+++ This bug was initially created as a clone of Bug #963769 +++

Description of problem:
When using simple English word list (from package words), cracklib-check doesn't recognize passwords with simple modifications as based on dictionary words.

Version-Release number of selected component (if applicable):
cracklib-2.8.16-4.el6.x86_64

How reproducible:
Always

Steps to Reproduce:
1. create-cracklib-dict /usr/share/dict/words
2. echo "d.ncing" | cracklib-check
3. echo "d@ncing" | cracklib-check
4. echo "glo8ulin" | cracklib-check
5. echo "dancingdancing" | cracklib-check
6. echo "ancingd" | cracklib-check # "dancing" barrel-shifted by one letter
7. echo "1dancing1" | cracklib-check
8. echo "aca7alectic" | cracklib-check

Actual results:
d.ncing: OK
d@ncing: OK
glo8ulin: OK
dancingdancing: OK
ancingd: OK
1dancing1: OK
aca7alectic: OK

Expected results:
d.ncing: it is based on a dictionary word
d@ncing: it is based on a dictionary word
glo8ulin: it is based on a dictionary word
dancingdancing: it is based on a dictionary word
ancingd: it is based on a dictionary word
1dancing1: it is based on a dictionary word
aca7alectic: it is based on a dictionary word

Additional info:
Considering that passwords such as "-*!dancing,-=" are rejected, not rejecting much simpler changes/substitutions is unexpected.

As most of those modifications are present in standard rules used by John The Ripper they should be rejected by cracklib.

While some of those cases are introducing completely new tests, the list of missing leetspeak substitutions is quite glaring.

Proposed list of substitutions:
@ -> a
4 -> a
8 -> b (B)
3 -> e (E)
6 -> g
9 -> g
! -> i or l
1 -> i or l
$ -> s
5 -> s
7 -> t (T)
+ -> t
| -> i or l

--- Additional comment from Nalin Dahyabhai on 2013-05-17 15:04:04 EDT ---

These aren't integrated properly, but doing it in a way that wouldn't explode the destructors list would require redoing a few layers of logic.

--- Additional comment from Hubert Kario on 2013-05-20 04:42:42 EDT ---

Regarding patch in Comment #2: I don't know how the code later searches if the word is in wordlist or not, but I know that pam_cracklib does check if a password doesn't contain words as a substring of "passwordpassword". This finds not only repetitions but such simple barrel shifts like in the "ancingd" case.

Regarding patch in Comment #3: I'm not sure if I read the patch correctly, but the a -> . substitution was more to show that obfuscating a single character is enough to render cracklib useless, not an actual "l33t speak" substitution.
And second "I'm not sure if I read the patch correctly" is the 8 -> B substitution: doesn't cracklib do a case insensitive search over its wordlist?

--- Additional comment from Nalin Dahyabhai on 2013-05-20 15:57:13 EDT ---

(In reply to Hubert Kario from comment #4)
> Regarding patch in Comment #2: I don't know how the code later searches if
> the word is in wordlist or not, but I know that pam_cracklib does check if a
> password doesn't contain words as a substring of "passwordpassword". This
> finds not only repetitions but such simple barrel shifts like in the
> "ancingd" case.

The pam_cracklib module (and the cracklib python module, and apparently everything else that uses cracklib) implements additional logic beyond what cracklib does, so cracklib-check doesn't benefit from those efforts.

> Regarding patch in Comment #3: I'm not sure if I read the patch correctly,
> but the a -> . substitution was more to show that obfuscating a single
> character is enough to render cracklib useless, not an actual "l33t speak"
> substitution.
> And second "I'm not sure if I read the patch correctly" is the 8 -> B
> substitution: doesn't cracklib do a case insensitive search over its
> wordlist?

So it does.  That rule should be discarded from the patch, then.  As for the larger point, you're right, cracklib's doesn't catch everything.