You Don't Need to Know Programming to Appreciate This Angry Rant

By Chris Mills on at

Linus Torvalds is a legendary software engineer and founder of the Linux kernel. He also can’t stand bad code, so sit down, grab the popcorn, and enjoy.

Torvalds posted this response to some new code that popped up in Linux version 4.3. It’s quite technical, but even if you don’t grasp the nuances, the fine swearing can be appreciated by all:

Christ people. This is just sh*t

.The conflict I get is due to stupid new gcc header file crap. But what makes me upset is that the crap is for completely bogus reasons.

This is the old code in net/ipv6/ip6_output.c:

mtu -= hlen + sizeof(struct frag_hdr);

and this is the new “improved” code that uses fancy stuff that wants magical built-in compiler support and has silly wrapper functions for when it doesn’t exist:

if (overflow_usub(mtu, hlen + sizeof(struct frag_hdr), &mtu) ||

mtu <= 7)

goto fail_toobig;

and anybody who thinks that the above is (a) legible (b) efficient (even with the magical compiler support) (c) particularly safe is just incompetent and out to lunch.

The above code is sh*t, and it generates shit code. It looks bad, and there’s no reason for it.

The code could *easily* have been done with just a single and understandable conditional, and the compiler would actually have generated better code, and the code would look better and moreunderstandable. Why is this not

if (mtu < hlen + sizeof(struct frag_hdr) + 8)

goto fail_toobig;

mtu -= hlen + sizeof(struct frag_hdr);

which is the same number of lines, doesn’t use crazy helper functions that nobody knows what they do, and is much more obvious what it actually does. I guarantee that the second more obvious version is easier to read and understand. Does anybody really want to dispute this?

Really. Give me *one* reason why it was written in that idiotic way with two different conditionals, and a shiny new nonstandard function that wants particular compiler support to generate even half-way sane code, and even then generates worse code? A shiny function that we have never ever needed anywhere else, and that is just compiler-masturbation.

And yes, you still could have overflow issues if the whole “hlen +xyz” expression overflows, but quite frankly, the “overflow_usub()“ code had that too. So if you worry about that, then you damn well didn’t do the right thing to begin with.

So I really see no reason for this kind of complete idiotic crap.

Tell me why. Because I’m not pulling this kind of completely insane stuff that generates conflicts at rc7 time, and that seems to have absolutely no reason for being an idiotic unreadable mess.

The code seems *designed* to use that new “overflow_usub()“ code. It seems to be an excuse to use that function.

And it’s a f*cking bad excuse for that braindamage.

I’m sorry, but we don’t add idiotic new interfaces like this for idiotic new code like that.

Yes, yes, if this had stayed inside the network layer I would never have noticed. But since I *did* notice, I really don’t want to pull this. In fact, I want to make it clear to *everybody* that code like this is completely unacceptable. Anybody who thinks that code like this is “safe” and “secure” because it uses fancy overflow detection functions is so far out to lunch that it’s not even funny. All this kind of crap does is to make the code an unreadable mess with code that no sane person will ever really understand what it actually does.

Get rid of it. And I don’t *ever* want to see that shit again.

Linus

Lesson learned: don’t mess with Linus’s code.

[The Register]