Add support for F2FS atomic writes. Untested at this point.

FossilOrigin-Name: 416973ede3bde8567d1f2699728f72352979e054ef988d1c1e1cfe4290f6f8b8

1 files changed, 77 insertions, 34 deletions

diff --git a/src/os_unix.c b/src/os_unix.c
index 7f0ebdba6..3397b3a21 100644
--- a/src/os_unix.c
+++ b/src/os_unix.c
@@ -90,6 +90,7 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
+#include <sys/ioctl.h>
 #include <unistd.h>
 #include <time.h>
 #include <sys/time.h>
@@ -220,10 +221,8 @@ struct unixFile {
   sqlite3_int64 mmapSizeMax;          /* Configured FCNTL_MMAP_SIZE value */
   void *pMapRegion;                   /* Memory mapped region */
 #endif
-#ifdef __QNXNTO__
   int sectorSize;                     /* Device sector size */
   int deviceCharacteristics;          /* Precomputed device characteristics */
-#endif
 #if SQLITE_ENABLE_LOCKING_STYLE
   int openFlags;                      /* The flags specified at open() */
 #endif
@@ -328,6 +327,13 @@ static pid_t randomnessPid = 0;
 # define lseek lseek64
 #endif
 
+#define F2FS_IOCTL_MAGIC        0xf5
+#define F2FS_IOC_START_ATOMIC_WRITE     _IO(F2FS_IOCTL_MAGIC, 1)
+#define F2FS_IOC_COMMIT_ATOMIC_WRITE    _IO(F2FS_IOCTL_MAGIC, 2)
+#define F2FS_IOC_START_VOLATILE_WRITE   _IO(F2FS_IOCTL_MAGIC, 3)
+#define F2FS_IOC_ABORT_VOLATILE_WRITE   _IO(F2FS_IOCTL_MAGIC, 5)
+
+
 /*
 ** Different Unix systems declare open() in different ways.  Same use
 ** open(const char*,int,mode_t).  Others use open(const char*,int,...).
@@ -500,6 +506,9 @@ static struct unix_syscall {
 #endif
 #define osLstat      ((int(*)(const char*,struct stat*))aSyscall[27].pCurrent)
 
+  { "ioctl",         (sqlite3_syscall_ptr)ioctl,          0 },
+#define osIoctl ((int(*)(int,int))aSyscall[28].pCurrent)
+
 }; /* End of the overrideable system calls */
 
 
@@ -3777,6 +3786,19 @@ static int unixGetTempname(int nBuf, char *zBuf);
 static int unixFileControl(sqlite3_file *id, int op, void *pArg){
   unixFile *pFile = (unixFile*)id;
   switch( op ){
+    case SQLITE_FCNTL_BEGIN_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_START_ATOMIC_WRITE);
+      return rc ? SQLITE_ERROR : SQLITE_OK;
+    }
+    case SQLITE_FCNTL_COMMIT_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_COMMIT_ATOMIC_WRITE);
+      return rc ? SQLITE_ERROR : SQLITE_OK;
+    }
+    case SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE: {
+      int rc = osIoctl(pFile->h, F2FS_IOC_ABORT_VOLATILE_WRITE);
+      return rc ? SQLITE_ERROR : SQLITE_OK;
+    }
+
     case SQLITE_FCNTL_LOCKSTATE: {
       *(int*)pArg = pFile->eFileLock;
       return SQLITE_OK;
@@ -3860,30 +3882,43 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
 }
 
 /*
-** 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.
+** If pFd->sectorSize is non-zero when this function is called, it is a
+** no-op. Otherwise, the values of pFd->sectorSize and 
+** pFd->deviceCharacteristics are set according to the file-system 
+** characteristics. 
 **
-** 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.
+** There are two versions of this function. One for QNX and one for all
+** other systems.
 */
-#ifndef __QNXNTO__ 
-static int unixSectorSize(sqlite3_file *NotUsed){
-  UNUSED_PARAMETER(NotUsed);
-  return SQLITE_DEFAULT_SECTOR_SIZE;
-}
-#endif
+#ifndef __QNXNTO__
+static void setDeviceCharacteristics(unixFile *pFd){
+  if( pFd->sectorSize==0 ){
+    int res;
+    assert( pFd->deviceCharacteristics==0 );
 
-/*
-** The following version of unixSectorSize() is optimized for QNX.
-*/
-#ifdef __QNXNTO__
+    /* Check for support for F2FS atomic batch writes. */
+    res = osIoctl(pFd->h, F2FS_IOC_START_VOLATILE_WRITE);
+    if( res==SQLITE_OK ){
+      osIoctl(pFd->h, F2FS_IOC_ABORT_VOLATILE_WRITE);
+      pFd->deviceCharacteristics = 
+        SQLITE_IOCAP_BATCH_ATOMIC |
+        SQLITE_IOCAP_ATOMIC |
+        SQLITE_IOCAP_SEQUENTIAL |
+        SQLITE_IOCAP_SAFE_APPEND;
+    }
+
+    /* Set the POWERSAFE_OVERWRITE flag if requested. */
+    if( pFd->ctrlFlags & UNIXFILE_PSOW ){
+      pFd->deviceCharacteristics |= SQLITE_IOCAP_POWERSAFE_OVERWRITE;
+    }
+
+    pFd->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
+  }
+}
+#else
 #include <sys/dcmd_blk.h>
 #include <sys/statvfs.h>
-static int unixSectorSize(sqlite3_file *id){
-  unixFile *pFile = (unixFile*)id;
+static void setDeviceCharacteristics(unixFile *pFile){
   if( pFile->sectorSize == 0 ){
     struct statvfs fsInfo;
        
@@ -3952,9 +3987,24 @@ static int unixSectorSize(sqlite3_file *id){
     pFile->deviceCharacteristics = 0;
     pFile->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
   }
-  return pFile->sectorSize;
 }
-#endif /* __QNXNTO__ */
+#endif
+
+/*
+** 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.
+*/
+static int unixSectorSize(sqlite3_file *id){
+  unixFile *pFd = (unixFile*)id;
+  setDeviceCharacteristics(pFd);
+  return pFd->sectorSize;
+}
 
 /*
 ** Return the device characteristics for the file.
@@ -3970,16 +4020,9 @@ static int unixSectorSize(sqlite3_file *id){
 ** available to turn it off and URI query parameter available to turn it off.
 */
 static int unixDeviceCharacteristics(sqlite3_file *id){
-  unixFile *p = (unixFile*)id;
-  int rc = 0;
-#ifdef __QNXNTO__
-  if( p->sectorSize==0 ) unixSectorSize(id);
-  rc = p->deviceCharacteristics;
-#endif
-  if( p->ctrlFlags & UNIXFILE_PSOW ){
-    rc |= SQLITE_IOCAP_POWERSAFE_OVERWRITE;
-  }
-  return rc;
+  unixFile *pFd = (unixFile*)id;
+  setDeviceCharacteristics(pFd);
+  return pFd->deviceCharacteristics;
 }
 
 #if !defined(SQLITE_OMIT_WAL) || SQLITE_MAX_MMAP_SIZE>0
@@ -7598,7 +7641,7 @@ int sqlite3_os_init(void){
 
   /* Double-check that the aSyscall[] array has been constructed
   ** correctly.  See ticket [bb3a86e890c8e96ab] */
-  assert( ArraySize(aSyscall)==28 );
+  assert( ArraySize(aSyscall)==29 );
 
   /* Register all VFSes defined in the aVfs[] array */
   for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){