shell utility to check if directory is an ancestor to a file

Question

How can I check that a given regular file of path f is a descendant of a directory of path d. If not for symlinks, I would just check if d (with a trailing slash) is prefix to f. Now with symlinks this gets marginally more complicated, as I could find the physical paths of the ancestors of f and compare each of them to the physical path of d. However, I believe there might be a userland utility for that, is there one?

Answer 1

1. If the ancestors of a file are defined as the directory containing it and the ancestors of that directory, i.e., the descendants of a directory are its entries and (for the entries that are directories) their own descendants:

   Assuming that $f is the path of a regular file (and not the path of a symbolic link), you can find the “real” directory containing it with $(cd -- "$(dirname "$f")" && command -p pwd). Similarly, find the “real” location of the directory $d with $(cd -- "$d" && command -p pwd). In bash/ksh/zsh, you can replace command -p pwd by pwd -P.

   If $f may be a symbolic link, it's more complicated. Some systems have a readlink command, for example you could use $(dirname "$(readlink -f -- "$f")) with reasonably recent GNU coreutils, but note that there are other systems with an incompatible readlink command or none at all.

   Note that the file may be moved between the time you make the check and the time you use the result. There are also corner cases where the question doesn't have a clear yes/no answer, such as when the path from the directory to the file traverses a mount point.

   If all you want to know is which filesystem $f is on, run df "$f".

2. If the descendants of a directory $d are defined as the files that can be given a name of the form $d/x1/x2/.../xn where the xi can be directories or symbolic links:

   Testing this property requires a traversal of the whole directory tree rooted at $d, following all symbolic links. The following command should do this (warning, typed directly into the browser; requires GNU find):

   [ -n "$(find -L "$d" -samefile "$f" -print -quit)" ]
   

   If $f has hard links, this may access the file $f through a different name. If you require the file to be found under the name $f, it gets more complicated; here's an attempt (again, typed directly into the browser; requires GNU find and GNU grep; assumes no newlines in file names):

   f_directory="$(cd -- "$(dirname "$f")" && command -p pwd)"
   f_truename="$f_directory/$(basename "$f")"
   [ -n "$(find -L "$(cd "$d" && command -p pwd)" -exec readlink -f \; |
           grep -F -q -x "$f_truename")" ]
   

   No matter what exact definition you choose, this property is not very robust since you can change its value by adding or removing a symbolic link deep underneath $d.

Answer 2

My 'ls' command has flag "--dereference-command-line-symlink-to-dir" that shows the full path to the file. This 'ls' is from GNU coreutils 6.9.