It appears that tar doesn't always recognize (and treat as hard links) files that are the same file, but appear in two different directories from bind mounts.
$ sudo mount -o bind /tmp/foo/bar /tmp/foo/baz $ echo "test data" >/tmp/foo/bar/bleh $ stat --printf "%d %i %n\n" /tmp/foo/bar/bleh /tmp/foo/baz/bleh 51 97105 /tmp/foo/bar/bleh 51 97105 /tmp/foo/baz/bleh Notice the same device and inode number Test 1 -- using tar with a single file path: $ tar -cf /tmp/testing.tar /tmp/foo $ tar -tvf /tmp/testing.tar drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/ drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/baz/ -rw-rw-r-- derekp/derekp 10 2021-12-29 19:00 tmp/foo/baz/bleh drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/bar/ -rw-rw-r-- derekp/derekp 10 2021-12-29 19:00 tmp/foo/bar/bleh Notice that "bleh" is repeated twice as regular files. Test 2 -- adding a second path to the command line: $ tar -cf /tmp/testing.tar /tmp/foo /tmp/test $ tar -tvf /tmp/testing.tar drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/ drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/baz/ -rw-rw-r-- derekp/derekp 10 2021-12-29 19:00 tmp/foo/baz/bleh drwxrwxr-x derekp/derekp 0 2021-12-29 18:59 tmp/foo/bar/ hrw-rw-r-- derekp/derekp 0 2021-12-29 19:00 tmp/foo/bar/bleh link to tmp/foo/baz/bleh -rw-rw-r-- derekp/derekp 0 2021-12-29 19:01 tmp/test In this case it successfully recognizes /tmp/foo/bar/bleh as a hard link to /tmp/foo/baz/bleh.
