>
> I'm going to attempt some source staring;
And there was much rejoicing in the streets.... ;-)
> if someone could confirm or deny the following assertions (and add
> appropriate additional information), that would help me.
>
> 1) The corruption always takes the form of a 4k aligned page of zeros
> showing up.
>
I believe this is true. Or more precisely, in my experience, this is
always the form it takes.
> 2) The corruption does not happen to directories.
>
True. All of the bad entries I have run across were parts of regular
files. I have not had a directory entry be corrupted.
> 3) The corruption happens to both existing and new files.
>
True. In fact, it very often seems to happen more often in the existing
files. For example, I saw the corruption most often when I had a single
Hurd partition. When I built ncurses I had two crashes during the build
cycle (at the time there were some memory leaks that have since been
corrected). After the second crash, some of the existing terminfo entries
were corrupted. But these files should not have been touched by the build
(this was not a "make install" situation -- everything should have stayed
in the build directory).
On another occassion, various system libraries were hit by the corruption.
> 4) The corruption happens only when files are being extended.
>
Unknown. I can say that the files that are corrupted are not generally
the ones being extended (as in the terminfo or system library cases above).
> 5) The corruption happens only in the middle of files.
>
This seems to be true.
> 6) If you have a corrupt filesystem, and shut it down sanely, and then
> fsck it, fsck reports no errors. (In other words, the corruption
> is not associated with block map errors or such.)
>
Not true. I see fsck errors fairly frequently with the corruption.
One interesting side-note is that the corruption seems to be limited to
the "ext2fs.static" server. My understanding is that ext2fs.static is used
for the root partition, but the "dynamic" ext2fs is used for all other
mounted partitions.
Once I started building on a separate (mounted) partition, I stopped seeing
the corruption. This may just be luck or coincidence, however I can state
that
I would experience corruption on any build where I ran out of memory,
whereas
with the newer "two-partition" setup, I do not get corruption, even in the
time
or two I have crashed during a build.
Hope that helps,
-Brent
> Thomas
>
