFile *pFile = (unixFile *)id;
1406 if( !pFile ) return SQLITE_OK;
1407 unixUnlock(id, NO_LOCK);
1408 if( pFile->dirfd>=0 ) close(pFile->dirfd);
1409 pFile->dirfd = -1;
1410 enterMutex();
1411
1412 if( pFile->pOpen->nLock ){
1413 /* If there are outstanding locks, do not actually close the file just
1414 ** yet because that would clear those locks. Instead, add the file
1415 ** descriptor to pOpen->aPending. It will be automatically closed when
1416 ** the last lock is cleared.
1417 */
1418 int *aNew;
1419 struct openCnt *pOpen = pFile->pOpen;
1420 aNew = realloc( pOpen->aPending, (pOpen->nPending+1)*sizeof(int) );
1421 if( aNew==0 ){
1422 /* If a malloc fails, just leak the file descriptor */
1423 }else{
1424 pOpen->aPending = aNew;
1425 pOpen->aPending[pOpen->nPending] = pFile->h;
1426 pOpen->nPending++;
1427 }
1428 }else{
1429 /* There are no outstanding locks so we can close the file immediately */
1430 close(pFile->h);
1431 }
1432 releaseLockInfo(pFile->pLock);
1433 releaseOpenCnt(pFile->pOpen);
1434
1435 leaveMutex();
1436 OSTRACE2("CLOSE %-3d\n", pFile->h);
1437 OpenCounter(-1);
1438 memset(pFile, 0, sizeof(unixFile));
1439 return SQLITE_OK;
1440}
1441
1442
1443#ifdef SQLITE_ENABLE_LOCKING_STYLE
1444#pragma mark AFP Support
1445
1446/*
1447 ** The afpLockingContext structure contains all afp lock specific state
1448 */
1449typedef struct afpLockingContext afpLockingContext;
1450struct afpLockingContext {
1451 unsigned long long sharedLockByte;
1452 char *filePath;
1453};
1454
1455struct ByteRangeLockPB2
1456{
1457 unsigned long long offset; /* offset to first byte to lock */
1458 unsigned long long length; /* nbr of bytes to lock */
1459 unsigned long long retRangeStart; /* nbr of 1st byte locked if successful */
1460 unsigned char unLockFlag; /* 1 = unlock, 0 = lock */
1461 unsigned char startEndFlag; /* 1=rel to end of fork, 0=rel to start */
1462 int fd; /* file desc to assoc this lock with */
1463};
1464
1465#define afpfsByteRangeLock2FSCTL _IOWR('z', 23, struct ByteRangeLockPB2)
1466
1467/*
1468** Return 0 on success, 1 on failure. To match the behavior of the
1469** normal posix file locking (used in unixLock for example), we should
1470** provide 'richer' return codes - specifically to differentiate between
1471** 'file busy' and 'file system error' results.
1472*/
1473static int _AFPFSSetLock(
1474 const char *path,
1475 int fd,
1476 unsigned long long offset,
1477 unsigned long long length,
1478 int setLockFlag
1479){
1480 struct ByteRangeLockPB2 pb;
1481 int err;
1482
1483 pb.unLockFlag = setLockFlag ? 0 : 1;
1484 pb.startEndFlag = 0;
1485 pb.offset = offset;
1486 pb.length = length;
1487 pb.fd = fd;
1488 OSTRACE5("AFPLOCK setting lock %s for %d in range %llx:%llx\n",
1489 (setLockFlag?"ON":"OFF"), fd, offset, length);
1490 err = fsctl(path, afpfsByteRangeLock2FSCTL, &pb, 0);
1491 if ( err==-1 ) {
1492 OSTRACE4("AFPLOCK failed to fsctl() '%s' %d %s\n", path, errno,
1493 strerror(errno));
1494 return 1; /* error */
1495 } else {
1496 return 0;
1497 }
1498}
1499
1500/*
1501 ** This routine checks if there is a RESERVED lock held on the specified
1502 ** file by this or any other process. If such a lock is held, return
1503 ** non-zero. If the file is unlocked or holds only SHARED locks, then
1504 ** return zero.
1505 */
1506static int afpUnixCheckReservedLock(sqlite3_file *id){
1507 int r = 0;
1508 unixFile *pFile = (unixFile*)id;
1509
1510 assert( pFile );
1511 afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
1512
1513 /* Check if a thread in this process holds such a lock */
1514 if( pFile->locktype>SHARED_LOCK ){
1515 r = 1;
1516 }
1517
1518 /* Otherwise see if some other process holds it.
1519 */
1520 if ( !r ) {
1521 /* lock the byte */
1522 int failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);
1523 if (failed) {
1524 /* if we failed to get the lock then someone else must have it */
1525 r = 1;
1526 } else {
1527 /* if we succeeded in taking the reserved lock, unlock it to restore
1528 ** the original state */
1529 _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1, 0);
1530 }
1531 }
1532 OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);
1533
1534 return r;
1535}
1536
1537/* AFP-style locking following the behavior of unixLock, see the unixLock
1538** function comments for details of lock management. */
1539static int afpUnixLock(sqlite3_file *id, int locktype)
1540{
1541 int rc = SQLITE_OK;
1542 unixFile *pFile = (unixFile*)id;
1543 afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
1544 int gotPendingLock = 0;
1545
1546 assert( pFile );
1547 OSTRACE5("LOCK %d %s was %s pid=%d\n", pFile->h,
1548 locktypeName(locktype), locktypeName(pFile->locktype), getpid());
1549 /* If there is already a lock of this type or more restrictive on the
1550 ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as
1551 ** enterMutex() hasn't been called yet.
1552 */
1553 if( pFile->locktype>=locktype ){
1554 OSTRACE3("LOCK %d %s ok (already held)\n", pFile->h,
1555 locktypeName(locktype));
1556 return SQLITE_OK;
1557 }
1558
1559 /* Make sure the locking sequence is correct
1560 */
1561 assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
1562 assert( locktype!=PENDING_LOCK );
1563 assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );
1564
1565 /* This mutex is needed because pFile->pLock is shared across threads
1566 */
1567 enterMutex();
1568
1569 /* Make sure the current thread owns the pFile.
1570 */
1571 rc = transferOwnership(pFile);
1572 if( rc!=SQLITE_OK ){
1573 leaveMutex();
1574 return rc;
1575 }
1576
1577 /* A PENDING lock is needed before acquiring a SHARED lock and before
1578 ** acquiring an EXCLUSIVE lock. For the SHARED lock, the PENDING will
1579 ** be released.
1580 */
1581 if( locktype==SHARED_LOCK
1582 || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
1583 ){
1584 int failed = _AFPFSSetLock(context->filePath, pFile->h,
1585 PENDING_BYTE, 1, 1);
1586 if (failed) {
1587 rc = SQLITE_BUSY;
1588 goto afp_end_lock;
1589 }
1590 }
1591
1592 /* If control gets to this point, then actually go ahead and make
1593 ** operating system calls for the specified lock.
1594 */
1595 if( locktype==SHARED_LOCK ){
1596 int lk, failed;
1597 int tries = 0;
1598
1599 /* Now get the read-lock */
1600 /* note that the quality of the randomness doesn't matter that much */
1601 lk = random();
1602 context->sharedLockByte = (lk & 0x7fffffff)%(SHARED_SIZE - 1);
1603 failed = _AFPFSSetLock(context->filePath, pFile->h,
1604 SHARED_FIRST+context->sharedLockByte, 1, 1);
1605
1606 /* Drop the temporary PENDING lock */
1607 if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)) {
1608 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1609 goto afp_end_lock;
1610 }
1611
1612 if( failed ){
1613 rc = SQLITE_BUSY;
1614 } else {
1615 pFile->locktype = SHARED_LOCK;
1616 }
1617 }else{
1618 /* The request was for a RESERVED or EXCLUSIVE lock. It is
1619 ** assumed that there is a SHARED or greater lock on the file
1620 ** already.
1621 */
1622 int failed = 0;
1623 assert( 0!=pFile->locktype );
1624 if (locktype >= RESERVED_LOCK && pFile->locktype < RESERVED_LOCK) {
1625 /* Acquire a RESERVED lock */
1626 failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);
1627 }
1628 if (!failed && locktype == EXCLUSIVE_LOCK) {
1629 /* Acquire an EXCLUSIVE lock */
1630
1631 /* Remove the shared lock before trying the range. we'll need to
1632 ** reestablish the shared lock if we can't get the afpUnixUnlock
1633 */
1634 if (!_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
1635 context->sharedLockByte, 1, 0)) {
1636 /* now attemmpt to get the exclusive lock range */
1637 failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST,
1638 SHARED_SIZE, 1);
1639 if (failed && _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
1640 context->sharedLockByte, 1, 1)) {
1641 rc = SQLITE_IOERR_RDLOCK; /* this should never happen */
1642 }
1643 } else {
1644 /* */
1645 rc = SQLITE_IOERR_UNLOCK; /* this should never happen */
1646 }
1647 }
1648 if( failed && rc == SQLITE_OK){
1649 rc = SQLITE_BUSY;
1650 }
1651 }
1652
1653 if( rc==SQLITE_OK ){
1654 pFile->locktype = locktype;
1655 }else if( locktype==EXCLUSIVE_LOCK ){
1656 pFile->locktype = PENDING_LOCK;
1657 }
1658
1659afp_end_lock:
1660 leaveMutex();
1661 OSTRACE4("LOCK %d %s %s\n", pFile->h, locktypeName(locktype),
1662 rc==SQLITE_OK ? "ok" : "failed");
1663 return rc;
1664}
1665
1666/*
1667 ** Lower the locking level on file descriptor pFile to locktype. locktype
1668 ** must be either NO_LOCK or SHARED_LOCK.
1669 **
1670 ** If the locking level of the file descriptor is already at or below
1671 ** the requested locking level, this routine is a no-op.
1672 */
1673static int afpUnixUnlock(sqlite3_file *id, int locktype) {
1674 struct flock lock;
1675 int rc = SQLITE_OK;
1676 unixFile *pFile = (unixFile*)id;
1677 afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
1678
1679 assert( pFile );
1680 OSTRACE5("UNLOCK %d %d was %d pid=%d\n", pFile->h, locktype,
1681 pFile->locktype, getpid());
1682
1683 assert( locktype<=SHARED_LOCK );
1684 if( pFile->locktype<=locktype ){
1685 return SQLITE_OK;
1686 }
1687 if( CHECK_THREADID(pFile) ){
1688 return SQLITE_MISUSE;
1689 }
1690 enterMutex();
1691 if( pFile->locktype>SHARED_LOCK ){
1692 if( locktype==SHARED_LOCK ){
1693 int failed = 0;
1694
1695 /* unlock the exclusive range - then re-establish the shared lock */
1696 if (pFile->locktype==EXCLUSIVE_LOCK) {
1697 failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST,
1698 SHARED_SIZE, 0);
1699 if (!failed) {
1700 /* successfully removed the exclusive lock */
1701 if (_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST+
1702 context->sharedLockByte, 1, 1)) {
1703 /* failed to re-establish our shared lock */
1704 rc = SQLITE_IOERR_RDLOCK; /* This should never happen */
1705 }
1706 } else {
1707 /* This should never happen - failed to unlock the exclusive range */
1708 rc = SQLITE_IOERR_UNLOCK;
1709 }
1710 }
1711 }
1712 if (rc == SQLITE_OK && pFile->locktype>=PENDING_LOCK) {
1713 if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)){
1714 /* failed to release the pending lock */
1715 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1716 }
1717 }
1718 if (rc == SQLITE_OK && pFile->locktype>=RESERVED_LOCK) {
1719 if (_AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1, 0)) {
1720 /* failed to release the reserved lock */
1721 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1722 }
1723 }
1724 }
1725 if( locktype==NO_LOCK ){
1726 int failed = _AFPFSSetLock(context->filePath, pFile->h,
1727 SHARED_FIRST + context->sharedLockByte, 1, 0);
1728 if (failed) {
1729 rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
1730 }
1731 }
1732 if (rc == SQLITE_OK)
1733 pFile->locktype = locktype;
1734 leaveMutex();
1735 return rc;
1736}
1737
1738/*
1739 ** Close a file & cleanup AFP specific locking context
1740 */
1741static int afpUnixClose(sqlite3_file *id) {
1742 unixFile *pFile = (unixFile*)pId;
1743
1744 if( !pFile ) return SQLITE_OK;
1745 afpUnixUnlock(*pId, NO_LOCK);
1746 /* free the AFP locking structure */
1747 if (pFile->lockingContext != NULL) {
1748 if (((afpLockingContext *)pFile->lockingContext)->filePath != NULL)
1749 sqlite3_free(((afpLockingContext*)pFile->lockingContext)->filePath);
1750 sqlite3_free(pFile->lockingContext);
1751 }
1752
1753 if( pFile->dirfd>=0 ) close(pFile->dirfd);
1754 pFile->dirfd = -1;
1755 close(pFile->h);
1756 OSTRACE2("CLOSE %-3d\n", pFile->h);
1757 OpenCounter(-1);
1758 return SQLITE_OK;
1759}
1760
1761
1762#pragma mark flock() style locking
1763
1764/*
1765 ** The flockLockingContext is not used
1766 */
1767typedef void flockLockingContext;
1768
1769static int flockUnixCheckReservedLock(sqlite3_file *id) {
1770 unixFile *pFile = (unixFile*)id;
1771
1772 if (pFile->locktype == RESERVED_LOCK) {
1773 return 1; /* already have a reserved lock */
1774 } else {
1775 /* attempt to get the lock */
1776 int rc = flock(pFile->h, LOCK_EX | LOCK_NB);
1777 if (!rc) {
1778 /* got the lock, unlock it */
1779 flock(pFile->h, LOCK_UN);
1780 return 0; /* no one has it reserved */
1781 }
1782 return 1; /* someone else might have it reserved */
1783 }
1784}
1785
1786static int flockUnixLock(sqlite3_file *id, int locktype) {
1787 unixFile *pFile = (unixFile*)id;
1788
1789 /* if we already have a lock, it is exclusive.
1790 ** Just adjust level and punt on outta here. */
1791 if (pFile->locktype > NO_LOCK) {
1792 pFile->locktype = locktype;
1793 return SQLITE_OK;
1794 }
1795
1796 /* grab an exclusive lock */
1797 int rc = flock(pFile->h, LOCK_EX | LOCK_NB);
1798 if (rc) {
1799 /* didn't get, must be busy */
1800 return SQLITE_BUSY;
1801 } else {
1802 /* got it, set the type and return ok */
1803 pFile->locktype = locktype;
1804 return SQLITE_OK;
1805 }
1806}
1807
1808static int flockUnixUnlock(sqlite3_file *id, int locktype) {
1809 unixFile *pFile = (unixFile*)id;
1810
1811 assert( locktype<=SHARED_LOCK );
1812
1813 /* no-op if possible */
1814 if( pFile->locktype==locktype ){
1815 return SQLITE_OK;
1816 }
1817
1818 /* shared can just be set because we always have an exclusive */
1819 if (locktype==SHARED_LOCK) {
1820 pFile->locktype = locktype;
1821 return SQLITE_OK;
1822 }
1823
1824 /* no, really, unlock. */
1825 int rc = flock(pFile->h, LOCK_UN);
1826 if (rc)
1827 return SQLITE_IOERR_UNLOCK;
1828 else {
1829 pFile->locktype = NO_LOCK;
1830 return SQLITE_OK;
1831 }
1832}
1833
1834/*
1835 ** Close a file.
1836 */
1837static int flockUnixClose(sqlite3_file *pId) {
1838 unixFile *pFile = (unixFile*)*pId;
1839
1840 if( !pFile ) return SQLITE_OK;
1841 flockUnixUnlock(*pId, NO_LOCK);
1842
1843 if( pFile->dirfd>=0 ) close(pFile->dirfd);
1844 pFile->dirfd = -1;
1845 enterMutex();
1846
1847 close(pFile->h);
1848 leaveMutex();
1849 OSTRACE2("CLOSE %-3d\n", pFile->h);
1850 OpenCounter(-1);
1851 return SQLITE_OK;
1852}
1853
1854#pragma mark Old-School .lock file based locking
1855
1856/*
1857 ** The dotlockLockingContext structure contains all dotlock (.lock) lock
1858 ** specific state
1859 */
1860typedef struct dotlockLockingContext dotlockLockingContext;
1861struct dotlockLockingContext {
1862 char *lockPath;
1863};
1864
1865
1866static int dotlockUnixCheckReservedLock(sqlite3_file *id) {
1867 unixFile *pFile = (unixFile*)id;
1868 dotlockLockingContext *context =
1869 (dotlockLockingContext *) pFile->lockingContext;
1870
1871 if (pFile->locktype == RESERVED_LOCK) {
1872 return 1; /* already have a reserved lock */
1873 } else {
1874 struct stat statBuf;
1875 if (lstat(context->lockPath,&statBuf) == 0)
1876 /* file exists, someone else has the lock */
1877 return 1;
1878 else
1879 /* file does not exist, we could have it if we want it */
1880 return 0;
1881 }
1882}
1883
1884static int dotlockUnixLock(sqlite3_file *id, int locktype) {
1885 unixFile *pFile = (unixFile*)id;
1886 dotlockLockingContext *context =
1887 (dotlockLockingContext *) pFile->lockingContext;
1888
1889 /* if we already have a lock, it is exclusive.
1890 ** Just adjust level and punt on outta here. */
1891 if (pFile->locktype > NO_LOCK) {
1892 pFile->locktype = locktype;
1893
1894 /* Always update the timestamp on the old file */
1895 utimes(context->lockPath,NULL);
1896 return SQLITE_OK;
1897 }
1898
1899 /* check to see if lock file already exists */
1900 struct stat statBuf;
1901 if (lstat(context->lockPath,&statBuf) == 0){
1902 return SQLITE_BUSY; /* it does, busy */
1903 }
1904
1905 /* grab an exclusive lock */
1906 int fd = open(context->lockPath,O_RDONLY|O_CREAT|O_EXCL,0600);
1907 if (fd < 0) {
1908 /* failed to open/create the file, someone else may have stolen the lock */
1909 return SQLITE_BUSY;
1910 }
1911 close(fd);
1912
1913 /* got it, set the type and return ok */
1914 pFile->locktype = locktype;
1915 return SQLITE_OK;
1916}
1917
1918static int dotlockUnixUnlock(sqlite3_file *id, int locktype) {
1919 unixFile *pFile = (unixFile*)id;
1920 dotlockLockingContext *context =
1921 (dotlockLockingContext *) pFile->lockingContext;
1922
1923 assert( locktype<=SHARED_LOCK );
1924
1925 /* no-op if possible */
1926 if( pFile->locktype==locktype ){
1927 return SQLITE_OK;
1928 }
1929
1930 /* shared can just be set because we always have an exclusive */
1931 if (locktype==SHARED_LOCK) {
1932 pFile->locktype = locktype;
1933 return SQLITE_OK;
1934 }
1935
1936 /* no, really, unlock. */
1937 unlink(context->lockPath);
1938 pFile->locktype = NO_LOCK;
1939 return SQLITE_OK;
1940}
1941
1942/*
1943 ** Close a file.
1944 */
1945static int dotlockUnixClose(sqlite3_file *id) {
1946 unixFile *pFile = (unixFile*)id;
1947
1948 if( !pFile ) return SQLITE_OK;
1949 dotlockUnixUnlock(*pId, NO_LOCK);
1950 /* free the dotlock locking structure */
1951 if (pFile->lockingContext != NULL) {
1952 if (((dotlockLockingContext *)pFile->lockingContext)->lockPath != NULL)
1953 sqlite3_free( ( (dotlockLockingContext *)
1954 pFile->lockingContext)->lockPath);
1955 sqlite3_free(pFile->lockingContext);
1956 }
1957
1958 if( pFile->dirfd>=0 ) close(pFile->dirfd);
1959 pFile->dirfd = -1;
1960 enterMutex();
1961
1962 close(pFile->h);
1963
1964 leaveMutex();
1965 OSTRACE2("CLOSE %-3d\n", pFile->h);
1966 OpenCounter(-1);
1967 return SQLITE_OK;
1968}
1969
1970
1971#pragma mark No locking
1972
1973/*
1974 ** The nolockLockingContext is void
1975 */
1976typedef void nolockLockingContext;
1977
1978static int nolockUnixCheckReservedLock(sqlite3_file *id) {
1979 return 0;
1980}
1981
1982static int nolockUnixLock(sqlite3_file *id, int locktype) {
1983 return SQLITE_OK;
1984}
1985
1986static int nolockUnixUnlock(sqlite3_file *id, int locktype) {
1987 return SQLITE_OK;
1988}
1989
1990/*
1991 ** Close a file.
1992 */
1993static int nolockUnixClose(sqlite3_file *id) {
1994 unixFile *pFile = (unixFile*)id;
1995
1996 if( !pFile ) return SQLITE_OK;
1997 if( pFile->dirfd>=0 ) close(pFile->dirfd);
1998 pFile->dirfd = -1;
1999 enterMutex();
2000
2001 close(pFile->h);
2002
2003 leaveMutex();
2004 OSTRACE2("CLOSE %-3d\n", pFile->h);
2005 OpenCounter(-1);
2006 return SQLITE_OK;
2007}
2008
2009#endif /* SQLITE_ENABLE_LOCKING_STYLE */
2010
2011
2012/*
2013** Information and control of an open file handle.
2014*/
2015static int unixFileControl(sqlite3_file *id, int op, void *pArg){
2016 switch( op ){
2017 case SQLITE_FCNTL_LOCKSTATE: {
2018 *(int*)pArg = ((unixFile*)id)->locktype;
2019 return SQLITE_OK;
2020 }
2021 }
2022 return SQLITE_ERROR;
2023}
2024
2025/*
2026** Return the sector size in bytes of the underlying block device for
2027** the specified file. This is almost always 512 bytes, but may be
2028** larger for some devices.
2029**
2030** SQLite code assumes this function cannot fail. It also assumes that
2031** if two files are created in the same file-system directory (i.e.
2032** a database and it's journal file) that the sector size will be the
2033** same for both.
2034*/
2035static int unixSectorSize(sqlite3_file *id){
2036 return SQLITE_DEFAULT_SECTOR_SIZE;
2037}
2038
2039/*
2040** Return the device characteristics for the file. This is always 0.
2041*/
2042static int unixDeviceCharacteristics(sqlite3_file *id){
2043 return 0;
2044}
2045
2046/*
2047** This vector defines all the methods that can operate on an sqlite3_file
2048** for unix.
2049*/
2050static const sqlite3_io_methods sqlite3UnixIoMethod = {
2051 1, /* iVersion */
2052 unixClose,
2053 unixRead,
2054 unixWrite,
2055 unixTruncate,
2056 unixSync,
2057 unixFileSize,
2058 unixLock,
2059 unixUnlock,
2060 unixCheckReservedLock,
2061 unixFileControl,
2062 unixSectorSize,
2063 unixDeviceCharacteristics
2064};
2065
2066#ifdef SQLITE_ENABLE_LOCKING_STYLE
2067/*
2068** This vector defines all the methods that can operate on an sqlite3_file
2069** for unix with AFP style file locking.
2070*/
2071static const sqlite3_io_methods sqlite3AFPLockingUnixIoMethod = {
2072 1, /* iVersion */
2073 unixClose,
2074 unixRead,
2075 unixWrite,
2076 unixTruncate,
2077 unixSync,
2078 unixFileSize,
2079 afpUnixLock,
2080 afpUnixUnlock,
2081 afpUnixCheckReservedLock,
2082 unixFileControl,
2083 unixSectorSize,
2084 unixDeviceCharacteristics
2085};
2086
2087/*
2088** This vector defines all the methods that can operate on an sqlite3_file
2089** for unix with flock() style file locking.
2090*/
2091static const sqlite3_io_methods sqlite3FlockLockingUnixIoMethod = {
2092 1, /* iVersion */
2093 flockUnixClose,
2094 unixRead,
2095 unixWrite,
2096 unixTruncate,
2097 unixSync,
2098 unixFileSize,
2099 flockUnixLock,
2100 flockUnixUnlock,
2101 flockUnixCheckReservedLock,
2102 unixFileControl,
2103 unixSectorSize,
2104 unixDeviceCharacteristics
2105};
2106
2107/*
2108** This vector defines all the methods that can operate on an sqlite3_file
2109** for unix with dotlock style file locking.
2110*/
2111static const sqlite3_io_methods sqlite3DotlockLockingUnixIoMethod = {
2112 1, /* iVersion */
2113 dotlockUnixClose,
2114 unixRead,
2115 unixWrite,
2116 unixTruncate,
2117 unixSync,
2118 unixFileSize,
2119 dotlockUnixLock,
2120 dotlockUnixUnlock,
2121 dotlockUnixCheckReservedLock,
2122 unixFileControl,
2123 unixSectorSize,
2124 unixDeviceCharacteristics
2125};
2126
2127/*
2128** This vector defines all the methods that can operate on an sqlite3_file
2129** for unix with dotlock style file locking.
2130*/
2131static const sqlite3_io_methods sqlite3NolockLockingUnixIoMethod = {
2132 1, /* iVersion */
2133 nolockUnixClose,
2134 unixRead,
2135 unixWrite,
2136 unixTruncate,
2137 unixSync,
2138 unixFileSize,
2139 nolockUnixLock,
2140 nolockUnixUnlock,
2141 nolockUnixCheckReservedLock,
2142 unixFileControl,
2143 unixSectorSize,
2144 unixDeviceCharacteristics
2145};
2146
2147#endif /* SQLITE_ENABLE_LOCKING_STYLE */
2148
2149/*
2150** Allocate memory for a new unixFile and initialize that unixFile.
2151** Write a pointer to the new unixFile into *pId.
2152** If we run out of memory, close the file and return an error.
2153*/
2154#ifdef SQLITE_ENABLE_LOCKING_STYLE
2155/*
2156** When locking extensions are enabled, the filepath and locking style
2157** are needed to determine the unixFile pMethod to use for locking operations.
2158** The locking-style specific lockingContext data structure is created
2159** and assigned here also.
2160*/
2161static int fillInUnixFile(
2162 int h, /* Open file descriptor of file being opened */
2163 int dirfd, /* Directory file descriptor */
2164 sqlite3_file *pId, /* Write completed initialization here */
2165 const char *zFilename, /* Name of the file being opened */
2166){
2167 sqlite3LockingStyle lockingStyle;
2168 unixFile *pNew = (unixFile *)pId;
2169 int rc;
2170
2171 memset(pNew, 0, sizeof(unixFile));
2172 lockingStyle = sqlite3DetectLockingStyle(zFilename, h);
2173 if ( lockingStyle == posixLockingStyle ) {
2174 enterMutex();
2175 rc = findLockInfo(h, &pNew->pLock, &pNew->pOpen);
2176 leaveMutex();
2177 if( rc ){
2178 close(h);
2179 unlink(zFilename);
2180 return SQLITE_NOMEM;
2181 }
2182 } else {
2183 /* pLock and pOpen are only used for posix advisory locking */
2184 pNew->pLock = NULL;
2185 pNew->pOpen = NULL;
2186 }
2187 pNew->dirfd = -1;
2188 pNew->h = h;
2189 SET_THREADID(pNew);
2190 pNew = sqlite3_malloc( sizeof(unixFile) );
2191 if( pNew==0 ){
2192 close(h);
2193 enterMutex();
2194 releaseLockInfo(pNew->pLock);
2195 releaseOpenCnt(pNew->pOpen);
2196 leaveMutex();
2197 return SQLITE_NOMEM;
2198 }else{
2199 switch(lockingStyle) {
2200 case afpLockingStyle: {
2201 /* afp locking uses the file path so it needs to be included in
2202 ** the afpLockingContext */
2203 int nFilename;
2204 pNew->pMethod = &sqlite3AFPLockingUnixIoMethod;
2205 pNew->lockingContext =
2206 sqlite3_malloc(sizeof(afpLockingContext));
2207 nFilename = strlen(zFilename)+1;
2208 ((afpLockingContext *)pNew->lockingContext)->filePath =
2209 sqlite3_malloc(nFilename);
2210 memcpy(((afpLockingContext *)pNew->lockingContext)->filePath,
2211 zFilename, nFilename);
2212 srandomdev();
2213 break;
2214 }
2215 case flockLockingStyle:
2216 /* flock locking doesn't need additional lockingContext information */
2217 pNew->pMethod = &sqlite3FlockLockingUnixIoMethod;
2218 break;
2219 case dotlockLockingStyle: {
2220 /* dotlock locking uses the file path so it needs to be included in
2221 ** the dotlockLockingContext */
2222 int nFilename;
2223 pNew->pMethod = &sqlite3DotlockLockingUnixIoMethod;
2224 pNew->lockingContext = sqlite3_malloc(
2225 sizeof(dotlockLockingContext));
2226 nFilename = strlen(zFilename) + 6;
2227 ((dotlockLockingContext *)pNew->lockingContext)->lockPath =
2228 sqlite3_malloc( nFilename );
2229 sqlite3_snprintf(nFilename,
2230 ((dotlockLockingContext *)pNew->lockingContext)->lockPath,
2231 "%s.lock", zFilename);
2232 break;
2233 }
2234 case posixLockingStyle:
2235 /* posix locking doesn't need additional lockingContext information */
2236 pNew->pMethod = &sqlite3UnixIoMethod;
2237 break;
2238 case noLockingStyle:
2239 case unsupportedLockingStyle:
2240 default:
2241 pNew->pMethod = &sqlite3NolockLockingUnixIoMethod;
2242 }
2243 OpenCounter(+1);
2244 return SQLITE_OK;
2245 }
2246}
2247#else /* SQLITE_ENABLE_LOCKING_STYLE */
2248static int fillInUnixFile(
2249 int h, /* Open file descriptor on file being opened */
2250 int dirfd,
2251 sqlite3_file *pId, /* Write to the unixFile structure here */
2252 const char *zFilename /* Name of the file being opened */
2253){
2254 unixFile *pNew = (unixFile *)pId;
2255 int rc;
2256
2257#ifdef FD_CLOEXEC
2258 fcntl(h, F_SETFD, fcntl(h, F_GETFD, 0) | FD_CLOEXEC);
2259#endif
2260
2261 enterMutex();
2262 rc = findLockInfo(h, &pNew->pLock, &pNew->pOpen);
2263 leaveMutex();
2264 if( rc ){
2265 close(h);
2266 return SQLITE_NOMEM;
2267 }
2268
2269 OSTRACE3("OPEN %-3d %s\n", h, zFilename);
2270 pNew->dirfd = -1;
2271 pNew->h = h;
2272 pNew->dirfd = dirfd;
2273 SET_THREADID(pNew);
2274
2275 pNew->pMethod = &sqlite3UnixIoMethod;
2276 OpenCounter(+1);
2277 return SQLITE_OK;
2278}
2279#endif /* SQLITE_ENABLE_LOCKING_STYLE */
2280
2281/*
2282** Open a file descriptor to the directory containing file zFilename.
2283** If successful, *pFd is set to the opened file descriptor and
2284** SQLITE_OK is returned. If an error occurs, either SQLITE_NOMEM
2285** or SQLITE_CANTOPEN is returned and *pFd is set to an undefined
2286** value.
2287**
2288** If SQLITE_OK is returned, the caller is responsible for closing
2289** the file descriptor *pFd using close().
2290*/
2291static int openDirectory(const char *zFilename, int *pFd){
2292 int ii;
2293 int fd = -1;
2294 char zDirname[MAX_PATHNAME+1];
2295
2296 sqlite3_snprintf(MAX_PATHNAME, zDirname, "%s", zFilename);
2297 for(ii=strlen(zDirname); ii>=0 && zDirname[ii]!='/'; ii--);
2298 if( ii>0 ){
2299 zDirname[ii] = '\0';
2300 fd = open(zDirname, O_RDONLY|O_BINARY, 0);
2301 if( fd>=0 ){
2302#ifdef FD_CLOEXEC
2303 fcntl(fd, F_SETFD, fcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
2304#endif
2305 OSTRACE3("OPENDIR %-3d %s\n", fd, zDirname);
2306 }
2307 }
2308 *pFd = fd;
2309 return (fd>=0?SQLITE_OK:SQLITE_CANTOPEN);
2310}
2311
2312/*
2313** Open the file zPath.
2314**
2315** Previously, the SQLite OS layer used three functions in place of this
2316** one:
2317**
2318** sqlite3OsOpenReadWrite();
2319** sqlite3OsOpenReadOnly();
2320** sqlite3OsOpenExclusive();
2321**
2322** These calls correspond to the following combinations of flags:
2323**
2324** ReadWrite() -> (READWRITE | CREATE)
2325** ReadOnly() -> (READONLY)
2326** OpenExclusive() -> (READWRITE | CREATE | EXCLUSIVE)
2327**
2328** The old OpenExclusive() accepted a boolean argument - "delFlag". If
2329** true, the file was configured to be automatically deleted when the
2330** file handle closed. To achieve the same effect using this new
2331** interface, add the DELETEONCLOSE flag to those specified above for
2332** OpenExclusive().
2333*/
2334static int unixOpen(
2335 sqlite3_vfs *pVfs,
2336 const char *zPath,
2337 sqlite3_file *pFile,
2338 int flags,
2339 int *pOutFlags
2340){
2341 int fd = 0; /* File descriptor returned by open() */
2342 int dirfd = -1; /* Directory file descriptor */
2343 int oflags = 0; /* Flags to pass to open() */
2344 int eType = flags&0xFFFFFF00; /* Type of file to open */
2345
2346 int isExclusive = (flags & SQLITE_OPEN_EXCLUSIVE);
2347 int isDelete = (flags & SQLITE_OPEN_DELETEONCLOSE);
2348 int isCreate = (flags & SQLITE_OPEN_CREATE);
2349 int isReadonly = (flags & SQLITE_OPEN_READONLY);
2350 int isReadWrite = (flags & SQLITE_OPEN_READWRITE);
2351
2352 /* If creating a master or main-file journal, this function will open
2353 ** a file-descriptor on the directory too. The first time unixSync()
2354 ** is called the directory file descriptor will be fsync()ed and close()d.
2355 */
2356 int isOpenDirectory = (isCreate &&
2357 (eType==SQLITE_OPEN_MASTER_JOURNAL || eType==SQLITE_OPEN_MAIN_JOURNAL)
2358 );
2359
2360 /* Check the following statements are true:
2361 **
2362 ** (a) Exactly one of the READWRITE and READONLY flags must be set, and
2363 ** (b) if CREATE is set, then READWRITE must also be set, and
2364 ** (c) if EXCLUSIVE is set, then CREATE must also be set.
2365 ** (d) if DELETEONCLOSE is set, then CREATE must also be set.
2366 */
2367 assert((isReadonly==0 || isReadWrite==0) && (isReadWrite || isReadonly));
2368 assert(isCreate==0 || isReadWrite);
2369 assert(isExclusive==0 || isCreate);
2370 assert(isDelete==0 || isCreate);
2371
2372
2373 /* The main DB, main journal, and master journal are never automatically
2374 ** deleted
2375 */
2376 assert( eType!=SQLITE_OPEN_MAIN_DB || !isDelete );
2377 assert( eType!=SQLITE_OPEN_MAIN_JOURNAL || !isDelete );
2378 assert( eType!=SQLITE_OPEN_MASTER_JOURNAL || !isDelete );
2379
2380 /* Assert that the upper layer has set one of the "file-type" flags. */
2381 assert( eType==SQLITE_OPEN_MAIN_DB || eType==SQLITE_OPEN_TEMP_DB
2382 || eType==SQLITE_OPEN_MAIN_JOURNAL || eType==SQLITE_OPEN_TEMP_JOURNAL
2383 || eType==SQLITE_OPEN_SUBJOURNAL || eType==SQLITE_OPEN_MASTER_JOURNAL
2384 || eType==SQLITE_OPEN_TRANSIENT_DB
2385 );
2386
2387 if( isReadonly ) oflags |= O_RDONLY;
2388 if( isReadWrite ) oflags |= O_RDWR;
2389 if( isCreate ) oflags |= O_CREAT;
2390 if( isExclusive ) oflags |= (O_EXCL|O_NOFOLLOW);
2391 oflags |= (O_LARGEFILE|O_BINARY);
2392
2393 memset(pFile, 0, sizeof(unixFile));
2394 fd = open(zPath, oflags, isDelete?0600:SQLITE_DEFAULT_FILE_PERMISSIONS);
2395 if( fd<0 && errno!=EISDIR && isReadWrite && !isExclusive ){
2396 /* Failed to open the file for read/write access. Try read-only. */
2397 flags &= ~(SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE);
2398 flags |= SQLITE_OPEN_READONLY;
2399 return unixOpen(pVfs, zPath, pFile, flags, pOutFlags);
2400 }
2401 if( fd<0 ){
2402 return SQLITE_CANTOPEN;
2403 }
2404 if( isDelete ){
2405 unlink(zPath);
2406 }
2407 if( pOutFlags ){
2408 *pOutFlags = flags;
2409 }
2410
2411 assert(fd!=0);
2412 if( isOpenDirectory ){
2413 int rc = openDirectory(zPath, &dirfd);
2414 if( rc!=SQLITE_OK ){
2415 close(fd);
2416 return rc;
2417 }
2418 }
2419 return fillInUnixFile(fd, dirfd, pFile, zPath);
2420}
2421
2422/*
2423** Delete the file at zPath. If the dirSync argument is true, fsync()
2424** the directory after deleting the file.
2425*/
2426static int unixDelete(sqlite3_vfs *pVfs, const char *zPath, int dirSync){
2427 int rc = SQLITE_OK;
2428 SimulateIOError(return SQLITE_IOERR_DELETE);
2429 unlink(zPath);
2430 if( dirSync ){
2431 int fd;
2432 rc = openDirectory(zPath, &fd);
2433 if( rc==SQLITE_OK ){
2434 if( fsync(fd) ){
2435 rc = SQLITE_IOERR_DIR_FSYNC;
2436 }
2437 close(fd);
2438 }
2439 }
2440 return rc;
2441}
2442
2443/*
2444** Test the existance of or access permissions of file zPath. The
2445** test performed depends on the value of flags:
2446**
2447** SQLITE_ACCESS_EXISTS: Return 1 if the file exists
2448** SQLITE_ACCESS_READWRITE: Return 1 if the file is read and writable.
2449** SQLITE_ACCESS_READONLY: Return 1 if the file is readable.
2450**
2451** Otherwise return 0.
2452*/
2453static int unixAccess(sqlite3_vfs *pVfs, const char *zPath, int flags){
2454 int amode = 0;
2455 switch( flags ){
2456 case SQLITE_ACCESS_EXISTS:
2457 amode = F_OK;
2458 break;
2459 case SQLITE_ACCESS_READWRITE:
2460 amode = W_OK|R_OK;
2461 break;
2462 case SQLITE_ACCESS_READ:
2463 amode = R_OK;
2464 break;
2465
2466 default:
2467 assert(!"Invalid flags argument");
2468 }
2469 return (access(zPath, amode)==0);
2470}
2471
2472/*
2473** Create a temporary file name in zBuf. zBuf must be allocated
2474** by the calling process and must be big enough to hold at least
2475** pVfs->mxPathname bytes.
2476*/
2477static int unixGetTempname(sqlite3_vfs *pVfs, int nBuf, char *zBuf){
2478 static const char *azDirs[] = {
2479 0,
2480 "/var/tmp",
2481 "/usr/tmp",
2482 "/tmp",
2483 ".",
2484 };
2485 static const unsigned char zChars[] =
2486 "abcdefghijklmnopqrstuvwxyz"
2487 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
2488 "0123456789";
2489 int i, j;
2490 struct stat buf;
2491 const char *zDir = ".";
2492
2493 /* It's odd to simulate an io-error here, but really this is just
2494 ** using the io-error infrastructure to test that SQLite handles this
2495 ** function failing.
2496 */
2497 SimulateIOError( return SQLITE_ERROR );
2498
2499 azDirs[0] = sqlite3_temp_directory;
2500 for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); i++){
2501 if( azDirs[i]==0 ) continue;
2502 if( stat(azDirs[i], &buf) ) continue;
2503 if( !S_ISDIR(buf.st_mode) ) continue;
2504 if( access(azDirs[i], 07) ) continue;
2505 zDir = azDirs[i];
2506 break;
2507 }
2508 do{
2509 assert( pVfs->mxPathname==MAX_PATHNAME );
2510 assert( nBuf>=MAX_PATHNAME );
2511 sqlite3_snprintf(MAX_PATHNAME-17, zBuf, "%s/"SQLITE_TEMP_FILE_PREFIX, zDir);
2512 j = strlen(zBuf);
2513 sqlite3Randomness(15, &zBuf[j]);
2514 for(i=0; i<15; i++, j++){
2515 zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
2516 }
2517 zBuf[j] = 0;
2518 }while( access(zBuf,0)==0 );
2519 return SQLITE_OK;
2520}
2521
2522
2523/*
2524** Turn a relative pathname into a full pathname. The relative path
2525** is stored as a nul-terminated string in the buffer pointed to by
2526** zPath.
2527**
2528** zOut points to a buffer of at least sqlite3_vfs.mxPathname bytes
2529** (in this case, MAX_PATHNAME bytes). The full-path is written to
2530** this buffer before returning.
2531*/
2532static int unixFullPathname(
2533 sqlite3_vfs *pVfs, /* Pointer to vfs object */
2534 const char *zPath, /* Possibly relative input path */
2535 int nOut, /* Size of output buffer in bytes */
2536 char *zOut /* Output buffer */
2537){
2538
2539 /* It's odd to simulate an io-error here, but really this is just
2540 ** using the io-error infrastructure to test that SQLite handles this
2541 ** function failing. This function could fail if, for example, the
2542 ** current working directly has been unlinked.
2543 */
2544 SimulateIOError( return SQLITE_ERROR );
2545
2546 assert( pVfs->mxPathname==MAX_PATHNAME );
2547 zOut[MAX_PATHNAME-1] = '\0';
2548 if( zPath[0]=='/' ){
2549 sqlite3_snprintf(MAX_PATHNAME, zOut, "%s", zPath);
2550 }else{
2551 int nCwd;
2552 if( getcwd(zOut, MAX_PATHNAME-1)==0 ){
2553 return SQLITE_CANTOPEN;
2554 }
2555 nCwd = strlen(zOut);
2556 sqlite3_snprintf(MAX_PATHNAME-nCwd, &zOut[nCwd], "/%s", zPath);
2557 }
2558 return SQLITE_OK;
2559
2560#if 0
2561 /*
2562 ** Remove "/./" path elements and convert "/A/./" path elements
2563 ** to just "/".
2564 */
2565 if( zFull ){
2566 int i, j;
2567 for(i=j=0; zFull[i]; i++){
2568 if( zFull[i]=='/' ){
2569 if( zFull[i+1]=='/' ) continue;
2570 if( zFull[i+1]=='.' && zFull[i+2]=='/' ){
2571 i += 1;
2572 continue;
2573 }
2574 if( zFull[i+1]=='.' && zFull[i+2]=='.' && zFull[i+3]=='/' ){
2575 while( j>0 && zFull[j-1]!='/' ){ j--; }
2576 i += 3;
2577 continue;
2578 }
2579 }
2580 zFull[j++] = zFull[i];
2581 }
2582 zFull[j] = 0;
2583 }
2584#endif
2585}
2586
2587
2588#ifndef SQLITE_OMIT_LOAD_EXTENSION
2589/*
2590** Interfaces for opening a shared library, finding entry points
2591** within the shared library, and closing the shared library.
2592*/
2593#include <dlfcn.h>
2594static void *unixDlOpen(sqlite3_vfs *pVfs, const char *zFilename){
2595 return dlopen(zFilename, RTLD_NOW | RTLD_GLOBAL);
2596}
2597
2598/*
2599** SQLite calls this function immediately after a call to unixDlSym() or
2600** unixDlOpen() fails (returns a null pointer). If a more detailed error
2601** message is available, it is written to zBufOut. If no error message
2602** is available, zBufOut is left unmodified and SQLite uses a default
2603** error message.
2604*/
2605static void unixDlError(sqlite3_vfs *pVfs, int nBuf, char *zBufOut){
2606 char *zErr;
2607 enterMutex();
2608 zErr = dlerror();
2609 if( zErr ){
2610 sqlite3_snprintf(nBuf, zBufOut, "%s", zErr);
2611 }
2612 leaveMutex();
2613}
2614static void *unixDlSym(sqlite3_vfs *pVfs, void *pHandle, const char *zSymbol){
2615 return dlsym(pHandle, zSymbol);
2616}
2617static void unixDlClose(sqlite3_vfs *pVfs, void *pHandle){
2618 dlclose(pHandle);
2619}
2620#else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */
2621 #define unixDlOpen 0
2622 #define unixDlError 0
2623 #define unixDlSym 0
2624 #define unixDlClose 0
2625#endif
2626
2627/*
2628** Write nBuf bytes of random data to the supplied buffer zBuf.
2629*/
2630static int unixRandomness(sqlite3_vfs *pVfs, int nBuf, char *zBuf){
2631
2632 assert(nBuf>=(sizeof(time_t)+sizeof(int)));
2633
2634 /* We have to initialize zBuf to prevent valgrind from reporting
2635 ** errors. The reports issued by valgrind are incorrect - we would
2636 ** prefer that the randomness be increased by making use of the
2637 ** uninitialized space in zBuf - but valgrind errors tend to worry
2638 ** some users. Rather than argue, it seems easier just to initialize
2639 ** the whole array and silence valgrind, even if that means less randomness
2640 ** in the random seed.
2641 **
2642 ** When testing, initializing zBuf[] to zero is all we do. That means
2643 ** that we always use the same random number sequence. This makes the
2644 ** tests repeatable.
2645 */
2646 memset(zBuf, 0, nBuf);
2647#if !defined(SQLITE_TEST)
2648 {
2649 int pid, fd;
2650 fd = open("/dev/urandom", O_RDONLY);
2651 if( fd<0 ){
2652 time_t t;
2653 time(&t);
2654 memcpy(zBuf, &t, sizeof(t));
2655 pid = getpid();
2656 memcpy(&zBuf[sizeof(t)], &pid, sizeof(pid));
2657 }else{
2658 read(fd, zBuf, nBuf);
2659 close(fd);
2660 }
2661 }
2662#endif
2663 return SQLITE_OK;
2664}
2665
2666
2667/*
2668** Sleep for a little while. Return the amount of time slept.
2669** The argument is the number of microseconds we want to sleep.
2670** The return value is the number of microseconds of sleep actually
2671** requested from the underlying operating system, a number which
2672** might be greater than or equal to the argument, but not less
2673** than the argument.
2674*/
2675static int unixSleep(sqlite3_vfs *pVfs, int microseconds){
2676#if defined(HAVE_USLEEP) && HAVE_USLEEP
2677 usleep(microseconds);
2678 return microseconds;
2679#else
2680 int seconds = (microseconds+999999)/1000000;
2681 sleep(seconds);
2682 return seconds*1000000;
2683#endif
2684}
2685
2686/*
2687** The following variable, if set to a non-zero value, becomes the result
2688** returned from sqlite3OsCurrentTime(). This is used for testing.
2689*/
2690#ifdef SQLITE_TEST
2691int sqlite3_current_time = 0;
2692#endif
2693
2694/*
2695** Find the current time (in Universal Coordinated Time). Write the
2696** current time and date as a Julian Day number into *prNow and
2697** return 0. Return 1 if the time and date cannot be found.
2698*/
2699static int unixCurrentTime(sqlite3_vfs *pVfs, double *prNow){
2700#ifdef NO_GETTOD
2701 time_t t;
2702 time(&t);
2703 *prNow = t/86400.0 + 2440587.5;
2704#else
2705 struct timeval sNow;
2706 gettimeofday(&sNow, 0);
2707 *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_usec/86400000000.0;
2708#endif
2709#ifdef SQLITE_TEST
2710 if( sqlite3_current_time ){
2711 *prNow = sqlite3_current_time/86400.0 + 2440587.5;
2712 }
2713#endif
2714 return 0;
2715}
2716
2717/*
2718** Return a pointer to the sqlite3DefaultVfs structure. We use
2719** a function rather than give the structure global scope because
2720** some compilers (MSVC) do not allow forward declarations of
2721** initialized structures.
2722*/
2723sqlite3_vfs *sqlite3OsDefaultVfs(void){
2724 static sqlite3_vfs unixVfs = {
2725 1, /* iVersion */
2726 sizeof(unixFile), /* szOsFile */
2727 MAX_PATHNAME, /* mxPathname */
2728 0, /* pNext */
2729 "unix", /* zName */
2730 0, /* pAppData */
2731
2732 unixOpen, /* xOpen */
2733 unixDelete, /* xDelete */
2734 unixAccess, /* xAccess */
2735 unixGetTempname, /* xGetTempName */
2736 unixFullPathname, /* xFullPathname */
2737 unixDlOpen, /* xDlOpen */
2738 unixDlError, /* xDlError */
2739 unixDlSym, /* xDlSym */
2740 unixDlClose, /* xDlClose */
2741 unixRandomness, /* xRandomness */
2742 unixSleep, /* xSleep */
2743 unixCurrentTime /* xCurrentTime */
2744 };
2745
2746 return &unixVfs;
2747}
2748
2749#endif /* OS_UNIX */