Re: Symlink indirection

[Amos Waterland <apw at us dot ibm dot com>]

I think that you and Richard are dicussing a slightly different issue
than the original poster asked about.  The original question was:

On Fri, 13 Dec 2002, Andrew Walrond wrote:
> Is the number of allowed levels of symlink indirection (if that is the 
> right phrase; I mean symlink -> symlink -> ... -> file) dependant on the 
> kernel, or libc ? Where is it defined, and can it be changed?

To which Richard replied:

> Since a symlink is just a file containing a name, the resulting path
> length is simply the maximum path length that user-space tools allow.
> This should be defined as "PATH_MAX". Posix defines this as 255
> characters but I think posix requires that this be the minimum and all
> file-name handling buffers must be at least PATH_MAX in length.
>
> A hard link is just another directory-entry for the same file. This,
> therefore follows the same rules. There must be enough space on the
> device to contain the number of directory entries, as well as enough
> buffer length in the tools necessary to manipulate these "nested"
> directories, which are not really "nested" at all. 

But Richard is not actually completely correct.  There is a limit of 5
levels of symlink indirection in vanilla 2.4 series Linux kernels.

  % touch 0
  % for i in `seq 1 10`; do ln -s `ls | sort | tail -1` $i; done
  % ls
  0  1  10  2  3  4  5  6  7  8  9
  % cat 5
  % cat 6
  cat: 6: Too many levels of symbolic links
  % strace cat 6 2>&1 | grep 'open("6",'
  open("6", O_RDONLY|O_LARGEFILE) = -1 ELOOP (Too many levels of symbolic links)

This has been discussed by Al Viro et al. many times on lkml.  I believe
that it is not a user-space or POSIX issue, but rather a kernel issue.
Thanks.

Amos Waterland