**
** * General-purpose declarations and utility functions.
** * Unique file ID logic used by VxWorks.
-** * Various locking primitive implementations:
+** * Various locking primitive implementations (all except proxy locking):
** + for Posix Advisory Locks
** + for no-op locks
** + for dot-file locks
** + for flock() locking
** + for named semaphore locks (VxWorks only)
** + for AFP filesystem locks (MacOSX only)
-** + for proxy locks (MacOSX only)
** * sqlite3_file methods not associated with locking.
** * Definitions of sqlite3_io_methods objects for all locking
** methods plus "finder" functions for each locking method.
** * sqlite3_vfs method implementations.
+** * Locking primitives for the proxy uber-locking-method. (MacOSX only)
** * Definitions of sqlite3_vfs objects for all locking methods
** plus implementations of sqlite3_os_init() and sqlite3_os_end().
**
-** $Id: os_unix.c,v 1.226 2008/12/03 21:26:20 aswift Exp $
+** $Id: os_unix.c,v 1.227 2008/12/03 22:32:45 drh Exp $
*/
#include "sqliteInt.h"
#if SQLITE_OS_UNIX /* This file is used on unix only */
** it a global so that the test code can change its value in order to verify
** that the right stuff happens in either case.
*/
-#ifndef SQLITE_THREAD_OVERRIDE_LOCK
-# define SQLITE_THREAD_OVERRIDE_LOCK -1
-#endif
-#ifdef SQLITE_TEST
+#if SQLITE_THREADSAFE && defined(__linux__)
+# ifndef SQLITE_THREAD_OVERRIDE_LOCK
+# define SQLITE_THREAD_OVERRIDE_LOCK -1
+# endif
+# ifdef SQLITE_TEST
int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
-#else
+# else
static int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
+# endif
#endif
/*
if( d.result!=0 ) return;
threadsOverrideEachOthersLocks = (d.lock.l_type==F_UNLCK);
}
-#elif SQLITE_THREADSAFE
-/*
-** On anything other than linux, assume threads override each others locks.
-*/
-static void testThreadLockingBehavior(int fd_orig){
- UNUSED_PARAMETER(fd_orig);
- threadsOverrideEachOthersLocks = 1;
-}
#endif /* SQLITE_THERADSAFE && defined(__linux__) */
/*
}
}
}
-end_afpunlock:
unixLeaveMutex();
if( rc==SQLITE_OK ) pFile->locktype = locktype;
return rc;
********************* End of the AFP lock implementation **********************
******************************************************************************/
+
/******************************************************************************
-************************** Begin Proxy Locking ********************************
-**
-**
-** The default locking schemes in SQLite use byte-range locks on the
-** database file to coordinate safe, concurrent access by multiple readers
-** and writers [http://sqlite.org/lockingv3.html]. The five file locking
-** states (UNLOCKED, PENDING, SHARED, RESERVED, EXCLUSIVE) are implemented
-** as POSIX read & write locks over fixed set of locations (via fsctl),
-** on AFP and SMB only exclusive byte-range locks are available via fsctl
-** with _IOWR('z', 23, struct ByteRangeLockPB2) to track the same 5 states.
-** To simulate a F_RDLCK on the shared range, on AFP a randomly selected
-** address in the shared range is taken for a SHARED lock, the entire
-** shared range is taken for an EXCLUSIVE lock):
-**
-** PENDING_BYTE 0x40000000
-** RESERVED_BYTE 0x40000001
-** SHARED_RANGE 0x40000002 -> 0x40000200
-**
-** This works well on the local file system, but shows a nearly 100x
-** slowdown in read performance on AFP because the AFP client disables
-** the read cache when byte-range locks are present. Enabling the read
-** cache exposes a cache coherency problem that is present on all OS X
-** supported network file systems. NFS and AFP both observe the
-** close-to-open semantics for ensuring cache coherency
-** [http://nfs.sourceforge.net/#faq_a8], which does not effectively
-** address the requirements for concurrent database access by multiple
-** readers and writers
-** [http://www.nabble.com/SQLite-on-NFS-cache-coherency-td15655701.html].
-**
-** To address the performance and cache coherency issues, proxy file locking
-** changes the way database access is controlled by limiting access to a
-** single host at a time and moving file locks off of the database file
-** and onto a proxy file on the local file system.
-**
-**
-** Using proxy locks
-** -----------------
-**
-** C APIs
-**
-** sqlite3_file_control(db, dbname, SQLITE_SET_LOCKPROXYFILE,
-** <proxy_path> | ":auto:");
-** sqlite3_file_control(db, dbname, SQLITE_GET_LOCKPROXYFILE, &<proxy_path>);
-**
-**
-** SQL pragmas
-**
-** PRAGMA [database.]lock_proxy_file=<proxy_path> | :auto:
-** PRAGMA [database.]lock_proxy_file
-**
-** Specifying ":auto:" means that if there is a conch file with a matching
-** host ID in it, the proxy path in the conch file will be used, otherwise
-** a proxy path based on the user's temp dir
-** (via confstr(_CS_DARWIN_USER_TEMP_DIR,...)) will be used and the
-** actual proxy file name is generated from the name and path of the
-** database file. For example:
-**
-** For database path "/Users/me/foo.db"
-** The lock path will be "<tmpdir>/sqliteplocks/_Users_me_foo.db:auto:")
-**
-** Once a lock proxy is configured for a database connection, it can not
-** be removed, however it may be switched to a different proxy path via
-** the above APIs (assuming the conch file is not being held by another
-** connection or process).
-**
-**
-** How proxy locking works
-** -----------------------
-**
-** Proxy file locking relies primarily on two new supporting files:
-**
-** * conch file to limit access to the database file to a single host
-** at a time
-**
-** * proxy file to act as a proxy for the advisory locks normally
-** taken on the database
-**
-** The conch file - to use a proxy file, sqlite must first "hold the conch"
-** by taking an sqlite-style shared lock on the conch file, reading the
-** contents and comparing the host's unique host ID (see below) and lock
-** proxy path against the values stored in the conch. The conch file is
-** stored in the same directory as the database file and the file name
-** is patterned after the database file name as ".<databasename>-conch".
-** If the conch file does not exist, or it's contents do not match the
-** host ID and/or proxy path, then the lock is escalated to an exclusive
-** lock and the conch file contents is updated with the host ID and proxy
-** path and the lock is downgraded to a shared lock again. If the conch
-** is held by another process (with a shared lock), the exclusive lock
-** will fail and SQLITE_BUSY is returned.
-**
-** The proxy file - a single-byte file used for all advisory file locks
-** normally taken on the database file. This allows for safe sharing
-** of the database file for multiple readers and writers on the same
-** host (the conch ensures that they all use the same local lock file).
-**
-** There is a third file - the host ID file - used as a persistent record
-** of a unique identifier for the host, a 128-byte unique host id file
-** in the path defined by the HOSTIDPATH macro (default value is
-** /Library/Caches/.com.apple.sqliteConchHostId).
-**
-** Requesting the lock proxy does not immediately take the conch, it is
-** only taken when the first request to lock database file is made.
-** This matches the semantics of the traditional locking behavior, where
-** opening a connection to a database file does not take a lock on it.
-** The shared lock and an open file descriptor are maintained until
-** the connection to the database is closed.
-**
-** The proxy file and the lock file are never deleted so they only need
-** to be created the first time they are used.
-**
-** Configuration options
-** ---------------------
-**
-** SQLITE_PREFER_PROXY_LOCKING
-**
-** Database files accessed on non-local file systems are
-** automatically configured for proxy locking, lock files are
-** named automatically using the same logic as
-** PRAGMA lock_proxy_file=":auto:"
-**
-** SQLITE_PROXY_DEBUG
-**
-** Enables the logging of error messages during host id file
-** retrieval and creation
-**
-** HOSTIDPATH
-**
-** Overrides the default host ID file path location
-**
-** LOCKPROXYDIR
-**
-** Overrides the default directory used for lock proxy files that
-** are named automatically via the ":auto:" setting
-**
-** SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
+**************** Non-locking sqlite3_file methods *****************************
**
-** Permissions to use when creating a directory for storing the
-** lock proxy files, only used when LOCKPROXYDIR is not set.
-**
-**
-** As mentioned above, when compiled with SQLITE_PREFER_PROXY_LOCKING,
-** setting the environment variable SQLITE_FORCE_PROXY_LOCKING to 1 will
-** force proxy locking to be used for every database file opened, and 0
-** will force automatic proxy locking to be disabled for all database
-** files (explicity calling the SQLITE_SET_LOCKPROXYFILE pragma or
-** sqlite_file_control API is not affected by SQLITE_FORCE_PROXY_LOCKING).
+** The next division contains implementations for all methods of the
+** sqlite3_file object other than the locking methods. The locking
+** methods were defined in divisions above (one locking method per
+** division). Those methods that are common to all locking modes
+** are gather together into this division.
*/
/*
-** Proxy locking is only available on MacOSX
+** Seek to the offset passed as the second argument, then read cnt
+** bytes into pBuf. Return the number of bytes actually read.
+**
+** NB: If you define USE_PREAD or USE_PREAD64, then it might also
+** be necessary to define _XOPEN_SOURCE to be 500. This varies from
+** one system to another. Since SQLite does not define USE_PREAD
+** any any form by default, we will not attempt to define _XOPEN_SOURCE.
+** See tickets #2741 and #2681.
+**
+** To avoid stomping the errno value on a failed read the lastErrno value
+** is set before returning.
*/
-#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
+static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
+ int got;
+ i64 newOffset;
+ TIMER_START;
+#if defined(USE_PREAD)
+ got = pread(id->h, pBuf, cnt, offset);
+ SimulateIOError( got = -1 );
+#elif defined(USE_PREAD64)
+ got = pread64(id->h, pBuf, cnt, offset);
+ SimulateIOError( got = -1 );
+#else
+ newOffset = lseek(id->h, offset, SEEK_SET);
+ SimulateIOError( newOffset-- );
+ if( newOffset!=offset ){
+ if( newOffset == -1 ){
+ ((unixFile*)id)->lastErrno = errno;
+ }else{
+ ((unixFile*)id)->lastErrno = 0;
+ }
+ return -1;
+ }
+ got = read(id->h, pBuf, cnt);
+#endif
+ TIMER_END;
+ if( got<0 ){
+ ((unixFile*)id)->lastErrno = errno;
+ }
+ OSTRACE5("READ %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
+ return got;
+}
/*
-** Forward reference
+** Read data from a file into a buffer. Return SQLITE_OK if all
+** bytes were read successfully and SQLITE_IOERR if anything goes
+** wrong.
*/
-static int fillInUnixFile(
- sqlite3_vfs *pVfs,
- int h,
- int dirfd,
- sqlite3_file *pId,
- const char *zFilename,
- int noLock,
- int isDelete
-);
-
-
-#ifdef SQLITE_TEST
-/* simulate multiple hosts by creating unique hostid file paths */
-int sqlite3_hostid_num = 0;
-#endif
+static int unixRead(
+ sqlite3_file *id,
+ void *pBuf,
+ int amt,
+ sqlite3_int64 offset
+){
+ int got;
+ assert( id );
+ got = seekAndRead((unixFile*)id, offset, pBuf, amt);
+ if( got==amt ){
+ return SQLITE_OK;
+ }else if( got<0 ){
+ /* lastErrno set by seekAndRead */
+ return SQLITE_IOERR_READ;
+ }else{
+ ((unixFile*)id)->lastErrno = 0; /* not a system error */
+ /* Unread parts of the buffer must be zero-filled */
+ memset(&((char*)pBuf)[got], 0, amt-got);
+ return SQLITE_IOERR_SHORT_READ;
+ }
+}
/*
-** The proxyLockingContext has the path and file structures for the remote
-** and local proxy files in it
+** Seek to the offset in id->offset then read cnt bytes into pBuf.
+** Return the number of bytes actually read. Update the offset.
+**
+** To avoid stomping the errno value on a failed write the lastErrno value
+** is set before returning.
*/
-typedef struct proxyLockingContext proxyLockingContext;
-struct proxyLockingContext {
- unixFile *conchFile; /* Open conch file */
- char *conchFilePath; /* Name of the conch file */
- unixFile *lockProxy; /* Open proxy lock file */
- char *lockProxyPath; /* Name of the proxy lock file */
- char *dbPath; /* Name of the open file */
- int conchHeld; /* True if the conch is currently held */
- void *oldLockingContext; /* Original lockingcontext to restore on close */
- sqlite3_io_methods const *pOldMethod; /* Original I/O methods for close */
-};
-
-/* HOSTIDLEN and CONCHLEN both include space for the string
-** terminating nul
-*/
-#define HOSTIDLEN 128
-#define CONCHLEN (MAXPATHLEN+HOSTIDLEN+1)
-#ifndef HOSTIDPATH
-# define HOSTIDPATH "/Library/Caches/.com.apple.sqliteConchHostId"
+static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
+ int got;
+ i64 newOffset;
+ TIMER_START;
+#if defined(USE_PREAD)
+ got = pwrite(id->h, pBuf, cnt, offset);
+#elif defined(USE_PREAD64)
+ got = pwrite64(id->h, pBuf, cnt, offset);
+#else
+ newOffset = lseek(id->h, offset, SEEK_SET);
+ if( newOffset!=offset ){
+ if( newOffset == -1 ){
+ ((unixFile*)id)->lastErrno = errno;
+ }else{
+ ((unixFile*)id)->lastErrno = 0;
+ }
+ return -1;
+ }
+ got = write(id->h, pBuf, cnt);
#endif
-
-/* basically a copy of unixRandomness with different
-** test behavior built in */
-static int proxyGenerateHostID(char *pHostID){
- int pid, fd, i, len;
- unsigned char *key = (unsigned char *)pHostID;
-
- memset(key, 0, HOSTIDLEN);
- len = 0;
- fd = open("/dev/urandom", O_RDONLY);
- if( fd>=0 ){
- len = read(fd, key, HOSTIDLEN);
- close(fd); /* silently leak the fd if it fails */
+ TIMER_END;
+ if( got<0 ){
+ ((unixFile*)id)->lastErrno = errno;
}
- if( len < HOSTIDLEN ){
- time_t t;
- time(&t);
- memcpy(key, &t, sizeof(t));
- pid = getpid();
- memcpy(&key[sizeof(t)], &pid, sizeof(pid));
+
+ OSTRACE5("WRITE %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
+ return got;
+}
+
+
+/*
+** Write data from a buffer into a file. Return SQLITE_OK on success
+** or some other error code on failure.
+*/
+static int unixWrite(
+ sqlite3_file *id,
+ const void *pBuf,
+ int amt,
+ sqlite3_int64 offset
+){
+ int wrote = 0;
+ assert( id );
+ assert( amt>0 );
+ while( amt>0 && (wrote = seekAndWrite((unixFile*)id, offset, pBuf, amt))>0 ){
+ amt -= wrote;
+ offset += wrote;
+ pBuf = &((char*)pBuf)[wrote];
}
-
-#ifdef MAKE_PRETTY_HOSTID
- /* filter the bytes into printable ascii characters and NUL terminate */
- key[(HOSTIDLEN-1)] = 0x00;
- for( i=0; i<(HOSTIDLEN-1); i++ ){
- unsigned char pa = key[i]&0x7F;
- if( pa<0x20 ){
- key[i] = (key[i]&0x80 == 0x80) ? pa+0x40 : pa+0x20;
- }else if( pa==0x7F ){
- key[i] = (key[i]&0x80 == 0x80) ? pa=0x20 : pa+0x7E;
+ SimulateIOError(( wrote=(-1), amt=1 ));
+ SimulateDiskfullError(( wrote=0, amt=1 ));
+ if( amt>0 ){
+ if( wrote<0 ){
+ /* lastErrno set by seekAndWrite */
+ return SQLITE_IOERR_WRITE;
+ }else{
+ ((unixFile*)id)->lastErrno = 0; /* not a system error */
+ return SQLITE_FULL;
}
}
-#endif
return SQLITE_OK;
}
-/* writes the host id path to path, path should be an pre-allocated buffer
-** with enough space for a path
-*/
-static int proxyGetHostIDPath(char *path, size_t len){
- strlcpy(path, HOSTIDPATH, len);
#ifdef SQLITE_TEST
- if( sqlite3_hostid_num>0 ){
- char suffix[2] = "1";
- suffix[0] = suffix[0] + sqlite3_hostid_num;
- strlcat(path, suffix, len);
- }
+/*
+** Count the number of fullsyncs and normal syncs. This is used to test
+** that syncs and fullsyncs are occurring at the right times.
+*/
+int sqlite3_sync_count = 0;
+int sqlite3_fullsync_count = 0;
#endif
- OSTRACE3("GETHOSTIDPATH %s pid=%d\n", path, getpid());
-}
-/* get the host ID from a sqlite hostid file stored in the
-** user-specific tmp directory, create the ID if it's not there already
+/*
+** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.
+** Otherwise use fsync() in its place.
*/
-static int proxyGetHostID(char *pHostID, int *pError){
- int fd;
- char path[MAXPATHLEN];
- size_t len;
- int rc=SQLITE_OK;
-
- proxyGetHostIDPath(path, MAXPATHLEN);
- /* try to create the host ID file, if it already exists read the contents */
- fd = open(path, O_CREAT|O_WRONLY|O_EXCL, 0644);
- if( fd<0 ){
- int err=errno;
-
- if( err!=EEXIST ){
-#ifdef SQLITE_PROXY_DEBUG /* set the sqlite error message instead */
- fprintf(stderr, "sqlite error creating host ID file %s: %s\n",
- path, strerror(err));
-#endif
- return SQLITE_PERM;
- }
- /* couldn't create the file, read it instead */
- fd = open(path, O_RDONLY|O_EXCL);
- if( fd<0 ){
- int err = errno;
-#ifdef SQLITE_PROXY_DEBUG /* set the sqlite error message instead */
- fprintf(stderr, "sqlite error opening host ID file %s: %s\n",
- path, strerror(err));
+#ifndef HAVE_FDATASYNC
+# define fdatasync fsync
#endif
- return SQLITE_PERM;
- }
- len = pread(fd, pHostID, HOSTIDLEN, 0);
- if( len<0 ){
- *pError = errno;
- rc = SQLITE_IOERR_READ;
- }else if( len<HOSTIDLEN ){
- *pError = 0;
- rc = SQLITE_IOERR_SHORT_READ;
- }
- close(fd); /* silently leak the fd if it fails */
- OSTRACE3("GETHOSTID read %s pid=%d\n", pHostID, getpid());
- return rc;
- }else{
- int i;
- /* we're creating the host ID file (use a random string of bytes) */
- proxyGenerateHostID(pHostID);
- len = pwrite(fd, pHostID, HOSTIDLEN, 0);
- if( len<0 ){
- *pError = errno;
- rc = SQLITE_IOERR_WRITE;
- }else if( len<HOSTIDLEN ){
- *pError = 0;
- rc = SQLITE_IOERR_WRITE;
- }
- close(fd); /* silently leak the fd if it fails */
- OSTRACE3("GETHOSTID wrote %s pid=%d\n", pHostID, getpid());
- return rc;
- }
-}
-
-static int proxyGetLockPath(const char *dbPath, char *lPath, size_t maxLen){
- int len;
- int dbLen;
- int i;
-#ifdef LOCKPROXYDIR
- len = strlcpy(lPath, LOCKPROXYDIR, maxLen);
+/*
+** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
+** the F_FULLFSYNC macro is defined. F_FULLFSYNC is currently
+** only available on Mac OS X. But that could change.
+*/
+#ifdef F_FULLFSYNC
+# define HAVE_FULLFSYNC 1
#else
-# ifdef _CS_DARWIN_USER_TEMP_DIR
- {
- char utdir[MAXPATHLEN];
-
- confstr(_CS_DARWIN_USER_TEMP_DIR, lPath, maxLen);
- len = strlcat(lPath, "sqliteplocks", maxLen);
- if( mkdir(lPath, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
- /* if mkdir fails, handle as lock file creation failure */
- int err = errno;
-# ifdef SQLITE_DEBUG
- if( err!=EEXIST ){
- fprintf(stderr, "proxyGetLockPath: mkdir(%s,0%o) error %d %s\n", lPath,
- SQLITE_DEFAULT_PROXYDIR_PERMISSIONS, err, strerror(err));
- }
-# endif
- }else{
- OSTRACE3("GETLOCKPATH mkdir %s pid=%d\n", lPath, getpid());
- }
-
- }
-# else
- len = strlcpy(lPath, "/tmp/", maxLen);
-# endif
+# define HAVE_FULLFSYNC 0
#endif
- if( lPath[len-1]!='/' ){
- len = strlcat(lPath, "/", maxLen);
- }
-
- /* transform the db path to a unique cache name */
- dbLen = strlen(dbPath);
- for( i=0; i<dbLen && (i+len+7)<maxLen; i++){
- char c = dbPath[i];
- lPath[i+len] = (c=='/')?'_':c;
- }
- lPath[i+len]='\0';
- strlcat(lPath, ":auto:", maxLen);
- return SQLITE_OK;
-}
/*
-** Create a new VFS file descriptor (stored in memory obtained from
-** sqlite3_malloc) and open the file named "path" in the file descriptor.
+** The fsync() system call does not work as advertised on many
+** unix systems. The following procedure is an attempt to make
+** it work better.
**
-** The caller is responsible not only for closing the file descriptor
-** but also for freeing the memory associated with the file descriptor.
+** The SQLITE_NO_SYNC macro disables all fsync()s. This is useful
+** for testing when we want to run through the test suite quickly.
+** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
+** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
+** or power failure will likely corrupt the database file.
*/
-static int proxyCreateUnixFile(const char *path, unixFile **ppFile) {
- int fd;
- int dirfd = -1;
- unixFile *pNew;
- int rc = SQLITE_OK;
+static int full_fsync(int fd, int fullSync, int dataOnly){
+ int rc;
- fd = open(path, O_RDWR | O_CREAT, SQLITE_DEFAULT_FILE_PERMISSIONS);
- if( fd<0 ){
- return SQLITE_CANTOPEN;
- }
-
- pNew = (unixFile *)sqlite3_malloc(sizeof(unixFile));
- if( pNew==NULL ){
- rc = SQLITE_NOMEM;
- goto end_create_proxy;
- }
- memset(pNew, 0, sizeof(unixFile));
-
- rc = fillInUnixFile(NULL, fd, dirfd, (sqlite3_file*)pNew, path, 0, 0);
- if( rc==SQLITE_OK ){
- *ppFile = pNew;
- return SQLITE_OK;
- }
-end_create_proxy:
- close(fd); /* silently leak fd if error, we're already in error */
- sqlite3_free(pNew);
- return rc;
-}
-static const sqlite3_io_methods afpIoMethods;
-static int unixWrite(
- sqlite3_file *id,
- const void *pBuf,
- int amt,
- sqlite3_int64 offset
- );
-static int unixRead(
- sqlite3_file *id,
- void *pBuf,
- int amt,
- sqlite3_int64 offset
- );
-/* takes the conch by taking a shared lock and read the contents conch, if
-** lockPath is non-NULL, the host ID and lock file path must match. A NULL
-** lockPath means that the lockPath in the conch file will be used if the
-** host IDs match, or a new lock path will be generated automatically
-** and written to the conch file.
-*/
-static int proxyTakeConch(unixFile *pFile){
- proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
-
- if( pCtx->conchHeld>0 ){
- return SQLITE_OK;
- }else{
- unixFile *conchFile = pCtx->conchFile;
- char testValue[CONCHLEN];
- char conchValue[CONCHLEN];
- char lockPath[MAXPATHLEN];
- char *tLockPath = NULL;
- int rc = SQLITE_OK;
- int readRc = SQLITE_OK;
- int syncPerms = 0;
-
- OSTRACE4("TAKECONCH %d for %s pid=%d\n", conchFile->h,
- (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"), getpid());
-
- rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, SHARED_LOCK);
- if( rc==SQLITE_OK ){
- int pError = 0;
- memset(testValue, 0, CONCHLEN); // conch is fixed size
- rc = proxyGetHostID(testValue, &pError);
- if( rc&SQLITE_IOERR==SQLITE_IOERR ){
- pFile->lastErrno = pError;
- }
- if( pCtx->lockProxyPath ){
- strlcpy(&testValue[HOSTIDLEN], pCtx->lockProxyPath, MAXPATHLEN);
- }
- }
- if( rc!=SQLITE_OK ){
- goto end_takeconch;
- }
-
- readRc = unixRead((sqlite3_file *)conchFile, conchValue, CONCHLEN, 0);
- if( readRc!=SQLITE_IOERR_SHORT_READ ){
- int match = 0;
- if( readRc!=SQLITE_OK ){
- if( rc&SQLITE_IOERR==SQLITE_IOERR ){
- pFile->lastErrno = conchFile->lastErrno;
- }
- rc = readRc;
- goto end_takeconch;
- }
- /* if the conch has data compare the contents */
- if( !pCtx->lockProxyPath ){
- /* for auto-named local lock file, just check the host ID and we'll
- ** use the local lock file path that's already in there */
- if( !memcmp(testValue, conchValue, HOSTIDLEN) ){
- tLockPath = (char *)&conchValue[HOSTIDLEN];
- goto end_takeconch;
- }
- }else{
- /* we've got the conch if conchValue matches our path and host ID */
- if( !memcmp(testValue, conchValue, CONCHLEN) ){
- goto end_takeconch;
- }
- }
- }else{
- /* a short read means we're "creating" the conch (even though it could
- ** have been user-intervention), if we acquire the exclusive lock,
- ** we'll try to match the current on-disk permissions of the database
- */
- syncPerms = 1;
- }
-
- /* either conch was emtpy or didn't match */
- if( !pCtx->lockProxyPath ){
- proxyGetLockPath(pCtx->dbPath, lockPath, MAXPATHLEN);
- tLockPath = lockPath;
- strlcpy(&testValue[HOSTIDLEN], lockPath, MAXPATHLEN);
- }
-
- /* update conch with host and path (this will fail if other process
- ** has a shared lock already) */
- rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, EXCLUSIVE_LOCK);
- if( rc==SQLITE_OK ){
- rc = unixWrite((sqlite3_file *)conchFile, testValue, CONCHLEN, 0);
- if( rc==SQLITE_OK && syncPerms ){
- struct stat buf;
- int err = fstat(pFile->h, &buf);
- if( err==0 ){
- mode_t mode = buf.st_mode & 0100666;
- /* try to match the database file permissions, ignore failure */
-#ifndef SQLITE_PROXY_DEBUG
- fchmod(conchFile->h, buf.st_mode);
+ /* The following "ifdef/elif/else/" block has the same structure as
+ ** the one below. It is replicated here solely to avoid cluttering
+ ** up the real code with the UNUSED_PARAMETER() macros.
+ */
+#ifdef SQLITE_NO_SYNC
+ UNUSED_PARAMETER(fd);
+ UNUSED_PARAMETER(fullSync);
+ UNUSED_PARAMETER(dataOnly);
+#elif HAVE_FULLFSYNC
+ UNUSED_PARAMETER(dataOnly);
#else
- if( fchmod(conchFile->h, buf.st_mode)!=0 ){
- int code = errno;
- fprintf(stderr, "fchmod %o FAILED with %d %s\n",
- buf.st_mode, code, strerror(code));
- } else {
- fprintf(stderr, "fchmod %o SUCCEDED\n",buf.st_mode);
- }
- }else{
- int code = errno;
- fprintf(stderr, "STAT FAILED[%d] with %d %s\n",
- err, code, strerror(code));
+ UNUSED_PARAMETER(fullSync);
#endif
- }
- }
- }
- conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, SHARED_LOCK);
-
-end_takeconch:
- OSTRACE2("TRANSPROXY: CLOSE %d\n", pFile->h);
- if( rc==SQLITE_OK && pFile->openFlags ){
- if( pFile->h>=0 ){
-#ifdef STRICT_CLOSE_ERROR
- if( close(pFile->h) ){
- pFile->lastErrno = errno;
- return SQLITE_IOERR_CLOSE;
- }
-#else
- close(pFile->h); /* silently leak fd if fail */
+
+ /* Record the number of times that we do a normal fsync() and
+ ** FULLSYNC. This is used during testing to verify that this procedure
+ ** gets called with the correct arguments.
+ */
+#ifdef SQLITE_TEST
+ if( fullSync ) sqlite3_fullsync_count++;
+ sqlite3_sync_count++;
#endif
- }
- pFile->h = -1;
- int fd = open(pCtx->dbPath, pFile->openFlags,
- SQLITE_DEFAULT_FILE_PERMISSIONS);
- OSTRACE2("TRANSPROXY: OPEN %d\n", fd);
- if( fd>=0 ){
- pFile->h = fd;
- }else{
- rc=SQLITE_CANTOPEN; // SQLITE_BUSY? proxyTakeConch called during locking
- }
- }
- if( rc==SQLITE_OK && !pCtx->lockProxy ){
- char *path = tLockPath ? tLockPath : pCtx->lockProxyPath;
- // ACS: Need to make a copy of path sometimes
- rc = proxyCreateUnixFile(path, &pCtx->lockProxy);
- }
- if( rc==SQLITE_OK ){
- pCtx->conchHeld = 1;
- if( tLockPath ){
- pCtx->lockProxyPath = sqlite3DbStrDup(0, tLockPath);
- if( pCtx->lockProxy->pMethod == &afpIoMethods ){
- ((afpLockingContext *)pCtx->lockProxy->lockingContext)->dbPath =
- pCtx->lockProxyPath;
- }
- }
- } else {
- conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK);
- }
- OSTRACE3("TAKECONCH %d %s\n", conchFile->h, rc==SQLITE_OK?"ok":"failed");
- return rc;
+ /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
+ ** no-op
+ */
+#ifdef SQLITE_NO_SYNC
+ rc = SQLITE_OK;
+#elif HAVE_FULLFSYNC
+ if( fullSync ){
+ rc = fcntl(fd, F_FULLFSYNC, 0);
+ }else{
+ rc = 1;
}
-}
+ /* If the FULLFSYNC failed, fall back to attempting an fsync().
+ ** It shouldn't be possible for fullfsync to fail on the local
+ ** file system (on OSX), so failure indicates that FULLFSYNC
+ ** isn't supported for this file system. So, attempt an fsync
+ ** and (for now) ignore the overhead of a superfluous fcntl call.
+ ** It'd be better to detect fullfsync support once and avoid
+ ** the fcntl call every time sync is called.
+ */
+ if( rc ) rc = fsync(fd);
-/*
-** If pFile holds a lock on a conch file, then release that lock.
-*/
-static int proxyReleaseConch(unixFile *pFile){
- int rc; /* Subroutine return code */
- proxyLockingContext *pCtx; /* The locking context for the proxy lock */
- unixFile *conchFile; /* Name of the conch file */
+#else
+ if( dataOnly ){
+ rc = fdatasync(fd);
+ if( OS_VXWORKS && rc==-1 && errno==ENOTSUP ){
+ rc = fsync(fd);
+ }
+ }else{
+ rc = fsync(fd);
+ }
+#endif /* ifdef SQLITE_NO_SYNC elif HAVE_FULLFSYNC */
- pCtx = (proxyLockingContext *)pFile->lockingContext;
- conchFile = pCtx->conchFile;
- OSTRACE4("RELEASECONCH %d for %s pid=%d\n", conchFile->h,
- (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"),
- getpid());
- pCtx->conchHeld = 0;
- rc = conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK);
- OSTRACE3("RELEASECONCH %d %s\n", conchFile->h,
- (rc==SQLITE_OK ? "ok" : "failed"));
+ if( OS_VXWORKS && rc!= -1 ){
+ rc = 0;
+ }
return rc;
}
/*
-** Given the name of a database file, compute the name of its conch file.
-** Store the conch filename in memory obtained from sqlite3_malloc().
-** Make *pConchPath point to the new name. Return SQLITE_OK on success
-** or SQLITE_NOMEM if unable to obtain memory.
+** Make sure all writes to a particular file are committed to disk.
**
-** The caller is responsible for ensuring that the allocated memory
-** space is eventually freed.
+** If dataOnly==0 then both the file itself and its metadata (file
+** size, access time, etc) are synced. If dataOnly!=0 then only the
+** file data is synced.
**
-** *pConchPath is set to NULL if a memory allocation error occurs.
+** Under Unix, also make sure that the directory entry for the file
+** has been created by fsync-ing the directory that contains the file.
+** If we do not do this and we encounter a power failure, the directory
+** entry for the journal might not exist after we reboot. The next
+** SQLite to access the file will not know that the journal exists (because
+** the directory entry for the journal was never created) and the transaction
+** will not roll back - possibly leading to database corruption.
*/
-static int proxyCreateConchPathname(char *dbPath, char **pConchPath){
- int i; /* Loop counter */
- int len = strlen(dbPath); /* Length of database filename - dbPath */
- char *conchPath; /* buffer in which to construct conch name */
-
- /* Allocate space for the conch filename and initialize the name to
- ** the name of the original database file. */
- *pConchPath = conchPath = (char *)sqlite3_malloc(len + 8);
- if( conchPath==0 ){
- return SQLITE_NOMEM;
- }
- memcpy(conchPath, dbPath, len+1);
-
- /* now insert a "." before the last / character */
- for( i=(len-1); i>=0; i-- ){
- if( conchPath[i]=='/' ){
- i++;
- break;
- }
- }
- conchPath[i]='.';
- while ( i<len ){
- conchPath[i+1]=dbPath[i];
- i++;
- }
-
- /* append the "-conch" suffix to the file */
- memcpy(&conchPath[i+1], "-conch", 7);
- assert( strlen(conchPath) == len+7 );
-
- return SQLITE_OK;
-}
+static int unixSync(sqlite3_file *id, int flags){
+ int rc;
+ unixFile *pFile = (unixFile*)id;
+ int isDataOnly = (flags&SQLITE_SYNC_DATAONLY);
+ int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL;
-/* Takes a fully configured proxy locking-style unix file and switches
-** the local lock file path
-*/
-static int switchLockProxyPath(unixFile *pFile, const char *path) {
- proxyLockingContext *pCtx = (proxyLockingContext*)pFile->lockingContext;
- char *oldPath = pCtx->lockProxyPath;
- int taken = 0;
- int rc = SQLITE_OK;
+ /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */
+ assert((flags&0x0F)==SQLITE_SYNC_NORMAL
+ || (flags&0x0F)==SQLITE_SYNC_FULL
+ );
- if( pFile->locktype!=NO_LOCK ){
- return SQLITE_BUSY;
- }
+ /* Unix cannot, but some systems may return SQLITE_FULL from here. This
+ ** line is to test that doing so does not cause any problems.
+ */
+ SimulateDiskfullError( return SQLITE_FULL );
- /* nothing to do if the path is NULL, :auto: or matches the existing path */
- if( !path || path[0]=='\0' || !strcmp(path, ":auto:") ||
- (oldPath && !strncmp(oldPath, path, MAXPATHLEN)) ){
- return SQLITE_OK;
- }else{
- unixFile *lockProxy = pCtx->lockProxy;
- pCtx->lockProxy=NULL;
- pCtx->conchHeld = 0;
- if( lockProxy!=NULL ){
- rc=lockProxy->pMethod->xClose((sqlite3_file *)lockProxy);
- if( rc ) return rc;
- sqlite3_free(lockProxy);
+ assert( pFile );
+ OSTRACE2("SYNC %-3d\n", pFile->h);
+ rc = full_fsync(pFile->h, isFullsync, isDataOnly);
+ SimulateIOError( rc=1 );
+ if( rc ){
+ pFile->lastErrno = errno;
+ return SQLITE_IOERR_FSYNC;
+ }
+ if( pFile->dirfd>=0 ){
+ int err;
+ OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,
+ HAVE_FULLFSYNC, isFullsync);
+#ifndef SQLITE_DISABLE_DIRSYNC
+ /* The directory sync is only attempted if full_fsync is
+ ** turned off or unavailable. If a full_fsync occurred above,
+ ** then the directory sync is superfluous.
+ */
+ if( (!HAVE_FULLFSYNC || !isFullsync) && full_fsync(pFile->dirfd,0,0) ){
+ /*
+ ** We have received multiple reports of fsync() returning
+ ** errors when applied to directories on certain file systems.
+ ** A failed directory sync is not a big deal. So it seems
+ ** better to ignore the error. Ticket #1657
+ */
+ /* pFile->lastErrno = errno; */
+ /* return SQLITE_IOERR; */
+ }
+#endif
+ err = close(pFile->dirfd); /* Only need to sync once, so close the */
+ if( err==0 ){ /* directory when we are done */
+ pFile->dirfd = -1;
+ }else{
+ pFile->lastErrno = errno;
+ rc = SQLITE_IOERR_DIR_CLOSE;
}
- sqlite3_free(oldPath);
- pCtx->lockProxyPath = sqlite3DbStrDup(0, path);
}
-
return rc;
}
-static const sqlite3_io_methods dotlockIoMethods;
/*
-** pFile is a file that has been opened by a prior xOpen call. dbPath
-** is a string buffer at least MAXPATHLEN+1 characters in size.
-**
-** This routine find the filename associated with pFile and writes it
-** int dbPath.
+** Truncate an open file to a specified size
*/
-static int proxyGetDbPathForUnixFile(unixFile *pFile, char *dbPath){
-#if defined(__DARWIN__)
- if( pFile->pMethod == &afpIoMethods ){
- /* afp style keeps a reference to the db path in the filePath field
- ** of the struct */
- assert( strlen((char*)pFile->lockingContext)<=MAXPATHLEN );
- strcpy(dbPath, ((afpLockingContext *)pFile->lockingContext)->dbPath);
- }else
-#endif
- if( pFile->pMethod == &dotlockIoMethods ){
- /* dot lock style uses the locking context to store the dot lock
- ** file path */
- int len = strlen((char *)pFile->lockingContext) - strlen(DOTLOCK_SUFFIX);
- memcpy(dbPath, (char *)pFile->lockingContext, len + 1);
+static int unixTruncate(sqlite3_file *id, i64 nByte){
+ int rc;
+ assert( id );
+ SimulateIOError( return SQLITE_IOERR_TRUNCATE );
+ rc = ftruncate(((unixFile*)id)->h, (off_t)nByte);
+ if( rc ){
+ ((unixFile*)id)->lastErrno = errno;
+ return SQLITE_IOERR_TRUNCATE;
}else{
- /* all other styles use the locking context to store the db file path */
- assert( strlen((char*)pFile->lockingContext)<=MAXPATHLEN );
- strcpy(dbPath, (char *)pFile->lockingContext);
+ return SQLITE_OK;
}
- return SQLITE_OK;
}
-static const sqlite3_io_methods proxyIoMethods;
+
/*
-** Takes an already filled in unix file and alters it so all file locking
-** will be performed on the local proxy lock file. The following fields
-** are preserved in the locking context so that they can be restored and
-** the unix structure properly cleaned up at close time:
-** ->lockingContext
-** ->pMethod
+** Determine the current size of a file in bytes
*/
-static int transformUnixFileForLockProxy(unixFile *pFile, const char *path) {
- proxyLockingContext *pCtx;
- char dbPath[MAXPATHLEN+1]; /* Name of the database file */
- char *lockPath=NULL;
- int rc = SQLITE_OK;
-
- if( pFile->locktype!=NO_LOCK ){
- return SQLITE_BUSY;
- }
- proxyGetDbPathForUnixFile(pFile, dbPath);
- if( !path || path[0]=='\0' || !strcmp(path, ":auto:") ){
- lockPath=NULL;
- }else{
- lockPath=(char *)path;
+static int unixFileSize(sqlite3_file *id, i64 *pSize){
+ int rc;
+ struct stat buf;
+ assert( id );
+ rc = fstat(((unixFile*)id)->h, &buf);
+ SimulateIOError( rc=1 );
+ if( rc!=0 ){
+ ((unixFile*)id)->lastErrno = errno;
+ return SQLITE_IOERR_FSTAT;
}
-
- OSTRACE4("TRANSPROXY %d for %s pid=%d\n", pFile->h,
- (lockPath ? lockPath : ":auto:"), getpid());
+ *pSize = buf.st_size;
- pCtx = sqlite3_malloc( sizeof(*pCtx) );
- if( pCtx==0 ){
- return SQLITE_NOMEM;
- }
- memset(pCtx, 0, sizeof(*pCtx));
+ /* When opening a zero-size database, the findLockInfo() procedure
+ ** writes a single byte into that file in order to work around a bug
+ ** in the OS-X msdos filesystem. In order to avoid problems with upper
+ ** layers, we need to report this file size as zero even though it is
+ ** really 1. Ticket #3260.
+ */
+ if( *pSize==1 ) *pSize = 0;
- rc = proxyCreateConchPathname(dbPath, &pCtx->conchFilePath);
- if( rc==SQLITE_OK ){
- rc = proxyCreateUnixFile(pCtx->conchFilePath, &pCtx->conchFile);
- }
- if( rc==SQLITE_OK && lockPath ){
- pCtx->lockProxyPath = sqlite3DbStrDup(0, lockPath);
- }
-end_transform_file:
- if( rc==SQLITE_OK ){
- /* all memory is allocated, proxys are created and assigned,
- ** switch the locking context and pMethod then return.
- */
- pCtx->dbPath = sqlite3DbStrDup(0, dbPath);
- pCtx->oldLockingContext = pFile->lockingContext;
- pFile->lockingContext = pCtx;
- pCtx->pOldMethod = pFile->pMethod;
- pFile->pMethod = &proxyIoMethods;
- }else{
- if( pCtx->conchFile ){
- rc = pCtx->conchFile->pMethod->xClose((sqlite3_file *)pCtx->conchFile);
- if( rc ) return rc;
- sqlite3_free(pCtx->conchFile);
- }
- sqlite3_free(pCtx->conchFilePath);
- sqlite3_free(pCtx);
- }
- OSTRACE3("TRANSPROXY %d %s\n", pFile->h,
- (rc==SQLITE_OK ? "ok" : "failed"));
- return rc;
+ return SQLITE_OK;
}
/*
-** Within this division (the proxying locking implementation) the procedures
-** above this point are all utilities. The lock-related methods of the
-** proxy-locking sqlite3_io_method object follow.
+** Handler for proxy-locking file-control verbs. Defined below in the
+** proxying locking division.
*/
+static int proxyFileControl(sqlite3_file*,int,void*);
/*
-** This routine checks if there is a RESERVED lock held on the specified
-** file by this or any other process. If such a lock is held, set *pResOut
-** to a non-zero value otherwise *pResOut is set to zero. The return value
-** is set to SQLITE_OK unless an I/O error occurs during lock checking.
+** Information and control of an open file handle.
*/
-static int proxyCheckReservedLock(sqlite3_file *id, int *pResOut) {
- unixFile *pFile = (unixFile*)id;
- int rc = proxyTakeConch(pFile);
- if( rc==SQLITE_OK ){
- proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
- unixFile *proxy = pCtx->lockProxy;
- return proxy->pMethod->xCheckReservedLock((sqlite3_file*)proxy, pResOut);
+static int unixFileControl(sqlite3_file *id, int op, void *pArg){
+ switch( op ){
+ case SQLITE_FCNTL_LOCKSTATE: {
+ *(int*)pArg = ((unixFile*)id)->locktype;
+ return SQLITE_OK;
+ }
+ case SQLITE_LAST_ERRNO: {
+ *(int*)pArg = ((unixFile*)id)->lastErrno;
+ return SQLITE_OK;
+ }
+#if SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__)
+ case SQLITE_SET_LOCKPROXYFILE:
+ case SQLITE_GET_LOCKPROXYFILE: {
+ return proxyFileControl(id,op,pArg);
+ }
+#endif /* SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__) */
}
- return rc;
+ return SQLITE_ERROR;
}
/*
-** Lock the file with the lock specified by parameter locktype - one
-** of the following:
-**
-** (1) SHARED_LOCK
-** (2) RESERVED_LOCK
-** (3) PENDING_LOCK
-** (4) EXCLUSIVE_LOCK
-**
-** Sometimes when requesting one lock state, additional lock states
-** are inserted in between. The locking might fail on one of the later
-** transitions leaving the lock state different from what it started but
-** still short of its goal. The following chart shows the allowed
-** transitions and the inserted intermediate states:
-**
-** UNLOCKED -> SHARED
-** SHARED -> RESERVED
-** SHARED -> (PENDING) -> EXCLUSIVE
-** RESERVED -> (PENDING) -> EXCLUSIVE
-** PENDING -> EXCLUSIVE
+** Return the sector size in bytes of the underlying block device for
+** the specified file. This is almost always 512 bytes, but may be
+** larger for some devices.
**
-** This routine will only increase a lock. Use the sqlite3OsUnlock()
-** routine to lower a locking level.
+** SQLite code assumes this function cannot fail. It also assumes that
+** if two files are created in the same file-system directory (i.e.
+** a database and its journal file) that the sector size will be the
+** same for both.
*/
-static int proxyLock(sqlite3_file *id, int locktype) {
- unixFile *pFile = (unixFile*)id;
- int rc = proxyTakeConch(pFile);
- if( rc==SQLITE_OK ){
- proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
- unixFile *proxy = pCtx->lockProxy;
- rc = proxy->pMethod->xLock((sqlite3_file*)proxy, locktype);
- pFile->locktype = proxy->locktype;
- }
- return rc;
+static int unixSectorSize(sqlite3_file *NotUsed){
+ UNUSED_PARAMETER(NotUsed);
+ return SQLITE_DEFAULT_SECTOR_SIZE;
}
+/*
+** Return the device characteristics for the file. This is always 0 for unix.
+*/
+static int unixDeviceCharacteristics(sqlite3_file *NotUsed){
+ UNUSED_PARAMETER(NotUsed);
+ return 0;
+}
/*
-** Lower the locking level on file descriptor pFile to locktype. locktype
-** must be either NO_LOCK or SHARED_LOCK.
+** Here ends the implementation of all sqlite3_file methods.
**
-** If the locking level of the file descriptor is already at or below
-** the requested locking level, this routine is a no-op.
+********************** End sqlite3_file Methods *******************************
+******************************************************************************/
+
+/*
+** This division contains definitions of sqlite3_io_methods objects that
+** implement various file locking strategies. It also contains definitions
+** of "finder" functions. A finder-function is used to locate the appropriate
+** sqlite3_io_methods object for a particular database file. The pAppData
+** field of the sqlite3_vfs VFS objects are initialized to be pointers to
+** the correct finder-function for that VFS.
+**
+** Most finder functions return a pointer to a fixed sqlite3_io_methods
+** object. The only interesting finder-function is autolockIoFinder, which
+** looks at the filesystem type and tries to guess the best locking
+** strategy from that.
+**
+**
+** Each instance of this macro generates two objects:
+**
+** * A constant sqlite3_io_methods object call METHOD that has locking
+** methods CLOSE, LOCK, UNLOCK, CKRESLOCK.
+**
+** * An I/O method finder function called FINDER that returns a pointer
+** to the METHOD object in the previous bullet.
*/
-static int proxyUnlock(sqlite3_file *id, int locktype) {
- unixFile *pFile = (unixFile*)id;
- int rc = proxyTakeConch(pFile);
- if( rc==SQLITE_OK ){
- proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
- unixFile *proxy = pCtx->lockProxy;
- rc = proxy->pMethod->xUnlock((sqlite3_file*)proxy, locktype);
- pFile->locktype = proxy->locktype;
- }
- return rc;
+#define IOMETHODS(FINDER, METHOD, CLOSE, LOCK, UNLOCK, CKLOCK) \
+static const sqlite3_io_methods METHOD = { \
+ 1, /* iVersion */ \
+ CLOSE, /* xClose */ \
+ unixRead, /* xRead */ \
+ unixWrite, /* xWrite */ \
+ unixTruncate, /* xTruncate */ \
+ unixSync, /* xSync */ \
+ unixFileSize, /* xFileSize */ \
+ LOCK, /* xLock */ \
+ UNLOCK, /* xUnlock */ \
+ CKLOCK, /* xCheckReservedLock */ \
+ unixFileControl, /* xFileControl */ \
+ unixSectorSize, /* xSectorSize */ \
+ unixDeviceCharacteristics /* xDeviceCapabilities */ \
+}; \
+static const sqlite3_io_methods *FINDER(const char *z, int h){ \
+ UNUSED_PARAMETER(z); UNUSED_PARAMETER(h); \
+ return &METHOD; \
}
/*
-** Close a file that uses proxy locks.
+** Here are all of the sqlite3_io_methods objects for each of the
+** locking strategies. Functions that return pointers to these methods
+** are also created.
*/
-static int proxyClose(sqlite3_file *id) {
- if( id ){
- unixFile *pFile = (unixFile*)id;
- proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
- unixFile *lockProxy = pCtx->lockProxy;
- unixFile *conchFile = pCtx->conchFile;
- int rc = SQLITE_OK;
-
- if( lockProxy ){
- rc = lockProxy->pMethod->xUnlock((sqlite3_file*)lockProxy, NO_LOCK);
- if( rc ) return rc;
- rc = lockProxy->pMethod->xClose((sqlite3_file*)lockProxy);
- if( rc ) return rc;
- sqlite3_free(lockProxy);
- pCtx->lockProxy = 0;
+IOMETHODS(
+ posixIoFinder, /* Finder function name */
+ posixIoMethods, /* sqlite3_io_methods object name */
+ unixClose, /* xClose method */
+ unixLock, /* xLock method */
+ unixUnlock, /* xUnlock method */
+ unixCheckReservedLock /* xCheckReservedLock method */
+);
+IOMETHODS(
+ nolockIoFinder, /* Finder function name */
+ nolockIoMethods, /* sqlite3_io_methods object name */
+ nolockClose, /* xClose method */
+ nolockLock, /* xLock method */
+ nolockUnlock, /* xUnlock method */
+ nolockCheckReservedLock /* xCheckReservedLock method */
+);
+IOMETHODS(
+ dotlockIoFinder, /* Finder function name */
+ dotlockIoMethods, /* sqlite3_io_methods object name */
+ dotlockClose, /* xClose method */
+ dotlockLock, /* xLock method */
+ dotlockUnlock, /* xUnlock method */
+ dotlockCheckReservedLock /* xCheckReservedLock method */
+);
+
+#if SQLITE_ENABLE_LOCKING_STYLE
+IOMETHODS(
+ flockIoFinder, /* Finder function name */
+ flockIoMethods, /* sqlite3_io_methods object name */
+ flockClose, /* xClose method */
+ flockLock, /* xLock method */
+ flockUnlock, /* xUnlock method */
+ flockCheckReservedLock /* xCheckReservedLock method */
+);
+#endif
+
+#if OS_VXWORKS
+IOMETHODS(
+ semIoFinder, /* Finder function name */
+ semIoMethods, /* sqlite3_io_methods object name */
+ semClose, /* xClose method */
+ semLock, /* xLock method */
+ semUnlock, /* xUnlock method */
+ semCheckReservedLock /* xCheckReservedLock method */
+);
+#endif
+
+#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
+IOMETHODS(
+ afpIoFinder, /* Finder function name */
+ afpIoMethods, /* sqlite3_io_methods object name */
+ afpClose, /* xClose method */
+ afpLock, /* xLock method */
+ afpUnlock, /* xUnlock method */
+ afpCheckReservedLock /* xCheckReservedLock method */
+);
+#endif
+
+/*
+** The proxy locking method is a "super-method" in the sense that it
+** opens secondary file descriptors for the conch and lock files and
+** it uses proxy, dot-file, AFP, and flock() locking methods on those
+** secondary files. For this reason, the division that implements
+** proxy locking is located much further down in the file. But we need
+** to go ahead and define the sqlite3_io_methods and finder function
+** for proxy locking here. So we forward declare the I/O methods.
+*/
+#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
+static int proxyClose(sqlite3_file*);
+static int proxyLock(sqlite3_file*, int);
+static int proxyUnlock(sqlite3_file*, int);
+static int proxyCheckReservedLock(sqlite3_file*, int*);
+IOMETHODS(
+ proxyIoFinder, /* Finder function name */
+ proxyIoMethods, /* sqlite3_io_methods object name */
+ proxyClose, /* xClose method */
+ proxyLock, /* xLock method */
+ proxyUnlock, /* xUnlock method */
+ proxyCheckReservedLock /* xCheckReservedLock method */
+);
+#endif
+
+
+#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
+/*
+** This "finder" function attempts to determine the best locking strategy
+** for the database file "filePath". It then returns the sqlite3_io_methods
+** object that implements that strategy.
+**
+** This is for MacOSX only.
+*/
+static const sqlite3_io_methods *autolockIoFinder(
+ const char *filePath, /* name of the database file */
+ int fd /* file descriptor open on the database file */
+){
+ static const struct Mapping {
+ const char *zFilesystem; /* Filesystem type name */
+ const sqlite3_io_methods *pMethods; /* Appropriate locking method */
+ } aMap[] = {
+ { "hfs", &posixIoMethods },
+ { "ufs", &posixIoMethods },
+ { "afpfs", &afpIoMethods },
+#ifdef SQLITE_ENABLE_AFP_LOCKING_SMB
+ { "smbfs", &afpIoMethods },
+#else
+ { "smbfs", &flockIoMethods },
+#endif
+ { "webdav", &nolockIoMethods },
+ { 0, 0 }
+ };
+ int i;
+ struct statfs fsInfo;
+ struct flock lockInfo;
+
+ if( !filePath ){
+ /* If filePath==NULL that means we are dealing with a transient file
+ ** that does not need to be locked. */
+ return &nolockIoMethods;
+ }
+ if( statfs(filePath, &fsInfo) != -1 ){
+ if( fsInfo.f_flags & MNT_RDONLY ){
+ return &nolockIoMethods;
}
- if( conchFile ){
- if( pCtx->conchHeld ){
- rc = proxyReleaseConch(pFile);
- if( rc ) return rc;
+ for(i=0; aMap[i].zFilesystem; i++){
+ if( strcmp(fsInfo.f_fstypename, aMap[i].zFilesystem)==0 ){
+ return aMap[i].pMethods;
}
- rc = conchFile->pMethod->xClose((sqlite3_file*)conchFile);
- if( rc ) return rc;
- sqlite3_free(conchFile);
}
- sqlite3_free(pCtx->lockProxyPath);
- sqlite3_free(pCtx->conchFilePath);
- sqlite3_free(pCtx->dbPath);
- /* restore the original locking context and pMethod then close it */
- pFile->lockingContext = pCtx->oldLockingContext;
- pFile->pMethod = pCtx->pOldMethod;
- sqlite3_free(pCtx);
- return pFile->pMethod->xClose(id);
}
- return SQLITE_OK;
-}
-
-
+ /* Default case. Handles, amongst others, "nfs".
+ ** Test byte-range lock using fcntl(). If the call succeeds,
+ ** assume that the file-system supports POSIX style locks.
+ */
+ lockInfo.l_len = 1;
+ lockInfo.l_start = 0;
+ lockInfo.l_whence = SEEK_SET;
+ lockInfo.l_type = F_RDLCK;
+ if( fcntl(fd, F_GETLK, &lockInfo)!=-1 ) {
+ return &posixIoMethods;
+ }else{
+ return &dotlockIoMethods;
+ }
+}
#endif /* defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE */
+
/*
-** The proxy locking style is intended for use with AFP filesystems.
-** And since AFP is only supported on MacOSX, the proxy locking is also
-** restricted to MacOSX.
-**
-**
-******************* End of the proxy lock implementation **********************
-******************************************************************************/
+** An abstract type for a pointer to a IO method finder function:
+*/
+typedef const sqlite3_io_methods *(*finder_type)(const char*,int);
-/******************************************************************************
-**************** Non-locking sqlite3_file methods *****************************
+/****************************************************************************
+**************************** sqlite3_vfs methods ****************************
**
-** The next division contains implementations for all methods of the
-** sqlite3_file object other than the locking methods. The locking
-** methods were defined in divisions above (one locking method per
-** division). Those methods that are common to all locking modes
-** are gather together into this division.
+** This division contains the implementation of methods on the
+** sqlite3_vfs object.
*/
/*
-** Seek to the offset passed as the second argument, then read cnt
-** bytes into pBuf. Return the number of bytes actually read.
-**
-** NB: If you define USE_PREAD or USE_PREAD64, then it might also
-** be necessary to define _XOPEN_SOURCE to be 500. This varies from
-** one system to another. Since SQLite does not define USE_PREAD
-** any any form by default, we will not attempt to define _XOPEN_SOURCE.
-** See tickets #2741 and #2681.
-**
-** To avoid stomping the errno value on a failed read the lastErrno value
-** is set before returning.
+** Initialize the contents of the unixFile structure pointed to by pId.
*/
-static int seekAndRead(unixFile *id, sqlite3_int64 offset, void *pBuf, int cnt){
- int got;
- i64 newOffset;
- TIMER_START;
-#if defined(USE_PREAD)
- got = pread(id->h, pBuf, cnt, offset);
- SimulateIOError( got = -1 );
-#elif defined(USE_PREAD64)
- got = pread64(id->h, pBuf, cnt, offset);
- SimulateIOError( got = -1 );
-#else
- newOffset = lseek(id->h, offset, SEEK_SET);
- SimulateIOError( newOffset-- );
- if( newOffset!=offset ){
- if( newOffset == -1 ){
- ((unixFile*)id)->lastErrno = errno;
- }else{
- ((unixFile*)id)->lastErrno = 0;
- }
- return -1;
- }
- got = read(id->h, pBuf, cnt);
+static int fillInUnixFile(
+ sqlite3_vfs *pVfs, /* Pointer to vfs object */
+ int h, /* Open file descriptor of file being opened */
+ int dirfd, /* Directory file descriptor */
+ sqlite3_file *pId, /* Write to the unixFile structure here */
+ const char *zFilename, /* Name of the file being opened */
+ int noLock, /* Omit locking if true */
+ int isDelete /* Delete on close if true */
+){
+ const sqlite3_io_methods *pLockingStyle;
+ unixFile *pNew = (unixFile *)pId;
+ int rc = SQLITE_OK;
+
+ assert( pNew->pLock==NULL );
+ assert( pNew->pOpen==NULL );
+
+ /* Parameter isDelete is only used on vxworks.
+ ** Express this explicitly here to prevent compiler warnings
+ ** about unused parameters.
+ */
+#if !OS_VXWORKS
+ UNUSED_PARAMETER(isDelete);
#endif
- TIMER_END;
- if( got<0 ){
- ((unixFile*)id)->lastErrno = errno;
+
+ OSTRACE3("OPEN %-3d %s\n", h, zFilename);
+ pNew->h = h;
+ pNew->dirfd = dirfd;
+ SET_THREADID(pNew);
+
+#if OS_VXWORKS
+ pNew->pId = vxworksFindFileId(zFilename);
+ if( pNew->pId==0 ){
+ noLock = 1;
+ rc = SQLITE_NOMEM;
}
- OSTRACE5("READ %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
- return got;
-}
+#endif
-/*
-** Read data from a file into a buffer. Return SQLITE_OK if all
-** bytes were read successfully and SQLITE_IOERR if anything goes
-** wrong.
-*/
-static int unixRead(
- sqlite3_file *id,
- void *pBuf,
- int amt,
- sqlite3_int64 offset
-){
- int got;
- assert( id );
- got = seekAndRead((unixFile*)id, offset, pBuf, amt);
- if( got==amt ){
- return SQLITE_OK;
- }else if( got<0 ){
- /* lastErrno set by seekAndRead */
- return SQLITE_IOERR_READ;
+ if( noLock ){
+ pLockingStyle = &nolockIoMethods;
}else{
- ((unixFile*)id)->lastErrno = 0; /* not a system error */
- /* Unread parts of the buffer must be zero-filled */
- memset(&((char*)pBuf)[got], 0, amt-got);
- return SQLITE_IOERR_SHORT_READ;
+ pLockingStyle = (*(finder_type)pVfs->pAppData)(zFilename, h);
+#if SQLITE_ENABLE_LOCKING_STYLE
+ /* Cache zFilename in the locking context (AFP and dotlock override) for
+ ** proxyLock activation is possible (remote proxy is based on db name)
+ ** zFilename remains valid until file is closed, to support */
+ pNew->lockingContext = (void*)zFilename;
+#endif
}
-}
-/*
-** Seek to the offset in id->offset then read cnt bytes into pBuf.
-** Return the number of bytes actually read. Update the offset.
-**
-** To avoid stomping the errno value on a failed write the lastErrno value
-** is set before returning.
-*/
-static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
- int got;
- i64 newOffset;
- TIMER_START;
-#if defined(USE_PREAD)
- got = pwrite(id->h, pBuf, cnt, offset);
-#elif defined(USE_PREAD64)
- got = pwrite64(id->h, pBuf, cnt, offset);
-#else
- newOffset = lseek(id->h, offset, SEEK_SET);
- if( newOffset!=offset ){
- if( newOffset == -1 ){
- ((unixFile*)id)->lastErrno = errno;
+ if( pLockingStyle == &posixIoMethods ){
+ unixEnterMutex();
+ rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
+ unixLeaveMutex();
+ }
+
+#if SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__)
+ else if( pLockingStyle == &afpIoMethods ){
+ /* AFP locking uses the file path so it needs to be included in
+ ** the afpLockingContext.
+ */
+ afpLockingContext *pCtx;
+ pNew->lockingContext = pCtx = sqlite3_malloc( sizeof(*pCtx) );
+ if( pCtx==0 ){
+ rc = SQLITE_NOMEM;
}else{
- ((unixFile*)id)->lastErrno = 0;
+ /* NB: zFilename exists and remains valid until the file is closed
+ ** according to requirement F11141. So we do not need to make a
+ ** copy of the filename. */
+ pCtx->dbPath = zFilename;
+ srandomdev();
+ unixEnterMutex();
+ rc = findLockInfo(pNew, NULL, &pNew->pOpen);
+ unixLeaveMutex();
}
- return -1;
}
- got = write(id->h, pBuf, cnt);
#endif
- TIMER_END;
- if( got<0 ){
- ((unixFile*)id)->lastErrno = errno;
- }
-
- OSTRACE5("WRITE %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED);
- return got;
-}
-
-/*
-** Write data from a buffer into a file. Return SQLITE_OK on success
-** or some other error code on failure.
-*/
-static int unixWrite(
- sqlite3_file *id,
- const void *pBuf,
- int amt,
- sqlite3_int64 offset
-){
- int wrote = 0;
- assert( id );
- assert( amt>0 );
- while( amt>0 && (wrote = seekAndWrite((unixFile*)id, offset, pBuf, amt))>0 ){
- amt -= wrote;
- offset += wrote;
- pBuf = &((char*)pBuf)[wrote];
- }
- SimulateIOError(( wrote=(-1), amt=1 ));
- SimulateDiskfullError(( wrote=0, amt=1 ));
- if( amt>0 ){
- if( wrote<0 ){
- /* lastErrno set by seekAndWrite */
- return SQLITE_IOERR_WRITE;
+#if SQLITE_ENABLE_LOCKING_STYLE
+ else if( pLockingStyle == &dotlockIoMethods ){
+ /* Dotfile locking uses the file path so it needs to be included in
+ ** the dotlockLockingContext
+ */
+ char *zLockFile;
+ int nFilename;
+ nFilename = strlen(zFilename) + 6;
+ zLockFile = (char *)sqlite3_malloc(nFilename);
+ if( zLockFile==0 ){
+ rc = SQLITE_NOMEM;
}else{
- ((unixFile*)id)->lastErrno = 0; /* not a system error */
- return SQLITE_FULL;
+ sqlite3_snprintf(nFilename, zLockFile, "%s" DOTLOCK_SUFFIX, zFilename);
}
+ pNew->lockingContext = zLockFile;
}
- return SQLITE_OK;
-}
-
-#ifdef SQLITE_TEST
-/*
-** Count the number of fullsyncs and normal syncs. This is used to test
-** that syncs and fullsyncs are occurring at the right times.
-*/
-int sqlite3_sync_count = 0;
-int sqlite3_fullsync_count = 0;
#endif
-/*
-** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.
-** Otherwise use fsync() in its place.
-*/
-#ifndef HAVE_FDATASYNC
-# define fdatasync fsync
+#if OS_VXWORKS
+ else if( pLockingStyle == &semIoMethods ){
+ /* Named semaphore locking uses the file path so it needs to be
+ ** included in the semLockingContext
+ */
+ unixEnterMutex();
+ rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
+ if( (rc==SQLITE_OK) && (pNew->pOpen->pSem==NULL) ){
+ char *zSemName = pNew->pOpen->aSemName;
+ int n;
+ sqlite3_snprintf(MAX_PATHNAME, zSemName, "%s.sem",
+ pNew->pId->zCanonicalName);
+ for( n=0; zSemName[n]; n++ )
+ if( zSemName[n]=='/' ) zSemName[n] = '_';
+ pNew->pOpen->pSem = sem_open(zSemName, O_CREAT, 0666, 1);
+ if( pNew->pOpen->pSem == SEM_FAILED ){
+ rc = SQLITE_NOMEM;
+ pNew->pOpen->aSemName[0] = '\0';
+ }
+ }
+ unixLeaveMutex();
+ }
+#endif
+
+ pNew->lastErrno = 0;
+#if OS_VXWORKS
+ if( rc!=SQLITE_OK ){
+ unlink(zFilename);
+ isDelete = 0;
+ }
+ pNew->isDelete = isDelete;
#endif
+ if( rc!=SQLITE_OK ){
+ if( dirfd>=0 ) close(dirfd); /* silent leak if fail, already in error */
+ close(h);
+ }else{
+ pNew->pMethod = pLockingStyle;
+ OpenCounter(+1);
+ }
+ return rc;
+}
/*
-** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
-** the F_FULLFSYNC macro is defined. F_FULLFSYNC is currently
-** only available on Mac OS X. But that could change.
+** Open a file descriptor to the directory containing file zFilename.
+** If successful, *pFd is set to the opened file descriptor and
+** SQLITE_OK is returned. If an error occurs, either SQLITE_NOMEM
+** or SQLITE_CANTOPEN is returned and *pFd is set to an undefined
+** value.
+**
+** If SQLITE_OK is returned, the caller is responsible for closing
+** the file descriptor *pFd using close().
*/
-#ifdef F_FULLFSYNC
-# define HAVE_FULLFSYNC 1
-#else
-# define HAVE_FULLFSYNC 0
-#endif
+static int openDirectory(const char *zFilename, int *pFd){
+ int ii;
+ int fd = -1;
+ char zDirname[MAX_PATHNAME+1];
+ sqlite3_snprintf(MAX_PATHNAME, zDirname, "%s", zFilename);
+ for(ii=strlen(zDirname); ii>=0 && zDirname[ii]!='/'; ii--);
+ if( ii>0 ){
+ zDirname[ii] = '\0';
+ fd = open(zDirname, O_RDONLY|O_BINARY, 0);
+ if( fd>=0 ){
+#ifdef FD_CLOEXEC
+ fcntl(fd, F_SETFD, fcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
+#endif
+ OSTRACE3("OPENDIR %-3d %s\n", fd, zDirname);
+ }
+ }
+ *pFd = fd;
+ return (fd>=0?SQLITE_OK:SQLITE_CANTOPEN);
+}
/*
-** The fsync() system call does not work as advertised on many
-** unix systems. The following procedure is an attempt to make
-** it work better.
-**
-** The SQLITE_NO_SYNC macro disables all fsync()s. This is useful
-** for testing when we want to run through the test suite quickly.
-** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
-** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
-** or power failure will likely corrupt the database file.
+** Create a temporary file name in zBuf. zBuf must be allocated
+** by the calling process and must be big enough to hold at least
+** pVfs->mxPathname bytes.
*/
-static int full_fsync(int fd, int fullSync, int dataOnly){
- int rc;
+static int getTempname(int nBuf, char *zBuf){
+ static const char *azDirs[] = {
+ 0,
+ 0,
+ "/var/tmp",
+ "/usr/tmp",
+ "/tmp",
+ ".",
+ };
+ static const unsigned char zChars[] =
+ "abcdefghijklmnopqrstuvwxyz"
+ "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+ "0123456789";
+ unsigned int i, j;
+ struct stat buf;
+ const char *zDir = ".";
- /* The following "ifdef/elif/else/" block has the same structure as
- ** the one below. It is replicated here solely to avoid cluttering
- ** up the real code with the UNUSED_PARAMETER() macros.
+ /* It's odd to simulate an io-error here, but really this is just
+ ** using the io-error infrastructure to test that SQLite handles this
+ ** function failing.
*/
-#ifdef SQLITE_NO_SYNC
- UNUSED_PARAMETER(fd);
- UNUSED_PARAMETER(fullSync);
- UNUSED_PARAMETER(dataOnly);
-#elif HAVE_FULLFSYNC
- UNUSED_PARAMETER(dataOnly);
-#else
- UNUSED_PARAMETER(fullSync);
-#endif
+ SimulateIOError( return SQLITE_IOERR );
- /* Record the number of times that we do a normal fsync() and
- ** FULLSYNC. This is used during testing to verify that this procedure
- ** gets called with the correct arguments.
- */
-#ifdef SQLITE_TEST
- if( fullSync ) sqlite3_fullsync_count++;
- sqlite3_sync_count++;
-#endif
+ azDirs[0] = sqlite3_temp_directory;
+ if (NULL == azDirs[1]) {
+ azDirs[1] = getenv("TMPDIR");
+ }
+
+ for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); i++){
+ if( azDirs[i]==0 ) continue;
+ if( stat(azDirs[i], &buf) ) continue;
+ if( !S_ISDIR(buf.st_mode) ) continue;
+ if( access(azDirs[i], 07) ) continue;
+ zDir = azDirs[i];
+ break;
+ }
- /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
- ** no-op
+ /* Check that the output buffer is large enough for the temporary file
+ ** name. If it is not, return SQLITE_ERROR.
*/
-#ifdef SQLITE_NO_SYNC
- rc = SQLITE_OK;
-#elif HAVE_FULLFSYNC
- if( fullSync ){
- rc = fcntl(fd, F_FULLFSYNC, 0);
- }else{
- rc = 1;
+ if( (strlen(zDir) + strlen(SQLITE_TEMP_FILE_PREFIX) + 17) >= (size_t)nBuf ){
+ return SQLITE_ERROR;
}
- /* If the FULLFSYNC failed, fall back to attempting an fsync().
- ** It shouldn't be possible for fullfsync to fail on the local
- ** file system (on OSX), so failure indicates that FULLFSYNC
- ** isn't supported for this file system. So, attempt an fsync
- ** and (for now) ignore the overhead of a superfluous fcntl call.
- ** It'd be better to detect fullfsync support once and avoid
- ** the fcntl call every time sync is called.
- */
- if( rc ) rc = fsync(fd);
-#else
- if( dataOnly ){
- rc = fdatasync(fd);
- if( OS_VXWORKS && rc==-1 && errno==ENOTSUP ){
- rc = fsync(fd);
+ do{
+ sqlite3_snprintf(nBuf-17, zBuf, "%s/"SQLITE_TEMP_FILE_PREFIX, zDir);
+ j = strlen(zBuf);
+ sqlite3_randomness(15, &zBuf[j]);
+ for(i=0; i<15; i++, j++){
+ zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
}
- }else{
- rc = fsync(fd);
- }
-#endif /* ifdef SQLITE_NO_SYNC elif HAVE_FULLFSYNC */
-
- if( OS_VXWORKS && rc!= -1 ){
- rc = 0;
- }
- return rc;
+ zBuf[j] = 0;
+ }while( access(zBuf,0)==0 );
+ return SQLITE_OK;
}
+
/*
-** Make sure all writes to a particular file are committed to disk.
+** Open the file zPath.
+**
+** Previously, the SQLite OS layer used three functions in place of this
+** one:
**
-** If dataOnly==0 then both the file itself and its metadata (file
-** size, access time, etc) are synced. If dataOnly!=0 then only the
-** file data is synced.
+** sqlite3OsOpenReadWrite();
+** sqlite3OsOpenReadOnly();
+** sqlite3OsOpenExclusive();
**
-** Under Unix, also make sure that the directory entry for the file
-** has been created by fsync-ing the directory that contains the file.
-** If we do not do this and we encounter a power failure, the directory
-** entry for the journal might not exist after we reboot. The next
-** SQLite to access the file will not know that the journal exists (because
-** the directory entry for the journal was never created) and the transaction
-** will not roll back - possibly leading to database corruption.
+** These calls correspond to the following combinations of flags:
+**
+** ReadWrite() -> (READWRITE | CREATE)
+** ReadOnly() -> (READONLY)
+** OpenExclusive() -> (READWRITE | CREATE | EXCLUSIVE)
+**
+** The old OpenExclusive() accepted a boolean argument - "delFlag". If
+** true, the file was configured to be automatically deleted when the
+** file handle closed. To achieve the same effect using this new
+** interface, add the DELETEONCLOSE flag to those specified above for
+** OpenExclusive().
*/
-static int unixSync(sqlite3_file *id, int flags){
- int rc;
- unixFile *pFile = (unixFile*)id;
+static int unixOpen(
+ sqlite3_vfs *pVfs, /* The VFS for which this is the xOpen method */
+ const char *zPath, /* Pathname of file to be opened */
+ sqlite3_file *pFile, /* The file descriptor to be filled in */
+ int flags, /* Input flags to control the opening */
+ int *pOutFlags /* Output flags returned to SQLite core */
+){
+ int fd = 0; /* File descriptor returned by open() */
+ int dirfd = -1; /* Directory file descriptor */
+ int openFlags = 0; /* Flags to pass to open() */
+ int eType = flags&0xFFFFFF00; /* Type of file to open */
+ int noLock; /* True to omit locking primitives */
+ int rc = SQLITE_OK;
- int isDataOnly = (flags&SQLITE_SYNC_DATAONLY);
- int isFullsync = (flags&0x0F)==SQLITE_SYNC_FULL;
+ int isExclusive = (flags & SQLITE_OPEN_EXCLUSIVE);
+ int isDelete = (flags & SQLITE_OPEN_DELETEONCLOSE);
+ int isCreate = (flags & SQLITE_OPEN_CREATE);
+ int isReadonly = (flags & SQLITE_OPEN_READONLY);
+ int isReadWrite = (flags & SQLITE_OPEN_READWRITE);
- /* Check that one of SQLITE_SYNC_NORMAL or FULL was passed */
- assert((flags&0x0F)==SQLITE_SYNC_NORMAL
- || (flags&0x0F)==SQLITE_SYNC_FULL
+ /* If creating a master or main-file journal, this function will open
+ ** a file-descriptor on the directory too. The first time unixSync()
+ ** is called the directory file descriptor will be fsync()ed and close()d.
+ */
+ int isOpenDirectory = (isCreate &&
+ (eType==SQLITE_OPEN_MASTER_JOURNAL || eType==SQLITE_OPEN_MAIN_JOURNAL)
);
- /* Unix cannot, but some systems may return SQLITE_FULL from here. This
- ** line is to test that doing so does not cause any problems.
+ /* If argument zPath is a NULL pointer, this function is required to open
+ ** a temporary file. Use this buffer to store the file name in.
*/
- SimulateDiskfullError( return SQLITE_FULL );
+ char zTmpname[MAX_PATHNAME+1];
+ const char *zName = zPath;
- assert( pFile );
- OSTRACE2("SYNC %-3d\n", pFile->h);
- rc = full_fsync(pFile->h, isFullsync, isDataOnly);
- SimulateIOError( rc=1 );
- if( rc ){
- pFile->lastErrno = errno;
- return SQLITE_IOERR_FSYNC;
+ /* Check the following statements are true:
+ **
+ ** (a) Exactly one of the READWRITE and READONLY flags must be set, and
+ ** (b) if CREATE is set, then READWRITE must also be set, and
+ ** (c) if EXCLUSIVE is set, then CREATE must also be set.
+ ** (d) if DELETEONCLOSE is set, then CREATE must also be set.
+ */
+ assert((isReadonly==0 || isReadWrite==0) && (isReadWrite || isReadonly));
+ assert(isCreate==0 || isReadWrite);
+ assert(isExclusive==0 || isCreate);
+ assert(isDelete==0 || isCreate);
+
+ /* The main DB, main journal, and master journal are never automatically
+ ** deleted
+ */
+ assert( eType!=SQLITE_OPEN_MAIN_DB || !isDelete );
+ assert( eType!=SQLITE_OPEN_MAIN_JOURNAL || !isDelete );
+ assert( eType!=SQLITE_OPEN_MASTER_JOURNAL || !isDelete );
+
+ /* Assert that the upper layer has set one of the "file-type" flags. */
+ assert( eType==SQLITE_OPEN_MAIN_DB || eType==SQLITE_OPEN_TEMP_DB
+ || eType==SQLITE_OPEN_MAIN_JOURNAL || eType==SQLITE_OPEN_TEMP_JOURNAL
+ || eType==SQLITE_OPEN_SUBJOURNAL || eType==SQLITE_OPEN_MASTER_JOURNAL
+ || eType==SQLITE_OPEN_TRANSIENT_DB
+ );
+
+ memset(pFile, 0, sizeof(unixFile));
+
+ if( !zName ){
+ assert(isDelete && !isOpenDirectory);
+ rc = getTempname(MAX_PATHNAME+1, zTmpname);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ zName = zTmpname;
}
- if( pFile->dirfd>=0 ){
- int err;
- OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,
- HAVE_FULLFSYNC, isFullsync);
-#ifndef SQLITE_DISABLE_DIRSYNC
- /* The directory sync is only attempted if full_fsync is
- ** turned off or unavailable. If a full_fsync occurred above,
- ** then the directory sync is superfluous.
- */
- if( (!HAVE_FULLFSYNC || !isFullsync) && full_fsync(pFile->dirfd,0,0) ){
- /*
- ** We have received multiple reports of fsync() returning
- ** errors when applied to directories on certain file systems.
- ** A failed directory sync is not a big deal. So it seems
- ** better to ignore the error. Ticket #1657
- */
- /* pFile->lastErrno = errno; */
- /* return SQLITE_IOERR; */
- }
+
+ if( isReadonly ) openFlags |= O_RDONLY;
+ if( isReadWrite ) openFlags |= O_RDWR;
+ if( isCreate ) openFlags |= O_CREAT;
+ if( isExclusive ) openFlags |= (O_EXCL|O_NOFOLLOW);
+ openFlags |= (O_LARGEFILE|O_BINARY);
+
+ fd = open(zName, openFlags, isDelete?0600:SQLITE_DEFAULT_FILE_PERMISSIONS);
+ OSTRACE4("OPENX %-3d %s 0%o\n", fd, zName, openFlags);
+ if( fd<0 && errno!=EISDIR && isReadWrite && !isExclusive ){
+ /* Failed to open the file for read/write access. Try read-only. */
+ flags &= ~(SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE);
+ flags |= SQLITE_OPEN_READONLY;
+ return unixOpen(pVfs, zPath, pFile, flags, pOutFlags);
+ }
+ if( fd<0 ){
+ return SQLITE_CANTOPEN;
+ }
+ if( isDelete ){
+#if OS_VXWORKS
+ zPath = zName;
+#else
+ unlink(zName);
#endif
- err = close(pFile->dirfd); /* Only need to sync once, so close the */
- if( err==0 ){ /* directory when we are done */
- pFile->dirfd = -1;
- }else{
- pFile->lastErrno = errno;
- rc = SQLITE_IOERR_DIR_CLOSE;
- }
}
- return rc;
-}
-
-/*
-** Truncate an open file to a specified size
-*/
-static int unixTruncate(sqlite3_file *id, i64 nByte){
- int rc;
- assert( id );
- SimulateIOError( return SQLITE_IOERR_TRUNCATE );
- rc = ftruncate(((unixFile*)id)->h, (off_t)nByte);
- if( rc ){
- ((unixFile*)id)->lastErrno = errno;
- return SQLITE_IOERR_TRUNCATE;
- }else{
- return SQLITE_OK;
+#if SQLITE_ENABLE_LOCKING_STYLE
+ else{
+ ((unixFile*)pFile)->openFlags = openFlags;
+ }
+#endif
+ if( pOutFlags ){
+ *pOutFlags = flags;
}
-}
-/*
-** Determine the current size of a file in bytes
-*/
-static int unixFileSize(sqlite3_file *id, i64 *pSize){
- int rc;
- struct stat buf;
- assert( id );
- rc = fstat(((unixFile*)id)->h, &buf);
- SimulateIOError( rc=1 );
- if( rc!=0 ){
- ((unixFile*)id)->lastErrno = errno;
- return SQLITE_IOERR_FSTAT;
+ assert(fd!=0);
+ if( isOpenDirectory ){
+ rc = openDirectory(zPath, &dirfd);
+ if( rc!=SQLITE_OK ){
+ close(fd); /* silently leak if fail, already in error */
+ return rc;
+ }
}
- *pSize = buf.st_size;
- /* When opening a zero-size database, the findLockInfo() procedure
- ** writes a single byte into that file in order to work around a bug
- ** in the OS-X msdos filesystem. In order to avoid problems with upper
- ** layers, we need to report this file size as zero even though it is
- ** really 1. Ticket #3260.
- */
- if( *pSize==1 ) *pSize = 0;
+#ifdef FD_CLOEXEC
+ fcntl(fd, F_SETFD, fcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
+#endif
+ noLock = eType!=SQLITE_OPEN_MAIN_DB;
- return SQLITE_OK;
-}
+#if SQLITE_PREFER_PROXY_LOCKING
+ if( zPath!=NULL && !noLock ){
+ char *envforce = getenv("SQLITE_FORCE_PROXY_LOCKING");
+ int useProxy = 0;
+ /* SQLITE_FORCE_PROXY_LOCKING==1 means force always use proxy,
+ ** 0 means never use proxy, NULL means use proxy for non-local files only
+ */
+ if( envforce!=NULL ){
+ useProxy = atoi(envforce)>0;
+ }else{
+ struct statfs fsInfo;
-/*
-** Information and control of an open file handle.
-*/
-static int unixFileControl(sqlite3_file *id, int op, void *pArg){
- switch( op ){
- case SQLITE_FCNTL_LOCKSTATE: {
- *(int*)pArg = ((unixFile*)id)->locktype;
- return SQLITE_OK;
- }
- case SQLITE_LAST_ERRNO: {
- *(int*)pArg = ((unixFile*)id)->lastErrno;
- return SQLITE_OK;
- }
-#if SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__)
- case SQLITE_GET_LOCKPROXYFILE: {
- unixFile *pFile = (unixFile*)id;
- if( pFile->pMethod == &proxyIoMethods ){
- proxyLockingContext *pCtx = (proxyLockingContext*)pFile->lockingContext;
- proxyTakeConch(pFile);
- if( pCtx->lockProxyPath ){
- *(const char **)pArg = pCtx->lockProxyPath;
- }else{
- *(const char **)pArg = ":auto: (not held)";
- }
- } else {
- *(const char **)pArg = NULL;
+ if( statfs(zPath, &fsInfo) == -1 ){
+ ((unixFile*)pFile)->lastErrno = errno;
+ if( dirfd>=0 ) close(dirfd); /* silently leak if fail, in error */
+ close(fd); /* silently leak if fail, in error */
+ return SQLITE_IOERR_ACCESS;
}
- return SQLITE_OK;
+ useProxy = !(fsInfo.f_flags&MNT_LOCAL);
}
- case SQLITE_SET_LOCKPROXYFILE: {
- unixFile *pFile = (unixFile*)id;
- int rc = SQLITE_OK;
- int isProxyStyle = (pFile->pMethod == &proxyIoMethods);
- if( pArg==NULL || (const char *)pArg==0 ){
- if( isProxyStyle ){
- /* turn off proxy locking - not supported */
- rc = SQLITE_ERROR /*SQLITE_PROTOCOL? SQLITE_MISUSE?*/;
- }else{
- /* turn off proxy locking - already off - NOOP */
- rc = SQLITE_OK;
- }
- }else{
- const char *proxyPath = (const char *)pArg;
- if( isProxyStyle ){
- proxyLockingContext *pCtx =
- (proxyLockingContext*)pFile->lockingContext;
- if( !strcmp(pArg, ":auto:")
- || (pCtx->lockProxyPath &&
- !strncmp(pCtx->lockProxyPath, proxyPath, MAXPATHLEN))
- ){
- rc = SQLITE_OK;
- }else{
- rc = switchLockProxyPath(pFile, proxyPath);
- }
- }else{
- /* turn on proxy file locking */
- rc = transformUnixFileForLockProxy(pFile, proxyPath);
- }
+ if( useProxy ){
+ rc = fillInUnixFile(pVfs, fd, dirfd, pFile, zPath, noLock, isDelete);
+ if( rc==SQLITE_OK ){
+ rc = proxyTransformUnixFile((unixFile*)pFile, ":auto:");
}
return rc;
}
-#endif /* SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__) */
}
- return SQLITE_ERROR;
+#endif
+
+ return fillInUnixFile(pVfs, fd, dirfd, pFile, zPath, noLock, isDelete);
}
/*
-** Return the sector size in bytes of the underlying block device for
-** the specified file. This is almost always 512 bytes, but may be
-** larger for some devices.
-**
-** SQLite code assumes this function cannot fail. It also assumes that
-** if two files are created in the same file-system directory (i.e.
-** a database and its journal file) that the sector size will be the
-** same for both.
+** Delete the file at zPath. If the dirSync argument is true, fsync()
+** the directory after deleting the file.
*/
-static int unixSectorSize(sqlite3_file *NotUsed){
+static int unixDelete(
+ sqlite3_vfs *NotUsed, /* VFS containing this as the xDelete method */
+ const char *zPath, /* Name of file to be deleted */
+ int dirSync /* If true, fsync() directory after deleting file */
+){
+ int rc = SQLITE_OK;
UNUSED_PARAMETER(NotUsed);
- return SQLITE_DEFAULT_SECTOR_SIZE;
+ SimulateIOError(return SQLITE_IOERR_DELETE);
+ unlink(zPath);
+#ifndef SQLITE_DISABLE_DIRSYNC
+ if( dirSync ){
+ int fd;
+ rc = openDirectory(zPath, &fd);
+ if( rc==SQLITE_OK ){
+#if OS_VXWORKS
+ if( fsync(fd)==-1 )
+#else
+ if( fsync(fd) )
+#endif
+ {
+ rc = SQLITE_IOERR_DIR_FSYNC;
+ }
+ if( close(fd)&&!rc ){
+ rc = SQLITE_IOERR_DIR_CLOSE;
+ }
+ }
+ }
+#endif
+ return rc;
}
/*
-** Return the device characteristics for the file. This is always 0 for unix.
+** Test the existance of or access permissions of file zPath. The
+** test performed depends on the value of flags:
+**
+** SQLITE_ACCESS_EXISTS: Return 1 if the file exists
+** SQLITE_ACCESS_READWRITE: Return 1 if the file is read and writable.
+** SQLITE_ACCESS_READONLY: Return 1 if the file is readable.
+**
+** Otherwise return 0.
*/
-static int unixDeviceCharacteristics(sqlite3_file *NotUsed){
+static int unixAccess(
+ sqlite3_vfs *NotUsed, /* The VFS containing this xAccess method */
+ const char *zPath, /* Path of the file to examine */
+ int flags, /* What do we want to learn about the zPath file? */
+ int *pResOut /* Write result boolean here */
+){
+ int amode = 0;
UNUSED_PARAMETER(NotUsed);
- return 0;
-}
+ SimulateIOError( return SQLITE_IOERR_ACCESS; );
+ switch( flags ){
+ case SQLITE_ACCESS_EXISTS:
+ amode = F_OK;
+ break;
+ case SQLITE_ACCESS_READWRITE:
+ amode = W_OK|R_OK;
+ break;
+ case SQLITE_ACCESS_READ:
+ amode = R_OK;
+ break;
+
+ default:
+ assert(!"Invalid flags argument");
+ }
+ *pResOut = (access(zPath, amode)==0);
+ return SQLITE_OK;
+}
+
/*
-** Here ends the implementation of all sqlite3_file methods.
+** Turn a relative pathname into a full pathname. The relative path
+** is stored as a nul-terminated string in the buffer pointed to by
+** zPath.
**
-********************** End sqlite3_file Methods *******************************
-******************************************************************************/
+** zOut points to a buffer of at least sqlite3_vfs.mxPathname bytes
+** (in this case, MAX_PATHNAME bytes). The full-path is written to
+** this buffer before returning.
+*/
+static int unixFullPathname(
+ sqlite3_vfs *pVfs, /* Pointer to vfs object */
+ const char *zPath, /* Possibly relative input path */
+ int nOut, /* Size of output buffer in bytes */
+ char *zOut /* Output buffer */
+){
+
+ /* It's odd to simulate an io-error here, but really this is just
+ ** using the io-error infrastructure to test that SQLite handles this
+ ** function failing. This function could fail if, for example, the
+ ** current working directory has been unlinked.
+ */
+ SimulateIOError( return SQLITE_ERROR );
+
+ assert( pVfs->mxPathname==MAX_PATHNAME );
+ UNUSED_PARAMETER(pVfs);
+
+ zOut[nOut-1] = '\0';
+ if( zPath[0]=='/' ){
+ sqlite3_snprintf(nOut, zOut, "%s", zPath);
+ }else{
+ int nCwd;
+ if( getcwd(zOut, nOut-1)==0 ){
+ return SQLITE_CANTOPEN;
+ }
+ nCwd = strlen(zOut);
+ sqlite3_snprintf(nOut-nCwd, &zOut[nCwd], "/%s", zPath);
+ }
+ return SQLITE_OK;
+}
+
+#ifndef SQLITE_OMIT_LOAD_EXTENSION
/*
-** This division contains definitions of sqlite3_io_methods objects that
-** implement various file locking strategies. It also contains definitions
-** of "finder" functions. A finder-function is used to locate the appropriate
-** sqlite3_io_methods object for a particular database file. The pAppData
-** field of the sqlite3_vfs VFS objects are initialized to be pointers to
-** the correct finder-function for that VFS.
-**
-** Most finder functions return a pointer to a fixed sqlite3_io_methods
-** object. The only interesting finder-function is autolockIoFinder, which
-** looks at the filesystem type and tries to guess the best locking
-** strategy from that.
-**
-**
-** Each instance of this macro generates two objects:
-**
-** * A constant sqlite3_io_methods object call METHOD that has locking
-** methods CLOSE, LOCK, UNLOCK, CKRESLOCK.
-**
-** * An I/O method finder function called FINDER that returns a pointer
-** to the METHOD object in the previous bullet.
+** Interfaces for opening a shared library, finding entry points
+** within the shared library, and closing the shared library.
*/
-#define IOMETHODS(FINDER, METHOD, CLOSE, LOCK, UNLOCK, CKLOCK) \
-static const sqlite3_io_methods METHOD = { \
- 1, /* iVersion */ \
- CLOSE, /* xClose */ \
- unixRead, /* xRead */ \
- unixWrite, /* xWrite */ \
- unixTruncate, /* xTruncate */ \
- unixSync, /* xSync */ \
- unixFileSize, /* xFileSize */ \
- LOCK, /* xLock */ \
- UNLOCK, /* xUnlock */ \
- CKLOCK, /* xCheckReservedLock */ \
- unixFileControl, /* xFileControl */ \
- unixSectorSize, /* xSectorSize */ \
- unixDeviceCharacteristics /* xDeviceCapabilities */ \
-}; \
-static const sqlite3_io_methods *FINDER(const char *z, int h){ \
- UNUSED_PARAMETER(z); UNUSED_PARAMETER(h); \
- return &METHOD; \
+#include <dlfcn.h>
+static void *unixDlOpen(sqlite3_vfs *NotUsed, const char *zFilename){
+ UNUSED_PARAMETER(NotUsed);
+ return dlopen(zFilename, RTLD_NOW | RTLD_GLOBAL);
}
/*
-** Here are all of the sqlite3_io_methods objects for each of the
-** locking strategies. Functions that return pointers to these methods
-** are also created.
+** SQLite calls this function immediately after a call to unixDlSym() or
+** unixDlOpen() fails (returns a null pointer). If a more detailed error
+** message is available, it is written to zBufOut. If no error message
+** is available, zBufOut is left unmodified and SQLite uses a default
+** error message.
*/
-IOMETHODS(
- posixIoFinder, /* Finder function name */
- posixIoMethods, /* sqlite3_io_methods object name */
- unixClose, /* xClose method */
- unixLock, /* xLock method */
- unixUnlock, /* xUnlock method */
- unixCheckReservedLock /* xCheckReservedLock method */
-);
-IOMETHODS(
- nolockIoFinder, /* Finder function name */
- nolockIoMethods, /* sqlite3_io_methods object name */
- nolockClose, /* xClose method */
- nolockLock, /* xLock method */
- nolockUnlock, /* xUnlock method */
- nolockCheckReservedLock /* xCheckReservedLock method */
-);
-IOMETHODS(
- dotlockIoFinder, /* Finder function name */
- dotlockIoMethods, /* sqlite3_io_methods object name */
- dotlockClose, /* xClose method */
- dotlockLock, /* xLock method */
- dotlockUnlock, /* xUnlock method */
- dotlockCheckReservedLock /* xCheckReservedLock method */
-);
+static void unixDlError(sqlite3_vfs *NotUsed, int nBuf, char *zBufOut){
+ char *zErr;
+ UNUSED_PARAMETER(NotUsed);
+ unixEnterMutex();
+ zErr = dlerror();
+ if( zErr ){
+ sqlite3_snprintf(nBuf, zBufOut, "%s", zErr);
+ }
+ unixLeaveMutex();
+}
+static void *unixDlSym(sqlite3_vfs *NotUsed, void *pHandle, const char*zSymbol){
+ UNUSED_PARAMETER(NotUsed);
+ return dlsym(pHandle, zSymbol);
+}
+static void unixDlClose(sqlite3_vfs *NotUsed, void *pHandle){
+ UNUSED_PARAMETER(NotUsed);
+ dlclose(pHandle);
+}
+#else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */
+ #define unixDlOpen 0
+ #define unixDlError 0
+ #define unixDlSym 0
+ #define unixDlClose 0
+#endif
-#if SQLITE_ENABLE_LOCKING_STYLE
-IOMETHODS(
- flockIoFinder, /* Finder function name */
- flockIoMethods, /* sqlite3_io_methods object name */
- flockClose, /* xClose method */
- flockLock, /* xLock method */
- flockUnlock, /* xUnlock method */
- flockCheckReservedLock /* xCheckReservedLock method */
-);
+/*
+** Write nBuf bytes of random data to the supplied buffer zBuf.
+*/
+static int unixRandomness(sqlite3_vfs *NotUsed, int nBuf, char *zBuf){
+ UNUSED_PARAMETER(NotUsed);
+ assert((size_t)nBuf>=(sizeof(time_t)+sizeof(int)));
+
+ /* We have to initialize zBuf to prevent valgrind from reporting
+ ** errors. The reports issued by valgrind are incorrect - we would
+ ** prefer that the randomness be increased by making use of the
+ ** uninitialized space in zBuf - but valgrind errors tend to worry
+ ** some users. Rather than argue, it seems easier just to initialize
+ ** the whole array and silence valgrind, even if that means less randomness
+ ** in the random seed.
+ **
+ ** When testing, initializing zBuf[] to zero is all we do. That means
+ ** that we always use the same random number sequence. This makes the
+ ** tests repeatable.
+ */
+ memset(zBuf, 0, nBuf);
+#if !defined(SQLITE_TEST)
+ {
+ int pid, fd;
+ fd = open("/dev/urandom", O_RDONLY);
+ if( fd<0 ){
+ time_t t;
+ time(&t);
+ memcpy(zBuf, &t, sizeof(t));
+ pid = getpid();
+ memcpy(&zBuf[sizeof(t)], &pid, sizeof(pid));
+ assert( sizeof(t)+sizeof(pid)<=(size_t)nBuf );
+ nBuf = sizeof(t) + sizeof(pid);
+ }else{
+ nBuf = read(fd, zBuf, nBuf);
+ close(fd);
+ }
+ }
#endif
+ return nBuf;
+}
+
+/*
+** Sleep for a little while. Return the amount of time slept.
+** The argument is the number of microseconds we want to sleep.
+** The return value is the number of microseconds of sleep actually
+** requested from the underlying operating system, a number which
+** might be greater than or equal to the argument, but not less
+** than the argument.
+*/
+static int unixSleep(sqlite3_vfs *NotUsed, int microseconds){
#if OS_VXWORKS
-IOMETHODS(
- semIoFinder, /* Finder function name */
- semIoMethods, /* sqlite3_io_methods object name */
- semClose, /* xClose method */
- semLock, /* xLock method */
- semUnlock, /* xUnlock method */
- semCheckReservedLock /* xCheckReservedLock method */
-);
+ struct timespec sp;
+
+ sp.tv_sec = microseconds / 1000000;
+ sp.tv_nsec = (microseconds % 1000000) * 1000;
+ nanosleep(&sp, NULL);
+ return microseconds;
+#elif defined(HAVE_USLEEP) && HAVE_USLEEP
+ usleep(microseconds);
+ return microseconds;
+#else
+ int seconds = (microseconds+999999)/1000000;
+ sleep(seconds);
+ return seconds*1000000;
#endif
+ UNUSED_PARAMETER(NotUsed);
+}
+/*
+** The following variable, if set to a non-zero value, is interpreted as
+** the number of seconds since 1970 and is used to set the result of
+** sqlite3OsCurrentTime() during testing.
+*/
+#ifdef SQLITE_TEST
+int sqlite3_current_time = 0; /* Fake system time in seconds since 1970. */
+#endif
+
+/*
+** Find the current time (in Universal Coordinated Time). Write the
+** current time and date as a Julian Day number into *prNow and
+** return 0. Return 1 if the time and date cannot be found.
+*/
+static int unixCurrentTime(sqlite3_vfs *NotUsed, double *prNow){
+#if defined(NO_GETTOD)
+ time_t t;
+ time(&t);
+ *prNow = t/86400.0 + 2440587.5;
+#elif OS_VXWORKS
+ struct timespec sNow;
+ clock_gettime(CLOCK_REALTIME, &sNow);
+ *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_nsec/86400000000000.0;
+#else
+ struct timeval sNow;
+ gettimeofday(&sNow, 0);
+ *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_usec/86400000000.0;
+#endif
+
+#ifdef SQLITE_TEST
+ if( sqlite3_current_time ){
+ *prNow = sqlite3_current_time/86400.0 + 2440587.5;
+ }
+#endif
+ UNUSED_PARAMETER(NotUsed);
+ return 0;
+}
+
+/*
+** We added the xGetLastError() method with the intention of providing
+** better low-level error messages when operating-system problems come up
+** during SQLite operation. But so far, none of that has been implemented
+** in the core. So this routine is never called. For now, it is merely
+** a place-holder.
+*/
+static int unixGetLastError(sqlite3_vfs *NotUsed, int NotUsed2, char *NotUsed3){
+ UNUSED_PARAMETER(NotUsed);
+ UNUSED_PARAMETER(NotUsed2);
+ UNUSED_PARAMETER(NotUsed3);
+ return 0;
+}
+
+/*
+************************ End of sqlite3_vfs methods ***************************
+******************************************************************************/
+
+/******************************************************************************
+************************** Begin Proxy Locking ********************************
+**
+** Proxy locking is a "uber-locking-method" in this sense: It uses the
+** other locking methods on secondary lock files. Proxy locking is a
+** meta-layer over top of the primitive locking implemented above. For
+** this reason, the division that implements of proxy locking is deferred
+** until late in the file (here) after all of the other I/O methods have
+** been defined - so that the primitive locking methods are available
+** as services to help with the implementation of proxy locking.
+**
+****
+**
+** The default locking schemes in SQLite use byte-range locks on the
+** database file to coordinate safe, concurrent access by multiple readers
+** and writers [http://sqlite.org/lockingv3.html]. The five file locking
+** states (UNLOCKED, PENDING, SHARED, RESERVED, EXCLUSIVE) are implemented
+** as POSIX read & write locks over fixed set of locations (via fsctl),
+** on AFP and SMB only exclusive byte-range locks are available via fsctl
+** with _IOWR('z', 23, struct ByteRangeLockPB2) to track the same 5 states.
+** To simulate a F_RDLCK on the shared range, on AFP a randomly selected
+** address in the shared range is taken for a SHARED lock, the entire
+** shared range is taken for an EXCLUSIVE lock):
+**
+** PENDING_BYTE 0x40000000
+** RESERVED_BYTE 0x40000001
+** SHARED_RANGE 0x40000002 -> 0x40000200
+**
+** This works well on the local file system, but shows a nearly 100x
+** slowdown in read performance on AFP because the AFP client disables
+** the read cache when byte-range locks are present. Enabling the read
+** cache exposes a cache coherency problem that is present on all OS X
+** supported network file systems. NFS and AFP both observe the
+** close-to-open semantics for ensuring cache coherency
+** [http://nfs.sourceforge.net/#faq_a8], which does not effectively
+** address the requirements for concurrent database access by multiple
+** readers and writers
+** [http://www.nabble.com/SQLite-on-NFS-cache-coherency-td15655701.html].
+**
+** To address the performance and cache coherency issues, proxy file locking
+** changes the way database access is controlled by limiting access to a
+** single host at a time and moving file locks off of the database file
+** and onto a proxy file on the local file system.
+**
+**
+** Using proxy locks
+** -----------------
+**
+** C APIs
+**
+** sqlite3_file_control(db, dbname, SQLITE_SET_LOCKPROXYFILE,
+** <proxy_path> | ":auto:");
+** sqlite3_file_control(db, dbname, SQLITE_GET_LOCKPROXYFILE, &<proxy_path>);
+**
+**
+** SQL pragmas
+**
+** PRAGMA [database.]lock_proxy_file=<proxy_path> | :auto:
+** PRAGMA [database.]lock_proxy_file
+**
+** Specifying ":auto:" means that if there is a conch file with a matching
+** host ID in it, the proxy path in the conch file will be used, otherwise
+** a proxy path based on the user's temp dir
+** (via confstr(_CS_DARWIN_USER_TEMP_DIR,...)) will be used and the
+** actual proxy file name is generated from the name and path of the
+** database file. For example:
+**
+** For database path "/Users/me/foo.db"
+** The lock path will be "<tmpdir>/sqliteplocks/_Users_me_foo.db:auto:")
+**
+** Once a lock proxy is configured for a database connection, it can not
+** be removed, however it may be switched to a different proxy path via
+** the above APIs (assuming the conch file is not being held by another
+** connection or process).
+**
+**
+** How proxy locking works
+** -----------------------
+**
+** Proxy file locking relies primarily on two new supporting files:
+**
+** * conch file to limit access to the database file to a single host
+** at a time
+**
+** * proxy file to act as a proxy for the advisory locks normally
+** taken on the database
+**
+** The conch file - to use a proxy file, sqlite must first "hold the conch"
+** by taking an sqlite-style shared lock on the conch file, reading the
+** contents and comparing the host's unique host ID (see below) and lock
+** proxy path against the values stored in the conch. The conch file is
+** stored in the same directory as the database file and the file name
+** is patterned after the database file name as ".<databasename>-conch".
+** If the conch file does not exist, or it's contents do not match the
+** host ID and/or proxy path, then the lock is escalated to an exclusive
+** lock and the conch file contents is updated with the host ID and proxy
+** path and the lock is downgraded to a shared lock again. If the conch
+** is held by another process (with a shared lock), the exclusive lock
+** will fail and SQLITE_BUSY is returned.
+**
+** The proxy file - a single-byte file used for all advisory file locks
+** normally taken on the database file. This allows for safe sharing
+** of the database file for multiple readers and writers on the same
+** host (the conch ensures that they all use the same local lock file).
+**
+** There is a third file - the host ID file - used as a persistent record
+** of a unique identifier for the host, a 128-byte unique host id file
+** in the path defined by the HOSTIDPATH macro (default value is
+** /Library/Caches/.com.apple.sqliteConchHostId).
+**
+** Requesting the lock proxy does not immediately take the conch, it is
+** only taken when the first request to lock database file is made.
+** This matches the semantics of the traditional locking behavior, where
+** opening a connection to a database file does not take a lock on it.
+** The shared lock and an open file descriptor are maintained until
+** the connection to the database is closed.
+**
+** The proxy file and the lock file are never deleted so they only need
+** to be created the first time they are used.
+**
+** Configuration options
+** ---------------------
+**
+** SQLITE_PREFER_PROXY_LOCKING
+**
+** Database files accessed on non-local file systems are
+** automatically configured for proxy locking, lock files are
+** named automatically using the same logic as
+** PRAGMA lock_proxy_file=":auto:"
+**
+** SQLITE_PROXY_DEBUG
+**
+** Enables the logging of error messages during host id file
+** retrieval and creation
+**
+** HOSTIDPATH
+**
+** Overrides the default host ID file path location
+**
+** LOCKPROXYDIR
+**
+** Overrides the default directory used for lock proxy files that
+** are named automatically via the ":auto:" setting
+**
+** SQLITE_DEFAULT_PROXYDIR_PERMISSIONS
+**
+** Permissions to use when creating a directory for storing the
+** lock proxy files, only used when LOCKPROXYDIR is not set.
+**
+**
+** As mentioned above, when compiled with SQLITE_PREFER_PROXY_LOCKING,
+** setting the environment variable SQLITE_FORCE_PROXY_LOCKING to 1 will
+** force proxy locking to be used for every database file opened, and 0
+** will force automatic proxy locking to be disabled for all database
+** files (explicity calling the SQLITE_SET_LOCKPROXYFILE pragma or
+** sqlite_file_control API is not affected by SQLITE_FORCE_PROXY_LOCKING).
+*/
+
+/*
+** Proxy locking is only available on MacOSX
+*/
#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
-IOMETHODS(
- afpIoFinder, /* Finder function name */
- afpIoMethods, /* sqlite3_io_methods object name */
- afpClose, /* xClose method */
- afpLock, /* xLock method */
- afpUnlock, /* xUnlock method */
- afpCheckReservedLock /* xCheckReservedLock method */
-);
-IOMETHODS(
- proxyIoFinder, /* Finder function name */
- proxyIoMethods, /* sqlite3_io_methods object name */
- proxyClose, /* xClose method */
- proxyLock, /* xLock method */
- proxyUnlock, /* xUnlock method */
- proxyCheckReservedLock /* xCheckReservedLock method */
-);
+
+#ifdef SQLITE_TEST
+/* simulate multiple hosts by creating unique hostid file paths */
+int sqlite3_hostid_num = 0;
#endif
+/*
+** The proxyLockingContext has the path and file structures for the remote
+** and local proxy files in it
+*/
+typedef struct proxyLockingContext proxyLockingContext;
+struct proxyLockingContext {
+ unixFile *conchFile; /* Open conch file */
+ char *conchFilePath; /* Name of the conch file */
+ unixFile *lockProxy; /* Open proxy lock file */
+ char *lockProxyPath; /* Name of the proxy lock file */
+ char *dbPath; /* Name of the open file */
+ int conchHeld; /* True if the conch is currently held */
+ void *oldLockingContext; /* Original lockingcontext to restore on close */
+ sqlite3_io_methods const *pOldMethod; /* Original I/O methods for close */
+};
-#if defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE
-/*
-** This "finder" function attempts to determine the best locking strategy
-** for the database file "filePath". It then returns the sqlite3_io_methods
-** object that implements that strategy.
-**
-** This is for MacOSX only.
+/* HOSTIDLEN and CONCHLEN both include space for the string
+** terminating nul
*/
-static const sqlite3_io_methods *autolockIoFinder(
- const char *filePath, /* name of the database file */
- int fd /* file descriptor open on the database file */
-){
- static const struct Mapping {
- const char *zFilesystem; /* Filesystem type name */
- const sqlite3_io_methods *pMethods; /* Appropriate locking method */
- } aMap[] = {
- { "hfs", &posixIoMethods },
- { "ufs", &posixIoMethods },
- { "afpfs", &afpIoMethods },
-#ifdef SQLITE_ENABLE_AFP_LOCKING_SMB
- { "smbfs", &afpIoMethods },
-#else
- { "smbfs", &flockIoMethods },
+#define HOSTIDLEN 128
+#define CONCHLEN (MAXPATHLEN+HOSTIDLEN+1)
+#ifndef HOSTIDPATH
+# define HOSTIDPATH "/Library/Caches/.com.apple.sqliteConchHostId"
#endif
- { "webdav", &nolockIoMethods },
- { 0, 0 }
- };
- int i;
- struct statfs fsInfo;
- struct flock lockInfo;
- if( !filePath ){
- /* If filePath==NULL that means we are dealing with a transient file
- ** that does not need to be locked. */
- return &nolockIoMethods;
+/* basically a copy of unixRandomness with different
+** test behavior built in */
+static int proxyGenerateHostID(char *pHostID){
+ int pid, fd, len;
+ unsigned char *key = (unsigned char *)pHostID;
+
+ memset(key, 0, HOSTIDLEN);
+ len = 0;
+ fd = open("/dev/urandom", O_RDONLY);
+ if( fd>=0 ){
+ len = read(fd, key, HOSTIDLEN);
+ close(fd); /* silently leak the fd if it fails */
}
- if( statfs(filePath, &fsInfo) != -1 ){
- if( fsInfo.f_flags & MNT_RDONLY ){
- return &nolockIoMethods;
- }
- for(i=0; aMap[i].zFilesystem; i++){
- if( strcmp(fsInfo.f_fstypename, aMap[i].zFilesystem)==0 ){
- return aMap[i].pMethods;
+ if( len < HOSTIDLEN ){
+ time_t t;
+ time(&t);
+ memcpy(key, &t, sizeof(t));
+ pid = getpid();
+ memcpy(&key[sizeof(t)], &pid, sizeof(pid));
+ }
+
+#ifdef MAKE_PRETTY_HOSTID
+ {
+ int i;
+ /* filter the bytes into printable ascii characters and NUL terminate */
+ key[(HOSTIDLEN-1)] = 0x00;
+ for( i=0; i<(HOSTIDLEN-1); i++ ){
+ unsigned char pa = key[i]&0x7F;
+ if( pa<0x20 ){
+ key[i] = (key[i]&0x80 == 0x80) ? pa+0x40 : pa+0x20;
+ }else if( pa==0x7F ){
+ key[i] = (key[i]&0x80 == 0x80) ? pa=0x20 : pa+0x7E;
}
}
}
-
- /* Default case. Handles, amongst others, "nfs".
- ** Test byte-range lock using fcntl(). If the call succeeds,
- ** assume that the file-system supports POSIX style locks.
- */
- lockInfo.l_len = 1;
- lockInfo.l_start = 0;
- lockInfo.l_whence = SEEK_SET;
- lockInfo.l_type = F_RDLCK;
- if( fcntl(fd, F_GETLK, &lockInfo)!=-1 ) {
- return &posixIoMethods;
- }else{
- return &dotlockIoMethods;
- }
+#endif
+ return SQLITE_OK;
}
-#endif /* defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE */
-
-/*
-** An abstract type for a pointer to a IO method finder function:
-*/
-typedef const sqlite3_io_methods *(*finder_type)(const char*,int);
-
-/****************************************************************************
-**************************** sqlite3_vfs methods ****************************
-**
-** This division contains the implementation of methods on the
-** sqlite3_vfs object.
+/* writes the host id path to path, path should be an pre-allocated buffer
+** with enough space for a path
*/
+static void proxyGetHostIDPath(char *path, size_t len){
+ strlcpy(path, HOSTIDPATH, len);
+#ifdef SQLITE_TEST
+ if( sqlite3_hostid_num>0 ){
+ char suffix[2] = "1";
+ suffix[0] = suffix[0] + sqlite3_hostid_num;
+ strlcat(path, suffix, len);
+ }
+#endif
+ OSTRACE3("GETHOSTIDPATH %s pid=%d\n", path, getpid());
+}
-/*
-** Initialize the contents of the unixFile structure pointed to by pId.
+/* get the host ID from a sqlite hostid file stored in the
+** user-specific tmp directory, create the ID if it's not there already
*/
-static int fillInUnixFile(
- sqlite3_vfs *pVfs, /* Pointer to vfs object */
- int h, /* Open file descriptor of file being opened */
- int dirfd, /* Directory file descriptor */
- sqlite3_file *pId, /* Write to the unixFile structure here */
- const char *zFilename, /* Name of the file being opened */
- int noLock, /* Omit locking if true */
- int isDelete /* Delete on close if true */
-){
- const sqlite3_io_methods *pLockingStyle;
- unixFile *pNew = (unixFile *)pId;
- int rc = SQLITE_OK;
-
- assert( pNew->pLock==NULL );
- assert( pNew->pOpen==NULL );
+static int proxyGetHostID(char *pHostID, int *pError){
+ int fd;
+ char path[MAXPATHLEN];
+ size_t len;
+ int rc=SQLITE_OK;
- /* Parameter isDelete is only used on vxworks. Parameter pVfs is only
- ** used if ENABLE_LOCKING_STYLE is defined. Express this explicitly
- ** here to prevent compiler warnings about unused parameters.
- */
-#if !OS_VXWORKS
- UNUSED_PARAMETER(isDelete);
-#endif
-#if !SQLITE_ENABLE_LOCKING_STYLE
- UNUSED_PARAMETER(pVfs);
-#endif
-#if !OS_VXWORKS && !SQLITE_ENABLE_LOCKING_STYLE
- UNUSED_PARAMETER(zFilename);
+ proxyGetHostIDPath(path, MAXPATHLEN);
+ /* try to create the host ID file, if it already exists read the contents */
+ fd = open(path, O_CREAT|O_WRONLY|O_EXCL, 0644);
+ if( fd<0 ){
+ int err=errno;
+
+ if( err!=EEXIST ){
+#ifdef SQLITE_PROXY_DEBUG /* set the sqlite error message instead */
+ fprintf(stderr, "sqlite error creating host ID file %s: %s\n",
+ path, strerror(err));
#endif
-
- OSTRACE3("OPEN %-3d %s\n", h, zFilename);
- pNew->h = h;
- pNew->dirfd = dirfd;
- SET_THREADID(pNew);
-
-#if OS_VXWORKS
- pNew->pId = vxworksFindFileId(zFilename);
- if( pNew->pId==0 ){
- noLock = 1;
- rc = SQLITE_NOMEM;
- }
+ return SQLITE_PERM;
+ }
+ /* couldn't create the file, read it instead */
+ fd = open(path, O_RDONLY|O_EXCL);
+ if( fd<0 ){
+#ifdef SQLITE_PROXY_DEBUG /* set the sqlite error message instead */
+ int err = errno;
+ fprintf(stderr, "sqlite error opening host ID file %s: %s\n",
+ path, strerror(err));
#endif
-
- if( noLock ){
- pLockingStyle = &nolockIoMethods;
+ return SQLITE_PERM;
+ }
+ len = pread(fd, pHostID, HOSTIDLEN, 0);
+ if( len<0 ){
+ *pError = errno;
+ rc = SQLITE_IOERR_READ;
+ }else if( len<HOSTIDLEN ){
+ *pError = 0;
+ rc = SQLITE_IOERR_SHORT_READ;
+ }
+ close(fd); /* silently leak the fd if it fails */
+ OSTRACE3("GETHOSTID read %s pid=%d\n", pHostID, getpid());
+ return rc;
}else{
- pLockingStyle = (*(finder_type)pVfs->pAppData)(zFilename, h);
-#if SQLITE_ENABLE_LOCKING_STYLE
- /* Cache zFilename in the locking context (AFP and dotlock override) for
- ** proxyLock activation is possible (remote proxy is based on db name)
- ** zFilename remains valid until file is closed, to support */
- pNew->lockingContext = (void*)zFilename;
-#endif
- }
-
- if( pLockingStyle == &posixIoMethods ){
- unixEnterMutex();
- rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
- unixLeaveMutex();
- }
-
-#if SQLITE_ENABLE_LOCKING_STYLE && defined(__DARWIN__)
- else if( pLockingStyle == &afpIoMethods ){
- /* AFP locking uses the file path so it needs to be included in
- ** the afpLockingContext.
- */
- afpLockingContext *pCtx;
- pNew->lockingContext = pCtx = sqlite3_malloc( sizeof(*pCtx) );
- if( pCtx==0 ){
- rc = SQLITE_NOMEM;
- }else{
- /* NB: zFilename exists and remains valid until the file is closed
- ** according to requirement F11141. So we do not need to make a
- ** copy of the filename. */
- pCtx->dbPath = zFilename;
- srandomdev();
- unixEnterMutex();
- rc = findLockInfo(pNew, NULL, &pNew->pOpen);
- unixLeaveMutex();
+ /* we're creating the host ID file (use a random string of bytes) */
+ proxyGenerateHostID(pHostID);
+ len = pwrite(fd, pHostID, HOSTIDLEN, 0);
+ if( len<0 ){
+ *pError = errno;
+ rc = SQLITE_IOERR_WRITE;
+ }else if( len<HOSTIDLEN ){
+ *pError = 0;
+ rc = SQLITE_IOERR_WRITE;
}
+ close(fd); /* silently leak the fd if it fails */
+ OSTRACE3("GETHOSTID wrote %s pid=%d\n", pHostID, getpid());
+ return rc;
}
-#endif
+}
-#if SQLITE_ENABLE_LOCKING_STYLE
- else if( pLockingStyle == &dotlockIoMethods ){
- /* Dotfile locking uses the file path so it needs to be included in
- ** the dotlockLockingContext
- */
- char *zLockFile;
- int nFilename;
- nFilename = strlen(zFilename) + 6;
- zLockFile = (char *)sqlite3_malloc(nFilename);
- if( zLockFile==0 ){
- rc = SQLITE_NOMEM;
+static int proxyGetLockPath(const char *dbPath, char *lPath, size_t maxLen){
+ int len;
+ int dbLen;
+ int i;
+
+#ifdef LOCKPROXYDIR
+ len = strlcpy(lPath, LOCKPROXYDIR, maxLen);
+#else
+# ifdef _CS_DARWIN_USER_TEMP_DIR
+ {
+ confstr(_CS_DARWIN_USER_TEMP_DIR, lPath, maxLen);
+ len = strlcat(lPath, "sqliteplocks", maxLen);
+ if( mkdir(lPath, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
+ /* if mkdir fails, handle as lock file creation failure */
+ int err = errno;
+# ifdef SQLITE_DEBUG
+ if( err!=EEXIST ){
+ fprintf(stderr, "proxyGetLockPath: mkdir(%s,0%o) error %d %s\n", lPath,
+ SQLITE_DEFAULT_PROXYDIR_PERMISSIONS, err, strerror(err));
+ }
+# endif
}else{
- sqlite3_snprintf(nFilename, zLockFile, "%s" DOTLOCK_SUFFIX, zFilename);
+ OSTRACE3("GETLOCKPATH mkdir %s pid=%d\n", lPath, getpid());
}
- pNew->lockingContext = zLockFile;
+
}
+# else
+ len = strlcpy(lPath, "/tmp/", maxLen);
+# endif
#endif
-#if OS_VXWORKS
- else if( pLockingStyle == &semIoMethods ){
- /* Named semaphore locking uses the file path so it needs to be
- ** included in the semLockingContext
- */
- unixEnterMutex();
- rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
- if( (rc==SQLITE_OK) && (pNew->pOpen->pSem==NULL) ){
- char *zSemName = pNew->pOpen->aSemName;
- int n;
- sqlite3_snprintf(MAX_PATHNAME, zSemName, "%s.sem",
- pNew->pId->zCanonicalName);
- for( n=0; zSemName[n]; n++ )
- if( zSemName[n]=='/' ) zSemName[n] = '_';
- pNew->pOpen->pSem = sem_open(zSemName, O_CREAT, 0666, 1);
- if( pNew->pOpen->pSem == SEM_FAILED ){
- rc = SQLITE_NOMEM;
- pNew->pOpen->aSemName[0] = '\0';
- }
- }
- unixLeaveMutex();
+ if( lPath[len-1]!='/' ){
+ len = strlcat(lPath, "/", maxLen);
}
-#endif
- pNew->lastErrno = 0;
-#if OS_VXWORKS
- if( rc!=SQLITE_OK ){
- unlink(zFilename);
- isDelete = 0;
- }
- pNew->isDelete = isDelete;
-#endif
- if( rc!=SQLITE_OK ){
- if( dirfd>=0 ) close(dirfd); /* silent leak if fail, already in error */
- close(h);
- }else{
- pNew->pMethod = pLockingStyle;
- OpenCounter(+1);
+ /* transform the db path to a unique cache name */
+ dbLen = strlen(dbPath);
+ for( i=0; i<dbLen && (i+len+7)<maxLen; i++){
+ char c = dbPath[i];
+ lPath[i+len] = (c=='/')?'_':c;
}
- return rc;
+ lPath[i+len]='\0';
+ strlcat(lPath, ":auto:", maxLen);
+ return SQLITE_OK;
}
/*
-** Open a file descriptor to the directory containing file zFilename.
-** If successful, *pFd is set to the opened file descriptor and
-** SQLITE_OK is returned. If an error occurs, either SQLITE_NOMEM
-** or SQLITE_CANTOPEN is returned and *pFd is set to an undefined
-** value.
+** Create a new VFS file descriptor (stored in memory obtained from
+** sqlite3_malloc) and open the file named "path" in the file descriptor.
**
-** If SQLITE_OK is returned, the caller is responsible for closing
-** the file descriptor *pFd using close().
+** The caller is responsible not only for closing the file descriptor
+** but also for freeing the memory associated with the file descriptor.
*/
-static int openDirectory(const char *zFilename, int *pFd){
- int ii;
- int fd = -1;
- char zDirname[MAX_PATHNAME+1];
+static int proxyCreateUnixFile(const char *path, unixFile **ppFile) {
+ int fd;
+ int dirfd = -1;
+ unixFile *pNew;
+ int rc = SQLITE_OK;
+ sqlite3_vfs dummyVfs;
- sqlite3_snprintf(MAX_PATHNAME, zDirname, "%s", zFilename);
- for(ii=strlen(zDirname); ii>=0 && zDirname[ii]!='/'; ii--);
- if( ii>0 ){
- zDirname[ii] = '\0';
- fd = open(zDirname, O_RDONLY|O_BINARY, 0);
- if( fd>=0 ){
-#ifdef FD_CLOEXEC
- fcntl(fd, F_SETFD, fcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
+ fd = open(path, O_RDWR | O_CREAT, SQLITE_DEFAULT_FILE_PERMISSIONS);
+ if( fd<0 ){
+ return SQLITE_CANTOPEN;
+ }
+
+ pNew = (unixFile *)sqlite3_malloc(sizeof(unixFile));
+ if( pNew==NULL ){
+ rc = SQLITE_NOMEM;
+ goto end_create_proxy;
+ }
+ memset(pNew, 0, sizeof(unixFile));
+
+ dummyVfs.pAppData = (void*)autolockIoFinder;
+ rc = fillInUnixFile(&dummyVfs, fd, dirfd, (sqlite3_file*)pNew, path, 0, 0);
+ if( rc==SQLITE_OK ){
+ *ppFile = pNew;
+ return SQLITE_OK;
+ }
+end_create_proxy:
+ close(fd); /* silently leak fd if error, we're already in error */
+ sqlite3_free(pNew);
+ return rc;
+}
+
+/* takes the conch by taking a shared lock and read the contents conch, if
+** lockPath is non-NULL, the host ID and lock file path must match. A NULL
+** lockPath means that the lockPath in the conch file will be used if the
+** host IDs match, or a new lock path will be generated automatically
+** and written to the conch file.
+*/
+static int proxyTakeConch(unixFile *pFile){
+ proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
+
+ if( pCtx->conchHeld>0 ){
+ return SQLITE_OK;
+ }else{
+ unixFile *conchFile = pCtx->conchFile;
+ char testValue[CONCHLEN];
+ char conchValue[CONCHLEN];
+ char lockPath[MAXPATHLEN];
+ char *tLockPath = NULL;
+ int rc = SQLITE_OK;
+ int readRc = SQLITE_OK;
+ int syncPerms = 0;
+
+ OSTRACE4("TAKECONCH %d for %s pid=%d\n", conchFile->h,
+ (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"), getpid());
+
+ rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, SHARED_LOCK);
+ if( rc==SQLITE_OK ){
+ int pError = 0;
+ memset(testValue, 0, CONCHLEN); // conch is fixed size
+ rc = proxyGetHostID(testValue, &pError);
+ if( (rc&0xff)==SQLITE_IOERR ){
+ pFile->lastErrno = pError;
+ }
+ if( pCtx->lockProxyPath ){
+ strlcpy(&testValue[HOSTIDLEN], pCtx->lockProxyPath, MAXPATHLEN);
+ }
+ }
+ if( rc!=SQLITE_OK ){
+ goto end_takeconch;
+ }
+
+ readRc = unixRead((sqlite3_file *)conchFile, conchValue, CONCHLEN, 0);
+ if( readRc!=SQLITE_IOERR_SHORT_READ ){
+ if( readRc!=SQLITE_OK ){
+ if( (rc&0xff)==SQLITE_IOERR ){
+ pFile->lastErrno = conchFile->lastErrno;
+ }
+ rc = readRc;
+ goto end_takeconch;
+ }
+ /* if the conch has data compare the contents */
+ if( !pCtx->lockProxyPath ){
+ /* for auto-named local lock file, just check the host ID and we'll
+ ** use the local lock file path that's already in there */
+ if( !memcmp(testValue, conchValue, HOSTIDLEN) ){
+ tLockPath = (char *)&conchValue[HOSTIDLEN];
+ goto end_takeconch;
+ }
+ }else{
+ /* we've got the conch if conchValue matches our path and host ID */
+ if( !memcmp(testValue, conchValue, CONCHLEN) ){
+ goto end_takeconch;
+ }
+ }
+ }else{
+ /* a short read means we're "creating" the conch (even though it could
+ ** have been user-intervention), if we acquire the exclusive lock,
+ ** we'll try to match the current on-disk permissions of the database
+ */
+ syncPerms = 1;
+ }
+
+ /* either conch was emtpy or didn't match */
+ if( !pCtx->lockProxyPath ){
+ proxyGetLockPath(pCtx->dbPath, lockPath, MAXPATHLEN);
+ tLockPath = lockPath;
+ strlcpy(&testValue[HOSTIDLEN], lockPath, MAXPATHLEN);
+ }
+
+ /* update conch with host and path (this will fail if other process
+ ** has a shared lock already) */
+ rc = conchFile->pMethod->xLock((sqlite3_file*)conchFile, EXCLUSIVE_LOCK);
+ if( rc==SQLITE_OK ){
+ rc = unixWrite((sqlite3_file *)conchFile, testValue, CONCHLEN, 0);
+ if( rc==SQLITE_OK && syncPerms ){
+ struct stat buf;
+ int err = fstat(pFile->h, &buf);
+ if( err==0 ){
+ /* try to match the database file permissions, ignore failure */
+#ifndef SQLITE_PROXY_DEBUG
+ fchmod(conchFile->h, buf.st_mode);
+#else
+ if( fchmod(conchFile->h, buf.st_mode)!=0 ){
+ int code = errno;
+ fprintf(stderr, "fchmod %o FAILED with %d %s\n",
+ buf.st_mode, code, strerror(code));
+ } else {
+ fprintf(stderr, "fchmod %o SUCCEDED\n",buf.st_mode);
+ }
+ }else{
+ int code = errno;
+ fprintf(stderr, "STAT FAILED[%d] with %d %s\n",
+ err, code, strerror(code));
#endif
- OSTRACE3("OPENDIR %-3d %s\n", fd, zDirname);
+ }
+ }
+ }
+ conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, SHARED_LOCK);
+
+end_takeconch:
+ OSTRACE2("TRANSPROXY: CLOSE %d\n", pFile->h);
+ if( rc==SQLITE_OK && pFile->openFlags ){
+ if( pFile->h>=0 ){
+#ifdef STRICT_CLOSE_ERROR
+ if( close(pFile->h) ){
+ pFile->lastErrno = errno;
+ return SQLITE_IOERR_CLOSE;
+ }
+#else
+ close(pFile->h); /* silently leak fd if fail */
+#endif
+ }
+ pFile->h = -1;
+ int fd = open(pCtx->dbPath, pFile->openFlags,
+ SQLITE_DEFAULT_FILE_PERMISSIONS);
+ OSTRACE2("TRANSPROXY: OPEN %d\n", fd);
+ if( fd>=0 ){
+ pFile->h = fd;
+ }else{
+ rc=SQLITE_CANTOPEN; // SQLITE_BUSY? proxyTakeConch called during locking
+ }
+ }
+ if( rc==SQLITE_OK && !pCtx->lockProxy ){
+ char *path = tLockPath ? tLockPath : pCtx->lockProxyPath;
+ // ACS: Need to make a copy of path sometimes
+ rc = proxyCreateUnixFile(path, &pCtx->lockProxy);
+ }
+ if( rc==SQLITE_OK ){
+ pCtx->conchHeld = 1;
+
+ if( tLockPath ){
+ pCtx->lockProxyPath = sqlite3DbStrDup(0, tLockPath);
+ if( pCtx->lockProxy->pMethod == &afpIoMethods ){
+ ((afpLockingContext *)pCtx->lockProxy->lockingContext)->dbPath =
+ pCtx->lockProxyPath;
+ }
+ }
+ } else {
+ conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK);
}
+ OSTRACE3("TAKECONCH %d %s\n", conchFile->h, rc==SQLITE_OK?"ok":"failed");
+ return rc;
}
- *pFd = fd;
- return (fd>=0?SQLITE_OK:SQLITE_CANTOPEN);
}
/*
-** Create a temporary file name in zBuf. zBuf must be allocated
-** by the calling process and must be big enough to hold at least
-** pVfs->mxPathname bytes.
+** If pFile holds a lock on a conch file, then release that lock.
*/
-static int getTempname(int nBuf, char *zBuf){
- static const char *azDirs[] = {
- 0,
- 0,
- "/var/tmp",
- "/usr/tmp",
- "/tmp",
- ".",
- };
- static const unsigned char zChars[] =
- "abcdefghijklmnopqrstuvwxyz"
- "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
- "0123456789";
- unsigned int i, j;
- struct stat buf;
- const char *zDir = ".";
-
- /* It's odd to simulate an io-error here, but really this is just
- ** using the io-error infrastructure to test that SQLite handles this
- ** function failing.
- */
- SimulateIOError( return SQLITE_IOERR );
-
- azDirs[0] = sqlite3_temp_directory;
- if (NULL == azDirs[1]) {
- azDirs[1] = getenv("TMPDIR");
- }
-
- for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); i++){
- if( azDirs[i]==0 ) continue;
- if( stat(azDirs[i], &buf) ) continue;
- if( !S_ISDIR(buf.st_mode) ) continue;
- if( access(azDirs[i], 07) ) continue;
- zDir = azDirs[i];
- break;
- }
-
- /* Check that the output buffer is large enough for the temporary file
- ** name. If it is not, return SQLITE_ERROR.
- */
- if( (strlen(zDir) + strlen(SQLITE_TEMP_FILE_PREFIX) + 17) >= (size_t)nBuf ){
- return SQLITE_ERROR;
- }
+static int proxyReleaseConch(unixFile *pFile){
+ int rc; /* Subroutine return code */
+ proxyLockingContext *pCtx; /* The locking context for the proxy lock */
+ unixFile *conchFile; /* Name of the conch file */
- do{
- sqlite3_snprintf(nBuf-17, zBuf, "%s/"SQLITE_TEMP_FILE_PREFIX, zDir);
- j = strlen(zBuf);
- sqlite3_randomness(15, &zBuf[j]);
- for(i=0; i<15; i++, j++){
- zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
- }
- zBuf[j] = 0;
- }while( access(zBuf,0)==0 );
- return SQLITE_OK;
+ pCtx = (proxyLockingContext *)pFile->lockingContext;
+ conchFile = pCtx->conchFile;
+ OSTRACE4("RELEASECONCH %d for %s pid=%d\n", conchFile->h,
+ (pCtx->lockProxyPath ? pCtx->lockProxyPath : ":auto:"),
+ getpid());
+ pCtx->conchHeld = 0;
+ rc = conchFile->pMethod->xUnlock((sqlite3_file*)conchFile, NO_LOCK);
+ OSTRACE3("RELEASECONCH %d %s\n", conchFile->h,
+ (rc==SQLITE_OK ? "ok" : "failed"));
+ return rc;
}
-
/*
-** Open the file zPath.
-**
-** Previously, the SQLite OS layer used three functions in place of this
-** one:
-**
-** sqlite3OsOpenReadWrite();
-** sqlite3OsOpenReadOnly();
-** sqlite3OsOpenExclusive();
-**
-** These calls correspond to the following combinations of flags:
+** Given the name of a database file, compute the name of its conch file.
+** Store the conch filename in memory obtained from sqlite3_malloc().
+** Make *pConchPath point to the new name. Return SQLITE_OK on success
+** or SQLITE_NOMEM if unable to obtain memory.
**
-** ReadWrite() -> (READWRITE | CREATE)
-** ReadOnly() -> (READONLY)
-** OpenExclusive() -> (READWRITE | CREATE | EXCLUSIVE)
+** The caller is responsible for ensuring that the allocated memory
+** space is eventually freed.
**
-** The old OpenExclusive() accepted a boolean argument - "delFlag". If
-** true, the file was configured to be automatically deleted when the
-** file handle closed. To achieve the same effect using this new
-** interface, add the DELETEONCLOSE flag to those specified above for
-** OpenExclusive().
+** *pConchPath is set to NULL if a memory allocation error occurs.
*/
-static int unixOpen(
- sqlite3_vfs *pVfs, /* The VFS for which this is the xOpen method */
- const char *zPath, /* Pathname of file to be opened */
- sqlite3_file *pFile, /* The file descriptor to be filled in */
- int flags, /* Input flags to control the opening */
- int *pOutFlags /* Output flags returned to SQLite core */
-){
- int fd = 0; /* File descriptor returned by open() */
- int dirfd = -1; /* Directory file descriptor */
- int openFlags = 0; /* Flags to pass to open() */
- int eType = flags&0xFFFFFF00; /* Type of file to open */
- int noLock; /* True to omit locking primitives */
- int rc = SQLITE_OK;
-
- int isExclusive = (flags & SQLITE_OPEN_EXCLUSIVE);
- int isDelete = (flags & SQLITE_OPEN_DELETEONCLOSE);
- int isCreate = (flags & SQLITE_OPEN_CREATE);
- int isReadonly = (flags & SQLITE_OPEN_READONLY);
- int isReadWrite = (flags & SQLITE_OPEN_READWRITE);
-
- /* If creating a master or main-file journal, this function will open
- ** a file-descriptor on the directory too. The first time unixSync()
- ** is called the directory file descriptor will be fsync()ed and close()d.
- */
- int isOpenDirectory = (isCreate &&
- (eType==SQLITE_OPEN_MASTER_JOURNAL || eType==SQLITE_OPEN_MAIN_JOURNAL)
- );
-
- /* If argument zPath is a NULL pointer, this function is required to open
- ** a temporary file. Use this buffer to store the file name in.
- */
- char zTmpname[MAX_PATHNAME+1];
- const char *zName = zPath;
-
- /* Check the following statements are true:
- **
- ** (a) Exactly one of the READWRITE and READONLY flags must be set, and
- ** (b) if CREATE is set, then READWRITE must also be set, and
- ** (c) if EXCLUSIVE is set, then CREATE must also be set.
- ** (d) if DELETEONCLOSE is set, then CREATE must also be set.
- */
- assert((isReadonly==0 || isReadWrite==0) && (isReadWrite || isReadonly));
- assert(isCreate==0 || isReadWrite);
- assert(isExclusive==0 || isCreate);
- assert(isDelete==0 || isCreate);
-
- /* The main DB, main journal, and master journal are never automatically
- ** deleted
- */
- assert( eType!=SQLITE_OPEN_MAIN_DB || !isDelete );
- assert( eType!=SQLITE_OPEN_MAIN_JOURNAL || !isDelete );
- assert( eType!=SQLITE_OPEN_MASTER_JOURNAL || !isDelete );
-
- /* Assert that the upper layer has set one of the "file-type" flags. */
- assert( eType==SQLITE_OPEN_MAIN_DB || eType==SQLITE_OPEN_TEMP_DB
- || eType==SQLITE_OPEN_MAIN_JOURNAL || eType==SQLITE_OPEN_TEMP_JOURNAL
- || eType==SQLITE_OPEN_SUBJOURNAL || eType==SQLITE_OPEN_MASTER_JOURNAL
- || eType==SQLITE_OPEN_TRANSIENT_DB
- );
-
- memset(pFile, 0, sizeof(unixFile));
-
- if( !zName ){
- assert(isDelete && !isOpenDirectory);
- rc = getTempname(MAX_PATHNAME+1, zTmpname);
- if( rc!=SQLITE_OK ){
- return rc;
- }
- zName = zTmpname;
- }
-
- if( isReadonly ) openFlags |= O_RDONLY;
- if( isReadWrite ) openFlags |= O_RDWR;
- if( isCreate ) openFlags |= O_CREAT;
- if( isExclusive ) openFlags |= (O_EXCL|O_NOFOLLOW);
- openFlags |= (O_LARGEFILE|O_BINARY);
-
- fd = open(zName, openFlags, isDelete?0600:SQLITE_DEFAULT_FILE_PERMISSIONS);
- OSTRACE4("OPENX %-3d %s 0%o\n", fd, zName, openFlags);
- if( fd<0 && errno!=EISDIR && isReadWrite && !isExclusive ){
- /* Failed to open the file for read/write access. Try read-only. */
- flags &= ~(SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE);
- flags |= SQLITE_OPEN_READONLY;
- return unixOpen(pVfs, zPath, pFile, flags, pOutFlags);
- }
- if( fd<0 ){
- return SQLITE_CANTOPEN;
- }
- if( isDelete ){
-#if OS_VXWORKS
- zPath = zName;
-#else
- unlink(zName);
-#endif
- }
-#if SQLITE_ENABLE_LOCKING_STYLE
- else{
- ((unixFile*)pFile)->openFlags = openFlags;
- }
-#endif
- if( pOutFlags ){
- *pOutFlags = flags;
- }
+static int proxyCreateConchPathname(char *dbPath, char **pConchPath){
+ int i; /* Loop counter */
+ int len = strlen(dbPath); /* Length of database filename - dbPath */
+ char *conchPath; /* buffer in which to construct conch name */
- assert(fd!=0);
- if( isOpenDirectory ){
- rc = openDirectory(zPath, &dirfd);
- if( rc!=SQLITE_OK ){
- close(fd); /* silently leak if fail, already in error */
- return rc;
- }
+ /* Allocate space for the conch filename and initialize the name to
+ ** the name of the original database file. */
+ *pConchPath = conchPath = (char *)sqlite3_malloc(len + 8);
+ if( conchPath==0 ){
+ return SQLITE_NOMEM;
}
-
-#ifdef FD_CLOEXEC
- fcntl(fd, F_SETFD, fcntl(fd, F_GETFD, 0) | FD_CLOEXEC);
-#endif
-
- noLock = eType!=SQLITE_OPEN_MAIN_DB;
-
-#if SQLITE_PREFER_PROXY_LOCKING
- if( zPath!=NULL && !noLock ){
- char *envforce = getenv("SQLITE_FORCE_PROXY_LOCKING");
- int useProxy = 0;
-
- /* SQLITE_FORCE_PROXY_LOCKING==1 means force always use proxy,
- ** 0 means never use proxy, NULL means use proxy for non-local files only
- */
- if( envforce!=NULL ){
- useProxy = atoi(envforce)>0;
- }else{
- struct statfs fsInfo;
-
- if( statfs(zPath, &fsInfo) == -1 ){
- ((unixFile*)pFile)->lastErrno = errno;
- if( dirfd>=0 ) close(dirfd); /* silently leak if fail, in error */
- close(fd); /* silently leak if fail, in error */
- return SQLITE_IOERR_ACCESS;
- }
- useProxy = !(fsInfo.f_flags&MNT_LOCAL);
- }
- if( useProxy ){
- rc = fillInUnixFile(pVfs, fd, dirfd, pFile, zPath, noLock, isDelete);
- if( rc==SQLITE_OK ){
- rc = transformUnixFileForLockProxy((unixFile*)pFile, ":auto:");
- }
- return rc;
+ memcpy(conchPath, dbPath, len+1);
+
+ /* now insert a "." before the last / character */
+ for( i=(len-1); i>=0; i-- ){
+ if( conchPath[i]=='/' ){
+ i++;
+ break;
}
}
-#endif
-
- return fillInUnixFile(pVfs, fd, dirfd, pFile, zPath, noLock, isDelete);
+ conchPath[i]='.';
+ while ( i<len ){
+ conchPath[i+1]=dbPath[i];
+ i++;
+ }
+
+ /* append the "-conch" suffix to the file */
+ memcpy(&conchPath[i+1], "-conch", 7);
+ assert( strlen(conchPath) == len+7 );
+
+ return SQLITE_OK;
}
-/*
-** Delete the file at zPath. If the dirSync argument is true, fsync()
-** the directory after deleting the file.
+
+/* Takes a fully configured proxy locking-style unix file and switches
+** the local lock file path
*/
-static int unixDelete(
- sqlite3_vfs *NotUsed, /* VFS containing this as the xDelete method */
- const char *zPath, /* Name of file to be deleted */
- int dirSync /* If true, fsync() directory after deleting file */
-){
+static int switchLockProxyPath(unixFile *pFile, const char *path) {
+ proxyLockingContext *pCtx = (proxyLockingContext*)pFile->lockingContext;
+ char *oldPath = pCtx->lockProxyPath;
int rc = SQLITE_OK;
- UNUSED_PARAMETER(NotUsed);
- SimulateIOError(return SQLITE_IOERR_DELETE);
- unlink(zPath);
-#ifndef SQLITE_DISABLE_DIRSYNC
- if( dirSync ){
- int fd;
- rc = openDirectory(zPath, &fd);
- if( rc==SQLITE_OK ){
-#if OS_VXWORKS
- if( fsync(fd)==-1 )
-#else
- if( fsync(fd) )
-#endif
- {
- rc = SQLITE_IOERR_DIR_FSYNC;
- }
- if( close(fd)&&!rc ){
- rc = SQLITE_IOERR_DIR_CLOSE;
- }
+
+ if( pFile->locktype!=NO_LOCK ){
+ return SQLITE_BUSY;
+ }
+
+ /* nothing to do if the path is NULL, :auto: or matches the existing path */
+ if( !path || path[0]=='\0' || !strcmp(path, ":auto:") ||
+ (oldPath && !strncmp(oldPath, path, MAXPATHLEN)) ){
+ return SQLITE_OK;
+ }else{
+ unixFile *lockProxy = pCtx->lockProxy;
+ pCtx->lockProxy=NULL;
+ pCtx->conchHeld = 0;
+ if( lockProxy!=NULL ){
+ rc=lockProxy->pMethod->xClose((sqlite3_file *)lockProxy);
+ if( rc ) return rc;
+ sqlite3_free(lockProxy);
}
+ sqlite3_free(oldPath);
+ pCtx->lockProxyPath = sqlite3DbStrDup(0, path);
}
-#endif
+
return rc;
}
/*
-** Test the existance of or access permissions of file zPath. The
-** test performed depends on the value of flags:
-**
-** SQLITE_ACCESS_EXISTS: Return 1 if the file exists
-** SQLITE_ACCESS_READWRITE: Return 1 if the file is read and writable.
-** SQLITE_ACCESS_READONLY: Return 1 if the file is readable.
+** pFile is a file that has been opened by a prior xOpen call. dbPath
+** is a string buffer at least MAXPATHLEN+1 characters in size.
**
-** Otherwise return 0.
+** This routine find the filename associated with pFile and writes it
+** int dbPath.
*/
-static int unixAccess(
- sqlite3_vfs *NotUsed, /* The VFS containing this xAccess method */
- const char *zPath, /* Path of the file to examine */
- int flags, /* What do we want to learn about the zPath file? */
- int *pResOut /* Write result boolean here */
-){
- int amode = 0;
- UNUSED_PARAMETER(NotUsed);
- SimulateIOError( return SQLITE_IOERR_ACCESS; );
- switch( flags ){
- case SQLITE_ACCESS_EXISTS:
- amode = F_OK;
- break;
- case SQLITE_ACCESS_READWRITE:
- amode = W_OK|R_OK;
- break;
- case SQLITE_ACCESS_READ:
- amode = R_OK;
- break;
-
- default:
- assert(!"Invalid flags argument");
+static int proxyGetDbPathForUnixFile(unixFile *pFile, char *dbPath){
+#if defined(__DARWIN__)
+ if( pFile->pMethod == &afpIoMethods ){
+ /* afp style keeps a reference to the db path in the filePath field
+ ** of the struct */
+ assert( strlen((char*)pFile->lockingContext)<=MAXPATHLEN );
+ strcpy(dbPath, ((afpLockingContext *)pFile->lockingContext)->dbPath);
+ }else
+#endif
+ if( pFile->pMethod == &dotlockIoMethods ){
+ /* dot lock style uses the locking context to store the dot lock
+ ** file path */
+ int len = strlen((char *)pFile->lockingContext) - strlen(DOTLOCK_SUFFIX);
+ memcpy(dbPath, (char *)pFile->lockingContext, len + 1);
+ }else{
+ /* all other styles use the locking context to store the db file path */
+ assert( strlen((char*)pFile->lockingContext)<=MAXPATHLEN );
+ strcpy(dbPath, (char *)pFile->lockingContext);
}
- *pResOut = (access(zPath, amode)==0);
return SQLITE_OK;
}
-
/*
-** Turn a relative pathname into a full pathname. The relative path
-** is stored as a nul-terminated string in the buffer pointed to by
-** zPath.
-**
-** zOut points to a buffer of at least sqlite3_vfs.mxPathname bytes
-** (in this case, MAX_PATHNAME bytes). The full-path is written to
-** this buffer before returning.
+** Takes an already filled in unix file and alters it so all file locking
+** will be performed on the local proxy lock file. The following fields
+** are preserved in the locking context so that they can be restored and
+** the unix structure properly cleaned up at close time:
+** ->lockingContext
+** ->pMethod
*/
-static int unixFullPathname(
- sqlite3_vfs *pVfs, /* Pointer to vfs object */
- const char *zPath, /* Possibly relative input path */
- int nOut, /* Size of output buffer in bytes */
- char *zOut /* Output buffer */
-){
+static int proxyTransformUnixFile(unixFile *pFile, const char *path) {
+ proxyLockingContext *pCtx;
+ char dbPath[MAXPATHLEN+1]; /* Name of the database file */
+ char *lockPath=NULL;
+ int rc = SQLITE_OK;
+
+ if( pFile->locktype!=NO_LOCK ){
+ return SQLITE_BUSY;
+ }
+ proxyGetDbPathForUnixFile(pFile, dbPath);
+ if( !path || path[0]=='\0' || !strcmp(path, ":auto:") ){
+ lockPath=NULL;
+ }else{
+ lockPath=(char *)path;
+ }
+
+ OSTRACE4("TRANSPROXY %d for %s pid=%d\n", pFile->h,
+ (lockPath ? lockPath : ":auto:"), getpid());
- /* It's odd to simulate an io-error here, but really this is just
- ** using the io-error infrastructure to test that SQLite handles this
- ** function failing. This function could fail if, for example, the
- ** current working directory has been unlinked.
- */
- SimulateIOError( return SQLITE_ERROR );
+ pCtx = sqlite3_malloc( sizeof(*pCtx) );
+ if( pCtx==0 ){
+ return SQLITE_NOMEM;
+ }
+ memset(pCtx, 0, sizeof(*pCtx));
- assert( pVfs->mxPathname==MAX_PATHNAME );
- UNUSED_PARAMETER(pVfs);
+ rc = proxyCreateConchPathname(dbPath, &pCtx->conchFilePath);
+ if( rc==SQLITE_OK ){
+ rc = proxyCreateUnixFile(pCtx->conchFilePath, &pCtx->conchFile);
+ }
+ if( rc==SQLITE_OK && lockPath ){
+ pCtx->lockProxyPath = sqlite3DbStrDup(0, lockPath);
+ }
- zOut[nOut-1] = '\0';
- if( zPath[0]=='/' ){
- sqlite3_snprintf(nOut, zOut, "%s", zPath);
+ if( rc==SQLITE_OK ){
+ /* all memory is allocated, proxys are created and assigned,
+ ** switch the locking context and pMethod then return.
+ */
+ pCtx->dbPath = sqlite3DbStrDup(0, dbPath);
+ pCtx->oldLockingContext = pFile->lockingContext;
+ pFile->lockingContext = pCtx;
+ pCtx->pOldMethod = pFile->pMethod;
+ pFile->pMethod = &proxyIoMethods;
}else{
- int nCwd;
- if( getcwd(zOut, nOut-1)==0 ){
- return SQLITE_CANTOPEN;
+ if( pCtx->conchFile ){
+ rc = pCtx->conchFile->pMethod->xClose((sqlite3_file *)pCtx->conchFile);
+ if( rc ) return rc;
+ sqlite3_free(pCtx->conchFile);
+ }
+ sqlite3_free(pCtx->conchFilePath);
+ sqlite3_free(pCtx);
+ }
+ OSTRACE3("TRANSPROXY %d %s\n", pFile->h,
+ (rc==SQLITE_OK ? "ok" : "failed"));
+ return rc;
+}
+
+
+/*
+** This routine handles sqlite3_file_control() calls that are specific
+** to proxy locking.
+*/
+static int proxyFileControl(sqlite3_file *id, int op, void *pArg){
+ switch( op ){
+ case SQLITE_GET_LOCKPROXYFILE: {
+ unixFile *pFile = (unixFile*)id;
+ if( pFile->pMethod == &proxyIoMethods ){
+ proxyLockingContext *pCtx = (proxyLockingContext*)pFile->lockingContext;
+ proxyTakeConch(pFile);
+ if( pCtx->lockProxyPath ){
+ *(const char **)pArg = pCtx->lockProxyPath;
+ }else{
+ *(const char **)pArg = ":auto: (not held)";
+ }
+ } else {
+ *(const char **)pArg = NULL;
+ }
+ return SQLITE_OK;
+ }
+ case SQLITE_SET_LOCKPROXYFILE: {
+ unixFile *pFile = (unixFile*)id;
+ int rc = SQLITE_OK;
+ int isProxyStyle = (pFile->pMethod == &proxyIoMethods);
+ if( pArg==NULL || (const char *)pArg==0 ){
+ if( isProxyStyle ){
+ /* turn off proxy locking - not supported */
+ rc = SQLITE_ERROR /*SQLITE_PROTOCOL? SQLITE_MISUSE?*/;
+ }else{
+ /* turn off proxy locking - already off - NOOP */
+ rc = SQLITE_OK;
+ }
+ }else{
+ const char *proxyPath = (const char *)pArg;
+ if( isProxyStyle ){
+ proxyLockingContext *pCtx =
+ (proxyLockingContext*)pFile->lockingContext;
+ if( !strcmp(pArg, ":auto:")
+ || (pCtx->lockProxyPath &&
+ !strncmp(pCtx->lockProxyPath, proxyPath, MAXPATHLEN))
+ ){
+ rc = SQLITE_OK;
+ }else{
+ rc = switchLockProxyPath(pFile, proxyPath);
+ }
+ }else{
+ /* turn on proxy file locking */
+ rc = proxyTransformUnixFile(pFile, proxyPath);
+ }
+ }
+ return rc;
+ }
+ default: {
+ assert( 0 ); /* The call assures that only valid opcodes are sent */
}
- nCwd = strlen(zOut);
- sqlite3_snprintf(nOut-nCwd, &zOut[nCwd], "/%s", zPath);
}
- return SQLITE_OK;
+ /*NOTREACHED*/
+ return SQLITE_ERROR;
}
-
-#ifndef SQLITE_OMIT_LOAD_EXTENSION
/*
-** Interfaces for opening a shared library, finding entry points
-** within the shared library, and closing the shared library.
+** Within this division (the proxying locking implementation) the procedures
+** above this point are all utilities. The lock-related methods of the
+** proxy-locking sqlite3_io_method object follow.
*/
-#include <dlfcn.h>
-static void *unixDlOpen(sqlite3_vfs *NotUsed, const char *zFilename){
- UNUSED_PARAMETER(NotUsed);
- return dlopen(zFilename, RTLD_NOW | RTLD_GLOBAL);
-}
+
/*
-** SQLite calls this function immediately after a call to unixDlSym() or
-** unixDlOpen() fails (returns a null pointer). If a more detailed error
-** message is available, it is written to zBufOut. If no error message
-** is available, zBufOut is left unmodified and SQLite uses a default
-** error message.
+** This routine checks if there is a RESERVED lock held on the specified
+** file by this or any other process. If such a lock is held, set *pResOut
+** to a non-zero value otherwise *pResOut is set to zero. The return value
+** is set to SQLITE_OK unless an I/O error occurs during lock checking.
*/
-static void unixDlError(sqlite3_vfs *NotUsed, int nBuf, char *zBufOut){
- char *zErr;
- UNUSED_PARAMETER(NotUsed);
- unixEnterMutex();
- zErr = dlerror();
- if( zErr ){
- sqlite3_snprintf(nBuf, zBufOut, "%s", zErr);
+static int proxyCheckReservedLock(sqlite3_file *id, int *pResOut) {
+ unixFile *pFile = (unixFile*)id;
+ int rc = proxyTakeConch(pFile);
+ if( rc==SQLITE_OK ){
+ proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
+ unixFile *proxy = pCtx->lockProxy;
+ return proxy->pMethod->xCheckReservedLock((sqlite3_file*)proxy, pResOut);
}
- unixLeaveMutex();
-}
-static void *unixDlSym(sqlite3_vfs *NotUsed, void *pHandle, const char*zSymbol){
- UNUSED_PARAMETER(NotUsed);
- return dlsym(pHandle, zSymbol);
-}
-static void unixDlClose(sqlite3_vfs *NotUsed, void *pHandle){
- UNUSED_PARAMETER(NotUsed);
- dlclose(pHandle);
+ return rc;
}
-#else /* if SQLITE_OMIT_LOAD_EXTENSION is defined: */
- #define unixDlOpen 0
- #define unixDlError 0
- #define unixDlSym 0
- #define unixDlClose 0
-#endif
/*
-** Write nBuf bytes of random data to the supplied buffer zBuf.
+** Lock the file with the lock specified by parameter locktype - one
+** of the following:
+**
+** (1) SHARED_LOCK
+** (2) RESERVED_LOCK
+** (3) PENDING_LOCK
+** (4) EXCLUSIVE_LOCK
+**
+** Sometimes when requesting one lock state, additional lock states
+** are inserted in between. The locking might fail on one of the later
+** transitions leaving the lock state different from what it started but
+** still short of its goal. The following chart shows the allowed
+** transitions and the inserted intermediate states:
+**
+** UNLOCKED -> SHARED
+** SHARED -> RESERVED
+** SHARED -> (PENDING) -> EXCLUSIVE
+** RESERVED -> (PENDING) -> EXCLUSIVE
+** PENDING -> EXCLUSIVE
+**
+** This routine will only increase a lock. Use the sqlite3OsUnlock()
+** routine to lower a locking level.
*/
-static int unixRandomness(sqlite3_vfs *NotUsed, int nBuf, char *zBuf){
- UNUSED_PARAMETER(NotUsed);
- assert((size_t)nBuf>=(sizeof(time_t)+sizeof(int)));
-
- /* We have to initialize zBuf to prevent valgrind from reporting
- ** errors. The reports issued by valgrind are incorrect - we would
- ** prefer that the randomness be increased by making use of the
- ** uninitialized space in zBuf - but valgrind errors tend to worry
- ** some users. Rather than argue, it seems easier just to initialize
- ** the whole array and silence valgrind, even if that means less randomness
- ** in the random seed.
- **
- ** When testing, initializing zBuf[] to zero is all we do. That means
- ** that we always use the same random number sequence. This makes the
- ** tests repeatable.
- */
- memset(zBuf, 0, nBuf);
-#if !defined(SQLITE_TEST)
- {
- int pid, fd;
- fd = open("/dev/urandom", O_RDONLY);
- if( fd<0 ){
- time_t t;
- time(&t);
- memcpy(zBuf, &t, sizeof(t));
- pid = getpid();
- memcpy(&zBuf[sizeof(t)], &pid, sizeof(pid));
- assert( sizeof(t)+sizeof(pid)<=(size_t)nBuf );
- nBuf = sizeof(t) + sizeof(pid);
- }else{
- nBuf = read(fd, zBuf, nBuf);
- close(fd);
- }
+static int proxyLock(sqlite3_file *id, int locktype) {
+ unixFile *pFile = (unixFile*)id;
+ int rc = proxyTakeConch(pFile);
+ if( rc==SQLITE_OK ){
+ proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
+ unixFile *proxy = pCtx->lockProxy;
+ rc = proxy->pMethod->xLock((sqlite3_file*)proxy, locktype);
+ pFile->locktype = proxy->locktype;
}
-#endif
- return nBuf;
+ return rc;
}
/*
-** Sleep for a little while. Return the amount of time slept.
-** The argument is the number of microseconds we want to sleep.
-** The return value is the number of microseconds of sleep actually
-** requested from the underlying operating system, a number which
-** might be greater than or equal to the argument, but not less
-** than the argument.
+** Lower the locking level on file descriptor pFile to locktype. locktype
+** must be either NO_LOCK or SHARED_LOCK.
+**
+** If the locking level of the file descriptor is already at or below
+** the requested locking level, this routine is a no-op.
*/
-static int unixSleep(sqlite3_vfs *NotUsed, int microseconds){
-#if OS_VXWORKS
- struct timespec sp;
-
- sp.tv_sec = microseconds / 1000000;
- sp.tv_nsec = (microseconds % 1000000) * 1000;
- nanosleep(&sp, NULL);
- return microseconds;
-#elif defined(HAVE_USLEEP) && HAVE_USLEEP
- usleep(microseconds);
- return microseconds;
-#else
- int seconds = (microseconds+999999)/1000000;
- sleep(seconds);
- return seconds*1000000;
-#endif
- UNUSED_PARAMETER(NotUsed);
+static int proxyUnlock(sqlite3_file *id, int locktype) {
+ unixFile *pFile = (unixFile*)id;
+ int rc = proxyTakeConch(pFile);
+ if( rc==SQLITE_OK ){
+ proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
+ unixFile *proxy = pCtx->lockProxy;
+ rc = proxy->pMethod->xUnlock((sqlite3_file*)proxy, locktype);
+ pFile->locktype = proxy->locktype;
+ }
+ return rc;
}
/*
-** The following variable, if set to a non-zero value, is interpreted as
-** the number of seconds since 1970 and is used to set the result of
-** sqlite3OsCurrentTime() during testing.
-*/
-#ifdef SQLITE_TEST
-int sqlite3_current_time = 0; /* Fake system time in seconds since 1970. */
-#endif
-
-/*
-** Find the current time (in Universal Coordinated Time). Write the
-** current time and date as a Julian Day number into *prNow and
-** return 0. Return 1 if the time and date cannot be found.
+** Close a file that uses proxy locks.
*/
-static int unixCurrentTime(sqlite3_vfs *NotUsed, double *prNow){
-#if defined(NO_GETTOD)
- time_t t;
- time(&t);
- *prNow = t/86400.0 + 2440587.5;
-#elif OS_VXWORKS
- struct timespec sNow;
- clock_gettime(CLOCK_REALTIME, &sNow);
- *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_nsec/86400000000000.0;
-#else
- struct timeval sNow;
- gettimeofday(&sNow, 0);
- *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_usec/86400000000.0;
-#endif
-
-#ifdef SQLITE_TEST
- if( sqlite3_current_time ){
- *prNow = sqlite3_current_time/86400.0 + 2440587.5;
+static int proxyClose(sqlite3_file *id) {
+ if( id ){
+ unixFile *pFile = (unixFile*)id;
+ proxyLockingContext *pCtx = (proxyLockingContext *)pFile->lockingContext;
+ unixFile *lockProxy = pCtx->lockProxy;
+ unixFile *conchFile = pCtx->conchFile;
+ int rc = SQLITE_OK;
+
+ if( lockProxy ){
+ rc = lockProxy->pMethod->xUnlock((sqlite3_file*)lockProxy, NO_LOCK);
+ if( rc ) return rc;
+ rc = lockProxy->pMethod->xClose((sqlite3_file*)lockProxy);
+ if( rc ) return rc;
+ sqlite3_free(lockProxy);
+ pCtx->lockProxy = 0;
+ }
+ if( conchFile ){
+ if( pCtx->conchHeld ){
+ rc = proxyReleaseConch(pFile);
+ if( rc ) return rc;
+ }
+ rc = conchFile->pMethod->xClose((sqlite3_file*)conchFile);
+ if( rc ) return rc;
+ sqlite3_free(conchFile);
+ }
+ sqlite3_free(pCtx->lockProxyPath);
+ sqlite3_free(pCtx->conchFilePath);
+ sqlite3_free(pCtx->dbPath);
+ /* restore the original locking context and pMethod then close it */
+ pFile->lockingContext = pCtx->oldLockingContext;
+ pFile->pMethod = pCtx->pOldMethod;
+ sqlite3_free(pCtx);
+ return pFile->pMethod->xClose(id);
}
-#endif
- UNUSED_PARAMETER(NotUsed);
- return 0;
+ return SQLITE_OK;
}
-/*
-** We added the xGetLastError() method with the intention of providing
-** better low-level error messages when operating-system problems come up
-** during SQLite operation. But so far, none of that has been implemented
-** in the core. So this routine is never called. For now, it is merely
-** a place-holder.
-*/
-static int unixGetLastError(sqlite3_vfs *NotUsed, int NotUsed2, char *NotUsed3){
- UNUSED_PARAMETER(NotUsed);
- UNUSED_PARAMETER(NotUsed2);
- UNUSED_PARAMETER(NotUsed3);
- return 0;
-}
+
+#endif /* defined(__DARWIN__) && SQLITE_ENABLE_LOCKING_STYLE */
/*
-************************ End of sqlite3_vfs methods ***************************
+** The proxy locking style is intended for use with AFP filesystems.
+** And since AFP is only supported on MacOSX, the proxy locking is also
+** restricted to MacOSX.
+**
+**
+******************* End of the proxy lock implementation **********************
******************************************************************************/
/*
projects / thirdparty / sqlite.git / commitdiff
