[Turn] discussion on synchronous Io:fsync and Fdatasync

Original link:

Http://www.cnblogs.com/promise6522/archive/2012/05/27/2520028.html

The writing is very good and very detailed. Summary:

Fsync need sync file modification and metadata; Fdatasync only sync metadata when necessary to save a file sync operation.

Berkeley DB ensures that Fdatasync does not go to sync metadata by pre-setting the log file size.

Here is a simple list of the corresponding BDB code:

OS/OS_FSYNC.C:

__os_fsync ()if(Db_global (j_fsync)! =NULL) ret= Db_global (J_fsync) (fhp->FD); User-defined sync functionElse {#ifDefined (F_fullfsync)//MAC OS XRetry_chk ((Fcntl (FHP->FD, F_fullfsync,0) ), ret);        if(ret = =enotsup) Retry_chk ((Fsync (FHP-FD)), ret);#elifDefined (HAVE_QNX)//QNX real-time OSret=__qnx_fsync (FHP);#elifDefined (Have_fdatasync)//priority use FdatasyncRetry_chk ((Fdatasync (FHP-FD)), ret);#elseRetry_chk ((Fsync (FHP-FD)), ret);#endif    }

LOG/LOG_PUT.C:

__log_write () #ifdef Have_filesystem_notzeroif(Lp->w_off = =0&&!__os_fs_notzero ()) {#else    if(Lp->w_off = =0) {//Lp->w_off write offset for the current log file#endif//Lp->w_off is 0, marking the first time this log file is written        (void) __db_file_extend (env, DBLP->LFHP, lp->log_size); if( f_isset (DBLP, Dblog_zero)) (void) __db_zero_extend (env, dblp->LFHP,0, Lp->log_size/lp->buffer_size, lp->buffer_size); }

ENV/ENV_FILE.C:

__db_file_extend ()//size is a log file, the last byte of the write file is    0 '  / ' ;     sizeof (BUF)) / megabyte);     sizeof (BUF)) % megabyte);     if 0 sizeof (buf), &NW);         

[Turn] discussion on synchronous Io:fsync and Fdatasync