POSIX find all local files

findposix

I'm trying to write a cross-platform command line that will find all local files on a file system.

My intention is to use the find command, using only POSIX parameters. Is there a way to skip files on remote filesystems, while restricted to only POSIX commands, and no programming?

My current command line is

find -type f -print

but that obviously lists every regular file, not just local ones.

I am aware of the -local flag, but as far as I can tell, it's not POSIX or otherwise part of a cross-platform standard. locate doesn't appear to be cross-platform either. I don't think the -xdev flag is what I'm looking for, either, as I want to search across all locally mounted file systems – if / and /home are both local, and I run find / -xdev -print, find should skip /home (unless I'm misunderstanding stuff).

For what it's worth, the following URLs have been useful resources…

Best Answer

I don't believe that there is any way to determine whether a filesystem is remote or local using only POSIX interfaces. POSIX doesn't even mandate a mount command, and all references to file systems, as far as I know, are replete with the words "implementation-defined".

The -local primary to the find command is not much help, since it is only available on Solaris, and it depends on the existence of a configuration file which lists remote filesystem types.

POSIX does mandate a df command and requires a specific output format if that command is used with the -P flag; in that format, there are six space-separated fields, of which the first is the "name of the file system, in an implementation-defined format" and the last is the mount point.

What you need is a list of all the mount points corresponding to local filesystems. Here, I think you're going to have to fall back to a heuristic; which is to exclude filesystem names which "look like" remote filesystems. Some examples:

//...     CIFS (Samba)
host:...  NFS
/afs      AFS

That's not an exhaustive list, and it comes completely without guarantees.

If you then filter those lines out of the output of df and then remove the first five columns, you could then feed the list to the find command with the -xdev primary.

Even if that sounds plausible, there is actually no guarantee that a filesystem is actually "mounted" (see AFS) or that a filesystem is either entirely local or entirely remote (a fuse-based filesystem can arbitrarily redirect individual files). So proceed with caution, is my advice.

Related Solutions

POSIX Find Parameters – Why They Differ from Other Program Styles

Note that POSIX find does support options like other commands, and -- to mark the end of them like other commands. Those are -H and -L.

The -print, -type... are not options, they're sometimes called predicates. They are arguments whose order matters that appear after the file paths which themselves appear after the options. You've also got ( and !. Together, they build an expression that is used to determine what files to find.

find is not the only one. [ (aka test) and expr are other commands whose arguments are used to build an expression.

Like find, [ has operators that start with - and are more than one-letter (-gt, -eq...).

Like find, test has issues where those operators may be confused with operands.

find -- "$file1" "$file2" -type f
[ -f "$file1" -a -f "$file2" ]

If $file2 is !, it's a problem with find. If it's =, it's a problem with (some) [.

For all of find, test and expr, using options to build the expression would not really have worked. Another option could have been to have one string evaluated as the expression like awk or sed. like for

find f1 f2 \( -type f -mtime -1 -o ! -type f -newer x \) -exec ls -ld {} +

Do:

find 'found = 0
      if (typeof($f) == "f") {
        if (age($f) > 1) found=1
      } else if (age($f) < age("x")) found = 1
      if ($found) exec_multi("ls -ld {}")' f1 f2

But that means implementing a grammar parser in find. That also means potential quoting nightmare for the "x" and command line above.

Actually AT&T Research did come up with such a command: tw (tree walker), but even though it's now open source, I'm not aware that it is really used out of AT&T.

Are “mostly POSIX-compliant” systems still considered POSIX systems

No. The Open Group has an actual POSIX certification process, so if an operating system hasn't been through that, it cannot be referred to as POSIX-compliant.

Best Answer

Related Solutions

POSIX Find Parameters – Why They Differ from Other Program Styles

Are “mostly POSIX-compliant” systems still considered POSIX systems

Related Question