iptables-restore: fix parameter parsing (shows up with gcc-4.7)
This patch fixes parameter parsing in iptables-restore since time ago. The
problem has shown up with gcc-4.7. This version of gcc seem to perform more
agressive memory management than previous.
Peter Lekensteyn provided the following sample code similar to the one
in iptables-restore:
Many may expect 0123 as output. But GCC 4.7 does not do that when compiling
with optimization enabled (-O1 and higher). It instead puts random data in the
first bytes of the character array, which becomes:
| 0 | 1 | 2 | 3 | 4 |
| RANDOM | '3' | '\0' |
Since the array is declared inside the scope of loop's body, you can think of
it as of a new array being allocated in the automatic storage area for each
loop iteration.
The correct code should be:
char x[5];
for (;;) {
x[i] = '0' + i;
if (++i == 4) {
x[i] = '\0'; /* terminate string with null byte */
printf("%s\n", x);
break;
}
}
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Apply instead a patch that really clarifies the bug in iptables-restore.
This should be good for the record (specifically, for distributors so
they can find the fix by googling).
for (curchar = parsestart; *curchar; curchar++) {
- char param_buffer[1024];
if (quote_open) {
if (escaped) {
But I have hard time to apply this patch in such a way. Instead, I came
up with the idea of this cleanup, which does not harm after all (and fixes
the issue for us).
Someone in:
https://bugzilla.redhat.com/show_bug.cgi?id=82579
put some light on this:
"Yes, I ran into this too. The issue is that the gcc optimizer is
optimizing out the code that collects quoted strings in
iptables-restore.c at line 396. If inside a quotemark and it hasn't
seen another one yet, it executes
param_buffer[param_len++] = *curchar;
continue;
At -O1 or higher, the write to param_buffer[] never happens. It just
increments param_len and continues.
Moving the definition of char param_buffer[1024]; outside the loop
fixes it. Why, I'm not sure. Defining the param_buffer[] inside the
loop should simply restrict its scope to inside the loop."
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Hans Schillstrom [Tue, 17 Jul 2012 16:27:24 +0000 (18:27 +0200)]
libxt_HMARK: correct a number of errors introduced by Pablo's rework
* Fix typo in --hmark-rnd description.
* Remove trailing -set from port and spi options.
* Take missing value for ports and spi from command line.
* Fix spi / port validation.
* Remove --hmark-offset as mandatory.
Signed-off-by: Hans Schillstrom <hans@schillstrom.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
libxt_hashlimit: add support for byte-based operation
allows --hashlimit-(upto|above) Xb/s [ --hashlimit-burst Yb ]
to make hashlimit match when X bytes/second are exceeded;
optionally, Y bytes will not be matched (i.e. bursted).
[ Pablo fixed minor compilation warning in this patch with gcc-4.6 and x86_64 ]
libxt_hashlimit.c: In function ‘parse_bytes’:
libxt_hashlimit.c:216:6: warning: format ‘%llu’ expects argument of type ‘long long unsigned int’, but argument 3 has type ‘uint64_t’ [-Wformat]
Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Hans Schillstrom [Mon, 23 Apr 2012 03:35:28 +0000 (03:35 +0000)]
extensions: add HMARK target
The target allows you to set mark packets based Jenkins' hash calculation:
h(t, rnd) = x
mark = (x % mod) + offset
where:
* t is a tuple that is used for the hashing:
t = [ src, dst, proto, sport, dport ]
Note that you can customize the tuple, thus, removing some component
that you don't want to use for the calculation. You can also use spi
instead of sport and dport, btw.
* rnd is the random seed that is explicitly passed via --hmark-rnd
* mod is the modulus, to determine the range of possible marks
* offset determines where the mark starts from
This target only works for the "raw" and "mangle" tables.
This can be used to distribute flows between a cluster of
systems and uplinks.
Initially based on work from Hans Schillingstrom. Pablo took it
over and introduced several improvements.
Signed-off-by: Hans Schillstrom <hans.schillstrom@ericsson.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch adds generic functions to return the mask in CIDR
notation whenever is possible.
This patch also simplifies xtables_ip[6]mask_to_numeric, that
now use these new two functions.
This patch also bumps libxtables_vcurrent and libxtables_vage
since we added a couple new interfaces (thanks to Jan Engelhardt
for his little reminder on this).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Florian Westphal [Thu, 17 May 2012 01:03:08 +0000 (01:03 +0000)]
extensions: libxt_rateest: output all options in save hook
ipt-restore fails to parse the ipt-save output:
zmatches -m rateest --rateest RE1 --rateest-pps --rateest-lt 5
(should be "--rateest-pps 5 --rateest-lt"). Also, the "delta" option
was never shown in -save output, but twice in some cases when using
"iptables -L".
Also, the "b/pps1" option must be shown when "delta" option is used with
relative mode.
Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
A bug has been accidentally introduced in --ulog-cprange, limiting
possible values from 1 to 50. However, that limit should be applied
to --ulog-qthreshold.
Reported-by: Gaurav Sinha <vgsinha@gmail.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
src: mark newly opened fds as FD_CLOEXEC (close on exec)
By default, Unix-like systems leak file descriptors after fork/exec
call. I think this seem to result in SELinux spotting a strange AVC
log messages according to what I can find on the web.
Fedora 18 iptables source includes this change.
Maciej says:
"iptables does potentially fork/exec modprobe to load modules.
That can cause a selinux 'domain'/'role'/whatever-it-is-called crossing.
You can do automated inspection of what gets carried across such
privilege changes and any unexpected open file descriptors flag
problems, patches like this cut down on the noise."
Signed-off-by: Maciej enczykowski <maze@google.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Franz Flasch [Thu, 8 Mar 2012 04:20:41 +0000 (04:20 +0000)]
iptables: missing free() in function delete_entry()
Fixed a memory leak in the dry run path of function delete_entry().
Signed-off-by: Franz Flasch <franz.flasch@frequentis.com> Signed-off-by: Christian Engelmayer <christian.engelmayer@frequentis.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Franz Flasch [Thu, 8 Mar 2012 04:20:37 +0000 (04:20 +0000)]
iptables: missing free() in function cache_add_entry()
Fixed a memory leak in the error path of function cache_add_entry().
Signed-off-by: Franz Flasch <franz.flasch@frequentis.com> Signed-off-by: Christian Engelmayer <christian.engelmayer@frequentis.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Jan Engelhardt [Fri, 30 Dec 2011 01:14:00 +0000 (02:14 +0100)]
extensions: link on libxtables and check symbols
Have each extension link against libxtables.so; with this, all home
symbols are known at link time and we can use ld's --no-undefined to
run the check, dropping the homebrew solution.
By having libxtables.so required by extensions, package managers'
automatic dependency discovery will become effective so that manual
dependencies for distros with split extension packages (e.g. OpenWRT)
will not be necessary anymore.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Florian Westphal [Fri, 16 Dec 2011 17:34:06 +0000 (18:34 +0100)]
libxt_connbytes: fix handling of --connbytes FROM
quoting man page:
match packets from a connection whose packets/bytes/average
packet size is more than FROM and less than TO bytes/packets. if
TO is omitted only FROM check is done.
But, when TO was omitted, we did treat it like "x:x" which is not
the same at all.
Before commit 09631dc60ce41bc484a42fcf4d4ddf7036820bd1
(libxt_connbytes: use guided option parser), we failed to parse
"--connbytes x" ('Bad range "x"'), but treated "x:" like "x:0xffffffff".
Also, restore the "from must be smaller than to" check.
Jan Engelhardt [Sun, 18 Sep 2011 13:06:05 +0000 (15:06 +0200)]
build: restore build order of modules
iptables(exe) requires libext.a, but extensions/ require libxtables.la
(in iptables/). This circular dependency does not work out, so
separate libxtables into its own directory and put it in front.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sat, 27 Aug 2011 09:12:49 +0000 (11:12 +0200)]
libiptc: combine common types
Make an xt_chainlabel type out of ipt_chainlabel and ip6t_chainlabel,
and add backward-API #defines. The ABI naturally does not change
either, so no soversion bump.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 11 Sep 2011 15:24:26 +0000 (17:24 +0200)]
libiptc: resolve compile failure
CC libip4tc.lo
In file included from libip4tc.c:118:0:
libiptc.c:70:8: error: redefinition of "struct xt_error_target"
../include/linux/netfilter/x_tables.h:69:8: note: originally defined here
Remove libiptc's duplicate definition and substitute names.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Thu, 8 Sep 2011 15:08:37 +0000 (17:08 +0200)]
build: sort file list before build
Manpage subsections are already sorted for obvious reasons. Since
$(wildcard) can actually return results unordered (just what the OS
can do) do the sorting with the .o file list too, for developer
comfort.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sat, 3 Sep 2011 12:27:55 +0000 (14:27 +0200)]
iptables: move kernel version find routing into libxtables
That way, the remaining unreferenced symbols that do appear in
libipt_DNAT and libipt_SNAT as part of the new check can be resolved,
and the ugly -rdynamic hack can finally be removed.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Tom Eastep [Thu, 18 Aug 2011 22:11:16 +0000 (15:11 -0700)]
libxt_conntrack: improve error message on parsing violation
Tom Eastep noted:
$ iptables -A foo -m conntrack --ctorigdstport 22
iptables v1.4.12: conntrack rev 2 does not support port ranges
Try `iptables -h' or 'iptables --help' for more information.
Commit v1.4.12-41-g1ad6407 takes care of the actual cause of the bug,
but let's include Tom's patch nevertheless for the better error
message in case one actually does specify a range with rev 2.
References: http://marc.info/?l=netfilter-devel&m=131370592105298&w=2 Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sat, 27 Aug 2011 15:59:52 +0000 (17:59 +0200)]
xtoptions: fill in fallback value for nvals
Parsing for libxt_conntrack rev 2 is done by using rev 2's option
structure, which specifies XTTYPE_PORT, and using rev 3's parser
skeleton, which uses cb->nvals. Reading cb->nvals when not using
XTTYPE_PORTRC (or any other multi-value type) is undefined behavior.
Make it defined. Since XTTYPE_NONE is the only type that can take
void, nvals logically ought to be 1.
References: http://marc.info/?l=netfilter-devel&m=131370592105298&w=2 Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 11:04:38 +0000 (13:04 +0200)]
libxt_hashlimit: observe new default gc-expire time when saving
Since a while, --htable-gc-expire defaults to the chosen time quantum
instead of 10 fixed seconds, which leads the expiry value to be always
printed, which is redundant.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 10:27:06 +0000 (12:27 +0200)]
libxt_physdev: restore inversion support
Bug origin is in commit v1.4.11~26^2~4.
References: Dave Täht via netfilter-devel on 2011-08-20 14:40:11 -0700
References: <CAA93jw6mpDL6rLXM+9SpAhafkDdKoSfhAxU8UM87vUqjuzjYJw@mail.gmail.com> Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 10:25:06 +0000 (12:25 +0200)]
libxt_owner: restore inversion support
Bug origin is in commit v1.4.11~16^2~7.
References: Dave Täht via netfilter-devel on 2011-08-20 14:40:11 -0700
References: <CAA93jw6mpDL6rLXM+9SpAhafkDdKoSfhAxU8UM87vUqjuzjYJw@mail.gmail.com> Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 09:41:13 +0000 (11:41 +0200)]
libip6t_frag: restore inversion support
--fraglen also was not printed since v1.4.11~26^2~22.
References: Dave Täht via netfilter-devel on 2011-08-20 14:40:11 -0700
References: <CAA93jw6mpDL6rLXM+9SpAhafkDdKoSfhAxU8UM87vUqjuzjYJw@mail.gmail.com> Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 08:06:18 +0000 (10:06 +0200)]
libxt_conntrack: fix --ctproto 0 output
First, we are missing XTOPT_PUT when trying to use XTOPT_POINTER.
(Next commit will flag this.) Furthermore, l4proto is of type
uint16_t, while XTTYPE_PROTOCOL wants a uint8_t so the idea would not
work => revert v1.4.12~1^2.
Bug goes back to v1.4.12~1^2.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 06:52:56 +0000 (08:52 +0200)]
libxt_dscp: restore inversion support
References: Dave Täht via netfilter-devel on 2011-08-20 14:40:11 -0700
References: <CAA93jw6mpDL6rLXM+9SpAhafkDdKoSfhAxU8UM87vUqjuzjYJw@mail.gmail.com> Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Jan Engelhardt [Sun, 21 Aug 2011 07:39:21 +0000 (09:39 +0200)]
libxt_dccp: fix random output of ! on --dccp-option
dccp-option tests info->typemask, but it really should look at
info->invflags instead.
This bug goes back to commit v1.3.4~11.
References: Dave Täht via netfilter-devel on 2011-08-20 14:40:11 -0700
References: <CAA93jw6mpDL6rLXM+9SpAhafkDdKoSfhAxU8UM87vUqjuzjYJw@mail.gmail.com> Signed-off-by: Jan Engelhardt <jengelh@medozas.de>