Slashdot Mirror
Why iptables (Linux 2.4 Firewalling) Rocks
Jay writes: "There's a story on Linux 2.4's new Stateful firewalling.
Rusty and friends have vastly improved Linux capabilities, allowing it
to filter against many stealth scans and DoS attacks. Their code rewrite brings powerful "stateful" firewalling that was originally impossible under Linux. The muchly improved packet mangling allows
not only better transparent proxies, but load-sharing clusters and stuff. Actually, the coolest thing about the new architecture is that it's designed so anybody can add filtering methods, for stuff like rate limiting and MAC address-based filtering."
Inserting a proxy when the client doesn't have knowledge isn't something to be proud of; it breaks browsers, violates the end-to-end principle at the IP layer, and brings some serious privacy/data integrity problems to light.
"Building Linux and OpenBSD Firewalls" by Wes Sonnenreich, Tom Yates is a good resource that explains all this.
BdosError
Complexity is Easy. Simplicity is Hard.
This just requires that the person inside your building be a little smarter. MAC addresses can be changed, so all that is required is to wait until a trusted box goes offline, and then just borrow its MAC address.
Of course, this eliminates a good portion of those who would try to hack in. Just don't rely on it to be fool-proof.
I was always tempted to use a laptop running Linux to spoof one of the main campus Unix systems at Dartmouth when it went down for backups, but never did. (I think they had some public systems on the same subnet as the machine room.)
As it stands now (or previously... in ipchains), rules were defined in a way that could be reduced to a regular expression, which as everyone knows are fast and don't require a stack to interpret.
However, as soon as you introduce "state" into a grammar, you require of yourself a stack, which means that additional memory is going to be needed to store these values to aid in parsing the grammar.
This doesn't sound too bad, even though it may slow the process of packet interpretation, but it seems like there still would be a signifigant performance hit on busy servers.
Also, since these rules have state, I wonder if there will be ways send packets that will cause something similar to a stack overflow (infinite recursion) by sending it a quasi-infite series of packets.
(Imaging sending a C++ compiler the following ..
[I can't send what I want, a string of open parens, so this will have to suffice. /. lameness filter]
String= (*
Eventually there will be a stack overflow from all the parens.
Someone will have to explain to me why they keep up with the cryptic syntax. That's one of the reasons why i switched to ipf (OpenBSD) some time ago.
Now both (iptables, ipf) have states. Can someone infom me about the advantage of iptables over ipf?
---------------------------
"What is the most effective Windows NT remote management tool?
It'd be cool to filter out junk on my Apache/Win2K box. :)
-
-Be a man. Insult me without using an AC.
- I don't care if they globalize against free speech. All my best free thoughts are done in my head.
Each has their strong points. Why not be a little more optimistic about what each can achieve?
Yep, OpenBSD has a very thorough out-of-the-box NAT firewalling solution. It's extremely powerful (and in my experience far superior to linux 2.2 ipmasq)... I use it at home and work with OLD Sparcs (25-40mhz) and still achieve near 10mbps on all 10mbps interfaces.
:)
Iptables under the 2.4 kernel has impressed me. I use it for a test lab at work, and the first thing I noticed was that it allows active FTP through its implementation of NAT. It's probably due to some funky extension they've added to FTP standards, i'm not sure, nor good enough of a programmer to figure it out from the code... Nonetheless, they've finally made a decent product that starts comparing to the cleanliness and quality of BSD's NAT.
The other BSDs also use the same BSD NAT, however, my last experience of using NAT on FreeBSD (back on 3.3) was that it was quite erratic in behavior compared to the same setup using OpenBSD. I might have been royally screwing something up, too
.... um, i lost you after "0110100001101001".
I havent used linux firewalling since the days of ipfwadm, but from what I see here, iptables has some rather extreme leaps here, the likes of which have usually only been seen in commercial firewall packages. What I want to know though, is exactly how *does* this compare to commercial firewall packages, inparticular: Checkpoint Firewall1
---
Video meliora proboque deteriora sequor - Ovidius
These are neat features, just when i was thinking of replacing my linux masq box with a FreeBSD box to use it's spiffier NAT support and all that good stuff, this looks like it takes care of that. I may still do it for security though...
Now: Does anybody here know offhand if the packet filtering support in FreeBSD supports these features? (stateful ftp support, DNS probe rejection, etc...?) i know it has rate-limiting to make DOS attacks not work as well...
Thanks
---
Play Six Pack Man. I
The real solution to open ethernet ports is to use authenticated VLANs - require the workstation user to authenticate before being put into a suitable VLAN, and just ignore the MAC address.
I needed to block a list of thousands of ip addresses at the firewall. Here's how I did it:
First, I read the sample code for how to implement user-space queueing. That's where you pass a packet to a user-space program so it can decide what to do with it. Then I made some trivial changes to that program so that it looks for a file in a certain directory with a name equal to the IP address in the packet header.
If the file exists (i.e. an open() call succeeds), then the packet is denied; otherwise it is allowed. Anytime I want to add a new IP address to the blocking list, I just create a file in that directory. Since I run reiserfs, the test for file existence is as efficient as a hashtable lookup, but much simpler to code.
Not as sophisticated as a commercial firewall program, but not as much overhead, either. (Simplistic) benchmarks show no significant increase in network latency. Works for me!
The Web is like Usenet, but
the elephants are untrained.
Check out Firewall Builder... sourceforge.net/projects/fwbuilder very similar to Check Point's config GUI...
This sounds really interesting to me. I do have a question about 2.4 in general though. How well does it run on pre-pentium hardware?
;-)
I remember when 2.2 series came out, many were saying to stay at 2.0.x for pre-pentium hardware because 2.2 would just not run adequately. Was this true in the first place, and does 2.4 'raise the bar' at all for hardware requirements?
I only ask because my linux firewall/masq box for my cable at home is running a 2.0.36 kernel on a 386 and there probably isn't any hope.
For a while I was using a P-150 running a 2.2.x kernel (RedHat 6.2 install, probably) and ipchains seemed to be much better than ipfwd.
At some point I decided I needed that box for another project, so I dusted off that crap old Compaq (a laptop, no less!) and pressed it into service as my masq box using a docking station (two ISA slots, WOOHOO!) and a couple of $7 SMC cards and slink. It works great as long as I don't try anything fancy.
But, the question is... what is a reasonable amount of hardware to expect to have to use for a 2.4-kernel firewall? I'd love to play with this. Think a 486-100 w 32Mb would be enough? It's not doing much, just sitting around for compiling kernels for the 386.
There is much cruelty in the universe, John.
Yeah, we seem to have the tour map.
You're getting bit by the fact that there's more to MS PPTP than meets the eye - the tunnel itself is actually a GRE tunnel with some funky MS-proprietary encryption, with a control connection at 1723/tcp (or something like that).
It can be done, but a 2.2 kernel needs to be patched to do it. I've not tried it, but the 2.4 kernel should allow you to "port-forward" the tcp connection and the GRE tunnel back to the Win2k system.
Just keep in mind that there's no connection tracking, so it has to be the same Windows system making the VPN each and every time, and one and only one VPN can be going at any one time.
"This has been in BSD forever..."
I bet by the time I'm done typing and hitting submit there will be several posts saying something like this. Why can't everyone stand back and go "great job guys for implementing it on a new platform!" instead of "you lamers...we had this X years ago". If they had implemented this on MacOS(random OS picked off the top of my head) would the BSD guys say the same thing?
Try out Firestarter (Disclaimer: I wrote the program, so I'm biased). Supports both ipchains and netfilter, several nics, NAT/Masq, ToS/ICMP filtering etc.
I, just like you, got really tired of editing scripts all day long, the end result is this GNOME program. First the program presents a wizard for quickly setting up a firewall, then you can monitor the firewall hits you get and close/open/stealth ports with just a few clicks. Easy of use was the goal with this program.
in that's big-endian, like say from Solaris (the most popular Internet platform) the translation must occur before any filtering gets done. So a saavy hacker can mount an attack from any Solaris box and easilt Ddos any Linux box by doing a reverse htons() packet flood.
Nope. Just store the targets in network order and do a normal compare. All that matters is that the target list (hash or whatever) and the packet be in the same order.
The point here is not "BSD is better than Linux." The point: "it's unfair that Linux gets good press for doing something everyone else has been doing for so long."
echo Prpv a\'rfg cnf har cvcr | tr Pacfghnrvp Cnpstuaeic
I keep hearing that *BSD's IPstack is faster, more robust, etc. But I never see hard numbers to back it up. I want a identical hardware benchmark that can push both stacks, and see which one is better. The nearest I've seen compare linux 1.2.* and FreeBSD 2.0. Ancient history in otherwords.
If you've got numbers, and the testing details that produced them, I'd like to see them.
Zapman
Packet filter is not supposed to look into the payload of the packet. That is done by something called proxy. Regarding FTP, well, let's face it, it is a badly broken protocol, period.
Also, "limit the number of SYN packets from a single source" doesn't really help in defending the DoS attack. The attacking machine doesn't use its real ip as the source ip nor does it use a single source ip.
I agree with other posters: iptable is playing catch up with ipfilter.
gd
I was eagerly waiting for Linux 2.4 from the day I heard somewhere it would support ipfilter. I run mostly BSD boxes, due to personal preference, while still keeping a few Linux systems up and running. All systems run perfectly well and cause me little grief, if any at all.
However, once again the Linux camp is "doing things their own way" and contributing even more to the separation between different Unix flavors (as well as making those sigs that go something like "Linux - the Unix defragmentation tool" even funnier). Don't get me wrong, iptables seems to have a few cool features that are not present in ipfilter, but just why was ipfilter not used instead ? It is present in a few platforms already (runs on all BSD's, as well as Solaris / SunOS and IRIX), it is tried and tested, and it does most of the stuff iptables does. And it can probably be extended to do everything iptables does as well, with some extra work. But the Linux folks, however talented and bright they undoubtedly are, and however good a kernel they have produced, have just taken a completely silly route this time. Not only have they duplicated effort towards the same (or at least, a very similar) end, they have also created another headache for busy sysadmins who maintain a few different systems. I had high hopes on ipfilter being "officially" supported under Linux (the ipfilter code mentions sketchy support for Linux, which I have admittedly not tested, but from the looks of it it has not been maintained for a long time --- the most recent kernel mentioned in the HISTORY file is 2.0.34, the last reference to Linux appears in 1998), but now I will be forced to get rid of the only Linux box I still maintain as a server (for various reasons, none of which related to the quality of the OS --- that little box has run great for ages, and causes me no grief at all) if I ever need firewalling for it. Yes, I could learn to work with iptables (and I probably will anyway, out of curiosity), but the Linux folks could probably learn to cooperate better with the rest of the world in certain aspects. It really is a shame, because I honestly think Linux is great --- the mindset of some of it's users (and by the looks of it, some developers as well) is however a different matter...
FreeBSD SMP support is akin to Linux 2.0 in 1996! FreeBSD uses a BFL (big fuckin' lock). Sure FreeBSD 5.0 might include experimental Linux 2.2 (1998) SMP support merged from BSDi code, but when will that be? 2002?
cpeterso
While it's true that it is usually 10 bits per byte on asynchronous links (1 start bit, 8 data bits, 1 stop bit, no parity), that's not true for things like leased lines. They use synchronous serial communication, so it is 8 bits per byte. IIRC, ISDN is also synchronous, but I don't know about some of the newer signaling schemes.
Just tryin' to keep the math straight.
stateful firewalls = excellent... But LRP will NOT readily support 2.4 kernal size!
(LRP is the Linux Router Project at http://lrp.c0wz.com/ a kick ass firewall/router/NAT distro that needs a single floppy, 486 or better, 2 nics, and NO harddrive!)
Joshua Jackson, of the impressive www.coyotelinux.com LRP distro,
said "I wish I could move to the 2.4 kernels for Coyote, but not in the floppy version I am afraid. By the time you get all of the necessary options built into the kernel, it is roughly 350kB larger than the 2.2 kernel... in addition, the full iptables and iptroute2 package are quite a bit larger than their 2.2 equivs. The Embedded project is already running on the 2.4.0 final kernel, but it does not run from a floppy."
-Nathaniel
Bummer for the rest of us.
I am trying to find out if iptables does remasqing: Masquerading a packet with a destination back inside the masq'd LAN.
I know ipchains doesnt do this, and its been a considerable headache. Some of the things I cannot do are:
I know there is a kernel patch for 2.2 that will introduce this ability, but it breaks the ipchains code's security.
Looking at the way iptables works, I think this should be possible (Port forwarding would be done before routing, and masqing done post-routing), but it is not addressed specifically anywhere that ive seen.
Anybody know?
Disclaimer: Ive asked this question before on K5, but nobody seemed to have a ready answer. Maybe the broader audience
If BSD/OS is sooo much faster than Linux 2.4, then why does NO ONE use any BSD for SPECweb benchmarks ? I see Linux 2.4, Windows 2000, AIX, Tru64, and HPUX.
cpeterso
The typical Linux user does NOT have a multiprocessor machine.
Actually dual Celeron and P2/P3 Linux boxes are pretty common. While certainly not everyone has one, I do, and I probably know at least 6 or 7 other people who do as well.
Deciding what a 'typical' user is, now that is the problem.
For those of you not familiar with stateful firewalling, it is an incredibly good thing. I've gone from using ipfwadm to using ipchains to finally using ipfilter on BSD. Because of ipf's support of stateful filtering, my firewall rulesets were *much* simpler. Given that many (most?) problems associated with firewalls is the complexity of the configuration, this is a Very Good Thing(tm). After migrating from linux/ipchains to BSD/ipf, I was able to add serious protection for my network, including my high ports (which often run the most vulnerable services, namely RPC services). I haven't used netfilter/iptables, but it looks to be a *huge* step forward for Linux.
All that being said, I have a major problem with this article. It seems to suggest to users that FTP monitoring to handle active FTP clients is a good idea. In fact, this is a *terrible* idea. I got to watch Dug Song et al at BlackHat walk right through a CheckPoint FW1 system like it wasn't even there by exploiting some assumptions that FW1 made when monitoring FTP for the PORT commands. It sounds to me like the netfilter/iptables support for FTP functions in a very similar manner as FW1. If you must support FTP through your firewall, make your users use passive FTP. Every modern FTP client and every modern FTP server that I've seen all support passive FTP. Of course, a better approach is to encourage secure communications, like scp or SSL.
Bottom line, the best firewall design is the simplest one possible that gets the job done. Adding neato features like payload monitoring to poke extra holes in the firewall is diametrically opposite this philosophy.
Also, please remember that a firewall only serves as a method of blocking traffic between network segments. It does not magically secure network traffic from viewing, spoofing, manipulating, or hijacking (you need to use protocols that support strong authentication and strong cryptography to achieve this). It does not secure the applications or systems which you allow traffic to touch (you need to use secure OSes under secure configurations to accomplish this). It does not magically protect your systems against insider threats (you need to have good people working for you, restrictions on outside connectivity, and thorough physical security to accomplish this). Remember, crunchy on the outside and soft and gooey on the inside is great for candy, bad for networks.
The MAC address can be set by software. Any hardware solution is more expensive.
Some cards even explicitly give you the option to change it (although they shouldn't).
The MAC address is part of the lowest and therefore hardware level of the ethernet.
Normally, a ethernet-adapter filters any packages which are not targeted for its MAC address(although even this behaviour can be altered).
Furthermore, any package send contains the MAC address of the sender. This all happens in hardware and is the ethernet-protocol.
The change of the MAC address in the OS should only affect the ARP request, which translate an IP-address into a MAC-address.
"Between strong and weak, between rich and poor [...], it is freedom which oppresses and the law which sets free"
access() is in POSIX, so portability isn't a concern.
cool hack, too.
"If ignorance is bliss, may I never be happy.
-- Veni, vidi, dormivi
'stat' may have more overhead than the 'open' call, depending on the implementation of your filesystem. So it's probably just fine to keep that 'open' call the way it is.
[
Actually, identical hardware benchmarks are very poor, because it assumes that each system exploits that hardware to its fullest. The best benchmarks is to have two camps, give them each X number of dollars, and build the best server, and then compare those. For example, if the system has a TULIP ethernet card, and OS A supports it well, and OS B doesn't, then the OS A will win, even if OS B is fater on an EEPro. So, for stuff like this, its better just to do it based on dollars.
Engineering and the Ultimate
Darren Reed's IP Filter ("ipf") is very cross-platform and not exclusive to BSD. It works with several other *nixes as well.
What's really interesting is that linux used to have stateful firewall support, via ipf! That's right, IPF used to work with Linux. But somewhere along the way with all of the changes to the Linux kernel IPF stopped working and nobody bothered to do the maintainence.
Instead of doing the required work on the mature stateful firewall package that is ipf, the linux folks put up with stateless ipchains for the longest time before finally adding state to ipchains. Shame!
This isn't exclusive to Linux though. FreeBSD's IPFW was made stateful even though IPF has always worked under FreeBSD. Reason? IPFW supports ethernet bridging, while IPF doesn't. No doubt there are similar reasons for ipchains over ipf: filtering based on MAC addresses was mentioned (although the desired results have always been do-able with static ARP).
Instead of re-inventing the square wheel, why didn't these people enhance IPF with the desired functionality? Isn't that what open source is all about? Must NIH prevail???
Wood and metal scraps. Paper. What luxury! We had to scratch the O.S. into stone tablets with just our fingernails, using only zeros and ones. And we never had enough ones. Whenever we ran out of ones, we had to make do with just zeros.
And we'd have given anything to have a gerbil. We had to get a group of half-starved cockroaches running in the same direction on an old 45 RPM turntable, and that's not easy, let me tell you. And for blinking lights, we had to rent fireflies, we couldn't afford to actually buy them outright, you know.
You kids today, you don't know how lucky you have it.
John
John
The MAC address is hardwired in, but it's the software layers that put the address in the packet or program the Ethernet MAC's registers with most cards. As I understand it, in many cases under Linux, you can change the effective MAC address for a card with ifconfig.
I believe this is the flag that does the magic:
hw class addressDoes that answer your question?
At any rate, I know it's possible to do this by hacking the driver in the kernel (at least for NE2K clones), as I've done this personally in a previous life...
--Joe--
Program Intellivision!
--
No it's not typical... but there are a lot of ignorant rednecks here in the U.S. Just like any other country. It's not like human nature makes sudden dramatic shifts when crossing artificial borders.
More than likely it's just some kid trying to get a rise out of you. The question is, why did it strike such a nerve with you? Why the hostility directed at U.S. citizens rather than ignorant and hateful human beings in general?
The U.S. governments do not place any blame on minorities, a public official who said any such thing would likely be canned within the week. Racists who do blame minorities for their problems are unfortunately common, but not generally in the majority (depending on where you live).
The U.S. is just like anywhere else, mostly normal people trying to get by, raise some kids and be happy... I think you're as guilty of prejudice and ignorance as the poster to whom you were responding.
"Oh twap!"
Changing the MAC address is typically not very easy. If you have the source to the Firmware and you know how to really tickle the LAN chip, then you might be able to do it. MAC address ranges are assigned to vendors. The chip (node, card, whatever) must identify itself to the local ethernet network. It get's this number from a ROM (firmware) database usually. I *suppose* it could be hardwired in, though. There's probably a way to program the chip to tell it has a different MAC address.
What you are missing is that the ROM is actually separate from the ethernet chip. This is rather more obvious on an older card. The only connection between what is in the ROM and the ethernet hardware is the driver. Even if the ethernet chip generates the souce MAC address all it has to go on is what is in it's own registers.
Prior "wire techs" labelled nothing. It would be a month long job to figure out. We do clean it up as we upgrade though.
:)
Well the obvious jumpers to pull would be those where you have no light on the hub/switch to indicate there is anything at the other end