bin/189139: BUG in jail(8) variable substitution, and PATCH

[Dirk Engling]

>Number:         189139
>Category:       bin
>Synopsis:       BUG in jail(8) variable substitution, and PATCH
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Apr 30 19:00:00 UTC 2014
>Closed-Date:
>Last-Modified:
>Originator:     Dirk Engling
>Release:        9.2
>Organization:
>Environment:
>Description:
The variable substitution of FreeBSD's jail tool yields unexpected results when a parameter has more than one variable to substitute and one of the later variables needs substitution as well.
>How-To-Repeat:
Consider the simple test case:

$A = "A_${B}_C_${D}";
$B = "BBBBB";
$D = "DDDDD_${E}_FFFFF";
$E = "EEEEE";

bar {
        exec.poststart = "touch /tmp/$A";
}

EXPECTED OUTCOME for running "jail -c bar" would be a file with the name /tmp/A_BBBBB_C_DDDDD_EEEEE_FFFFF to be touched (and, of course, the jail bar being created).

OBSERVED OUTCOME is a file with the name /tmp/A_BBBDDDDD_EEEEE_FFFFFBB_C_ being created.

The reason is the way jail(8) resolves recursive substitutions. In head/usr.sbin/jail/config.c:193 a varoff variable is introduced that handles a shifting offset for multiple variable substitutions per parameter. This varoff is updated after each substitution in line 239 to reflect a new offset into the parameter's string. This ensures that all other variables are substituted at [their insertion point plus varoff] which is the accumulated length of all previously substituted variables.

Now in our example, if $A is to be expanded, first ${B} is inserted at offset 2 and varoff becomes 10. When substituting ${D}, the recursion check at line 216 detects that variable $D also needs expansion. It reorders the parameter list, so that the algorithm works on variable $D now. Then it jumps to find_vars at line 191 and properly expands DDDDD_${E}_FFFFF to DDDDD_EEEEE_FFFFF.

When the algorithm now returns to expanding $A by entering the loop body again, it finds a re-set varoff variable leading to (the now expanded) variable $D being inserted at the offset 5, where the parser initially would find it (the internal format for $A is approx: { "A__C_", {2, "B"}, {5, "D"}}) and not at the corrected offset 10.
>Fix:
Get rid of the varoff and replace line 239 with:

[struct cfvar *]vv = v;
while ((vv = STAILQ_NEXT(vv, tq)))
    v->pos += vs->len;

to make the offset permanent. Find a patch here https://erdgeist.org/arts/software/jail/usr.sbin.jail-substitution.patch

>Release-Note:
>Audit-Trail:
>Unformatted: