fdatasync hack, again

Check for ext3/ext4 fs, then check kernel version.

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index cc6eaa20bb9791354b4f9c2430770b08fea81562..b6703d6857890a9de4776255d4e14a0489b9cc3a 100644 (file)
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -1180,6 +1180,9 @@ struct MDB_env {
 #ifdef _WIN32
        int             me_pidquery;            /**< Used in OpenProcess */
 #endif
+#ifdef __linux
+       int             me_fsynconly;           /**< fdatasync is unreliable */
+#endif
 #if defined(_WIN32) || defined(MDB_USE_SYSV_SEM)
        /* Windows mutexes/SysV semaphores do not reside in shared mem */
        mdb_mutex_t     me_rmutex;
@@ -2356,6 +2359,12 @@ mdb_env_sync(MDB_env *env, int force)
                                rc = ErrCode();
 #endif
                } else {
+#ifdef __linux
+                       if (env->me_fsynconly) {
+                               if (fsync(env->me_fd))
+                                       rc = ErrCode();
+                       } else
+#endif
                        if (MDB_FDATASYNC(env->me_fd))
                                rc = ErrCode();
                }
@@ -3918,6 +3927,11 @@ mdb_fsize(HANDLE fd, size_t *size)
        return MDB_SUCCESS;
 }
 
+#ifdef __linux
+#include <sys/utsname.h>
+#include <sys/vfs.h>
+#endif
+
 /** Further setup required for opening an LMDB environment
  */
 static int ESECT
@@ -3936,6 +3950,54 @@ mdb_env_open2(MDB_env *env)
                env->me_pidquery = PROCESS_QUERY_INFORMATION;
 #endif /* _WIN32 */
 
+#ifdef __linux
+       /* ext3/ext4 fdatasync is broken on some older Linux kernels.
+        * https://lkml.org/lkml/2012/9/3/83
+        * Kernels after 3.6-rc6 are known good.
+        * https://lkml.org/lkml/2012/9/10/556
+        * See if the DB is on ext3/ext4, then check for new enough kernel
+        * Kernels 2.6.32.60, 2.6.34.15, 3.2.30, and 3.5.4 are also known
+        * to be patched.
+        */
+       {
+               struct statfs st;
+               fstatfs(env->me_fd, &st);
+               while (st.f_type == 0xEF53) {
+                       struct utsname uts;
+                       int i;
+                       uname(&uts);
+                       if (uts.release[0] < '3') {
+                               if (!strncmp(uts.release, "2.6.32.", 7)) {
+                                       i = atoi(uts.release+7);
+                                       if (i >= 60)
+                                               break;  /* 2.6.32.60 and newer is OK */
+                               } else if (!strncmp(uts.release, "2.6.34.", 7)) {
+                                       i = atoi(uts.release+7);
+                                       if (i >= 15)
+                                               break;  /* 2.6.34.15 and newer is OK */
+                               }
+                       } else if (uts.release[0] == '3') {
+                               i = atoi(uts.release+2);
+                               if (i > 5)
+                                       break;  /* 3.6 and newer is OK */
+                               if (i == 5) {
+                                       i = atoi(uts.release+4);
+                                       if (i >= 4)
+                                               break;  /* 3.5.4 and newer is OK */
+                               } else if (i == 2) {
+                                       i = atoi(uts.release+4);
+                                       if (i >= 30)
+                                               break;  /* 3.2.30 and newer is OK */
+                               }
+                       } else {        /* 4.x and newer is OK */
+                               break;
+                       }
+                       env->me_fsynconly = 1;
+                       break;
+               }
+       }
+#endif
+
        if ((i = mdb_env_read_header(env, &meta)) != 0) {
                if (i != ENOENT)
                        return i;