On Sun, 25 Jan 2004 11:44:01 +0100 (CET)
Lothar Hansche
You can read it in the SuSE handbook, which properties the types of file system have. Reiser cannot restore the blocks on the harddisk self, but the Reiser journal only. The physical data can be corrupt and reiserfs can not repair them!!!
I think you misunderstood the handbook. The journaling implementation of reiserfs in the SuSE kernel uses the same algorithms as the ext3 one, although implemented by different code (in the SuSE kernel both default to "ordered data" and use simple block based write ahead metadata logging) As for physical data safety, no file system (including full data journaled ones) can guarantee not corrupted user data over crashes. That's because the file system has no idea how the user data must look like to be consistent. Only the application can do that by doing a kind of journaling using O_SYNC/fsync() on its own. (many serious applications like mail servers do that) However for performance all file system IO happens in write behind mode, so it's quite possible that you see older data after a crash. If the file system wouldn't do write behind you wouldn't be satisfied with its performance. Also in addition the disks are doing their own buffering, which may also lose data on hardware problems or failing power outside the control of the file system. What the file system can do is to prevent you from ever seeing data belonging to other files and both reiserfs (in the SuSE kernel) and ext3 and XFS and JFS guarantee do that. But of course they cannot do anything when data corrupts on the disk for reasons beyond the control of the file system. -Andi