Re: /hurd/tmpfs problem

[Michael Teichgraeber]

Using tmpfs as an active translator, its behaviour is different on my
machine: A tmpfs process is running, but the underlying node does not
get translated into a directory.

According to `manual bootstrapping a translator' of Neal Walfield I
started a gdb session on tmpfs. For instance, i tried to translate the
file node `./fs' into a tmp file system. I realized that, when gdb is
waiting for input after finishing fsys_startup (and therefore tmpfs is
stopped), `fs' is seen as a directory:

E.g. `test -d fs && echo "dir"' first is blocking, since tmpfs is
stopped by gdb; a `cont' in gdb then causes "dir" to be echoed. So for
one moment tmpfs seems to correctly translate `fs'. (Similarly, a
`ls --color=auto fs' instead of using `test -d' will show `fs' as a
blue directory entry)

After that, tmpfs is running (main thread has exited), but `fs' is a
file node again (that it was before), and no translation occurs.

Maybe somewhat happens in tmfs/{dir,node}.c or in libdiskfs. Any
suggestions where to set breakpoints and which information to look at?