: On Wed, 10 Dec 1997, Guest section DW wrote:
: >: From: Andrea Arcangeli <arcangeli@mbox.queen.it>
: >
: >: #!/bin/sh
: >: while :; do mkdir foo; cd foo; done
: >
: >: create a full pathname that exceed a page memory (do_getname() return
: >: -ENAMETOOLONG), so can' t be referenced in any way and so can' t be
: >: deleted from rm -r etc...
: >
: >: PS. How can force the deleting of a tree created without the patch?
: >
: > An old and well-known problem.
: > The tree can be deleted by deltree (ftp.win.tue.nl:/pub/linux/misc).
: > Maybe recent GNU versions of rm will do as well.
: So a kernel patch is uneeded?
I have not seen this (your?) patch, only pointed out
that there is a program to remove deep trees.
Pathnames can be arbitrarily long, but there is a limit
to the size of pathnames that the kernel will handle.
This is how it always has been. Nothing wrong here. No patch needed.
: So why James Mastros had a filesystem panic
: after running the offending script?
How should I know? You want to cure symptoms instead of bugs?
If it is reproducible, someone should investigate this panic,
and fix whatever is wrong. Having long names is not wrong.
: And how can be useful a tree that can't be reached in any way?
It can be reached.
: At least for my own system I like my patch to be allowed to run the script
: and delete the PAGE_SIZE long tree with rm -r. For my system 4096
: character of path are well.
Ah, so it is your patch, and I understand that you want to
limit the maximum length of pathnames. That might well be a
violation of POSIX. Anyway, there is no good reason to impose
such a restriction. But on your own machine you can, of course.
I think the latest development version of rm is able to remove
very deep trees, so this deltree utility will soon be obsolete.