I said:

> It occured to me after failing with both the doreallocblks and doasyncfree
> suggestions that I have been having this problem since long before the
> integration of the new 4.4 file system.

To which Mark replied:

> I had similar problems that resulted in a few disk deaths, always lots of
> duplicate inodes and corrupt dirs ... so I dumped, redid a low-level format,
> restored, and since then I haven't had the problems.

Thanks for this suggestion, but this doesn't sound like my situation since:

> SCSI disks are supposed to remap automagically, but more than once I've had
> cases where borderline sectors don't get remapped, ...  but a fresh low level
> format cleans up the the problem.

1) I have ide disks (though I guess they could be similar to scsi in that they
   claim to automagically handle bad sectors but might not for marginal ones).

> I noticed it was always the same
> inodes (in general) that at least would start off the spew from fsck.

2) In my case, the inodes are not always they same.  I am also seeing
   corruption where blocks from one file get written to another, which fsck
   doesn't pick up...

3) Finally, I can reproduce the problem using a script that exercises the
   file system on at least three different machines, two of which have the
   same brand motherboard and disks, but the third is completely different
   in all respects ...

As I explained earlier, when I first encountered this problem back in Jan/Feb I
figured it had to be my hardware, since no one else seemed to be having these
problems.  When this "duplicated inodes" thread started, I thought maybe it was
a more widespread problem.  But now I seem to be the only one having problems
when doreallocblks and/or doayncfree are turned off so I'm not sure what to
think :-(

Duncan