From 76865258cf4b82c8ffcc04293b5aa3940ed55d5a Mon Sep 17 00:00:00 2001
From: Benguang Zhao <bgzhao@taosdata.com>
Date: Fri, 30 Sep 2022 15:49:23 +0800
Subject: [PATCH] enh: protect WAL meta by flushing to a tmpfile at first and
 then renaming

---
 source/libs/tdb/src/db/tdbDb.c |  4 +--
 source/libs/wal/src/walMeta.c  | 54 +++++++++++++++++++++++++++++-----
 2 files changed, 48 insertions(+), 10 deletions(-)

diff --git a/source/libs/tdb/src/db/tdbDb.c b/source/libs/tdb/src/db/tdbDb.c
index a7fc50dc86..6c01348bc2 100644
--- a/source/libs/tdb/src/db/tdbDb.c
+++ b/source/libs/tdb/src/db/tdbDb.c
@@ -121,7 +121,7 @@ int32_t tdbCommit(TDB *pDb, TXN *pTxn) {
   for (pPager = pDb->pgrList; pPager; pPager = pPager->pNext) {
     ret = tdbPagerCommit(pPager, pTxn);
     if (ret < 0) {
-      tdbError("failed to commit pager. dbName:%s, txnId:%d", pDb->dbName, pTxn->txnId);
+      tdbError("failed to commit pager since %s. dbName:%s, txnId:%d", tstrerror(terrno), pDb->dbName, pTxn->txnId);
       return -1;
     }
   }
@@ -136,7 +136,7 @@ int32_t tdbAbort(TDB *pDb, TXN *pTxn) {
   for (pPager = pDb->pgrList; pPager; pPager = pPager->pNext) {
     ret = tdbPagerAbort(pPager, pTxn);
     if (ret < 0) {
-      tdbError("failed to abort pager. dbName:%s, txnId:%d", pDb->dbName, pTxn->txnId);
+      tdbError("failed to abort pager since %s. dbName:%s, txnId:%d", tstrerror(terrno), pDb->dbName, pTxn->txnId);
       return -1;
     }
   }
diff --git a/source/libs/wal/src/walMeta.c b/source/libs/wal/src/walMeta.c
index de9c115067..597fd0000b 100644
--- a/source/libs/wal/src/walMeta.c
+++ b/source/libs/wal/src/walMeta.c
@@ -35,8 +35,12 @@ int64_t FORCE_INLINE walGetCommittedVer(SWal* pWal) { return pWal->vers.commitVe
 
 int64_t FORCE_INLINE walGetAppliedVer(SWal* pWal) { return pWal->vers.appliedVer; }
 
-static FORCE_INLINE void walBuildMetaName(SWal* pWal, int metaVer, char* buf) {
-  sprintf(buf, "%s/meta-ver%d", pWal->path, metaVer);
+static FORCE_INLINE int walBuildMetaName(SWal* pWal, int metaVer, char* buf) {
+  return sprintf(buf, "%s/meta-ver%d", pWal->path, metaVer);
+}
+
+static FORCE_INLINE int walBuildTmpMetaName(SWal* pWal, char* buf) {
+  return sprintf(buf, "%s/meta-ver.tmp", pWal->path);
 }
 
 static FORCE_INLINE int64_t walScanLogGetLastVer(SWal* pWal) {
@@ -578,22 +582,51 @@ static int walFindCurMetaVer(SWal* pWal) {
 int walSaveMeta(SWal* pWal) {
   int  metaVer = walFindCurMetaVer(pWal);
   char fnameStr[WAL_FILE_LEN];
-  walBuildMetaName(pWal, metaVer + 1, fnameStr);
-  TdFilePtr pMetaFile = taosOpenFile(fnameStr, TD_FILE_CREATE | TD_FILE_WRITE);
+  char tmpFnameStr[WAL_FILE_LEN];
+  int n;
+
+  // flush to a tmpfile
+  n = walBuildTmpMetaName(pWal, tmpFnameStr);
+  ASSERT(n < sizeof(tmpFnameStr) && "Buffer overflow of file name");
+
+  TdFilePtr pMetaFile = taosOpenFile(tmpFnameStr, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_TRUNC);
   if (pMetaFile == NULL) {
+    wError("failed to open file due to %s. file:%s", strerror(errno), tmpFnameStr);
+    terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
   }
+
   char* serialized = walMetaSerialize(pWal);
   int   len = strlen(serialized);
   if (len != taosWriteFile(pMetaFile, serialized, len)) {
     // TODO:clean file
+    wError("failed to write file due to %s. file:%s", strerror(errno), tmpFnameStr);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
 
-    taosCloseFile(&pMetaFile);
-    taosRemoveFile(fnameStr);
-    return -1;
+  if (taosFsyncFile(pMetaFile) < 0) {
+    wError("failed to sync file due to %s. file:%s", strerror(errno), tmpFnameStr);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  if (taosCloseFile(&pMetaFile) < 0) {
+    wError("failed to close file due to %s. file:%s", strerror(errno), tmpFnameStr);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  // rename it
+  n = walBuildMetaName(pWal, metaVer + 1, fnameStr);
+  ASSERT(n < sizeof(fnameStr) && "Buffer overflow of file name");
+
+  if (taosRenameFile(tmpFnameStr, fnameStr) < 0) {
+    wError("failed to rename file due to %s. dest:%s", strerror(errno), fnameStr);
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
   }
 
-  taosCloseFile(&pMetaFile);
   // delete old file
   if (metaVer > -1) {
     walBuildMetaName(pWal, metaVer, fnameStr);
@@ -601,6 +634,11 @@ int walSaveMeta(SWal* pWal) {
   }
   taosMemoryFree(serialized);
   return 0;
+
+_err:
+  taosCloseFile(&pMetaFile);
+  taosMemoryFree(serialized);
+  return -1;
 }
 
 int walLoadMeta(SWal* pWal) {
-- 
GitLab