On 09/27/2010 12:40 PM, Stefan Berger wrote:
V2 changes:
following Eric's comments:
- commands in the script are assigned using cmd='...' rather than
cmd="..."
- invoke commands using eval res=... rather than res=eval ...
- rewrote function escaping ` and single quotes (which became more tricky)
In this patch I am extending the rule instantiator to create the comment
node where supported, which is the case for iptables and ip6tables.
Since commands are written in the format
cmd='iptables ...-m comment --comment \"\" '
certain characters ('`) in the comment need to be escaped to
prevent comments from becoming commands themselves or cause other
forms of (bash) substitutions. I have tested this with various input and in
my tests the input made it straight into the comment. A test case for TCK
will be provided separately that tests this.
At first, I failed to see how ` needs escaping inside '', unless you
aren't uniformly using '' like you think you are. Then it hit me - yuck
- we aren't uniformly using '' - we really do have to escape ` inside
``, or come up with an alternate solution. That is:
ret=`iptables -m comment --comment '`'`
is indeed ill-formed. But if you are going to escape `, then you also
have to escape \ and $ for the same duration. Yuck again.
But fear not - I have a slicker solution:
comment='comment with lone '\'', ", `, \, $x, and two spaces'
cmd='iptables ...-m comment --comment "$comment"'
eval ret=\`"$cmd"\`
which at expansion time results in:
eval ret=\`"$cmd"\`
ret=`iptables ...-m comment --comment "$comment"`
iptables ...-m comment --comment \
'comment with lone '\'', `, ", `, \, $x, and two spaces'
That is, rather than trying to pass the comment literally through $cmd
(and thus having to carefully escape $cmd for its eventual use inside
``), it might be nicer to stick the comment in an intermediate variable
(where you only need to escape ') and make $cmd reference the
intermediate variable (where you won't have any problematic uses of ",
`, or ' to contend with, and where your only $ is one where you
intentionally want variable expansion).
@@ -52,10 +53,10 @@
#define CMD_SEPARATOR "\n"
-#define CMD_DEF_PRE "cmd=\""
-#define CMD_DEF_POST "\""
+#define CMD_DEF_PRE "cmd='"
+#define CMD_DEF_POST "'"
#define CMD_DEF(X) CMD_DEF_PRE X CMD_DEF_POST
-#define CMD_EXEC "res=`${cmd}`" CMD_SEPARATOR
+#define CMD_EXEC "eval res=\\`\"${cmd}\"\\`" CMD_SEPARATOR
This part seems okay - the ` is quoted to protect it from evaluation
until after eval has had a chance to collect its arguments.
+static char *
+escapeComment(const char *buf)
+{
+ char *res;
+ size_t i, j, add = 0, len = strlen(buf);
+
+ static const char SINGLEQUOTE_REPLACEMENT[12] =
"'\\'\\\"\\'\\\"\\''";
That seems rather long to me. Broken into chunks with C-string escaping
eliminated:
' \'\"\'\"\' '
end the current single quoting
the 5 literal shell chars '"'"'
resume single quoting
I'm not following why we need 5 literal shell characters, instead of 1.
+
+ if (len > IPTABLES_MAX_COMMENT_SIZE)
+ len = IPTABLES_MAX_COMMENT_SIZE;
+
+ for (i = 0; i < len; i++) {
+ if (buf[i] == '`')
+ add++;
+ else if (buf[i] == '\'')
+ add += sizeof(SINGLEQUOTE_REPLACEMENT);
+ }
Back to my observation that this doesn't help for \ or $, the other two
characters that need escaping inside ``. It would be so much easier if
we could rely on $() instead of ``. Wait a minute - this is only done
for Linux hosts, where we are guaranteed a sane shell (it's pretty much
just Solaris where /bin/sh is too puny to do $()) - using $() instead of
`` would solve a lot of escaping issues.
@@ -993,6 +1034,16 @@ iptablesHandleIpHdr(virBufferPtr buf,
}
}
+ if (HAS_ENTRY_ITEM(&ipHdr->dataComment)) {
+ char *cmt = escapeComment(ipHdr->dataComment.u.string);
+ if (!cmt)
+ goto err_exit;
+ virBufferVSprintf(buf,
+ " -m comment --comment
'\\''%s'\\''",
+ cmt);
+ VIR_FREE(cmt);
OK, maybe I see why your comment had such a long replacement for ',
because you aren't adding any escaping to cmt here. But I still think
we can come up with a more elegant solution, by using the intermediate
variable. Thanks for forcing me to explain myself - it's an interesting
process thinking about this.
[Do I even dare mention that at an even higher layer, it might be nicer
to avoid /bin/sh in the first place, and instead put effort into my
pending virCommand API patches to make it easier to directly invoke all
the iptables commands from C?]
--
Eric Blake eblake(a)redhat.com +1-801-349-2682
Libvirt virtualization library
http://libvirt.org