Re: [fuse-devel] fuseiso: Transport endpoint not connected

[Jonathan B. <jbr...@ea...>]

On Thu, 2006-01-05 at 10:41 +0100, Miklos Szeredi wrote:
> > I know that you folks aren't responsible for developing fuseiso, but I'm
> > hoping that someone here knows how it is supposed to work, or may be a
> > developer for fuseiso.
> > 
> > I am currently using fuse 2.4.0 on kernel 2.6.14 (Debian) on an amd64
> > system running in 64 bit mode (but the same issue exists in IA32 mode,
> > too).  I am trying to use fuseiso version 20051029, but it isn't quite
> > working.  When used as directed (according to fuseiso --help):
> > $ fuseiso debian-31r0a-amd64-netinst.iso ./test
> > 
> > I get this error whenever I try to ls below the mount point:
> > Transport endpoint is not connected
> > 
> > It appears that fuseiso is exiting immediately after invocation without
> > any warning or error messages, while leaving an entry in the mount table
> > that the fs is mounted.  How is it supposed to work?
> 
> Probably not like that ;)
> 
> You can try running in the foreground (-f) or with debug (-d) to see
> how it crashes.

When run with either -f or -d, it works!  That suggests to me that the
problem is in the backgrounding.  After a little digging, I found that
if fuseiso is given the full path to the .iso image file rather than the
relative path, it works.

Guess: useiso only opens the image file in its .init function.  This
function is called after libfuse calls daemon().  daemon(0, 0) changes
the working directory to /.  Therefore, a relative path is no longer
valid, and the program exits.

Is this fuseiso's fault for not changing the filename to an absolute
path reference before calling fuse_main() (or just not open()ing the
file earlier)?  Or is it libfuse's fault for calling daemon() before the
client program's .init() function?  I can submit a patch in either case.

-Jonathan